Search

Nowadays, the sequential information, i.e., the ordering of the recorded feedback, is one of the most frequently used auxiliary information for developing recommendation algorithms. However, there are still some challenges in collaborative filtering with sequential implicit feedback. For one thing, there may be some co-occurrence and "leap-occurrence" phenomena in users' behavior sequences. For another, a weak point of utilizing implicit feedback is that the unobserved behaviors do not always indicate "dislike". In this paper, we focus on these two fundamental challenges in users' preference learning, i.e., the uncertainty of the ordering of the next items and the uneven quality of the negative samples, and propose a novel solution named COFIS (short for co llaborative fi ltering with s equential implicit feedback via learning users' preferences over item-sets). Specifically, our COFIS incorporates two learning paradigms (i.e., pairwise and pointwise) and two expression strategies (i.e., "MOO" and "MOS", representing two different ways to treat the negative item-sets when compared with the positive item-sets). Our COFIS can be applied to existing methods such as many factorization- and deep learning-baseed models. We derive four specific algorithms of our COFIS based on Fossil (denoted as COFIS(-,-) for short) in particular, and further show the results of our COFIS-improved methods (including the factorization-based COFIS, and the deep learning-based SASRec(COFIS) and FMLP-Rec(COFIS)). Notice that we treat the seminal works and their pointwise version as special cases of our COFIS. Empirical studies on four real-world datasets show the superiority of our proposed COFIS compared with the state-of-the-art baselines. Moreover, we also study the impact of some key parameters in our COFIS. [ABSTRACT FROM AUTHOR]

In the context of recommender systems, collaborative filtering is the method of predicting the ratings of a set of items given by a set of users based on partial knowledge of the ratings. Commonly, items and users are represented via vectors, and to predict ratings, approaches such as vector inner-product (aka matrix factorization) or more advanced nonlinear functions are applied. In this paper, while we adopt the common vectorial representation, we consider a general model in which the ratings are smooth functions of the item representations. Smoothness ensures similar items with nearby vectors will also get similar ratings as we expect from a human rater. We represent user smooth scoring functions in a so-called frequency domain and learn their representations alongside item representations using 1) an iterative optimization approach that maps items and users alternatively, and 2) a feedforward neural network consisting of interpretable layers. We also address the challenge of the distribution shift from observed to unobserved ratings (aka missing-not-at-random) with insights from the frequency domain. We evaluate the predictive power of our method and its robustness in missed-not-at-random settings on four popular benchmarks. Despite its simplicity and interpretability, our method yields a remarkable performance compared to the state-of-the-art.1 [ABSTRACT FROM AUTHOR]

Behaviorally similar neighbors in the interaction graph have been actively explored to facilitate the collaboration between users and items and address the interaction sparsity issue. We investigate homogenous neighbors between users or items to mine collaborative signals for embedding learning. In the case of multiple and complex composition of user interests and item attributes, traditional uniform embedding is insufficient to depict matching between a specific user-item pair. Therefore, we propose a siamese graph-based dynamic matching (SGDM) model for collaborative filtering. A target-aware dual attention module is introduced to update neighbors with their varying contributions to unveil users' interests, which explicitly encodes the clue of the candidate matching target. For a target user-item pair, homogeneous neighbors participate in the interest propagation through graph convolution, which learns user/item embeddings respectively on the siamese homogeneous graphs. The dual dynamic aggregation in graph convolution endows a specific user-item pair with dynamic matching, which is expected to meet the needs of fine-grained filtering and promote the performance of collaborative filtering. Extensive experimental results confirm the collaboration between siamese homogeneous graphs of users and items. It further illustrates the effectiveness of the proposed SGDM in mining homogeneous collaborative signals for embedding learning and collaborative filtering. [ABSTRACT FROM AUTHOR]

With the increasing popularity of recommendation techniques and social networks, social network recommendation has become a significant research field, i.e., predicting a user's preferences based on her or his historical interaction data, because the social relationships of users can not only enrich their interaction information, but also imply the accurate characterization of their preferences. Although there are some existing studies that predict users' preferences by using the characteristics of social relationships, their time complexity and hardware resource consumption are often very expensive, which limits the feasibility in large-scale real-world scenarios. To address these issues, in this paper, we propose a simple and effective K -core Graph Collaborative Filtering (KGCF) model to incorporate the user's social features into a recommendation framework. The user's interaction information is auto-encoded linearly by incorporating the social relationship graph into the user's interaction information and constructing a multi-layer graph filter. The linear autoencoder graph filter alleviates the highly sparse data problem and dramatically reduces the training time and hardware resource consumption. Experimental results on several real-world datasets show the superior effectiveness and spatiotemporal efficiency compared with the existing baselines, especially the training speed is significantly accelerated for large-scale datasets. [ABSTRACT FROM AUTHOR]

