Search

An intelligent recommendation system facilitates users by recommending items based on the user's preferences. Most of the existing recommender systems employ collaborative filtering which consider homogenous data only and does not exploit different knowledge sources. Such approaches encounter many challenges such as Cold Start problem, Sparsity, and Scalability. To overcome these setbacks, a novel deep learning & semantic fusion-based recommendation system named DLSF is proposed in this paper which collaborates multiple knowledge sources by employing three modules to generate top-N recommendations. The first module precludes sparsity by using latent representations and deep learning for understanding user-item interactions. The second module exploits semantic information provided to similar items by the users for dealing with cold start problem. And the third module works for new users and items which don't have any prior information. A comprehensive experimental evaluation is conducted over benchmark datasets. The proposed DLSF showed on average 16.02% and 16.71% more efficient over the baselines recommendation systems taken into account in terms of MAE and RMSE respectively. DLSF has found to have significant efficiency in terms of Precision, Recall and overall accuracy. [ABSTRACT FROM AUTHOR]

The exploration of non-covalent interactions between drugs and proteins (NCIs) has significantly improved the performance of drug–target interactions (DTIs) prediction models. However, existing methods for simulating NCIs are limited to single or multiple interactions, ignoring the combinatorial effects (i.e., high-order crossover between drugs and proteins (HOC-D&P)) that can arise when multiple NCIs coexist. To overcome this limitation, a multi-channel parallel high-order prediction model (PHCDTI) is proposed, aiming to simulate the complex interactions between drugs and proteins, thereby modeling HOC-D&P. Firstly, a tri-channel parallel model structure is constructed, allowing each channel to independently learn the interaction patterns between drug molecules and amino acid groups. Secondly, the three channels are designed from low-order to high-order, incorporating linear, inner-product, and pair-interaction feature crossover methods to simulate the complex interactions between drug molecules and amino acid groups. The stacking of residual connections enables the model to effectively model HOC-D&P. Additionally, by utilizing multi-head attention mechanisms for inner product and squeeze-excitation networks (SENET) for pair-interaction, the interpretability of the NCIs modeling process and the specific representation of drug molecules or amino acid groups are achieved. PHCDTI was evaluated on three benchmark datasets in four experimental settings, and the results indicated that it outperformed the most recent baseline model. In all cases, the average accuracy, precision, recall, AUC, and AUPR of PHCDTI are improved over the best baseline model by 3.56%, 6.06%, 3.99%, 2.50%, and 8.11%, respectively. Moreover, strong information filtering capabilities were exhibited by PHCDTI, with the effectiveness of the three-channel design confirmed by ablation studies. Finally, a case study on the human γ -aminobutyric acid receptors (GABARs) demonstrated that PHCDTI can serve as a powerful tool for real-world drug screening and design. [ABSTRACT FROM AUTHOR]

In recommender system, the proportion of interacted items with users is extremely sparse compared to the total number of items. This data sparsity problem particularly affects collaborative filtering methods. Even in modern collaborative filtering recommendation models based on deep learning, effectively acquiring more neighbor information and exploring how to utilize this information efficiently remains a promising research area. This paper not only explores heuristic algorithms for neighbor selection but also extends the application of neighbor selection to the GAN framework and negative sampling. To capture user neighbors that contain more potential feature information, this paper proposes a neighbor strategy that integrates explicit and implicit user information. It aggregates reciprocal neighbors of users based on the fusion of their explicit and implicit information, aiming to capture user preference information through neighborhood information. Furthermore, by designing an MRNGAN model based on a twin-tower generator, this paper successfully embeds user neighborhood information as feature embedding into the generator of GAN. In addition, this paper also proposes a heuristic negative sampling mechanism in the recall phase. It uses explicit and implicit information to fuse the nearest neighbors of items with the farthest neighbors of users. Negative samples that contain user neighborhood information are embedded into the discriminator of MRNGAN. The deep relationships between nearest and farthest neighbors are explored through the adversarial training of GAN, and their impact on recommendation performance is validated. The experimental results on three real-world datasets show that the MRNGAN model significantly improves the accuracy of Top-N recommendation. [ABSTRACT FROM AUTHOR]

The tremendous growth the pharmaceutical field witnessed especially concerning drug-development, caused drug-related experiments such as those concerned with estimating their interactions with targets harder to conduct. This issue led researchers to transition from vivo and vitro to selico experimenting methods, which deploy artificial intelligence (AI) techniques such as recommender systems (RS) to solve problems at hand. However, most RSs depend on global features such as indices, which render them incapable of fully understanding the behaviors of individuals. In this paper, we introduce our Convolutional Recurrent Deep Learning model-based RS approach, or 'RSCRDL', which tackles the interaction prediction problem from a natural language processing approach's (NLP) point of view. RSCDRL processes the textual representations of both protein sequences and drugs in order to learn the implicit features of the tokens forming them, rendering it able to process unseen and non-reactive individuals indirectly. We also present a greedier variant of RSCRDL that can exploit additional drugs explicit features for better predictive abilities. Experiments performed on two real-world datasets show significant outperformance achieved by our models over two general-purpose state-of-the-art RSs, while other experiments conducted on orphan protein sequences helped deduce drug candidates that are assumed to be capable of inhibiting their activities. [ABSTRACT FROM AUTHOR]

The cross-domain recommendation (CDR) model addresses challenges such as data sparsity, the long tail distribution of user–item interactions, and the cold start of items or users. However, solely transferring domain-shared knowledge based on the co-occurrence patterns, without considering user preferences, leads to negative transfer in CDR. To overcome these limitations, we propose an advanced deep learning CDR model called the Domain Adversarial Deep Interest Network (DADIN) aims to facilitate smooth knowledge transfer from the source domain to the target domain and effectively alleviate negative transfer. Firstly, the joint distribution alignment of user preference in DADIN is realized by introducing a skip-connection-based domain agnostic layer, and then the domain classifier is artificially designed to distinguish between the information coming from the source domain or the target domain. Additionally, DADIN combines prediction loss, global domain confusion loss, and intra-class domain confusion losses through the Min-Max game and gradient reverse layer to achieve collaborative optimization. Two real-world experiments show the area under curve (AUC) of DADIN is 0.78 on the Huawei dataset, and it outperforms its competitors by 0.71% on the Amazon dataset, showcasing its state-of-the-art performance. Moreover, our ablation studies further demonstrate that domain adversarial technique increases the AUC by 2.34% on the Huawei dataset and 16.67% on the Amazon dataset, respectively. • DADIN first employing domain adversarial method in Cross-Domain Recommendation (CDR). • With label information in source domain, DADIN addresses negative transfer in CDR. • The Joint distribution alignment of user preferences is achieved in DADIN. • Hierarchical attention is employed to enhance feature representation in DADIN. • DADIN achieving competitive performance in comparison with popular baseline models. [ABSTRACT FROM AUTHOR]

Agriculture plays an imperative role in many countries' economies and is a substantive source of survival. The variation in a soil nutrient decreases crop yield. An accurate soil fertility classification and application of fertilizers are essential for enhancing crop productivity. Currently, soil fertility levels are assessed through laboratory testing of soil samples, and fertilizers are applied randomly. This traditional practice increases fertilization costs and causes environmental pollution. Thus, it is necessary to develop robust and inexpensive soil fertility classification and fertilizer application. This study identifies the machine learning (ML) or deep learning-based soil fertility classifications. A comprehensive review is conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The purpose of this study is to examine different approaches that researchers use to predict or classify soil fertility. It also discusses the fertilizer recommendation developed by the researchers. The earlier research showed that ML-based approaches could accurately classify soil fertility. Furthermore, this study discusses the importance of soil nutrients and preventive measures to be taken on the imbalance of soil nutrients. This study explores research gaps and challenges in soil fertility classification and fertilizer recommendation systems. Most studies predicted the fertility levels of soil parameters, whereas a few researchers classified soil fertility. Few researchers recommended fertilizers for soil nutrient depletion. Most studies relied on expensive laboratory measurements or regional soil data collected from satellites. Based on the identified research gaps, this study suggests potential future research possibilities in soil fertility classification and the recommendation of fertilizers. It aims to develop a low-cost soil fertility classifier to prescribe fertilizers. The developed model can help farmers to enhance soil fertility with reduced fertilization costs. • Precise soil fertilization is essential to increase sustainable agricultural yield. • A systematic study on soil fertility estimation is conducted using PRISMA guidelines. • An overview of 42 prior research on soil fertility estimation is provided. • Highlights research gaps and future directions in the estimation of soil fertility. • Discusses various machine learning-based regression and classification approaches. [ABSTRACT FROM AUTHOR]

