Search

Recommender systems are information filtering systems used in many online applications like music and video broadcasting and e-commerce platforms. They are also increasingly being applied to facilitate software engineering activities. Following this trend, we are witnessing a growing research interest on recommendation approaches that assist with modelling tasks and model-based development processes. In this paper, we report on a systematic mapping review (based on the analysis of 66 papers) that classifies the existing research work on recommender systems for model-driven engineering (MDE). This study aims to serve as a guide for tool builders and researchers in understanding the MDE tasks that might be subject to recommendations, the applicable recommendation techniques and evaluation methods, and the open challenges and opportunities in this field of research, This work has been funded by the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 813884 (Lowcomote [134]), by the Spanish Ministry of Science (projects MASSIVE, RTI2018-095255-B-I00, and FIT, PID2019-108965GB-I00) and by the R&D programme of Madrid (Project FORTE, P2018/TCS-4314

In recommender systems, measuring user similarity is essential for predicting a user's ratings on items. Most traditional works calculate the similarity based on historical ratings shared between two users, without considering the probability of users' different behaviors. To address this issue, our work designs a similarity measure based on the users' historical behavior probabilities. Based on the similarity method, a complex network of user relationships is modeled. The user degree and community information of the modeled network, as well as the number of shared ratings between users, are used with the proposed similarity measure to design a rating prediction algorithm for recommender systems. Experiments based on MovieLens and Netflix data sets show that this method is effective and can successfully improve the accuracy of rating predictions and reduce the number of neighbors required to achieve the optimal prediction accuracy. Our research shows that in a complex system, the relationship between users can be effectively measured by the users' historical behavior probability, providing a new perspective for future research on similarity measurement.

Various recommender systems (RSs) have been developed over recent years, and many of them have concentrated on English content. Thus, the majority of RSs from the literature were compared on English content. However, the research investigations about RSs when using contents in other languages such as Arabic are minimal. The researchers still neglect the field of Arabic RSs. Therefore, we aim through this study to fill this research gap by leveraging the benefit of recent advances in the English RSs field. Our main goal is to investigate recent RSs in an Arabic context. For that, we firstly selected five state-of-the-art RSs devoted originally to English content, and then we empirically evaluated their performance on Arabic content. As a result of this work, we first build four publicly available large-scale Arabic datasets for recommendation purposes. Second, various text preprocessing techniques have been provided for preparing the constructed datasets. Third, our investigation derived well-argued conclusions about the usage of modern RSs in the Arabic context. The experimental results proved that these systems ensure high performance when applied to Arabic content.

Recently, web service recommender systems have attracted much attention due to the popularity of Service-Oriented Computing and Cloud Computing. Memory-based collaborative filtering approaches which mainly rely on the similarity calculation are widely studied to realize the recommendation. In these research works, the similarity between two users is computed based on the QoS data of their commonly-invoked services and the similarity between two services is computed based on the common users who invoked them. However, most approaches ignore that the similarity calculation is not always accurate under a sparse data condition. To address this problem, we propose a similarity propagation method to accurately evaluate the similarities between users or services. Similarity propagation means that "if A and B are similar, and B and C are similar, then A and C will be similar to some extent". Firstly, the similarity graph of users or services is constructed according to the QoS data. Then, the similarity propagation paths between two nodes on the similarity graph are discovered. Finally, the similarity along each propagation path is measured and the indirect similarity between two users or services is evaluated by aggregating the similarities of different paths connecting them. Comprehensive experiments on real-world datasets demonstrate that our similarity propagation method can outstandingly improve the QoS prediction accuracy of memory-based collaborative filtering approaches.

Collaborative filtering (CF) is one of the most practical approaches on recommendation systems by predicting users' preferences for items based on the user-item interaction information. Besides the connections between users and items, social networks among users can provide auxiliary information to improve the performance of recommender systems. Here, we propose an end-to-end deep learning framework by learning latent social features to embed in a CF approach. First, representation learning is employed on the rating matrix to extract the latent social features. Then, a novel deep learning approach based on cascade tree forest is used in the recommendation process. Experiments on real-world datasets from different domains demonstrate that the proposed Collaborative Deep Forest Learning (CDFL) outperforms the state-of-the-art CF recommendation methods.

Background: Health recommender systems (HRSs) are intelligent systems that can be used to tailor digital health interventions. We compared two HRSs to assess their impact providing smoking cessation support messages. Methods: Smokers who downloaded a mobile app to support smoking abstinence were randomly assigned to two interventions. They received personalized, ratable motivational messages on the app. The first intervention had a knowledge-based HRS (n = 181): it selected random messages from a subset matching the users’ demographics and smoking habits. The second intervention had a hybrid HRS using collective intelligence (n = 190): it selected messages applying the knowledge-based filter first, and then chose the ones with higher ratings provided by other similar users in the system. Both interventions were compared on: (a) message appreciation, (b) engagement with the system, and (c) one’s own self-reported smoking cessation status, as indicated by the last seven-day point prevalence report in different time intervals during a period of six months. Results: Both interventions had similar message appreciation, number of rated messages, and abstinence results. The knowledge-based HRS achieved a significantly higher number of active days, number of abstinence reports, and better abstinence results. The hybrid algorithm led to more quitting attempts in participants who completed their user profiles.

Recommender systems have become a popular and effective means to manage the ever-increasing amount of multimedia content available today and to help users discover interesting new items. Today’s recommender systems suggest items of various media types, including audio, text, visual (images), and videos. In fact, scientific research related to the analysis of multimedia content has made possible effective content-based recommender systems capable of suggesting items based on an analysis of the features extracted from the item itself. The aim of this survey is to present a thorough review of the state-of-the-art of recommender systems that leverage multimedia content, by classifying the reviewed papers with respect to their media type, the techniques employed to extract and represent their content features, and the recommendation algorithm. Moreover, for each media type, we discuss various domains in which multimedia content plays a key role in human decision-making and is therefore considered in the recommendation process. Examples of the identified domains include fashion, tourism, food, media streaming, and e-commerce.