Recommender systems are intrinsically tied to a reliability/coverage dilemma: The more reliable we desire the forecasts, the more conservative the decision will be and thus, the fewer items will be recommended. This causes a detriment to the predictive capability of the system, as it is only able to estimate potential interest in items for which there is a consensus in their evaluation, rather than being able to estimate potential interest in any item. In this paper, we propose the inclusion of a new term in the learning process of matrix factorization-based recommender systems, called recklessness, that takes into account the variance of the output probability distribution of the predicted ratings. In this way, gauging this recklessness measure we can force more spiky output distribution, enabling the control of the risk level desired when making decisions about the reliability of a prediction. Experimental results demonstrate that recklessness not only allows for risk regulation but also improves the quantity and quality of predictions provided by the recommender system. • Adjusting the variance in probability-based recommendation systems modifies the shape of the generated probabilities. • High variance produces many low-precision predictions, while low variance yields fewer high-precision predictions. • The variance can be adjusted by adding it to the cost function and tuning its weight with a hyper-parameter. [ABSTRACT FROM AUTHOR]

• A sequential graph is introduced to reorganize all user-item interactions. • An algorithm is proposed to vectorize graph attention networks on complex graphs. • A personalized and transition-aware attention mechanism is proposed. • The proposed SGAT model explicitly captures collaborative and sequential information. Understanding collaborative and sequential information plays a major role in next-item recommendation. Although previous models have achieved considerable results, their capabilities have been limited by the basic paradigm of characterizing each user's profile with only his/her sequential behaviors. This paper considers collaborative information over users to solve the problem. Specifically, we first reorganize all interactions as a multiplex directed heterogeneous graph (named sequential graph) to depict each user's sequential behaviors and all users' collaborative interactions. Second, we propose a generic vectorization algorithm to address the challenge of multiplex edges. The algorithm can implement arbitrary graph attention networks on complex graphs without graph simplification. Finally, we propose a novel sequential graph attention network (SGAT) that controls information to attentively propagate through directed heterogeneous edges. Based on the sequential graph and the vectorization algorithm, SGAT infers each user's dynamic preference by capturing high-order sequential information among items and exploiting both latitudinally and longitudinally collaborative information over users. Extensive experiments and ablation studies on various datasets demonstrate the superiority of SGAT. The sequential graph provides an alternative way to organize users' sequential behaviors and the vectorization algorithm can promote the development of graph mining by retaining topological structures of complex graphs. [ABSTRACT FROM AUTHOR]

• We propose an alternative modeling for the input feature space for the CF problem. • We investigate the how the complexity of input feature space impacts the performance. • Our proposal presents competitive results against the state-of-the-art CF methods. Collaborative Filtering (CF) has been extensively studied over the last decade. Among the most successful methods, neural-based methods have changed the research landscape by bringing cutting-edge artificial neural networks (ANN) and big data to offer the best personalization users have ever experienced. However, a significant issue arose from these neural-based methods is that the increase of the model complexity directly impacts the computational cost. In general, this drawback is a consequence of the high dimensionality and sparsity of the input feature space. We believe that reducing sparsity and dimensionality of the input features is essencial to enhance accuracy while keeping computational costs low. This paper investigates an alternative modeling for the CF setting to represent users or items by using landmarks. This modeling drastically decreases the input feature space and eliminate sparsity while maintaining the underlying information needed to achieve great accuracy. Based on that, we propose a novel neural-based CF method via landmark modeling. Experiments on six real-world benchmark CF datasets were conducted by comparing the proposed method to well-known and widely-used CF methods of the state-of-the-art. The results show that the Neural Landmark method outperforms the other methods in both accuracy and computational performance. [ABSTRACT FROM AUTHOR]

• We propose a new rating prediction method in memory-based collaborative filtering. • We aim at the weights that are estimated in detail for each ordered triple (the neighbor, the active user, the target item). • We effectively model the fact that a neighbor of the active user did not rate the target item. • We implement the proposed approach for the combination of user-based and item-based views. Recommender systems play an indispensable role in today's online businesses. In these systems, memory-based (neighborhood-based) collaborative filtering is an important strategy to predict items as expected by users. It consists of two phases: computing the preference similarity between each pair of users in the offline phase and predicting the rating of an active user for a target item in the online phase by aggregating ratings of his/her neighbors for the target item. Previous studies on memory-based collaborative filtering have heavily concentrated on proposing methods for the computation of user preference similarity. To further improve the performance of memory-based collaborative filtering, this paper is aimed at the rating prediction phase. By optimizing a proposed objective function, the method we used in the rating prediction phase helps more accurately estimate the weight between the active user and each of his/her neighbors. The experimental results show that the proposed method outperforms others, especially in the case of a small and medium number of selected neighbors. [ABSTRACT FROM AUTHOR]

Collaborative filtering (CF) is one of the dominant techniques used in recommender systems. Most CF-based methods treat every user (or item) as an isolated existence, without explicitly modeling potential mutual relations among users (or items), which are latent in user-item interactions. In this paper, we design a novel strategy to mine user-user and item-item implicit relations and propose a natural way of utilizing the implicit relations for recommendation. Specifically, our method contains two major phases: neighbor construction and recommendation framework. The first phase constructs an implicit neighbor set for each user and item according to historical user-item interaction. In the second phase, based on the constructed neighbor sets, we propose a deep framework to generate recommendations. We conduct extensive experiments with four datasets on the movie, business, book, and restaurant recommendations and compare our methods with seven baselines, e.g., feature-based, neighborhood-based, and graph-based models. The experiment results demonstrate that our method achieves superior performance in rating prediction and top- k recommendation. [ABSTRACT FROM AUTHOR]