Due to the requirements of privacy protection and data asset ownership, recommender systems (RSs) based on centralized training process may be infeasible in some practical situations. In these situations, federated recommender systems (FedRec) based on deep learning are regarded as the substitutes with extensive application backgrounds and enormous potential business values. However, which deep learning algorithm and federated learning aggregation strategy perform best in the training of FedRec is still an unsolved question. To answer this question, experiments are conducted for comparing the performances of four popular deep learning algorithms and three federated learning aggregation strategies in FedRec. The experiments are conducted on 8 public datasets to obtain 96 FedRec precisions (4 deep learning algorithms × 3 federated learning aggregation strategies × 8 datasets). Then, the Wilcoxon test is used to verify whether there is a significant difference between the FedRec combinations and to derive the best combination. The result shows that, in terms of recommendation precision, the best FedRec is the combination of neural FM and federated averaging. In terms of execution iterations, the best FedRec is the combination of deep crossing and attentive federated aggregation. The obtained results would be valuable for further promoting the development of RSs under the situations that a centralized training process is infeasible. [ABSTRACT FROM AUTHOR]

In the age of information overload, modern recommendation systems provide an important role in helping people screen massive information. With the development of deep learning, personalized recommendations are further promoted to enhance user engagement and improve user satisfaction. Meanwhile, when the content provider has high-priority content to peddle, it often just exposes it to a conspicuous location without considering different users' preferences, which may even lead to negative impressions. To tackle this problem, in this paper, we propose a target-driven user preference transferring (TDUPTrans) module that will gradually orientate the user's preferences toward the target content. It is a brand new attempt that applies the user preference transfer process on personalized recommendations so that users don't just "see" the target item but also "like" it. To be specific, we propose a greedy-strategy-based intermediate content choosing algorithm, which can recommend different intermediate contents to users with different initial interests, and eventually lead users' interests to the target content. We give a clear interpretation of how our proposed process works through a news-based case study. Moreover, extensive experiments on real-world data sets demonstrate the interpretability of the guiding process, the abilities of our proposed TDUPTrans fusing with pre-existed recommendation algorithms, and the significantly improved target recommendation success rate with this TDUPTrans module plugged in. [ABSTRACT FROM AUTHOR]

With the proliferation and deepening of service-oriented architecture, more and more enterprises and organizations are exposing their computing functions and big data to the Internet in the form of cloud APIs to support service-oriented software development. This has resulted in a plethora of cloud APIs with similar functionality appearing on the Web, drowning users in a sea of cloud API choices. To solve this problem, quality of service (QoS)-aware recommender system is then widely applied to the selection of cloud APIs. Due to the dynamic and open network environment, the QoS-aware cloud API recommender systems are vulnerable to data poisoning attacks, where attackers inject poisoned data to skew the recommender system and make the recommendation direction follow the attacker's will. Given the lack of data poisoning attack methods and robustness analysis for existing QoS-aware cloud API recommender systems, in this work, we first built a general poisoning attack framework for QoS-aware cloud API recommender systems to elucidate and standardize the attack process. Then, we proposed a deep learning-based poison attack approach, which uses generative adversarial network (GAN) to learn the cloud API QoS data distribution of real users in an adversarial way, so as to generate high-quality fake user attack vectors. We conducted extensive experiments on real-world QoS datasets, and the experimental results show that our proposed GAN-based poisoning attack is effective and can better hide itself from being detected. In addition, we analyzed the data poisoning attack mechanism and the robustness of the cloud API recommender system based on four categories of twelve recommendation methods, thereby raising awareness about the security of cloud API recommendation and helping the recommender system defenders to develop more targeted defense strategies. [ABSTRACT FROM AUTHOR]

While searching for consumer goods, users frequently ask for suggestions from their peers by writing short free-form textual requests. For example, when searching for movies users may ask for " Drama movies with a mind-bending story and a surprise ending, such as Fight Club " in one of the many online discussion boards. Despite the recent developments in large language models (LLMs) and natural language processing (NLP), modern recommender systems still struggle to process such requests. Therefore, in this paper we evaluate several approaches for annotating structured information from such short, free-form natural language user texts to calculate recommendations. We set up this evaluation as a two phase processes including (a) identification of the best NLP approach to identify key elements of users' requests, and (b) assessment of the quality of recommendations computed with such elements. For our evaluation, we use a gold-standard reddit movie recommendation dataset consisting of annotations, manually created by crowdworkers who extracted keywords, actor names and movie titles. Using this dataset we evaluate a collection of more than 30 NLP and five recommender approaches. In addition, we perform an ablation study to assess relative annotation importance for movie recommendations. We find that domain-specific deep learning models, trained on a subset of data as well as embedding-based recommendation approaches are able to match the recommendation performance of recommendations computed from manual annotations. These promising results warrant further investigation in automatic processing of short free-form texts for computation of recommendations. Specifically, we provide insights into which NLP models and configurations work best for automatically annotating free text to compute (movie) recommendations, hence substantially reducing the search space for combinations of NLP and recommendation algorithms in the movie and potentially other domains. • Deep learning models in combination with external embedding techniques work best. • Rec. performance based on NLP matches the performance based on manual annotations. • Sentiment of NLP annotations does not significantly improve recommender performance. • Movie titles and movie keywords are the most important annotations for accurate rec. [ABSTRACT FROM AUTHOR]

Textual reviews contain fine-grained information that can effectively infer user preferences over the items. Accordingly, the latest studies in the field of recommender systems exploit content-rich review texts to complement user and item representations and improve the ability to make personalized recommendations. Furthermore, the interactive deep learning mechanism can better model the user-item interaction from fine-grained textual reviews compared to traditional recommendation approaches improving the predictive performances. Therefore, it becomes important to investigate the design of existing deep learning methods for review-based recommender systems and innovate to make them capable of meeting desired recommendation schemes. The purpose of this research is to explore the performance of deep learning networks and deep transformer models in review-based recommender systems. In this paper, we conduct a compendious survey of the latest deep learning techniques in review-based recommender systems. Then investigation calls to employ and analyze deep transformer models for the review-based recommender systems. The wide range of experiments shows that deep transformer models can extract interpretable and relevant user/item representations than traditional deep learning networks. The findings indicate that the best deep transformer performance gains the maximum relative improvement (RMSE = 4.6%, MAE = 7.4%) with Amazon electronics, compared to the best outcome from traditional deep learning networks. In the end, this article highlights research gaps and outlines research opportunities for future research in this field. [ABSTRACT FROM AUTHOR]

Highlights • A segmentation recommender based on crowdsourcing and transfer learning. • Two pre-trained architectures (VGG16 and ResNet50) were studied. • The method predicts dynamically the most appropriate segmenter for any input image. • Creation of new ground -truth dataset by using the ISIC2017 segmentation results. • Proposed image-based method improves the performance of skin lesion segmentation. Abstract Deep learning is widely used in medical applications regarding the high performance it can achieve. In this paper, we propose a segmentation recommender based on crowdsourcing and transfer learning for skin lesion extraction. In fact, after collecting and pre-processing data from the ISIC2017 segmentation challenge, we tested two pre-trained architectures (VGG16 and ResNet50) to extract features from the convolutional parts. Then, a classifier with an output layer, composed of five nodes representing the segmentation methods' classes, was built. Thus, the proposed architecture is able to dynamically predict the most appropriate segmentation technique for the detection of skin lesions in any input image. Experimental results prove the capability of the proposed image-based method to improve the segmentation performance comparatively to the state of the art methods. [ABSTRACT FROM AUTHOR]