In real-world recommender systems, users' interest and products' characteristics tend to go through a distinct series of changes over time. Thus, designing a recommender system that can simultaneously track the temporal dynamics of both drifts becomes a significant task. However, most of the existing temporal recommender systems only focus on users' dynamics, ignoring changes in products' characteristics. In this study, we propose a Multi-Trans matrix factorization (MTMF) model with improved time weight to capture temporal dynamics. Firstly, we introduce a personalized time weight that combines the forgetting curve and item similarity to reduce the impact of outdated information and retain the influence of users' stable preferences. Then, we model user and item dynamics by learning the multiple transitions at the user-factor and factor-item latent space between the ongoing time period and all past time periods. Accordingly, we formulate a joint objective function and take a gradient-based alternating optimization algorithm to solve this joint problem. Experimental results on historical datasets MovieLens show that the recommendation accuracy of MTMF with improved time weight is superior to the existing temporal recommendation methods.

Massive Open Online Courses (MOOCs) have witnessed a fast emergence in recent years thanks to their open and massive nature. A huge number of learners with different profiles, knowledge and learning objectives follow the same online courses with different personalized experiences. With this rapid development, course recommendation based on traits and characteristics becomes a necessity during the learning process to determine potentially suitable learning objects while analyzing user’s behaviours. Context-Aware Recommender Systems (CARS) have been shown to be effective in recommending items according to user’s interests and affect. In this paper, we propose a novel approach for Context-Aware Recommender Systems based on Factorization Machines (FMs). We propose a new algorithm called Unsupervised Graph Predictor Factorization Machines (UGPFMs) that models feature interactions during the recommendation process. UGPFMs apply Convolutional Neural Networks (CNNs) in the graph predictor to model interactions between context features in the rating matrice, then it employs the Factorization Machine (FM) to select the suitable items for the recommendation. UGPFM is a generic model that extends FM to improve the accuracy of recommendation, it outperforms state-of-the-art feature interactions-based FM techniques using MSE, RMSE, MAE and R-Squared metrics. We applied UGPFMs on MOOC while considering the sentiment, the cognition and the confusion as contexts for the recommendation. We compared different context combinations’ impact on the recommendation accuracy, then we computed each context-to-context feature interactions to better understand each context efficiency on e-learning recommendation.

Recommender Systems (RS) currently represent a fundamental tool in online services, especially with the advent of Online Social Networks (OSN). In this case, users generate huge amounts of contents and they can be quickly overloaded by useless information. At the same time, social media represent an important source of information to characterize contents and users’ interests. RS can exploit this information to further personalize suggestions and improve the recommendation process. In this paper we present a survey of Recommender Systems designed and implemented for Online and Mobile Social Networks, highlighting how the use of social context information improves the recommendation task, and how standard algorithms must be enhanced and optimized to run in a fully distributed environment, as opportunistic networks. We describe advantages and drawbacks of these systems in terms of algorithms, target domains, evaluation metrics and performance evaluations. Eventually, we present some open research challenges in this area.

The design of recommender systems (RSs) to support software development encompasses the fulfillment of different steps, including data preprocessing, choice of the most appropriate algorithms, item delivery. Though RSs can alleviate the curse of information overload, existing approaches resemble black-box systems, in which the end-user is not expected to fine-tune or personalize the overall process. In this work, we propose LEV4REC, a low-code environment to assist developers in designing, configuring, and delivering recommender systems. The first step supported by the proposed tool includes defining an initial model that allows for the configuration of the crucial components of the wanted RS. Then, a subsequent phase is performed to finalize the RS design, e.g., to specify configuration parameters. LEV4REC is eventually capable of generating source code for the desired RS. To evaluate the capabilities of the approach, we used LEV4REC to specify two existing RSs built on top of two different recommendation algorithms, i.e., collaborative filtering and supervised machine learning.

We are witnessing an increasing interest in building recommender systems (RSs) for all sorts of Software Engineering activities. Modelling is no exception to this trend, as modelling environments are being enriched with RSs that help building models by providing recommendations based on previous solutions to similar problems in the same domain. However, building a RS from scratch requires considerable effort and specialized knowledge. To alleviate this problem, we propose an automated approach to the generation of RSs for modelling languages. Our approach is model-based, and we provide a domain-specific language called Droid to configure every aspect of the RS (like the type and features of the recommended items, the recommendation method, and the evaluation metrics). The RS so configured can be deployed as a service, and we offer out-of-the-box integration of this service with the EMF tree editor. To assess the usefulness of our proposal, we present a case study on the integration of a generated RS with a modelling chatbot, and report on an offline experiment measuring the precision and completeness of the recommendations, This project has received funding from the EU Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 813884, the Spanish Ministry of Science (RTI2018-095255-B-I00) and the R&D programme of Madrid (P2018/TCS-4314)

Most NN-RSs focus on accuracy by building representations from the direct user-item interactions (e.g., user-item rating matrix), while ignoring the underlying relatedness between users and items (e.g., users who rate the same ratings for the same items should be embedded into similar representations), which is an ideological disadvantage. On the other hand, ME models directly employ inner products as a default loss function metric that cannot project users and items into a proper latent space, which is a methodological disadvantage. In this paper, we propose a supervised collaborative representation learning model - Magnetic Metric Learning (MML) - to map users and items into a unified latent vector space, enhancing the representation learning for NN-RSs. Firstly, MML utilizes dual triplets to model not only the observed relationships between users and items, but also the underlying relationships between users as well as items to overcome the ideological disadvantage. Specifically, a modified metric-based dual loss function is proposed in MML to gather similar entities and disperse the dissimilar ones. With MML, we can easily compare all the relationships (user to user, item to item, user to item) according to the weighted metric, which overcomes the methodological disadvantage. We conduct extensive experiments on four real-world datasets with large item space. The results demonstrate that MML can learn a proper unified latent space for representations from the user-item matrix with high accuracy and effectiveness, and lead to a performance gain over the state-of-the-art RS models by an average of 17%., Accepted by IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING(TKDE)

Collaborative Filtering (CF) filters the flow of data that can be recommended, by a Recommender System (RS), to a target user according to his taste and his preferences. The target user’s profile is built based on his similarity with other users. For this reason, CF technique is very sensitive to the similarity measure used to quantify the dependency strength between two users (or two items). In this paper we provide an in-depth review on similarity measures used for CF-based RS. For each measure, we outline its fundamental background and we test its performance through an experimental study. Experiments are carried out on three standard datasets (MovieLens100k, MovieLens1M and Jester) and reveal many important conclusions. In fact, results show that ITR and IPWR are the most suitable similarity measures for a user-based RS while AMI is the best choice for an item-based RS. Evaluation metrics show that under the user-based approach, ITR obtains an MAE equal to 0.786 and 0.731 on MovieLens100k and MovieLens1M, respectively. Whereas, IPWR reach an MAE equal to 3.256 on Jester. Also, AMI gets under the item-based approach an MAE equal to 0.745, 0.724 and 3.281 on MovieLens100k, MovieLens1M and Jester, respectively.