To explore the robustness of recommender systems, researchers have proposed various shilling attack models and analyzed their adverse effects. Primitive attacks are highly feasible but less effective due to simplistic handcrafted rules, while upgraded attacks are more powerful but costly and difficult to deploy because they require more knowledge from recommendations. In this paper, we explore a novel shilling attack called Graph cOnvolution-based generative shilling ATtack (GOAT) to balance the attacks' feasibility and effectiveness. GOAT adopts the primitive attacks' paradigm that assigns items for fake users by sampling and the upgraded attacks' paradigm that generates fake ratings by a deep learning-based model. It deploys a generative adversarial network (GAN) that learns the real rating distribution to generate fake ratings. Additionally, the generator combines a tailored graph convolution structure that leverages the correlations between co-rated items to smoothen the fake ratings and enhance their authenticity. The extensive experiments on two public datasets evaluate GOAT's performance from multiple perspectives. Our study of the GOAT demonstrates technical feasibility for building a more powerful and intelligent attack model with a much-reduced cost, enables analysis the threat of such an attack and guides for investigating necessary prevention measures. [ABSTRACT FROM AUTHOR]

Since the rise of E-commerce companies and many other web services, the applications of recommender systems have been adopted more broadly than ever before. Although collaborative filtering is the most well-known approach which utilizes customer's preference to discover their interest, the problems of data sparsity and similarities selection still exist in it. Thus, this study intends to propose a hybrid metaheuristic with perturbation-based K -nearest neighbors and densest imputation for collaborative filtering (KDI-KNN) algorithm to reduce the effects of data sparsity. A similarities union function is proposed to determine the fittest similarity and enhance the prediction performance. Eventually, the experimental results indicate that hybrid metaheuristics with perturbation-based KDI-KNN algorithms are superior to basic KNN, original KDI-KNN, and most single metaheuristic-based KDI-KNN. In addition, a real-world dataset, fund transaction dataset is adopted in the case study. The analysis reveals that the similarity is seriously affected by the different content of the dataset. [ABSTRACT FROM AUTHOR]

• The membership function of ratings is devised to build fuzzy profile of ratings. • Fuzzy profiles of users and items are proposed based on fuzzy profile of rating. • The new criteria of detecting natural noise in recommender system is given. • The noisy ratings are corrected based on maximum membership degree principle. A high-quality recommender system (RS) can effectively alleviate information overload by producing recommendations. The quality of the recommender system not only depends on the recommendation algorithm but also on the quality of collected data. Since users are often affected by environmental and accidental factors during the rating process, natural noise is probably brought into the data of RS by non-malicious users, which will lead to deviations in prediction results. In this paper, we propose a scheme based on fuzzy theory to manage the natural noise in RS. We first classify the ratings into three fuzzy categories with variable boundaries. Then, the fuzzy profiles of users and items are built to detect the natural noise in ratings. Finally, once the noisy ratings are detected, we replace them with the rating threshold values according to the Maximum membership principle. The proposed scheme is tested in two benchmark datasets and experimental results verify that the scheme can significantly improve the recommendation quality and has higher efficiency than the schemes based on re-predication. [ABSTRACT FROM AUTHOR]

• Extracting dominant opinion patterns of frequent favored-unfavored items. • Assigning users to the patterns to detect high confidence relations between them. • An assigned user to some pattern mediates between users of those patterns. • For each user, an impact value is calculated as the role in rating prediction. • Experimental results on four datasets show the proper performance of our method. In this paper, a new collaborative filtering method is proposed based on finding similar users directly and indirectly to overcome sparsity challenge. Moreover, selecting these users through extracting dominant opinion patterns leads to tackling scalability. In this method, frequent opinions between users are extracted to be used as dominant patterns. Then, users corresponding to the same dominant pattern are considered as direct similar users. Direct similar users who have seen more items and have corresponded to more than one pattern are regarded as reference users. Each reference user mediates between users who may have no/few commonly seen items and they are considered as indirect similar users. Utilizing indirect similar user helps the method to predict opinion of query user about items, which have not been seen by any direct similar user before. Clearly, indirect users are selected based on speculation using available information about direct users' preferences. Thus, the effect of indirect users is considered to be stricter than that of direct users on the final prediction step. Experiments conducted on MovieLens small, MovieLens 100 k, MovieLens 1 M, and Jester datasets showed that the proposed method outperforms the previously introduced methods, especially on sparse data. [ABSTRACT FROM AUTHOR]

Cross-domain recommendation can alleviate the data sparsity problem in the target domain and has become a promising research area. Recently, various models have been proposed to provide recommendation across domains. Some models successfully embedded adversarial learning to reduce domain discrepancies between the source and target domains; however, these models only focuse on extracting the domain-shared features among multiple domains. In this paper, we devise a novel framework that considers both domain-shared and domain-specific knowledge across domains, namely, DAAN. Specifically, we tightly couple matrix factorization-based collaborative filtering with deep adversarial domain adaptation via an attention network. In this framework, we first learn the domain-specific representations for each user and each item from the source and target user-item interaction matrices. Then, we capture the domain-shared features between two domains with common user (or item) embeddings in a domain-adversarial paradigm. Additionally, an attention network is used to adjust the degree of importance between domain-shared and domain-specific knowledge. Extensive experiments on six tasks demonstrate that our method outperforms various baselines in terms of two ranking metrics. To the best of our knowledge, our proposed approach is the first deep model that considers both domain-shared and domain-specific knowledge across domains within a domain-adversarial training paradigm for cross-domain recommendation. [ABSTRACT FROM AUTHOR]