Under the current paradigm, POI (Point-of-interest) recommendation tasks are mainly focused on representation learning. Therefore, the quality of the trajectory embeddings plays a key role in prediction. Existing methods mainly focus on learning the contextual information of check-in sequences, however, the contextual information is not the only medium for establishing state transitions between check-ins. Semantic information has been proven a powerful medium that can be learned with semantic data and injected into the source sequential embeddings for better prediction in other recommendation tasks. Unfortunately, for POI recommendation tasks, most of the key elements in the trajectory are discrete and there is no explicit semantic information. Therefore we argue that the deep and rich semantic information hidden in trajectories has not been fully exploited currently and how learning the deep semantic information from discrete trajectory data to improve the quality of the trajectory embedding is the key to further improving recommendation performance. We propose DSMR, a deep semantic recommender model for the next POI recommendation task to mitigate the issue. Specifically, we use prompt engineering to carry out continuous semantic modeling of discrete trajectory data and use the pre-trained language model to extract its implicit deep semantic information to establish causal transfer constraints between check-ins through the medium of semantics. Meanwhile, we propose a new position encoding function, temporal interval encoding, to avoid the neglect of temporal information of the check-ins sequence under the self-attention mechanism. Extensive experiments on two real-world datasets demonstrate the superior performance of our model to state-of-the-art approaches. • Establish causal transfer between check-ins through the medium of semantics. • Learning the deep semantic feature of trajectory. • Propose temporal interval encoding to avoid losing temporal information. • Fusing deep semantic features to improve the recommender model performance. [ABSTRACT FROM AUTHOR]

With the rapid development of service-oriented computing and cloud computing, an increasing number of Web services have been published on the Internet, which makes it difficult to select relevant Web services manually to satisfy complex user requirements. Many machine learning methods, especially matrix factorization based collaborative filtering models, have been widely employed in Web service recommendation. However, as a linear model of latent factors, matrix factorization is challenging to capture complex interactions between Web applications (or mashups) and their component services within an extremely sparse interaction matrix, which will result in poor service recommendation performance. Towards this problem, in this paper, we propose a novel deep learning based hybrid approach for Web service recommendation by combining collaborative filtering and textual content. The invocation interactions between mashups and services as well as their functionalities are seamlessly integrated into a deep neural network, which can be used to characterize the complex relations between mashups and services. Experiments conducted on a real-world Web service dataset demonstrate that our approach can achieve better recommendation performance than several state-of-the-art methods, which indicates the effectiveness of our proposed approach in service recommendation. [ABSTRACT FROM AUTHOR]

Sequential recommendation (SR) is an essential component of modern recommender systems. It models the dynamic interest of users based on their sequential interactions. Recently, several studies have utilized sequential deep learning models such as Recurrent Neural Networks and Transformers to facilitate sequential recommendation, which shows promising results. Inspired by the rise of contrastive learning techniques, some methods are devoted to enhance sequential deep learning models by designing contrastive learning loss on self-supervised signals. However, there are still a number of obstacles that make it challenging to efficiently learn user representations by contrastive learning. These issues include, but are not limited to, data sparsity, noisy data, and sampling bias (e.g., false negative), particularly in complex, parameter-intensive models. In light of these challenges, we examine how to deal with data sparsity and noisy data by implementing contrastive Self-Supervised Learning (SSL) and Momentum Contrast (MoCo) to the sequential recommendation. Except typical in-batch negatives, our basic idea is to maintain a dynamic queue to expand negative samples with a moving-averaged encoder. After being augmented by sequence-level and embedding-level methods, the representations from all historical encoder output are pushed into the dynamic queue, which usually leads to sampling bias when potential positives in the queue are used as expanded negative samples in contrastive learning. To tackle this issue, we integrate momentum updating mechanism with a novel instance weighting mechanism to penalize false negatives and guarantee the model's efficacy. To this end, we introduce a fresh framework called Mo mentum Co ntrastive Learning Framework for S equential Rec ommendation (MoCo4SRec). Experiments on eight real-world datasets demonstrate how the suggested strategy might boost model performance over current benchmarks by adopting improved user representations. • A new framework(MoCo4SRec) that brings together momentum and contrastive SSL with SR. • Samples are perturbed not just at the sequence level but also at the embedding level. • An instance weighting method is introduced to penalize the sampled false negatives. [ABSTRACT FROM AUTHOR]

While implementing software projects, developers do not reinvent the wheel but try to reuse existing API calls and source code. In recent years, the problems related to recommending APIs and code snippets have been intensively investigated. Although current approaches have achieved encouraging performance, there is still the need to improve the recommendation process's effectiveness and efficiency. In this work, we reformulate the problem of API recommendations by proposing learning and recommending API sequences relevant to a given coding context. We present LUPE, a novel approach to API and code recommendation, exploiting cutting-edge deep learning techniques. Thanks to the underlying Encoder–Decoder architecture specialized in transforming sequences, LUPE can effectively learn the order in which invocations occur. The approach has been evaluated on two Android datasets and compared with GAPI and FACER , two state-of-the-art API recommender systems. Being fed with augmented training data, our conceived approach can obtain a high prediction accuracy, and produce a perfect match in several cases, hence outperforming the baselines. • API recommendation can be formulated as a machine translation task. • The order in which APIs appear determines the relationship between them. • Having multiple APIs occurring in the same embedding also determines API synonymy. • A system built on top of neural networks to recommend additional API calls. [ABSTRACT FROM AUTHOR]

• A Deep Neural Collaborative Filtering method is proposed using WMF and FC-MLP. • A new Wide Matrix Factorization is developed by incorporating side features. • NDCF captures complex structure of Nutrients-Fertilizers using Deep Network. • Embedding side features-based ablation study revealed the robustness of the NDCF. With the advent of the modern era, deep neural networks have dominated recommender systems, as they can effectively capture complex interactions. Nevertheless, there is still a research gap in fertilizer recommendations based on soil nutrients content. In this work, we propose nutrient-centered deep collaborative filtering technique to determine the required amount of fertilizers for sustainable crop growth. The undetermined fertilizer's amount is treated as a data sparsity problem that is solved primarily by adding side features such as soil fertilizer level, land size, and soil chemical properties. We specifically introduce a new method of wide matrix factorization that maps the embedded land and nutrient vectors linearly to obtain the feature representation. Furthermore, the fully connected multi-layer perceptron captures the non-linear higher-order interaction between land and nutrients to deeply learn the historical fertilizer recommendations of the land. An exploratory outcome on two real-time datasets indicates the superior performance of the proposed method over recent state-of-the-art approaches. It reveals that the proposed deep network predicts well on nutrients data for precise fertilizer recommendation, which support farmers to increase the crop yield. [ABSTRACT FROM AUTHOR]

• This work included sentiment analysis into recommender systems. • Reviews and ratings for products are combined using deep learning. • This work justified that sufficient number of hidden layers may cut training time. • Backpropagation neural networks is combined for better results. • Results are better compare to other recommender systems. Identifying user preferences is a complex operation, which makes its automa- tion challenging, and existing recommendation systems that rely on one of the parameters ratings or reviews are incapable of performing effectively. In this work, an ensemble based deep learning method is used to develop a recommen- dation model in which both ratings and reviews are analyzed simultaneously. The first step is to identify the features required in our work which is reviews and ratings of a product. Next step is to conduct sentiment analysis on reviews to get numerical values for subsequent processing which gives output in terms of polarity and subjectivity, where polarity identifies the emotions expressed in reviews ranging from −1 to +1 and subjectivity identifies the relevance of par- ticular review. Reviews and ratings as features used in this work as inputs for ensemble based deep neural network. For this, both the inputs need to be in same range so scaling is done. The recommendation method used in this work is based on deep learning which employs back propagation neural networks with many hidden layers and varying nodes, facilitating rapid learning. In this pa- per, few selected representative deep learning architectures with varied amounts of layers concealed to improve the learning capability of this model. However, model of the recommendation system can yet be improved, such as the inabil- ity to explain the deep learning recommendation system, which diminishes its credibility. We assessed our work based on hidden layer architectural variations using measures such as precision, loss, and execution time. [ABSTRACT FROM AUTHOR]

This aims to propose a collaborative filtering algorithm based on embedding representation and word embedding techniques, namely UI2vec. According to the joint feature extraction network designed in this paper, UI2vec embeds users and items on the potential space at the same time, and uses the item similarity between them to predict the user's content of interest. Then a generative model VUI2vec with more stable performance is proposed based on UI2vec, which maps users and items as independent Gaussian distributions and obtains the approximate posterior distribution of both by variational inference. The recommendation performance of UI2vec and VUI2vec is evaluated on TaFeng, Movielens, and Netflix datasets. The impact of important superparameters within the model on performance is investigated. The experimental results show that compared with the baseline model, the proposed methods performs consistently well. • UI2vec and VUI2vec as collaborative filtering-based recommendation systems. • It is based on embedding representation of users or items. • Performance is competitive with the state of the art on real-world datasets. • Results encourage integrating collaborative filtering with deep learning. [ABSTRACT FROM AUTHOR]