A recommender system (RS) is highly efficient in filtering people's desired information from high-dimensional and sparse (HiDS) data. To date, a latent factor (LF)-based approach becomes highly popular when implementing a RS. However, current LF models mostly adopt single distance-oriented Loss like an L₂ norm-oriented one, which ignores target data's characteristics described by other metrics like an L₁ norm-oriented one. To investigate this issue, this article proposes an L₁-and-L₂-norm-oriented LF (L³F) model. It adopts twofold ideas: 1) aggregating L₁ norm's robustness and L₂ norm's stability to form its Loss and 2) adaptively adjusting weights of L₁ and L₂ norms in its Loss. By doing so, it achieves fine aggregation effects with L₁ norm-oriented Loss's robustness and L₂ norm-oriented Loss's stability to precisely describe HiDS data with outliers. Experimental results on nine HiDS datasets generated by real systems show that an L³F model significantly outperforms state-of-the-art models in prediction accuracy for missing data of an HiDS dataset. Its computational efficiency is also comparable with the most efficient LF models. Hence, it has good potential for addressing HiDS data from real applications.

In e-commerce, online retailers are usually suffering from professional malicious users (PMUs), who utilize negative reviews and low ratings to their consumed products on purpose to threaten the retailers for illegal profits. Specifically, there are three challenges for PMU detection: 1) professional malicious users do not conduct any abnormal or illegal interactions (they never concurrently leave too many negative reviews and low ratings at the same time), and they conduct masking strategies to disguise themselves. Therefore, conventional outlier detection methods are confused by their masking strategies. 2) the PMU detection model should take both ratings and reviews into consideration, which makes PMU detection a multi-modal problem. 3) there are no datasets with labels for professional malicious users in public, which makes PMU detection an unsupervised learning problem. To this end, we propose an unsupervised multi-modal learning model: MMD, which employs Metric learning for professional Malicious users Detection with both ratings and reviews. MMD first utilizes a modified RNN to project the informational review into a sentiment score, which jointly considers the ratings and reviews. Then professional malicious user profiling (MUP) is proposed to catch the sentiment gap between sentiment scores and ratings. MUP filters the users and builds a candidate PMU set. We apply a metric learning-based clustering to learn a proper metric matrix for PMU detection. Finally, we can utilize this metric and labeled users to detect PMUs. Specifically, we apply the attention mechanism in metric learning to improve the model's performance. The extensive experiments in four datasets demonstrate that our proposed method can solve this unsupervised detection problem. Moreover, the performance of the state-of-the-art recommender models is enhanced by taking MMD as a preprocessing stage., Comment: Accepted by IEEE Transactions on Knowledge and Data Engineering (TKDE)

Recommender systems aim to accurately and actively provide users with potentially interesting items (products, information or services). Deep reinforcement learning has been successfully applied to recommender systems, but still heavily suffer from data sparsity and cold-start in real-world tasks. In this work, we propose an effective way to address such issues by leveraging the pervasive social networks among users in the estimation of action-values (Q). Specifically, we develop a Social Attentive Deep Q-network (SADQN) to approximate the optimal action-value function based on the preferences of both individual users and social neighbors, by successfully utilizing a social attention layer to model the influence between them. Further, we propose an enhanced variant of SADQN, termed SADQN++, to model the complicated and diverse trade-offs between personal preferences and social influence for all involved users, making the agent more powerful and flexible in learning the optimal policies. The experimental results on real-world datasets demonstrate that the proposed SADQNs remarkably outperform the state-of-the-art deep reinforcement learning agents, with reasonable computation cost.

As the use of recommender systems becomes generalized in society, the interest in varying the orientation of their recommendations is increasing. There are shilling attacks' strategies that introduce malicious profiles in collaborative filtering recommender systems in order to promote the own products or services or to discredit those of the competition. Academic research against shilling attacks has been focused in statistical approaches to detect the unusual patterns in user ratings. Nowadays, there is a growing research area focused on the design of robust machine learning methods to neutralize the malicious profiles inserted into the system. This paper proposes an innovative robust method, based on matrix factorization, to neutralize the shilling attacks. Our method obtains the reliability value associated with each prediction of a user to an item. By monitoring the unusual reliability variations in the items prediction, we can avoid promoting the shilling predictions to the erroneous recommendations. This paper openly provides more than 13 000 individual experiments involving a wide range of attack strategies, both push, and nuke, in order to test the proposed approach. The results show that the proposed method is able to neutralize most of the existing attacks; its performance only decreases in the not relevant situations: when the attack size is not large enough to effectively affect the recommendations provided by the system.

Part 16: Recommendation Systems; International audience; Multi-stakeholder recommender systems (RSs) are a major paradigm shift from current RSs because recommendations affect not only item consumers (end-users) but also item providers (owners). They also motivate the need for new performance metrics beyond recommendation quality that explicitly affect the latter. In this work, we introduce a framework for optimizing multi-stakeholder RSs under constraints on diversity and coverage. Our goal is to make recommendations to end-users while treating each item provider equally, by ensuring sufficient user base coverage and diverse profiles of users to which items are recommended. Namely, items of each provider should be recommended to a certain number of users that are also diverse enough in their preferences. The optimization objective is that the total average rating of recommended items is as close as possible to that of a baseline RS. The problem is formulated as a quadratically constrained integer program, which is NP-Hard and impractical to solve in the presence of big data and many providers. Interestingly, we show that when only the coverage constraint exists, an instance of the problem can be solved optimally in polynomial time through its Linear Programming relaxation, and this solution can be used to initialize a low-complexity heuristic algorithm. Data experiments show good performance and demonstrate the impact of these constraints on average rating of recommended items.