Existing recommender systems rely on user and item representations in a fixed continuous low-dimensional latent space. To predict ratings, they use only an implicit feedback matrix, whereas user and item side information is ignored. Furthermore, they use the same arbitrary priors for the user and item latent vectors, reducing the ability of the model to identify the actual latent vectors. Currently, the latent parameters should be learned directly for every user and movie. This is problematic, as it would require both model retraining and learning the latent vector representations when users or items are added to the underlying dataset. To address these issues, we propose a two-step hybrid variational Bayesian autoencoder to characterize the uncertainty of predicted ratings. An encoder is first trained to map data vectors to the latent space so that the latent representations can be dynamically computed. Subsequently, we use the generative process of users and items with their priors as side-specific information to handle matrix sparsity and better learn their latent vectors. Finally, we consider stochastic variational inference to approximate the posterior density of intractable user-item latent vectors. Experiments conducted on two real-world datasets demonstrate the effectiveness of the proposed method compared with state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Sequential recommendations play a crucial role in recommender systems. Existing methods commonly focus on extracting sequential patterns within individual item sequences to make personalized recommendations. However, this paper argues that such a closure strategy has limitations. Firstly, the inherent uncertainty and sparsity within these sequences pose considerable challenges in accurately capturing sequential patterns. Secondly, ignoring group behaviors beyond the sequence hinders the modeling of user similarities, which are essential for recommendations. To address these limitations, this paper proposes a novel collaborative sequential recommendation approach, called G roup- a ware G raph N eural N etworks (GaGNN). GaGNN seeks to incorporate group behaviors within sequential recommendations, thereby enhancing the modeling of user preferences through the assimilation of collective wisdom. GaGNN constructs two transition graphs, each offering a distinct behavioral perspective. The group transition graph is designed to capture the collaborative effects and learn collaborative information, whereas the individual transition graph aims to articulate individual behaviors, thereby extracting sequential information. Additionally, a multi-source information fusion module is designed to bridge semantic gaps between the two types of information and yield a holistic understanding of user preferences. Through rigorous experimental comparisons and ablation studies, the superiority of GaGNN and the beneficial integration of collaborative information are emphatically validated. • Two types of transition graphs are introduced to depict group behaviors and individual behaviors. • Two information encoders are proposed to model collaborative and sequential information. • A multi-source information fusion module is proposed to achieve information complementarily. • The proposed GaGNN model aims to incorporate group behaviors into sequential recommendations. [ABSTRACT FROM AUTHOR]

Beyond accuracy, quality measures are gaining importance in modern recommender systems, with reliability being one of the most important indicators in the context of collaborative filtering. This paper proposes Bernoulli Matrix Factorization (BeMF), which is a matrix factorization model, to provide both prediction values and reliability values. BeMF is a very innovative approach from several perspectives: a) it acts on model-based collaborative filtering rather than on memory-based filtering, b) it does not use external methods or extended architectures, such as existing solutions, to provide reliability, c) it is based on a classification-based model instead of traditional regression-based models, and d) matrix factorization formalism is supported by the Bernoulli distribution to exploit the binary nature of the designed classification model. The experimental results show that the more reliable a prediction is, the less liable it is to be wrong: recommendation quality improves after the most reliable predictions are selected. State-of-the-art quality measures for reliability have been tested, which shows that BeMF outperforms previous baseline methods and models. [ABSTRACT FROM AUTHOR]

Accurately forecasting user's purchase intention over time is a huge challenge for personalized recommend systems, of which a critical problem is how to model the changes of user preference and temporal correlation of items. In this paper, aiming at addressing this question, we first introduce attention flow network to model users' purchase records by leveraging attention flow that describes the changing process of purchase intention. Then based on the attention flow network and individuals' attention flows, we propose a novel personalized recommendation algorithm named Attention Flow Network based Personalized Recommendation (AFNPR). Our method integrates all the purchase sequences of users into a weighted attention flow network, and recommends items based on transition probabilities related to attention flow network with user's attention decay, which is efficient in linear time. The experiments demonstrate its superior performance on several real datasets. [ABSTRACT FROM AUTHOR]

• We propose a probabilistic model to exploit co-pairwise ranking for item recommendation. • We devise a position-aware sampler to choose high-quality negative instances. • We design an efficient learning algorithm to optimize the model parameters. • Extensive experiments show that BSPR consistently surpasses other competitors. Product recommendation has played an important role in improving user experiences and obtaining more profits. To optimize recommendation models, pairwise learning has become a mainstream method for modeling user preferences from implicit feedback. Nevertheless, most existing pairwise methods optimize the relative order among products from only the user perspective. In fact, the purchase decision making of a given user depends on not only individual taste, but also the complementary relationships between the recommended products and the products in his/her historical basket. We argue that it is challenging to uncover meaningful user and product representations by only utilizing the user-side pairwise ranking. Towards this end, we propose a novel probabilistic pairwise method named BSPR, short for basket-sensitive personalized ranking , which solves both user- and product-side pairwise ranking problems in a unified manner. Specifically, BSPR discovers mutual correlations among users and products by exploiting co-pairwise ranking, alleviating the inherent flaw in existing pairwise methods. Considering that the negative sampler is one of the key components for pairwise learning, we devise a position-aware sampling strategy for the proposed method. To solve the optimization problem in BSPR, we further design an alternative optimization algorithm to efficiently learn the model parameters. Extensive experiments on multiple real-world datasets demonstrate significant improvements of our method over a series of state-of-the-art methods. Our implementation of BSPR is publicly available at: https://github.com/wubinzzu/BSPR. [ABSTRACT FROM AUTHOR]