Tiny object detection is one of the challenges in the field of object detection, which can be applied in a variety of fields. Thanks to the advances in deep learning, significant improvement has been made in image object detection. However, the performance of tiny object detection still needs to be considerably enhanced. In this paper, we proposed a novel feature pyramid composite neural network structure comprising two modules: the context enhancement module (CEM) and feature purification module (FPM). The top-to-bottom input of the feature pyramid network into the multi-scale dilated convolution features in the CEM can augment the context information. When fusing multi-scale features in the FPM, the feature purification procedures for channel and space dimensions are employed to eliminate conflicting information, and tiny objects are more noticeable in contradictory information. Furthermore, a new data-enhancement strategy is introduced to increase the contribution of tiny objects in the loss function, which is named copy-reduce-paste and improves the balance of training samples. Overall, the experiments on the VOC dataset show that the AP s score can reach 16.9% and the IOU is 0.5:0.95 for the suggested method. The AP s score is more than 3.9% that of YOLOV4, more than 7.7% that of CenterNet, and more than 5.3% that of RefineDet. On TinyPerson dataset, our AP s score is more than 0.8% that of YOLOV5, providing a new alternative solution for the tiny-object detection research community. • Tiny target data enhancement is used to improve the contribution of its loss in training. • Context enhancement is used to supplement the context information for tiny targets. • Feature purification is used to filter the conflict information after feature fusion. [ABSTRACT FROM AUTHOR]

Recently, Deep Neural Networks (DNNs) have proved their capability to model nonlinear relationships between users and items in recommender systems. Therefore, many researchers have exploited the DNN to model the non-linearity of individual interactions between users and items. However, it has been barely studied to model the non-linearity of a user's sequential item interactions that may be the fundamental clue to represent a user's interest naturally. This paper proposes a novel Deep Sequential Embedding for a single domain Recommendation based on implicit feedback, named DSER. It exploits Doc2vec techniques to convert sequential user–item interactions, rather than textual information, to the feature vectors of users and items. In our framework, a user's interaction is like a document containing items as words in the word embedding techniques. The obtained user and item vectors are then fed to Multi-Layer Perceptron (MLP) to model a high level of non-linearity for user–item interactions. Finally, the non-linearity of MLP is fused with the linear user–item relationships trained by matrix factorization to predict the possibility that a target user likes items. Extensive experiments on four real-world datasets demonstrate significant improvements of our DSER over state-of-the-art approaches. Moreover, empirical evidence shows that using word embedding offers considerably better predictive performance on sparse data of the smart tourism field, whose new user ratio is relatively high. • It is the first study using Doc2vec to model sequential interaction and fuse with DNN. • DSER outperforms the other SOTA approaches in HR and NDCG on four real-world datasets. • It shows significant superiority on the tourism datasets having a high new user ratio. • We empirically find out how Doc2vec works with DSER in terms of data characteristics. [ABSTRACT FROM AUTHOR]

Recommender Systems are a subclass of machine learning systems that employ sophisticated information filtering strategies to reduce the search time and suggest the most relevant items to any particular user. Hybrid recommender systems combine multiple recommendation strategies in different ways to benefit from their complementary advantages. Some hybrid recommender systems have combined collaborative filtering and content-based approaches to build systems that are more robust. In this paper, we propose a hybrid recommender system, which combines Alternating Least Squares (ALS) based collaborative filtering with deep learning to enhance recommendation performance as well as overcome the limitations associated with the collaborative filtering approach, especially concerning its " cold-start " problem. In essence, we use the outputs from ALS (collaborative filtering) to influence the recommendations from a Deep Neural Network (DNN), which combines characteristic, contextual, structural and sequential information, in a big data processing framework. We have conducted several experiments in testing the efficacy of the proposed hybrid architecture in recommending smartphones to prospective customers and compared its performance with other open-source recommenders. The results have shown that the proposed system has outperformed several existing hybrid recommender systems. [ABSTRACT FROM AUTHOR]

• We proposed a novel recommendation system based on collaborative filtering. • It is a combination of explicit (Intra & Inter) and implicit feedback interaction couplings. • It solves the cold start and sparsity problems of collaborative filtering methods. Due to the expansion of e-business, the availability of products on the internet has massively increased. Finding suitable stuff from the vast array of products available on the internet is a time-consuming task. Collaborative Filtering (CF) is the most effective recommendation method for providing users with the ability to identify relevant content and, therefore, increase engagement. However, CF has several flaws, including data sparsity and cold start problems. These are ongoing research questions that pose major hurdles to the precision of the algorithms. Therefore, in this work, a novel neural recommendation model is proposed based on non-independent and identically distributed (Non-IID) for CF by incorporating explicit and implicit coupling interaction. The explicit interactions consist of two models, namely Intra-coupling interactions within users and items, and Inter-coupling interactions between different users and items concerning the attributes of users and items. The Intra-coupled model learns using deep learning convolutional neural networks and is combined with the Inter-coupled model. Besides explicit coupling interactions, we present a Generalized Matrix Factorization Bias (GMFB) model that systematically trains the implicit user-item coupling. Finally, we combined with explicit and implicit coupling interactions within and between users and items accompanying the extra information about users and items under a framework called "IntegrateCF." Extensive experiments on two large real-world datasets have shown that the proposed model performs better than existing methods. [ABSTRACT FROM AUTHOR]

A session-based recommender system (RS) forecasts the item to be clicked next by taking advantage of the item sequence in the current session. The deep learning models, such as the recurrent neural network (RNN) and the graph neural network (GNN), are recently applied in the session-based recommendation. However, to the best of our knowledge, the existing methods ignore the fusion of the frequent sub-sequences which refer to those item sub-sequences that appear frequently in different sessions. Intuitively, the more frequently an item sequence appears, the more important it could be. In this paper, we propose the frequent sub-sequence modeling (FSM) method, by utilizing both the local and global information of item sequences. More specifically, we use the item sequence in the current session to build a local session graph and the frequent sub-sequences to construct a global session graph. In addition, we design a gated layer to control the contributions of these two session graphs. Our method is capable of learning both the session-level and global-level item embedding, and thus supports accurate predictions. It is tested on four benchmark datasets and the results show that our method is superior to state-of-the-art approaches consistently. • Investigate the distribution of the frequent sub-sequences. • Devise the FSM method to capture the importance of the frequent sub-sequences. • Verify the FSM's superiority in session-based RS by comparing it with baselines. [ABSTRACT FROM AUTHOR]

• Multi-view feature interactive learning was introduced for sequential recommendation. • Interactive feature learning balanced the contradiction between static and dynamic features. • DeepInteract was proposed to improve performance via multi-view feature learning. • Interactive features were demonstrated to be important for sequential recommendation. • DeepInteract outperformed state-of-the-art deep models significantly. Deep learning models have been successfully applied in sequential recommendations. However, previous studies ignored the interaction between static and dynamic features of both items and users, thus fail to exactly capture users' current preferences. To overcome this limitation, we first conducted feature representations from static, dynamic and interactive views and constructed corresponding feature mining modules. Then, based on the multi-view feature mining modules, we proposed a deep learning-based model, namely DeepInteract, to learn the interaction of multi-view features of both item profiles and user behaviors for sequential recommendation. Experimental results on three real-world datasets demonstrated that DeepInteract outperforms state-of-the-art methods not only on recommendation performance but also on stability and robustness. Furthermore, we used ablation experiments to investigate the importance of three feature mining modules on various measures of recommendation performance. It was demonstrated that interactive features play the most important role for sequential recommendation. [ABSTRACT FROM AUTHOR]

Nowadays, there has been a rapidly increasing number of scientific submissions in multiple research domains. A large number of journals have various acceptance rates, impact factors, and rankings in different publishers. It becomes time-consuming for many researchers to select the most suitable journal to submit their work with the highest acceptance rate. A paper submission recommendation system is more critical for the research community and publishers as it gives scientists another support to complete their submission conveniently. This paper investigates the submission recommendation system for two main research topics: computer science and applied mathematics. Unlike the previous works (Wang et al., 2018; Son et al., 2020) that extract TF–IDF and statistical features as well as utilize numerous machine learning algorithms (logistics regression and multiple perceptrons) for building the recommendation engine, we present an efficient paper submission recommendation algorithm by using different bidirectional transformer encoders and the Mixture of Transformer Encoders technique. We compare the performance between our methodology and other approaches by one dataset from Wang et al. (2018) with 14012 papers in computer science and another dataset collected by us with 223,782 articles in 178 Springer applied mathematics journals in terms of top K accuracy (K = 1 , 3 , 5 , 10). The experimental results show that our proposed method extensively outperforms other state-of-the-art techniques with a significant margin in all top K accuracy for both two datasets. We publish all datasets collected and our implementation codes for further references. 1 1 https://github.com/BinhMisfit/PSRMTE. • Bidirectional transformer encoders can improve the performance of the paper submission recommendation system. • The Mixture of Transformer Encoders framework shows the efficiency in the paper submission recommendation problem. • Proposed techniques can surpass other recent techniques on two datasets related. [ABSTRACT FROM AUTHOR]