In this study we explore the use of recommender systems as a means of providing automated and personalized feedback to students following summative assessment. The intended feedback is a personalized set of test questions (items) for each student that they could benefit from practicing with. Recommended items can be beneficial for students as they can support their learning process by targeting specific gaps in their knowledge, especially when there is little time to get feedback from instructors. The items are recommended using several commonly used recommender system algorithms, and are based on the students' scores in a summative assessment. The results show that in the context of the Dutch secondary education final examinations, item recommendations can be made to students with an acceptable level of model performance. Furthermore, it does not take a computationally complex model to do so: a simple baseline model which takes into account global, student-specific, and item-specific averages obtained similar performance to more complex models. Overall, we conclude that recommender systems are a promising tool for helping students in their learning process by combining multiple data sources and new methodologies, without putting additional strain on their instructors.

To promote online businesses and sales, e-commerce industry focuses to fulfill users' demands by giving them top set of recommendations which are ranked through different ranking measures.Deep learning based auto-encoder models have further improved the performance of recommender systems. A state-of-the-art collaborative denoising auto-encoder (CDAE) models user-item interactions as a corrupted version of users rating inputs. However, this architecture still lacks users' ratings-trend information which is an important parameter to recommend top-N items to users. In this paper, building upon CDAE characteristics, we propose a novel users rating-trend based collaborative denoising auto-encoder (UT-CDAE) which determines user-item correlations by evaluating rating-trend(High or Low) of a user towards a set of items. This inclusion of a user's rating-trend provides additional regularization flexibility which helps to predict improved top-N recommendations. The correctness of the suggested method is verified through different ranking evaluation metrics i.e., (mean reciprocal rank, mean average precision and normalized discounted gain), for various input corruption values, learning rates and regularization parameters.Experiments on standard ML-100K and ML-1M datasets show that suggested model has improved performance overstate-of-the-art denoising auto-encodermodels.

As we all known, transfer learning is an effective way to alleviate the sparsity problem in recommender systems by transferring the shared knowledge cross multiple related domains. However, additional related domain is not always available, and auxiliary data may be noisy and this leads to negative transfer. In this paper, we suppose that different parts of one domain also have the shared knowledge and put forward a novel in-domain collaborative filtering framework, which utilizes contextual information to divide an original user-item interaction matrix into some smaller sub-matrices and regards the selected sub-matrices as “multiple domains” to establish transfer learning. The proposed framework no longer needs additional domain information and has a better adaptive ability. Also, considering more actual situation that users may have multiple personalities and items may have diverse attributes, we resort to Rating-Matrix Generative Model (RMGM) to generate the shared cluster-level rating pattern. Experiments on dataset Douban with three different categories demonstrate that the proposed framework can improve the prediction accuracy as well as the top-N recommendation performance.

Recommender systems have been widely applied to several domains and applications to assist decision making by recommending items tailored to user preferences. One of the popular recommendation algorithms is the model-based approach which optimizes a specific objective to improve the recommendation performance. These traditional recommendation models usually deal with a single objective, such as minimizing the prediction errors or maximizing the ranking quality of the recommendations. In recent years, there is an emerging demand for multi-objective recommender systems in which multiple objectives are considered and the recommendations can be optimized by the multi-objective optimization. For example, a recommendation model may be built by optimizing multiple metrics, such as accuracy, novelty and diversity of the recommendations. The multi-objective optimization methodologies have been well developed and applied to the area of recommender systems. In this article, we provide a comprehensive literature review of the multi-objective recommender systems. Particularly, we identify the circumstances in which a multi-objective recommender system could be useful, summarize the methodologies and evaluation approaches in these systems, point out existing challenges or weaknesses, finally provide the guidelines and suggestions for the development of multi-objective recommender systems.

In the past decades recommender systems have become a powerful tool to improve personalization on the Web. Yet, many popular websites lack such functionality, its implementation usually requires certain technical skills, and, above all, its introduction is beyond the scope and control of end-users. To alleviate these problems, this paper presents a novel tool to empower end-users without programming skills, without any involvement of website providers, to embed personalized recommendations of items into arbitrary websites on client-side. For this we have developed a generic meta-model to capture recommender system configuration parameters in general as well as in a web augmentation context. Thereupon, we have implemented a wizard in the form of an easy-to-use browser plug-in, allowing the generation of so-called user scripts, which are executed in the browser to engage collaborative filtering functionality from a provided external rest service. We discuss functionality and limitations of the approach, and in a study with end-users we assess the usability and show its suitability for combining recommender systems with web augmentation techniques, aiming to empower end-users to implement controllable recommender applications for a more personalized browsing experience., Laboratorio de Investigación y Formación en Informática Avanzada

This paper discusses the valuable role recommender systems may play in cybersecurity. First, a comprehensive presentation of recommender system types is presented, as well as their advantages and disadvantages, possible applications and security concerns. Then, the paper collects and presents the state of the art concerning the use of recommender systems in cybersecurity; both the existing solutions and future ideas are presented. The contribution of this paper is two-fold: to date, to the best of our knowledge, there has been no work collecting the applications of recommenders for cybersecurity. Moreover, this paper attempts to complete a comprehensive survey of recommender types, after noticing that other works usually mention two–three types at once and neglect the others.

Recommender systems (RSs) have attained exceptional performance in learning users' preferences and finding the most suitable products. Recent advances in adversarial machine learning (AML) in computer vision have raised interests in recommenders' security.It has been demonstrated that widely adopted model-based recommenders, e.g., BPR-MF, are not robust to adversarial perturbations added on the learned parameters, e.g., users' embeddings, which can cause drastic reduction of recommendation accuracy.However, the state-of-the-art adversarial method, named fast gradient sign method (FGSM), builds the perturbation with a single-step procedure. In this work, we extend the FGSM method proposing multi-step adversarial perturbation (MSAP) procedures to study the recommenders' robustness under powerful methods. Letting fixed the perturbation magnitude, we illustrate that MSAP is much more harmful than FGSM in corrupting the recommendation performance of BPR-MF. Then, we assess the MSAP efficacy on a robustified version of BPR-MF, i.e., AMF. Finally, we analyze the variations of fairness measurements on each perturbed recommender. Code and data are available at https://github.com/sisinflab/MSAP.