Collaborative filtering is a fundamental technique in recommender systems, for which memory-based and matrix-factorization-based collaborative filtering are the two types of widely used methods. However, the performance of these two types of methods is limited in the case of sparse data, particularly with extremely sparse data. To improve the effectiveness of the methods in a sparse scenario, this paper proposes a multi-factor similarity measure that captures linear and nonlinear correlations between users resulting from extreme behavior. Subsequently, a fusion method that simultaneously considers the multi-factor similarity and the global rating information in a probability matrix factorization framework is proposed. In our framework, users' local relations are integrated into the global ratings optimization process, so that prediction accuracy and robustness are improved in sparse data, particularly in extremely sparse data. To verify the performance of the proposed methods, we conduct experiments on four public datasets. The experimental results show that the fusion method is superior to the typical matrix factorization models used in collaborative filtering and significantly improves both the prediction results and robustness in sparse data. [ABSTRACT FROM AUTHOR]

Recently, the prediction of retweet behaviors has attracted significant attention, as it can facilitate with a number of tasks, such as popular tweet prediction, personalized recommendation and business intelligence. However, in existing studies, two main problems exists in the prediction of retweet behaviors. (1) The relationship between users is extremely simple when social influences are used for prediction. (2) An effective framework that unifies the effects of both heterogeneous social relations of users and multidimensional similarities of tweets does not exist. Therefore, we propose a unified factorization model that incorporates social influence and tweet similarity into a traditional Bayesian Poisson factorization (BPF) model, named BPF++. Specifically, we utilize a variety of social influence and tweet similarity jointly to improve performance. Furthermore, we integrate trust strengths between users and degrees of similarity between tweets to the framework. We adopt an efficient coordinate ascent algorithm to learn the parameters of the BPF++ model. Extensive experiments are conducted to evaluate the performance of our model on the Sina Weibo dataset. Results demonstrate improvements of 113.64% and 116.28% in the NDCG@3 and precision@3 scores, respectively, compared with BPF. [ABSTRACT FROM AUTHOR]

The ever-increasing popularity of web service sharing communities have produced a considerable amount of web services that share similar functionalities but vary in Quality of Services (QoS) performances. To alleviate the heavy service selection burden on users, lightweight recommendation ideas, e.g., Collaborative Filtering (CF) have been developed to aid users to select their preferred services. However, existing CF methods often face two challenges. First, service QoS is often context-aware and hence depends on the spatial and temporal information of service invocations heavily. While it requires challenging efforts to integrate both spatial and temporal information into service recommendation decision-making process simultaneously. Second, the location-aware and time-aware QoS data often contain partial sensitive information of users, which raise an emergent privacy-preservation requirement when performing service recommendations. In view of above two challenges, in this paper, we integrate the spatial-temporal information of QoS data and Locality-Sensitive Hashing (LSH) into recommendation domain and bring forth a location-aware and time-aware recommendation approach considering privacy concerns. At last, a set of experiments conducted on well-known WS-DREAM dataset show the feasibility of our approach. [ABSTRACT FROM AUTHOR]

Memory-based Collaborative Filtering (CF) has been a widely used approach for personalised recommendation with considerable success in many applications. An important issue regarding memory-based CF lies in similarity computation: the sparsity of the rating matrix leads to similarity computations based on few co-rated items between users, resulting in high sensitive predictions. Additionally, the 'sparse' similarity computation has high computational cost, due to the dimensionality of the item space. In this paper, we pursue both these issues. We propose a new model to compute similarity by representing users (or items) through their distances to preselected users, named landmarks. Such user modelling allows the introduction of more ratings into similarity computations through transitive relations created by the landmarks. Unlike conventional memory-based CF, the proposal builds a new user space defined by distances to landmarks, avoiding sensitivity in similarity computations. Findings from our experiments show that the proposed modelling achieves better accuracy than the 'sparse' similarity representation in all tested datasets, and has also yielded competitive accuracy results against the compared model-based CF algorithms. Furthermore, the proposed implementation has beaten all compared methods in terms of computational performance, becoming a promising alternative to memory-based CF algorithms for large datasets. [ABSTRACT FROM AUTHOR]