Research about recommender systems emerges over the last decade and comprises valuable services to increase different companies' revenue. While most existing recommender systems rely either on a content-based approach or a collaborative approach, there are hybrid approaches that can improve recommendation accuracy using a combination of both approaches. Even though many algorithms are proposed using such methods, it is still necessary for further improvement. This paper proposes a recommender system method using a graph-based model associated with the similarity of users' ratings in combination with users' demographic and location information. By utilizing the advantages of Autoencoder feature extraction, we extract new features based on all combined attributes. Using the new set of features for clustering users, our proposed approach (GHRS) outperformed many existing recommendation algorithms on recommendation accuracy. Also, the method achieved significant results in the cold-start problem. All experiments have been performed on the MovieLens dataset due to the existence of users' side information. • We propose a recommendation system using graph-based features and Autoencoders. • We investigate and analyze each step of the method to find the best setup. • We examine the method with the most popular evaluation metrics and datasets. • Results show an improvement in comparison with basic and state of the art methods. [ABSTRACT FROM AUTHOR]

Explainable rating predication becomes challenging with the largely growing number of information and items. Of particular interest is to capture users' preferences for various items by using textual reviews to achieve accurate and interpretable recommendations. In this paper, we report an aspect-aware explainable neural attentional recommender model for rating predication (AENAR) and this model enables intelligent predication and recommendation by capturing the varying aspect attentions that users pay to different items. The experimental results based on six public datasets reveals that the designed model consistently outperforms five existing state-of-the-art alternatives. Furthermore, the designed attention network allows to highlight the context-aware information in textual reviews that unambiguously suggest users' aspect-level preference for their desired items, improving the interpretability of the rating prediction. • Applied CNN to extract fine-grained textual feature information. • An attention mechanism designed for capturing the diverse user preferences. • Explainable model that can trace back to keywords in review via attention mechanism. [ABSTRACT FROM AUTHOR]

• A probabilistic model is proposed to evaluate the rating profiles of users. • A reliable mechanism is proposed to enhance the users' rating profiles. • A deep sparse autoencoder is used to extract latent features of users. • An integration approach is introduced to make recommendations. Deep neural networks have been extensively employed in many applications such as natural language processing and computer vision. They have attracted a lot of attention in designing recommender systems due to their recent significant achievements. These models are mainly used in recommender systems to generate the users' latent features according to the input data such as user-item rating and trust matrixes. However, such input data are significantly sparse leading to reduce the effectiveness of deep learning models in generating the latent features. In addition, the existing deep learning based recommendation models ignore the reliability measure in their recommendation process. To address these challenges, in this paper, a trust-aware recommendation method based on deep sparse autoencoder is proposed. Specifically, an effective probabilistic model is proposed to determine how many ratings are required for each user to produce an accurate prediction. Moreover, for enhancing the rating profiles of users with insufficient ratings, an efficient mechanism is developed based on the utilization of implicit ratings. In this mechanism, the implicit ratings are selected based on a reliability measure to guarantee their effectiveness. Then, the reliable enhanced rating profiles and trust statements are considered as the input data of deep sparse autoencoder to generate the users' latent features. Finally, a similarity function that utilizes the generated latent features is developed to obtain similarity values between users and make recommendations. Massive experiments are carried out on two datasets to assess the effectiveness of the proposed method in comparison to the-state-of-the-art models. The results of experiments demonstrate that the proposed method can significantly beat other models in producing more accurate recommendations. [ABSTRACT FROM AUTHOR]

• Accurate and scalable CTR prediction model for real-time recommendations. • Showcase effectiveness of model distillation framework. • Significance of scalability in real-time recommendations. Accuracy and scalability are critical to the efficiency and effectiveness of real-time recommender systems. Recent deep learning-based click-through rate prediction models are improving in accuracy but at the expense of computational complexity. The purpose of this study is to propose an accurate and scalable click-through rate (CTR) prediction model for real-time recommendations. This study investigates the complexity, accuracy, and scalability aspects of various CTR models. This work ensembles top CTR models using a gated network and distill into a deep neural network (DNN) using a knowledge distillation framework. Distilled DNN model is more accurate and 20x scalable than any of the individual CTR models. The low latency of distilled model makes it scalable and fit for deployment in real-time recommender systems. The proposed distillation framework is extensible to integrate any CTR models to the ensemble and can be distilled to any neural architecture. [ABSTRACT FROM AUTHOR]

• Representation Learning for capturing user's intention of the basket. • Specific neural network layer design provides the scalability to large scale data. • Linear function combines different components without nonlinear transformation. Within-basket recommendation, which predicts the related item to be added to the basket from the item corpus, is prevalent in grocery shopping and e-commerce. Besides user-item collaborative filtering information, the retail platform also needs to combine the items that the user currently owns to make a recommendation. Previous work solves the task by rule mining or incorporating various types of associations. However, the representation of the basket and the high-order feature interaction is hardly investigated previously. In this work, we propose a deep learning-based model named DBFM (Deep Basket-Sensitive Factorization Machine) to address the task. We first make a personalized representation for a basket based on its constituent items instead of ID by latent factor learning, which improves the generalization of the model to baskets. Then we combine both low-order and high-order feature patterns to capture the sophisticated structures from inputs. Finally, a linear function is used to integrate with the results of different components. Experiments on three real-world datasets demonstrate higher performance of our model over state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Financial services are increasingly personalized due to the rapid development of artificial intelligence. In particular, precision marketing is the ultimate goal in providing financial services. However, precision marketing is hindered by problems in model scalability, cold starts, and insufficient transaction data points for many customers and products. Some researchers have attempted to solve these problems by incorporating deep learning into the collaborative filtering algorithm. This has improved the performance of the traditional matrix factorization algorithm, but it comes at the cost of imposing a limit on the capacity of the recommendation model to capture the complex interaction features. In this paper, we propose a graphical deep collaborative filtering (GraphDCF) algorithm for providing personalized mutual fund recommendations. The graph-structured network is constructed by connecting customer nodes with similar purchases and redeemed trading orders in the sequential view. In this manner, we can model different latent relationships among customers who have similar shopping habits. Subsequently, the corresponding embedding vector for each customer node is generated through an aggregate function based on similarity in transaction behaviors. Finally, to provide personalized recommendations, in addition to customer features and mutual fund attributes, the proposed deep embedded collaborative filtering framework predicts how willing a customer is to purchase a mutual fund. Experimental results on a real-world data set from Taiwan Commercial bank demonstrated indicated that DECF approaches outperformed deep learning methods such as DCF and NCF. The proposed GraphDCF algorithm outperformed (by up to 2.3%) other frequently used approaches. • A personalized mutual fund recommender system is developed for precision marketing. • The mutual funds are trained and represented by their periodic performance. • The dynamic trading behaviors of customers are also modeled in this framework. • The proposed graphical deep collaborative filtering can achieve high accuracy. [ABSTRACT FROM AUTHOR]

As linear TV remains a major source for media consumption, multiple stakeholders such as broadcasters and advertisers are interested in the prediction of the next programs to be watched by TV viewers. Such predictions are quite challenging given the nature of the domain, where viewing TV is not just an individual activity but is also influenced by various contextual factors. We aim to address this challenge by integrating temporal aspects of linear TV – in the form of sequences – into the next program prediction process. A user profile is built from sequences of 24-hour TV programs' views, at intervals of 15 min. Such profiles allow us to capture viewing preferences and sequential patterns, for predicting the next program/genre/category to be viewed at any time. We conducted several experiments using naive approaches, hidden Markov models and deep learning, juggling between accuracy and interpretability of the model. The precision@1 results were extremely promising (0.836 for next category prediction with LSTM, comparing to 0.57 an 0.34 with Naive approaches, and 0.366 with hidden Markov models). • Modeling user viewing habits in the TV domain as sequences of items. • The user profile is represented by a vector of a 24-hour 15-minute slot views. • Data shifting as a data augmentation method to predict any program at any time. • Experiments were conducted with LSTM, Hidden Markov Models and a naive approach. [ABSTRACT FROM AUTHOR]