Recommender Systems have shown to be an effective way to alleviate the over-choice problem and provide accurate and tailored recommendations. However, the impressive number of proposed recommendation algorithms, splitting strategies, evaluation protocols, metrics, and tasks, has made rigorous experimental evaluation particularly challenging. Puzzled and frustrated by the continuous recreation of appropriate evaluation benchmarks, experimental pipelines, hyperparameter optimization, and evaluation procedures, we have developed an exhaustive framework to address such needs. Elliot is a comprehensive recommendation framework that aims to run and reproduce an entire experimental pipeline by processing a simple configuration file. The framework loads, filters, and splits the data considering a vast set of strategies (13 splitting methods and 8 filtering approaches, from temporal training-test splitting to nested K-folds Cross-Validation). Elliot optimizes hyperparameters (51 strategies) for several recommendation algorithms (50), selects the best models, compares them with the baselines providing intra-model statistics, computes metrics (36) spanning from accuracy to beyond-accuracy, bias, and fairness, and conducts statistical analysis (Wilcoxon and Paired t-test). The aim is to provide the researchers with a tool to ease (and make them reproducible) all the experimental evaluation phases, from data reading to results collection. Elliot is available on GitHub (https://github.com/sisinflab/elliot)., Published at SIGIR 2021, 10 pages, 1 figure

Providing recommendations to groups of users has become a promising research area, since many items tend to be consumed by groups of people. Various techniques have been developed aiming at making recommendations to a group as a whole. Most works use aggregation techniques to combine preferences, recommendations or profiles. However, satisfying all group members in an even way still remains as a challenge. To deal with this problem, we propose an extension of a multi-agent approach based on negotiation techniques for group recommendation. In the approach, we use the multilateral Monotonic Concession Protocol (MCP) to combine individual recommendations into a group recommendation. In this work, we extend the MCP protocol to allow users to personalize the behavior of the agents. This extension was evaluated in two different domains (movies and points of interest) with satisfactory results. We compared our approach against different baselines, namely: a preference aggregation algorithm, a recommendation aggregation algorithm, and a simple one-step negotiation. The results show evidence that, when using our negotiation approach, users in the groups are more uniformly satisfied than with traditional aggregation approaches. Fil: Villavicencio, Christian Paulo. Consejo Nacional de Investigaciones Científicas y Técnicas. Centro Científico Tecnológico Conicet - Tandil. Instituto Superior de Ingeniería del Software. Universidad Nacional del Centro de la Provincia de Buenos Aires. Instituto Superior de Ingeniería del Software; Argentina Fil: Schiaffino, Silvia Noemi. Consejo Nacional de Investigaciones Científicas y Técnicas. Centro Científico Tecnológico Conicet - Tandil. Instituto Superior de Ingeniería del Software. Universidad Nacional del Centro de la Provincia de Buenos Aires. Instituto Superior de Ingeniería del Software; Argentina Fil: Diaz Pace, Jorge Andres. Consejo Nacional de Investigaciones Científicas y Técnicas. Centro Científico Tecnológico Conicet - Tandil. Instituto Superior de Ingeniería del Software. Universidad Nacional del Centro de la Provincia de Buenos Aires. Instituto Superior de Ingeniería del Software; Argentina Fil: Monteserin, Ariel José. Consejo Nacional de Investigaciones Científicas y Técnicas. Centro Científico Tecnológico Conicet - Tandil. Instituto Superior de Ingeniería del Software. Universidad Nacional del Centro de la Provincia de Buenos Aires. Instituto Superior de Ingeniería del Software; Argentina

The objective of this paper is to evaluate multi-label classification techniques and recommender systems for cross-sell purposes in the financial services sector. We carried out three analyses using data obtained from an international financial services provider. First, we tested four multi-label classification techniques, of which the two problem transformation methods were combined with several base classifiers. Second, we benchmarked the performance of five state-of-the-art recommender approaches. Third, we compared the best performing multi-label classification and recommender approaches with each other. The results identify user-based collaborative filtering as the top performing recommender system, with a cross-validated F1 measure of 42.20% and G-mean of 42.64%. Classifier chains binary relevance with adaboost and binary relevance with random forest are the top performing multi-label classification algorithms for respectively F1 measure and G-mean, yielding a cross-validated F1 measure of 53.33% and G-mean of 54.37%. The statistical comparison between the best performing approaches confirms the superiority of multi-label classification techniques. Our study provides important recommendations for financial services providers, who are interested in the most effective methods to determine cross-sell opportunities. In previous studies, multi-label classification techniques and recommender systems were always investigated independently of each other. To the best of our knowledge, our study is therefore the first to compare both techniques in the financial services sector.

This paper proposes two types of recommender systems based on sparse dictionary coding. Firstly, a novel predictive recommender system that attempts to predict a user’s future rating of a specific item. Secondly, a top-n recommender system which finds a list of items predicted to be most relevant for a given user. The proposed methods are assessed using a variety of different metrics and are shown to be competitive with existing collaborative filtering recommender systems. Specifically, the sparse dictionary-based predictive recommender has advantages over existing methods in terms of a lower computational cost and not requiring parameter tuning. The sparse dictionary-based top-n recommender system has advantages over existing methods in terms of the accuracy of the predictions it makes and not requiring parameter tuning. An open-source software implemented and used for the evaluation in this paper is also provided for reproducibility.