• We research the issues and challenges in context-aware recommender systems based on collaborative filtering techniques. • We propose a solution that exploit useful contextual information to improve the prediction accuracy of recommender systems. • We implement two improved context-aware rating prediction methods based on different collaborative filtering techniques to achieve the proposed goals. • Two large contextual datasets are constructed to upgrade the performance of our methods in large scale system. • Experimental results, with respect to rating prediction quality and recommendation performance on both public available and large created contextual datasets, show that our proposal outperforms the existing recommender systems especially on the created datasets. Many researchers have realized the importance of contextual information and focused on designing systems that predict user's contextual preferences. In this respect, several researches have been devoted to Context-Aware Recommender Systems (CARS). One of the remaining issues in these systems especially the collaborative filtering based ones, is determining which contextual information can be adopted to make effective rating prediction. In fact, many contextual dimensions (e.g., location, time, mood etc.) may affect the user's preferences, but not all of these dimensions are equally important for the rating prediction effectiveness. Many existing CARS approaches cannot fully capture the influence of relevant contextual dimensions and their interaction on the rating, and furthermore cannot obtain a better recommendation performance. To address these issues, we highlight contextual dimensions weighting, study the correlation between them to elicit the most useful ones, and propose two improved rating prediction methods based on collaborative filtering techniques, involving relevant and dependent contextual dimensions. Experimental results, with respect to rating prediction quality and recommendation performance on both public available and large created contextual datasets, show that our proposal outperforms the existing recommender systems especially on the created datasets. [ABSTRACT FROM AUTHOR]

Recurrent neural networks (RNN) based recommendation algorithms have been introduced recently as sequence information plays an increasingly important role when modeling user preferences. However, these methods have numerous limitations: they usually give undue importance to sequential changes and place insufficient emphasis on the correlation between adjacent items; additionally, they typically ignore the impacts of context information. To address these issues, we propose an attention-based context-aware sequential recommendation model using Gated Recurrent Unit (GRU), abbreviated as ACA-GRU. First, we consider the impact of context information on recommendations and classify them into four categories, including input context, correlation context, static interest context, and transition context. Then, by redefining the update and reset gate of the GRU unit, we calculate the global sequential state transition of the RNN determined by these contexts, to model the dynamics of user interest. Finally, by leveraging the attention mechanism in the correlation context, the model is able to distinguish the importance of each item in the rating sequence. The impact of outliers that are less informative or less predictive decreases or is ignored. Experimental results indicate that ACA-GRU outperforms state-of-the-art context-aware models as well as sequence recommendation algorithms, demonstrating the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

Collaborative Filtering (CF) is one of the most popular technologies used in online recommendation systems. Most of the existing CF studies focus on the offline algorithms, a major drawback of these algorithms is the lack of ability to use the latest user feedbacks to update the learned model in realtime, due to the high cost of the offline training procedure. In this work, we propose Logo , an online CF algorithm. Our proposed method is based on a hierarchical generative model, with which, we derive a set of local and global consistency constraints for the prediction targets, and eventually obtain the design of the learning algorithm. We conduct comprehensive experiments to evaluate the proposed algorithm, the results show that: (1) Under the online setting, our algorithm achieves notably better prediction results than the benchmark algorithms; (2) Under the offline setting, our algorithm attains comparable accurate prediction results with the best performed competitors; (3) In all the experiments, our algorithm performs tens or even hundreds of times faster than the comparison algorithms. [ABSTRACT FROM AUTHOR]

With the emergence of the ever-increasing vehicular applications and booming Internet services, the requirements of low-latency and high efficient transmission among vehicles become urgent to meet, and their corresponding solutions need to be well investigated. To resolve the above challenges, we propose a fog computing-based content transmission scheme with collective filtering in edge of vehicles. We first provide a system model based on fog-based rode side units by considering location-awareness, content-caching and decentralized computing. Then, a content-caching strategy in RSUs is designed to minimize the downloading latency. Specifically, we model the moving vehicles with the two-dimensional Markov chains, and calculate the probabilities of file caching in RSUs to minimize the latency in file downloading. Each vehicle can also maintain a neighbor list to record the encounters with high similarities, and update it based on the historic and real-time contacts. Finally, we carry on the experiments based on the real-world taxi trajectories in Beijing and Shanghai, China. Simulation results demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Collaborative filtering (CF) approaches are widely applied in recommender systems. Traditional CF approaches have high costs to train the models and cannot capture changes in user interests and item popularity. Most CF approaches assume that user interests remain unchanged throughout the whole process. However, user preferences are always evolving and the popularity of items is always changing. Additionally, in a sparse matrix, the amount of known rating data is very small. In this paper, we propose a method of online collaborative filtering with dynamic regularization (OCF-DR), that considers dynamic information and uses the neighborhood factor to track the dynamic change in online collaborative filtering (OCF). The results from experiments on the MovieLens100K, MovieLens1M, and HetRec2011 datasets show that the proposed methods are significant improvements over several baseline approaches. [ABSTRACT FROM AUTHOR]

Recently, item-based collaborative filtering has attracted a lot of attention. It recommends to users new items which may be of interests to them, based on their reported historical data (i.e., the items they have already been interested in). The reported historical data leads to significant privacy risks in case that the recommending service is not fully trusted. Many researches have focused on developing differential privacy mechanisms to protect personal data in various recommendations. However, most of these mechanisms can not ensure accuracy of the recommendations. The main reason for this problem is that these methods compute similarity directly from the perturbation data. The computed similarity is thus always inaccurate and this inaccurate similarity finally leads to inaccurate recommendation results. In this paper, we propose a locally differentially private item-based collaborative filtering framework, which protects users' private historical data on the user-side, and on the server-side reconstructs the similarity to ensure recommendation accuracy. The similarities are reconstructed for every pair of items, by estimating the number of users who have rated neither, either one, or both of them. The final recommendation is generated by the reconstructed similarities. Experimental results show that our proposed method significantly outperforms the state-of-the-art methods in terms of the recommendation accuracy and the trade-off between privacy and accuracy. [ABSTRACT FROM AUTHOR]