• The proposed method addresses the unbalanced and data sparsity problems. • Sentiment bias and temporal dynamics are explored to improve recommendation quality. • We design a deep learning method based on long- and short-term user preferences. • The performance of the proposed models is evaluated on several Amazon datasets. Recommendation systems rely on the historic data of users' purchases and their feedbacks to profile their preferences and make future recommendations. Most of these systems usually employ Collaborative Filtering (CF) models to analyze users' ratings and infer the latent factors which represent the user and item features in k -dimensional latent space. However, the historical rating data used for recommendations are usually sparsed and unbalanced. Various approaches have been used to resolve these issues by combining the user's ratings and reviews to better capture the user's sentiments and make accurate recommendations. Other challenges comprise changes in users' preferences and items' perceptions over time. Therefore, this paper presents a new Sentiment Scoring Model (SSM) based on Long-/Short-Term Memory and a combination function that catches the sentiment bias between user rating and review to relieve the sparsity and unbalanced dataset. Next, we proposed an Adaptive LSTM (ALSTM) method that can model the drifting of user and item features to improve the recommendation accuracy. We show the performance of our model on the three real-world rating datasets from Amazon reviews, which comprises Fine Food, Baby, and Cell-phone & Accessories categories. The result shows the superiority of our proposed model over the existing static and dynamic models. The statistical test shows that all the performance gains are significant at p < 0.05. [ABSTRACT FROM AUTHOR]

• a time-aware attention mechanism to model dynamic user preferences. • a two-stage DL model with representation learning and matching function learning. • a CF model with low-order and high-order feature interaction learning ability. Deep learning (DL) techniques have been widely used in recommender systems for user modeling and matching function learning based on historical interaction matrix. However, existing DL-based recommendation methods usually perform static user preference modeling by using historical interacted items of the user. In this article, we present a time-aware deep CF framework which contains two stages: dynamic user preference modeling based on attention mechanism and matching score prediction based on DL. In the first stage, short-term user preferences are modeled by the time-aware attention mechanism that fully considered the predicted item, the recent interacted items and their interaction time. The resulting short-term preferences are combined with long-term preferences for dynamic user preference modeling. In the second stage, high-order user-item feature interactions are learned by two types of DL models, Deep Matrix Factorization (DMF) and Multiple-Layer Perception (MLP), and the feature interaction vectors of the two models are fused in the last layer of the model to predict the matching score. Extensive experiments on five datasets indicate that our method is superior to the existing time-aware and DL-based recommendation methods in top-k recommendations significantly and consistently. [ABSTRACT FROM AUTHOR]

In the past decades, recommendation systems have provided lots of valuable personalized suggestions for the users to address the problem of information over-loaded. Collaborative Filtering (CF) is one of the most commonly applied and successful recommendation approaches, which refers to using the preferences of groups with similar interests to recommend information to other users. Recently, in addition to the traditional matrix factorization techniques, deep learning methods have been proposed to learn more abstract and higher-level representations for recommendation. However, most previous deep recommendation methods learn the higher-level feature representations of users and items through an identical model structure, which ignores the different characteristics of the user-based and item-based data. In addition, the rating matrix is usually sparse which may result in a significant degradation of recommendation performance. To address these problems, we propose a representation learning method with Collaborative Autoencoder for Personalized Recommendation (CAPR for short). In this method, user-based and item-based feature representations are learned by two different autoencoders for capturing different features of the data. Meanwhile, items' attributions are combined into the feature representations with semi-autoencoder for alleviating the sparsity problem. Extensive experimental results confirm the effectiveness of our proposed method compared to other state-of-the-art matrix factorization methods and deep recommendation methods. • Two different autoencoders are used to capture characteristics for users and items. • Manifold regularization is integrated into autoencoder for user's features learning. • The comprehensive experiments evaluate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Personalized recipe recommender systems help users mine certain dishes they want to find and even really desire, which play a significant role in matching dishes, balancing nutrients, and preventing non-communicable diseases. Generally, customer's preferences or needs vary from person to person, and people are often reluctant to accept recommended food without reasonable explanation, especially when their demands are not explicitly addressed. In this paper, we are devoted to providing recipe suggestions accompanied by rational interpretations generated from images or videos. First, we construct a recipe knowledge graph (RcpKG) through the use of multi-modality and hierarchical thought, which focuses on the underlying demands of users and the consideration of multiple fine-grained factors. On this basis, a novel multi-modal recipe recommendation method via the knowledge graph (RcpMKR) is proposed, which represents nodes in multiple aspects and performs multi-relational graph structure extraction of the RcpKG. It not only takes into account local associations within the graph but also global information, and incorporates user concerns at different levels. Then, we adopt BERT-based multi-modal models and generative adversarial networks to generate interpretations. Additionally, dynamic convolution and random synthetic attention are utilized in our work to discriminate among features. Experimental results show that the proposed method and BERT-based fusion models improve recipe recommendation performance and explanation generation. Specifically, the precision of the RcpMKR method through RcpKG, user concerns and graph convolutional network improves by 7.82%, and the viExpCBTBERT method via 2D&3D convolutional neural networks for developing text interpretations enhances the F1-score by 10% compared with the baseline. • Propose an extendable multi-modal recipe fusion framework based on knowledge graph. • Constructing a recipe knowledge graph based on user preferences. • Use multi-modality to improve recommendation results. • Utilizing deep learning to generate explanations from multimedia data. [ABSTRACT FROM AUTHOR]

Recently, deep learning has dominated the recommender system, as it is able to effectively capture nonlinear and nontrivial user–item relationships, and perform complex nonlinear transformations. However, there are still some issues with respects to the existing methods. Firstly, they always treat user–item interactions independently, and may fail to cover more complex and hidden information that is inherently implicit in the local neighborhood surrounding an interaction sample. Secondly, by quantifying the dependence degree of user–item sequences, it demonstrates that both short-term and long-term dependent behavioral patterns co-exist. Unfortunately, typical deep learning methods might be problematic when coping with very long-term sequential dependencies. To address these issues, we propose a novel unified neural collaborative recommendation algorithm that capitalizes on memory networks for learning attention embedding from implicit interaction (NCRAE). Particularly, the attention is capable of learning the relative importance of different users and items from user–item interaction sequences, which provides a better solution for concentrating on inputs and helps to better memorize long-term sequential dependencies. Extensive experiments on three real-world datasets show significant improvements of our proposed NCRAE algorithm over the competitive methods. Empirical evidence shows that using memory networks for learning attention embeddings of users' implicit interaction yields better recommendation performance. • We design a neural network to learn co-attention from user–item interactions. • The co-attention provides a better solution to memorize long-term dependencies. • We design a neural collaborative algorithm for incorporating co-attention. • Our algorithm is augmented with an external memory and neural attention. • Our algorithm can capture short-term and long-term dependence well. [ABSTRACT FROM AUTHOR]

[Display omitted] • PGRec is a novel graph-based ranking-oriented recommendation framework. • PGRec models the user preferences, by a novel heterogeneous graph called PrefGraph. • PGRec embeds nodes in graph with a hybrid factorization and deep learning method. • Unknown preferences are predicted by applying Neural Network on node embeddings. • PGRec outperforms the baseline methods in terms of the NDCG in several datasets. Graph-based recommender systems (GRSs) analyze the structural information available in the graphical representation of data to make better recommendations, especially when direct user-item relation data is sparse. Ranking-oriented GRSs mostly use graphical representation of preference (or rank) data for measuring node similarities, from which they can infer recommendations using neighborhood-based methods. In this paper, we propose PGRec, a novel model-based ranking-oriented recommendation framework. Unlike many other graph-based methods, PGRec extracts vector representations for users and preferences from a novel graph structure called PrefGraph , which models entity relations, feedbacks, and content. A general graph-embedding process is improved and applied to extract vector representations for entities. The resulting embeddings are then used for predicting the target user's unknown pairwise preferences by a neural network based on which a recommendation list is generated for the target user. We have evaluated the proposed method's performance against the state of the art model-based and neighborhood-based recommendation algorithms. Our experiments show that PGRec outperforms the baseline algorithms in terms of the NDCG metric in several datasets. [ABSTRACT FROM AUTHOR]