Recommender systems provide users with content they might be interested in. Conventionally, recommender systems are evaluated mostly by using prediction accuracy metrics only. But, the ultimate goal of a recommender system is to increase user satisfaction. Therefore, evaluations that measure user satisfaction should also be performed before deploying a recommender system in a real target environment. Such evaluations are laborious and complicated compared to the traditional, data-centric evaluations, though. In this study, we carried out a user-centric evaluation of state-of-the-art recommender systems as well as a graph-based approach in the ecologically valid setting of an authentic social learning platform. We also conducted a data-centric evaluation on the same data to investigate the added value of user-centric evaluations and how user satisfaction of a recommender system is related to its performance in terms of accuracy metrics. Our findings suggest that user-centric evaluation results are not necessarily in line with data-centric evaluation results. We conclude that the traditional evaluation of recommender systems in terms of prediction accuracy only does not suffice to judge performance of recommender systems on the user side. Moreover, the user-centric evaluation provides valuable insights in how candidate algorithms perform on each of the five quality metrics for recommendations: usefulness, accuracy, novelty, diversity, and serendipity.

Recommender systems (RSs) are widely utilised across industries as tools to provide users with recommendations based on their preferences. This paper reports on an examination of the influence of trusting beliefs on behavioural intentions to reuse RSs, emphasising the effects of trust propensity, perceived usefulness and product type. A distinctive contribution of this study is the research model, which integrated perceived usefulness as an antecedent of trusting beliefs. Data collected in Australia with 366 participants were used. A new approach (MICOM and PLS-MGA) was performed to assess the moderating effect of product type. The research results indicate that trust propensity has a positive and significant direct effect on consumers’ trusting beliefs in an ongoing relationship. The results also suggest that consumers’ trusting beliefs and perceived usefulness of RSs positively and significantly affect their intention to reuse RSs. Perceived usefulness of RSs is more important compared to trusting beliefs in predicting consumers’ behavioural intention to reuse RSs. Trusting beliefs partially mediate the impact of perceived usefulness on behavioural intentions to reuse RSs. The results also showed an insignificant difference between the effect of different product types on Australian consumers’ behavioural intention to reuse RSs. These results have implications for the design of RSs.

Recommender systems (RSs) have widely grown thanks to the outstanding capability of providing users with accurate and tailored recommendations. Recently, public awareness and new regulations forced RS researchers and practitioners to study solutions to user privacy endangerment. This tutorial will guide the attendees through the possible threats and the solutions towards private RSs.

An effective recommender system can significantly help customers to find desired products and assist business owners to earn more income. Nevertheless, the decision-making process of users is highly complex, not only dependent on the personality and preference of a user, but also complicated by the characteristics of a specific product. For example, for products of different domains (e.g., clothing vs. office products), the product aspects that affect a user's decision are very different. As such, traditional collaborative filtering methods that model only user-item interactions would deliver unsatisfactory recommendation results. In this work, we first divide a product's characteristics into visual and functional aspects. We then contribute a novel probabilistic model, named Visual and Functional Probabilistic Matrix Factorization (VFPMF), to unify the two factors to estimate user preferences on products. Nevertheless, such an expressive model poses efficiency challenge in parameter learning from implicit feedback. To address the technical challenge, we devise a computationally efficient learning algorithm for VFPMF. Furthermore, we provide an online updating procedure of the algorithm, shedding some light on how to adapt our method to real-world recommendation scenario where data continuously streams in. Extensive experiments on four datasets demonstrate the effectiveness of our method.

To protect recommender systems against shilling attacks, a variety of detection methods have been proposed over the past decade. However, these methods focus mainly on individual features and rarely consider the lockstep behaviours among attack users, which suffer from low precision in detecting group shilling attacks. In this work, we propose a three-stage detection method based on strong lockstep behaviours among group members and group behaviour features for detecting group shilling attacks. First, we construct a weighted user relationship graph by combining direct and indirect collusive degrees between users. Second, we find all dense subgraphs in the user relationship graph to generate a set of suspicious groups by introducing a topological potential method. Finally, we use a clustering method to detect shilling groups by extracting group behaviour features. Extensive experiments on the Netflix and sampled Amazon review datasets show that the proposed approach is effective for detecting group shilling attacks in recommender systems, and the F1-measure on two datasets can reach over 99 percent and 76 percent, respectively.

Recommender systems have been exhaustively implemented in the e-commerce industry for providing accurate, efficient, and effective personalized recommendations to candidate users. The variants of matrix factorization (MF) techniques incorporating the stochastic gradient descent (SGD) are exploited to improve the efficacy of recommender systems through effectively dealing the fuzzy behavior. The iterative update mechanism of MF-based SGD techniques involves current but limited information for providing related recommendations. The strength of sliding window and multi-innovation-based approximations with memory can improve the accuracy of the recommender systems through prior knowledge by utilizing the fuzziness among ratings. In this work, a moving-information-based computing paradigm is presented to effectively handle the fuzzy nature of preferences by recommender systems with the ability to capture the collective effect of ratings of previous update history obtained for a defined information length to provide fast and precise recommendations.

Recently, various deep learning-based models have been applied in the study of recommender systems. Some researches have combined the classic collaborative filtering method with deep learning frameworks in order to obtain more accurate recommendations. However, these models either add additional features, but still recommend in the original linear manner, or only extract the global latent factors of the rating matrices in a non-linear way without considering some local special relationships. In this paper, we propose a deep learning framework for explicit recommender systems, named Attention Collaborative Autoencoder (ACAE). Based on the denoising autoencoder, our model can extract the global latent factors in a non-linear fashion from the sparse rating matrices. In ACAE, attention units are introduced during back propagation, enabling discovering potential relationships between users and items in the neighborhood, which makes the model obtain better results in the rating prediction tasks. In addition, we propose how to optimize the training process of the model by proposing a new loss function. Experiments on two public datasets demonstrate the effectiveness of ACAE and its outperformance of competitive baselines.

Conversational Recommender Systems (CoRSs) implement a paradigm in which users can interact with the system to define their preferences and discover items that best fit their needs. When the CoRS is implemented as a dialog agent, user and recommender interact by exchanging text messages. However, there is little evidence on how effective the interaction is when the CoRS is implemented through a Social Humanoid Robot. In this paper, we evaluate the possibility of introducing an interface based on a Social Humanoid Robot in ConveRSE, a domain-independent framework for the development of Conversational Recommender Systems. The novel interface will be compared against the existing chatbot-based one. The objective is to discover whether the framework can adapt to the new interface without worsening user experience and accuracy. We carried out a preliminary study, which involved 20 subjects. Results proved that, even though there are differences in how users approach the system using the two interfaces, there is no significant difference in its performance.