Abstract A majority of the collaborative filtering techniques exploit user-item rating information to generate recommendations of unseen items for a user. However, a user's preference also depends on some extra information like item features, user attributes and others, which is known as side information. Further, according to recent studies, using preference relation as an alternative to absolute ratings often produces quality recommendations. This study proposes a collaborative filtering technique using Preference Relation based Restricted Boltzmann Machine for recommender system. The proposed method takes the preference relations of items as input and generates a ranking of items for any user. Using Conditional Restricted Boltzmann Machine, the side information of items along with preference relations are integrated into the model. Besides side information, the proposed method is also able to capture second order and higher order user-item interactions. Experimental verification of the proposed model is done using three datasets: MovieLens-1M, MovieLens-20M, and Book-Crossing, which are the most widely used datasets for testing recommender systems. Results obtained at different positions using standard ranking measures like, NDCG and MAP, indicate that the performance of the proposed method is better compared to related state-of-the-art collaborative filtering models for Top-N recommendation task. [ABSTRACT FROM AUTHOR]

Abstract Collaborative filtering (CF) is one of the most successful approaches for an online store to make personalized recommendations through its recommender systems. A neighborhood-based CF method makes recommendations to a target customer based on the similar preference of the target customer and those in the database. Similarity measuring between users directly contributes to an effective recommendation. In this paper, we propose a sub-one quasi-norm-based similarity measure for collaborative filtering in a recommender system. The proposed similarity measure shows its advantages over those commonly used similarity measures in the literature by making better use of rating values and deemphasizing the dissimilarity between users. Computational experiments using various real-life datasets clearly indicate the superiority of the proposed similarity measure, no matter in fully co-rated, sparsely co-rated or cold-start scenarios. [ABSTRACT FROM AUTHOR]

Abstract In the literature, various collaborative filtering approaches have been developed to perform an efficient recommendation on top of reducing the search cost of the customers. The recommender system methods are concentrated on improving the accuracy and to achieve that goal they focused on formulating complex similarity approaches and neglect the computation time in their model. Furthermore, in order to compute the similarity metric, most of traditional similarity measures have only considered co-rated items and overlooked the total rating vector of the user or item. However, considering only co-rated items to measure similarity metrics in recommender system is an insignificant approach to identifying appropriate nearest neighbors in relatively sparse datasets. Therefore, in this research, two new simple but effective similarity models have been developed by considering all rating vectors of users to classify relevant neighborhoods and generate recommendations in a lower computation time. Moreover, MovieLens, a well-known dataset used in recommender system domain, is involved here to validate the performance of the proposed model. It seems that the proposed relevant Jaccard similarity perform more accurately and effectively to generate well recommendation than other traditional similarity models. [ABSTRACT FROM AUTHOR]

Abstract The increasing popularity of O2O service make more and more people begin seeking and booking services online. After that, they experience the services in brick-and-mortar stores. This new business model has marketing potential and offer various opportunities to different industries. Consequently, various O2O services starting to appear, which results in difficult service selections for customers. Therefore, in this paper, we proposed a novel rating-based O2O service recommendation model considering user activity. In this method, the traditional similarity estimations are substituted by user activity which can better reflect the differentiations of customers' behavioral characteristics. Therefore, recommendations are more accurate. The experimental results show that proposed method outperforms rating-based methods, including widely used collaborative filtering methods and state-of-the-art matrix methods. In addition, we find the optimal parameter values of our model, and explore the influence of Top- k on rating-based recommendation. [ABSTRACT FROM AUTHOR]

Highlights • The first work to use the content CF-based methods to do item-tag co-recommendation. • The recommended tags serve as explanations to the recommended items. • Item latent factor is learned from user's implicit feedback, item's content, tags. • Experimental results show the improvement on accuracy and explanation. Abstract Tag-based recommendation has become increasingly important in recent years owing to the popularization of social tagging systems. Related studies on this subject can be categorized into item recommendation and tag recommendation based on the objective of the recommendation. In both categories, most existing recommendation approaches focus on improving prediction accuracy. But they ignore that the explanations of recommendations also greatly affect the decision-making of users. In social tagging systems, tags not only behave as auxiliary information of items but also show the implicit preferences of users. Therefore, they can be used to improve prediction accuracy and provide explanations to recommended items. Items and tags have interrelation and mutual effects. By focusing only on either item recommendation or tag recommendation, users may miss some information and can only achieve marginal gains. Fusing the two types of recommendations can improve the performance of both approaches. On the basis of the above ideas, in this study, we propose an EXPLainable item-tag CO-REcommendation (EXPLORE) framework that jointly recommends items and the corresponding tags. Different from conventional recommendations that utilize a single source of content, EXPLORE takes advantage of users' interests, item contents, and item tags. The experiments conducted on three real-world datasets demonstrate that EXPLORE outperforms the state-of-the-art methods. Importantly, the recommended tags can provide explanations to the recommended items, thus making the recommendation results explainable. [ABSTRACT FROM AUTHOR]