• We construct a Heterogeneous Bibliographic Network, which exploits semantics among the network objects. • We propose a GAN-based network embedding model to address the network sparsity problem. • We propose a citation recommendation model to produce personalized results corresponding to researchers' preferences. • We conduct extensive experiments on two real-world datasets and prove the significance of the proposed model. The variety and plethora of research papers available on the Web motivated researchers to propose models that could assist users with personalized citation recommendations. In recent years, citation recommendation models using Network Representation Learning (NRL) methods have shown promising results. Nevertheless, existing NRL-based models are limited in terms of exploiting semantic relations and contextual information between the objects of bibliographic papers' networks. Additionally, these models cannot adequately explore the structure of heterogeneous information networks, topical relevance, and relevant semantics. Consequently, they suffer from network sparsity and inadequate personalization problems. To overcome these shortcomings, we present a network embedding model termed as Global Citation Recommendation employing Generative Adversarial Network (GCR-GAN). The proposed model exploits the Heterogeneous Bibliographic Network (HBN) to generate personalized citation recommendations. In particular, the proposed model utilizes semantic relations corresponding to the objects of the heterogeneous bibliographic network and captures network structure proximity employing the Scientific Paper Embeddings using Citation-informed Transformers (SPECTER) and Denoising Auto-encoder networks to learn semantic-preserving graph representations. Compared to baseline models, the recommendations generated by our model over the DBLP and ACM datasets prove that it outperforms baseline methods by gaining almost 11% and 12% improvement in terms of Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (nDCG) metrics, respectively. Furthermore, we analyzed the effectiveness of the proposed model considering network sparsity issue, where our model gains almost 7% better recall@100 score against the second-best counterpart. [ABSTRACT FROM AUTHOR]

• Variational CF-NADE generative model for intractable density latent vectors. • Reusing the shared weights for fast sampling and quality of ratings produced. • User-item ratings and metadata features utilized with no independent assumptions. • User-item interaction function learned using abstraction and generation phases. • Experiments on MovieLens and Netflix datasets validate our approach. Generative models in some way reflect probability distributions over multiple variables. For latent variable generative models, the density function is intractable and these models thus make some independent assumptions to minimize the number of factors and, in turn, the number of parameters in the model. Autoregressive models such as Neural Autoregressive Distribution Estimator (NADE) have a tractable density with no latent variables and the goal is to learn a joint distribution over input variables. Autoregressive models have several advantages over latent variable models based on samples (ratings in our case) produced, independent assumptions, and tractable density functions. NADE models as such have seen less scrutiny in recommendation literature. Few papers have utilized NADE for collaborative filtering tasks but have limited capability as they could not perform abstraction, which restricts the model from learning accurate latent representations and the sampling rate was also slow. Further, the rating prediction process is calculative and not generative and model retraining is also required whenever a new user or new item gets added to the underlying dataset, limiting the model performance. Existing neural network based methods are deterministic and they can't account for the uncertainty in user-item latent representations. To overcome these issues, we propose a hybrid neural Variational CF-NADE generative model that extends NADE for collaborative filtering tasks using variational Bayesian autoencoder to characterize the uncertainty in ratings. We do not make any independence assumption for the users or items rating vectors but use the default factorization to learn a joint distribution over ratings. Therefore, the proposed model uses a tractable density distribution to have good quality ratings generated. The model learns comprehensive user-item interaction function using abstraction and generation phases. The variational autoencoder performs abstraction on user-item metadata information which acts as a regularizer to the rating generated. By reusing the shared weights for fast sampling among different predicted ratings, we speed up the computation with an increased performance by decomposing the rating vector into conditionals by the chain rule and thus reducing the number of parameters. Experiments on three real-world datasets show that the proposed model beats recent state-of-the-art approaches for collaborative filtering tasks. [ABSTRACT FROM AUTHOR]

• A novel graph embedding using dominating-set theory and deep learning is proposed. • A strategy to calculate experts' scores based on expert embeddings is proposed. • We present data for a co-author network by crawling author profiles from Scopus. A collaborative network is a social network that is comprised of experts who cooperate with each other to fulfill a special goal. Analyzing this network yields meaningful information about the expertise of these experts and their subject areas. To perform the analysis, graph embedding techniques have emerged as an effective and promising tool. Graph embedding attempts to represent graph nodes as low-dimensional vectors. In this paper, we propose a graph embedding method, called ExEm , that uses dominating-set theory and deep learning approaches to capture node representations. ExEm finds dominating nodes of the collaborative network and constructs intelligent random walks that comprise of at least two dominating nodes. One dominating node should appear at the beginning of each path sampled to characterize the local neighborhoods. Moreover, the second dominating node reflects the global structure information. To learn the node embeddings, ExEm exploits three embedding methods including Word2vec, fastText and the concatenation of these two. The final result is the low-dimensional vectors of experts, called expert embeddings. The extracted expert embeddings can be applied to many applications. In order to extend these embeddings into the expert recommendation system, we introduce a novel strategy that uses expert vectors to calculate experts' scores and recommend experts. At the end, we conduct extensive experiments to validate the effectiveness of ExEm through assessing its performance over multi-label classification, link prediction, and recommendation tasks on common datasets and our collected data formed by crawling the vast author Scopus profiles. The experiments show that ExEm outperforms the baselines especially in dense networks. [ABSTRACT FROM AUTHOR]

• Solving cold start problem in cross domain recommender systems. • Trust Aware Cross Domain Deep Neural Matrix Factorization model for rating prediction. • Modeling of linear and non-linear features using deep learning. • Incorporating implicit trust relation as trust degree via Ant Colony Optimization. • Evaluating the proposed model on ecommerce dataset that is AliExpress. Over the years, cross-domain and trust-based recommendation systems are proven to be very helpful in solving issues pertaining to data sparsity and cold start. Many e-commerce sites used recommender systems as business tools for increasing their sale productivity and help their customers in finding suitable products. However, due to sparse rating and lack of historical information, such systems cannot generate effective recommendations. Matrix factorization and deep learning techniques have been the focus of research community for the last few years to solve the problem of data sparsity and cold start. In this paper, we have proposed a model called Trust Aware Cross-Domain Deep Neural Matrix Factorization (TCrossDNMF) that predicts rating of an item for an active user and solves user cold start problem in the cross-domain scenario of 'User Overlap' in e-commerce system. TCrossDNMF model is divided into four main steps: i) Features learning that learns the users' features using a latent factor model and then finds the similarity between users of source and target domains. As the users are shared between two domains, the proposed model learns the common information and transfers the knowledge from a source to target domain. ii) Ranking that finds set of similar users (neighbors), and then filters out the dissimilar users based on similarity threshold θ , and then generates a bipartite trust graph from these reduced set of users and executes Ant Colony Optimization, to find trustworthy neighbors for an active user. iii) Weighting computes the trust degree between an active user and his or her top-k neighbors. iv) Prediction that trains the TCrossDNMF model using multilayer perceptron (MLP) and generalized matrix factorization (GMF) by representing user-item interactions in higher dimensions and ensembles the GMF and MLP with trust information for rating prediction. We evaluated proposed model on a real dataset collected from a popular e-commerce retail service 'AliExpress'. We used categories available in 'AliExpress' as source domain and a target domain. For observing the performance of proposed model, we took six domains that have a higher ratio of sparsity. The proposed model is evaluated by using MAE, RMSE and F-measure metrics and compared it with baselines. The experiments show that the proposed model is a viable solution for the mentioned problem with significant improvements in results. [ABSTRACT FROM AUTHOR]

• Heterogenous information network based recommendation is investigated. • An attention-aware metapath-based network embedding approach is proposed. • Each metapath is modeled as a weighted homogenous information network. • A self-attention mechanism generates integrated representations of users and items. • Deep neural network methods are used in the final stage of prediction. Heterogeneous information network (HIN) attracts increasing attention from the communities of recommender systems. HIN based recommendation methods can help overcome the difficulties of data sparsity and cold start. The majority of the existing HIN based recommendation methods use path-based semantic similarity between users and/or between items on HINs. However, the existing HIN based recommendation methods using metapath disregard the semantic differences among multiple metapaths (i.e., inter-metapaths) and the influence differences among neighbor pairs in each individual metapath (i.e., intra-metapaths). To solve these problems, we propose an attention-aware metapath-based network embedding for HIN based recommendation. To obtain additional semantic information, our method generates multiple metapath-based weighted homogeneous networks to model the auxiliary information of users and items of HIN. Thereafter, we design a novel self-attention integration to integrate multiple semantic information from multiple weighted homogenous information networks. Lastly, we utilize three deep neural network methods to model the implicit relations between users and items for the rating prediction task. Experimental results of three real-world datasets demonstrate that the proposed model outperforms existing state-of-the-art recommendation methods, solves the data sparsity problem, and models the multiple semantic information of users and items. [ABSTRACT FROM AUTHOR]