Recommender systems (RSs) have recently gained significant attention from both research and industrial communities. These systems generate the recommendations of items in one of two ways, namely collaborative or content-based filtering. Collaborative filtering is a technique used by recommender systems in order to suggest to the user a set of items based on the opinions of other users who share with him the same preferences. One of the key issues in collaborative filtering systems (CFSs) is how to generate adequate recommendations for newcomers who rate only a small number of items, a problem known as cold start user. Another interesting problem is the cold start item when a new item is introduced in the system and cannot be recommended. In this paper, we present a clustering-based approach SCOL that aims to alleviate the cold start challenges; by identifying the most effective opinion leaders among the social network of the CFS. SCOL clustering focuses on the credibility and correlation similarity concepts. 12 01 34 50

Conversational Recommender Systems (CoRSs) implement a paradigm that allows users to interact in natural language with the system for defining their preferences and discovering items that best fit their needs. CoRSs can be straightforwardly implemented as chatbots that, nowadays, are becoming more and more popular for several applications, such as customer care, health care, and medical diagnoses. Chatbots implement an interaction based on natural language, buttons, or both. The implementation of a chatbot is a challenging task since it requires knowledge about natural language processing and human–computer interaction. A CoRS might be particularly useful in the music domain since music is generally enjoyed in contexts when a standard interface cannot be exploited (driving, doing homeworks, running). However, there is no work in the literature that analytically compares different interaction modes for a conversational music recommender system. In this paper, we focus on the design and implementation of a CoRS for the music domain. Our CoRS consists of different components. The system implements content-based recommendation, critiquing and adaptive strategies, as well as explanation facilities. The main innovative contribution is that the user can interact through different interaction modes: natural language, buttons, and mixed. Due to the lack of available datasets for testing CoRSs, we carried out an in vivo experimental evaluation with the goal of investigating the impact of the different interaction modes on the recommendation accuracy and on the cost of interaction for the final user. The experiment involved 110 people, and 54 completed the whole process. The analysis of the results shows that the best interaction mode is based on a mixed strategy that combines buttons and natural language. In addition, the results allow to clearly understand which are the steps in the dialog that are particularly strenuous for the user.

Deep learning classifiers are hugely vulnerable to adversarial examples, and their existence raised cybersecurity concerns in many tasks with an emphasis on malware detection, computer vision, and speech recognition. While there is a considerable effort to investigate attacks and defense strategies in these tasks, only limited work explores the influence of targeted attacks on input data (e.g., images, textual descriptions, audio) used in multimedia recommender systems (MR). In this work, we examine the consequences of applying targeted adversarial attacks against the product images of a visual-based MR. We propose a novel adversarial attack approach, called Target Adversarial Attack against Multimedia Recommender Systems (TAaMR), to investigate the modification of MR behavior when the images of a category of low recommended products (e.g., socks) are perturbed to misclassify the deep neural classifier towards the class of more recommended products (e.g., running shoes) with human-level slight images alterations. We explore the TAaMR approach studying the effect of two targeted adversarial attacks (i.e., FGSM and PGD) against input pictures of two state-of-the-art MR (i.e., VBPR and AMR). Extensive experiments on two real-world recommender fashion datasets confirmed the effectiveness of TAaMR in terms of recommendation lists changing while keeping the original human judgment on the perturbed images.

Recommender systems have focused on algorithms for a recommendation for individuals. However, in many domains, it may be recommending an item, for example, movies, restaurants etc. for a group of persons for which some remarkable group recommender systems (GRSs) has been developed. GRSs satisfy a group of people optimally by considering the equal weighting of the individual preferences. We have proposed a multi-expert scheme (MES) for group recommendation using genetic algorithm (GA) MES-GRS-GA that depends on consensus techniques to further improve group recommendations. In order to deal with this problem of GRS, we also propose a consensus scheme for GRSs where consensus from multiple experts are brought together to make a single recommended list of items in which each expert represents an individual inside the group. The proposed GA based consensus scheme is modeled as many consensus schemes within two phases. In the consensus phase, we have applied GA to obtain the maximum utility offer for each expert and generated the most appropriate rating for each item in the group. In the recommendation generation phase, again GA has been employed to produce the resulting group profile, i.e. the list of ratings with the minimum sum of distances from the group members. Finally, the results of computational experiments that bear close resemblance to real-world scenarios are presented and compared to baseline GRS techniques that illustrate the superiority of the proposed model.

Privacy-preserving recommender systems have been an active research topic for many years. However, until today, it is still a challenge to design an efficient solution without involving a fully trusted third party or multiple semi-trusted third parties. The key obstacle is the large underlying user populations (i.e., huge input size) in the systems. In this paper, we revisit the concept of friendship-based recommender systems, proposed by Jeckmans et al. and Tang and Wang. These solutions are very promising because recommendations are computed based on inputs from a very small subset of the overall user population (precisely, a user's friends and some randomly chosen strangers). We first clarify the single prediction protocol and Top-n protocol by Tang and Wang, by correcting some flaws and improving the efficiency of the single prediction protocol. We then design a decentralized single protocol by getting rid of the semi-honest service provider. In order to validate the designed protocols, we crawl Twitter and construct two datasets (FMT and 10-FMT) which are equipped with auxiliary friendship information. Based on 10-FMT and MovieLens 100k dataset with simulated friendships, we show that even if our protocols use a very small subset of the datasets, their accuracy can still be equal to or better than some baseline algorithm. Based on these datasets, we further demonstrate that the outputs of our protocols leak very small amount of information of the inputs, and the leakage decreases when the input size increases. We finally show that he single prediction protocol is quite efficient but the Top-n is not. However, we observe that the efficiency of the Top-n protocol can be dramatically improved if we slightly relax the desired security guarantee.