Abstract We propose a novel recommendation approach based on chronological cohesive units (CCUs) of content consuming logs. Chronological cohesive units are defined as sub-sequences of logs in which items are highly related to each other. We first generate rules for splitting consuming logs into CCUs. We select features which are effective for splitting of consuming logs and combine them into a binary decision tree to generate splitting rules with genetic programming. With the rules, we split content consuming logs into CCUs, and identify strongly associated items in the CCUs. Next items are recommended with an association rule-based approach. The proposed method is evaluated using two-real datasets: web page navigation logs and movie consuming logs. The experiments confirm that the proposed approach is superior to the existing methods in various aspects such as hit ratio, click-soon ratio, sparsity, diversity and serendipity. [ABSTRACT FROM AUTHOR]

Applying a graph convolutional network (GCN) or its variants to user-item interaction graphs is one of the most commonly used approaches for learning the representation of users and items in modern recommender systems. However, these models perform poorly in recommending high-order items due to the over-smoothing when the GCN deepens and the lack of connectivity between the cold-start items and the target user. To improve the ability to recommend higher-order items, we propose our Knowledge graph-aware Deep Graph Convolutional Network with Initial Residual Connection (KDGCN-IC) and Knowledge graph-aware Deep Graph Convolutional Network with Dense Connection (KDGCN-DC) methods. First, we introduce an item knowledge graph to rebuild the connectivity between nodes in the user-item graph, especially the users and the cold-start items. Then, we present our methods for improving the information flow by reusing features to alleviate the over-smoothing issue to deepen the GCN to capture the higher-order collaborative signal. More specifically, KDGCN-IC reuses the initial feature information, and KDGCN-DC reuses the feature information of each layer. Extensive experiments demonstrated that our models were able to achieve substantial improvement over state-of-the-art models in so far as their recommendation accuracy and higher-order recommendations. [ABSTRACT FROM AUTHOR]

Abstract Stochastic gradient descent (SGD) is a widely-used technique to implement matrix factorization. SGD-based matrix factorization involves many iterative computations. Therefore, according to the sequential composition theory of differential privacy, conventional implementation strategies of differentially private matrix factorization may lead to significant error accumulation, no matter whether the Laplace noise is added to the original matrix or to the factorized matrices. In fact, the implementation of differentially private matrix factorization is so challenging that results proposed to date have the problem of inefficient privacy and data utility. In this paper, we employ the objective perturbation method to address the challenge; this method dramatically alleviates error accumulation by perturbing the objective function instead of perturbing the results. Our method outperforms the state-of-the-art methods since it only requires a scalar noise rather than a vector noise to achieve the same magnitude of privacy. Furthermore, our method may learn the resulted matrices by joint optimization, which follows the conventional learning procedure of SGD and optimizes its convergence speed and accuracy as much as possible. In addition to the differential privacy guarantee, we also empirically show the way that the novel model works together with k -coRating, a k -anonymity-like privacy preserving model, to enhance data utility. [ABSTRACT FROM AUTHOR]

Nowadays, recommender systems face the problem of time heterogeneous feedback recommendation, in which items are recommended according to several kinds of user feedback with time stamps. Previously proposed recurrent neural network based recommendation method (RNNRec) cannot analyze feedback sequences on multiple time scales, and gradient vanishing may occur when the model is trained through back propagation through time (BPTT) algorithm. To address these issues, we propose a gated recurrent units (GRU) based neural network to predict which items users will access in the future. The GRU layer in the model can analyze feedback sequences on multiple time scales and can avoid gradient vanishing during training. The proposed approach is verified on three large-scale real-life datasets, and the comparison indicates that the proposed approach outperforms several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The problem of information overload motivated the appearance of Recommender Systems. From the several open problems in this area, the decision of which is the best recommendation algorithm for a specific problem is one of the most important and less studied. The current trend to solve this problem is the experimental evaluation of several recommendation algorithms in a handful of datasets. However, these studies require an extensive amount of computational resources, particularly processing time. To avoid these drawbacks, researchers have investigated the use of Metalearning to select the best recommendation algorithms in different scopes. Such studies allow to understand the relationships between data characteristics and the relative performance of recommendation algorithms, which can be used to select the best algorithm(s) for a new problem. The contributions of this study are two-fold: 1) to identify and discuss the key concepts of algorithm selection for recommendation algorithms via a systematic literature review and 2) to perform an experimental study on the Metalearning approaches reviewed in order to identify the most promising concepts for automatic selection of recommendation algorithms. [ABSTRACT FROM AUTHOR]

Collaborative Filtering (CF) is one of the most successful recommendation techniques. Recently, implicit trust-based recommendation approaches have emerged that incorporate implicit trust information into CF in order to improve recommendation performance. Previous implicit trust models assume that all users have the same perception of ratings. However, although all users employ members of the same rating domain (e.g. ratings on a 1–5 scale), each individual has his own interpretations about a rating domain in order to express his preferences. Thus, it is reasonable that a user's rating vector has some degree of uncertainty, depending upon the rating usage trend of that user. In this paper, we present a new approach for confidence modeling in the context of recommender systems. The idea of this modeling is that confidence in a particular user depends not only on the trust in the opinions of that user but also on the certainty of these opinions. Based on this idea, we propose a new Confidence-Based Recommendation (CBR) approach. This approach employs four different confidence models that derive the users’ and items’ confidence values from both local and global perspectives. Experimental results on real-world data sets demonstrate the effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]