• A sequence-based deep learning model for movie recommendation is proposed. • User Movie Interest Space (UMIS) is introduce to reflect user similarities. • Three indexes are developed to describe UMIS from multiple points of interest. • Rating sequence predictions based on UMIS perform better than those without UMIS. • The method greatly improves dynamic recommendation performance than other methods. Exploring user interest behind massive user behaviors is essential for online recommendations. Although recommendation models have been proposed recently with great success, existing studies ignore not only the timeliness of online users' behaviors in terms of their interest, but also the sequential characteristics of their behaviors. To overcome this limitation, we construct a User Movie Interest Space (UMIS) model based on the sequential ratings of users. We define three indexes to elucidate the features of the interest of users for UMIS, which describe different patterns of behaviors of users related to their interests. Based on UMIS we propose a deep learning model named Dynamic Interest Flow (DIF) to provide dynamic movie recommendations. The DIF model achieves intelligently multi-dimensional observations on a user's interest space and to predict simultaneously a variety of their future interests. Experimental results indicate that DIF outperforms traditional rating-based models and other state-of-the-art deep learning models. Results also demonstrate that modeling a dynamic recommendation as a sequential prediction is supposed to obtain outstanding advantages. [ABSTRACT FROM AUTHOR]

The worldwide adoption of mobile devices is raising the value of Mobile Performance Marketing, which is supported by Demand-Side Platforms (DSP) that match mobile users to advertisements. In these markets, monetary compensation only occurs when there is a user conversion. Thus, a key DSP issue is the design of a data-driven model to predict user conversion. To handle this nontrivial task, we propose a novel Multi-objective Optimization (MO) approach to evolve Decision Trees (DT) using a Grammatical Evolution (GE), under two main variants: a pure GE method (MGEDT) and a GE with Lamarckian Evolution (MGEDTL). Both variants evolve variable-length DTs and perform a simultaneous optimization of the predictive performance and model complexity. To handle big data, the GE methods include a training sampling and parallelism evaluation mechanism. The algorithms were applied to a recent database with around 6 million records from a real-world DSP. Using a realistic Rolling Window (RW) validation, the two GE variants were compared with a standard DT algorithm (CART), a Random Forest and a state-of-the-art Deep Learning (DL) model. Competitive results were obtained by the GE methods, which present affordable training times and very fast predictive response times. • We propose two novel methods (MGEDT and MGEDTL) to evolve decision trees. • The methods optimize the classification performance and model interpretability. • The methods were designed for the Mobile Performance Marketing domain. • A realistic experimentation was conducted using big data (6 million records). • Competitive results were obtained by the proposed MGEDT and MGEDTL methods. [ABSTRACT FROM AUTHOR]

The huge amount of research papers on the web makes finding a relevant manuscript a difficult task. In recent years many models were introduced to support researchers by providing personalized citation recommendations. Moreover, deep learning methods have been employed in this domain to improve the quality of the final recommendations. However, a thorough study that classifies citation recommendation models and examines their (a) strengths and weaknesses, (b) evaluation metrics used, (c) popular datasets, and challenges faced is missing. Therefore, with this survey, we present a new classification approach for deep learning models that provide citation recommendation. Our approach uses the following six criteria: data factors, data representation methods, methodologies, types of recommendations used, problems addressed, and personalization. Additionally, we present a comparative analysis of those models that use the same set of evaluation metrics and datasets. Moreover, we examine hot upcoming issues and solutions in light of explored literature. Also, the survey discusses and analyzes the evaluation metrics and datasets adopted by the explored models. Finally, we conclude our survey with trends and future directions to further assist research on that domain. • We propose a novel taxonomy for deep learning-based citation recommendation models. • We examine and analyze the most popular evaluation metrics and datasets. • We present a comparative analysis among the explored models. • We highlight prominent research issues and recommendations. • We uncover research trends and present future research directions. [ABSTRACT FROM AUTHOR]

Recently, deep learning techniques have been widely used in recommendation tasks and have attained record performance. However, the input quality of the deep learning model has a great influence on the recommendation performance. In this work, an efficient and effective input optimization method is proposed. Specifically, we propose an integrated recommendation framework based on two-stage deep learning. In the first stage, with user and item features as the original input, a low-cost marginalized stacked denoising auto-encoder (mSDA) model is used to learn the latent factors of users and items. In the second stage, the resulting latent factors are combined and used as input vector to the DNN model for fast and accurate prediction. Using the latent factor vector as the input to the deep learning-based recommendation model not only captures the high-order feature interaction, but also reduces the burden of the hidden layer, and also avoids the model training falling into local optimum. Extensive experiments with real-world datasets show that the proposed model shows much better performance than the state-of-the-art recommendation methods in terms of prediction accuracy, parameter space and training speed. [ABSTRACT FROM AUTHOR]

• A novel hybrid deep learning based recommender system 'DNNRec' is proposed. • DNNRec leverages embeddings, combines side information and a very deep network. • DNNRec addresses cold start case and learns of non-linear latent factors. • Proposed solution is benchmarked against existing methods on accuracy and run time. • DNNRec outperforms state-of-the-art methods overall and in cold start case. We propose a novel deep learning hybrid recommender system to address the gaps in Collaborative Filtering systems and achieve the state-of-the-art predictive accuracy using deep learning. While collaborative filtering systems are popular with many state-of-the-art achievements in recommender systems, they suffer from the cold start problem, when there is no history about the users and items. Further, the latent factors learned by these methods are linear in nature. To address these gaps, we describe a novel hybrid recommender system using deep learning. The solution uses embeddings for representing users and items to learn non-linear latent factors. The solution alleviates the cold start problem by integrating side information about users and items into a very deep neural network. The proposed solution uses a decreasing learning rate in conjunction with increasing weight decay, the values cyclically varied across epochs to further improve accuracy. The proposed solution is benchmarked against existing methods on both predictive accuracy and running time. Predictive Accuracy is measured by Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and R-squared. Running time is measured by the mean and standard deviation across seven runs. Comprehensive experiments are conducted on several datasets such as the MovieLens 100 K, FilmTrust, Book-Crossing and MovieLens 1 M. The results show that the proposed technique outperforms existing methods in both non-cold start and cold start cases. The proposed solution framework is generic from the outperformance on four different datasets and can be leveraged for other ratings prediction datasets in recommender systems. [ABSTRACT FROM AUTHOR]

• Deep learning-based aspect extraction model for building recommender system. • Exploiting wordembedding and POS tag embedding for aspect extraction. • Using LDA and lexicon-based method to generate aspect-based ratings. • Integration of the aspect ratings into tensor factorization technique for rating prediction. The main goal of Aspect-Based Opinion Mining is to extract product's aspects and the associated user opinions from the user text review. Although this serves as vital source information for enhancing rating prediction performance, few studies have attempted to fully utilize it for better accuracy of recommendation systems. Most of these studies typically assign equal weights to all aspects in the opinion mining process, however, in practices; users tend to give different priority on different aspects of the product when reaching overall ratings. In addition, most of the existing methods typically rely on handcrafted, rule-based or double propagation methods in the opinion mining process which are known to be time-consuming and often inclined to errors. This could affect the reliability and performance of the recommender systems (RS). Therefore, in this paper, we propose a weighted A spect-based O pinion mining using D eep learning method for R ecommender system (AODR) that can extract product's aspects and the underlying weighted user opinions from the review text using a deep learning method and then fuse them into extended collaborative filtering (CF) technique for improving the RS. The proposed method is basically comprised of two components: (1) Aspect-based opinion mining module which aims to extract the product aspects from the review text to generate aspect rating matrix. (2) Recommendation generation component that uses tensor factorization (TF) technique to compute weighted aspect ratings and finally infer the overall rating prediction. We evaluate the proposed model in terms of both aspect extraction and recommendation performance. Experiment results on different datasets show that our AODR model achieves better results compared to the baselines. [ABSTRACT FROM AUTHOR]