Recommender systems play an increasingly important role in on-line web services for the personalization and recommendation of content to individual users. The quantity and quality of user-based information has progressed presenting the opportunity to further tailor recommendations to users based on feature view integration. In this work, we propose a hybrid framework which combines a collaborative filtering recommendation system with fuzzy decision-making approaches (based on the use of aggregation functions) to improve the accuracy of domain-specific recommendations. We extend upon the classical, neighborhood-based collaborative filtering process by conflating preference information with user-profile data in the recommendation process. This is performed using intelligent information fusion techniques whereby Ordered Weighted Averaging (OWA) operators and uninorm aggregation functions are implemented in the fusion of multiple views of pairwise similarity degrees between users. To address the shortcoming of generating sensible recommendations to cold users, we incorporate a novel weighting scheme based on fuzzy set modeling within the uninorm-based aggregation of similarity views. We finally outline the application of the proposed approach through an empirical study based in the Urban Resilience domain, along with an example to movie recommendation.

Recommender systems have emerged as a class of essential tools in the success of modern e-commerce applications. These applications typically handle large datasets and often face challenges like data sparsity and scalability. Clustering techniques help to reduce the computational time needed for recommendation as well as handle the sparsity problem more efficiently. Traditional clustering based recommender systems create partitions (clusters) of the user-item rating matrix and execute the recommendation algorithm in the clusters separately in order to decrease the overall runtime of the system. Each user or item generally belong to at most one cluster. However, it may so happen that some users (boundary users) present in a particular cluster exhibit higher similarity with the preferences of the users residing in the nearby clusters than the ones present in their own cluster. Therefore, we propose a clustering based scalable recommendation algorithm that has a provision for switching a user from its original cluster to another cluster in order to provide more accurate recommendations. For a user belonging to multiple clusters, we aggregate recommendations from those clusters to which the user belongs in order to produce the final set of recommendations to that user. In this work, we propose two types of clustering, one on the basis of rating and the other on the basis of frequency and then compare their performances. Finally, we explore the applicability of cluster ensembles techniques in the proposed method. Our aim is to develop a recommendation framework that can scale well to handle large datasets without much affecting the recommendation quality. The outcomes of our experiments clearly demonstrate the scalability as well as efficacy of our method. It reduces the runtime of the baseline CF algorithm by a minimum of 58% and a maximum of 90% for MovieLens-10M dataset, and a minimum of 42% and a maximum of 84% for MovieLens-20M dataset. The accuracies of recommendations in terms of F1, MAP and NDCG metrics are also better than the existing clustering based recommender systems.

Memory-based collaborative filtering constitutes an important technique of recommender systems mainly due to its simplicity and efficiency. However, it suffers from several fundamentally critical problems when its system makes recommendations based on ratings records of similar users. This study addresses data sparsity and scalability problems that are major drawbacks of the memory-based system. In order to take care of the data sparsity problem, we deduce user interest in movie genres from the user ratings and devise a similarity measure based on the genre preference. Then clusters of users are built based on the genre preference similarity by employing a fuzzy clustering technique, which not only reflects the subjectivity of user ratings but reduces the data scalability problem. Furthermore, we apply an optimization method to the proposed technique to resolve shortcomings of the fuzzy clustering algorithm by using the genetic algorithm. Extensive experiments are conducted to find that the proposed method demonstrates its performance superior or comparable to the previous methods in terms of various metrics. Moreover, the proposed approach turns out to yield the highest prediction accuracy among the experimented methods, thus proving to overcome the serious problem of low prediction encountered with clustering-based collaborative filtering.

Providing useful information to the users by recommending highly demanded products and services is a fundamental part of the business of many top tier companies. Recommender Systems make use of many sources of information to provide users with accurate predictions and novel recommendations of items. Here we propose, DeepMF, a novel collaborative filtering method that combines the Deep Learning paradigm with Matrix Factorization (MF) to improve the quality of both predictions and recommendations made to the user. Specifically, DeepMF performs successive refinements of a MF model with a layered architecture that uses the acquired knowledge in a layer as input for subsequent layers. Experimental results showed that the quality of both the predictions and recommendations of DeepMF overcome the baselines., This work has been supported by Spanish Ministry of Science and Education and Competitivity (MINECO) and European Regional Development Fund (FEDER) under grants TIN2017-85727-C4-3-P (DeepBio).

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.

In e-business, recommender systems have been instrumental in guiding users through their online experiences. However, these systems are often limited by the lack of labels data and data sparsity. Increasingly, data-mining techniques are utilized to address this issue. In most research, recommendations to be made are achieved via supervised learning that typically employs the k-nearest neighbor learner. However, supervised learning relies on labeled data, which may not be available at the time of learning. Data sparsity, which refers to situations where the number of items that have been recommended represents only a small subset of all available items, further affects model performance. One suggested solution is to apply cluster analysis as a preprocessing step and thus guide the learning process from natural grouping, typically using similar customer profiles, to improve predictive accuracy. In this paper, we study the benefits of applying cluster analysis as a preprocessing step prior to constructing classification models. Our HCC-Learn framework combines content-based analysis in the preprocessing stage and collaborative filtering in the final prediction stage. Our results show the value of our HCC-Learn framework applied to real-world data sets, especially when combining soft clustering and ensembles based on feature subspaces., 10th International Joint Conference, IC3K 2018, September 18-20, 2018, Seville, Spain, Series: Communications in Computer and Information Science

E-commerce platforms heavily rely on automatic personalized recommender systems, e.g., collaborative filtering models, to improve customer experience. Some hybrid models have been proposed recently to address the deficiency of existing models. However, their performances drop significantly when the dataset is sparse. Most of the recent works failed to fully address this shortcoming. At most, some of them only tried to alleviate the problem by considering either user side or item side content information. In this article, we propose a novel recommender model called Hybrid Variational Autoencoder (HVAE) to improve the performance on sparse datasets. Different from the existing approaches, we encode both user and item information into a latent space for semantic relevance measurement. In parallel, we utilize collaborative filtering to find the implicit factors of users and items, and combine their outputs to deliver a hybrid solution. In addition, we compare the performance of Gaussian distribution and multinomial distribution in learning the representations of the textual data. Our experiment results show that HVAE is able to significantly outperform state-of-the-art models with robust performance.