Successful Recommender System Research Articles

Privacy-preserving recommendation system based on social relationships

In the era of internet-based data, recommendation systems are crucial for helping users access personalized content and facilitating business promotion and sales. Recommendation systems based on social relationships have gained popularity due to their ability to use social influence and enhance the impact and credibility of recommendations. However, building a successful recommendation system faces the challenge of protecting user data privacy, as it requires a significant amount of sensitive user data. For this challenge, we propose a privacy-preserving recommendation system based on social relationships in the dual-cloud model. To achieve the secure computation of this system, we also design and improve some underlying secure protocols such as secure equality protocol and secure comparison protocol. Our underlying protocol has the advantages of high efficiency and provable security. As a result, our recommendation system can provide excellent recommendations while ensuring privacy protection. Experiments show that our system can perform secure recommendation calculations on 128-bit data with semi-honest security in approximately 5.5 s.

Computational Model of Recommender System Intervention

A recommender system is an information selection system that offers preferences to users and enhances their decision-making. This system is commonly implemented in human-computer-interaction (HCI) intervention because of its information filtering and personalization. However, its success rate in decision-making intervention is considered low and the rationale for this is associated with users’ psychological reactance which is causing unsuccessful recommender system interventions. This paper employs a computational model to depict factors that lead to recommender system rejection by users and how these factors can be enhanced to achieve successful recommender system interventions. The study made use of design science research methodology by executing a computational analysis based on an agent-based simulation approach for the model development and implementation. A total of sixteen model concepts were identified and formalized which were implemented in a Matlab environment using three major case conditions as suggested in previous studies. The result of the study provides an explicit comprehension on interplaying of recommender system that generate psychological reactance which is of great importance to recommender system developers and designers to depict how successful recommender system interventions can be achieved without users experiencing reactance and rejection on the system.

A differentially private matrix factorization based on vector perturbation for recommender system

Matrix factorization (MF) techniques have yielded immense success in recommender systems (RSs). Since a huge amount of user data is collected and used in RS, it raises concerns about data privacy. As a strict privacy protection framework, many efforts attempt to apply Differential Privacy (DP) to MF. However, there are still some challenges or problems in designing MF with privacy preservation, such as error accumulation in the multiple iterations of MF, introduction of unnecessary noises, and difficult sensitivity analysis. To overcome these problems, we devise a vector perturbation-based differentially private matrix factorization (VP-DPMF). Our scheme can prevent error accumulation by perturbing the objective function of MF rather than its factorization process or results. It also addresses the difficulty of analyzing sensitivity by exploiting the polynomial representation of the objective function. Furthermore, our scheme can reduce unnecessary noises by controlling the perturbation within the vector term of the polynomial, and can preserve the convexity property of the original function. Theoretical analysis demonstrates that our scheme can achieve good performance in a large-scale recommender system. Experimental results on some benchmark datasets show that the proposed scheme can provide both rigid privacy guarantee and satisfactory recommendation quality.

Fused Matrix Factorization with Geographical and Social Influence in Location-Based Social Networks

Recently, location-based social networks (LBSNs), such as Gowalla, Foursquare, Facebook, and Brightkite, etc., have attracted millions of users to share their social friendship and their locations via check-ins. The available check-in information makes it possible to mine users’ preference on locations and to provide favorite recommendations. Personalized Point-of-interest (POI) recommendation is a significant task in LBSNs since it can help targeted users explore their surroundings as well as help third-party developers to provide personalized services. To solve this task, matrix factorization is a promising tool due to its success in recommender systems. However, previously proposed matrix factorization (MF) methods do not explore geographical influence, e.g., multi-center check-in property, which yields suboptimal solutions for the recommendation. In this paper, to the best of our knowledge, we are the first to fuse MF with geographical and social influence for POI recommendation in LBSNs. We first capture the geographical influence via modeling the probability of a user’s check-in on a location as a Multi-center Gaussian Model (MGM). Next, we include social information and fuse the geographical influence into a generalized matrix factorization framework. Our solution to POI recommendation is efficient and scales linearly with the number of observations. Finally, we conduct thorough experiments on a large-scale real-world LBSNs dataset and demonstrate that the fused matrix factorization framework with MGM utilizes the distance information sufficiently and outperforms other state-of-the-art methods significantly.

Recommender systems: Past, present, future

AbstractThe origins of modern recommender systems date back to the early 1990s when they were mainly applied experimentally to personal email and information filtering. Today, 30 years later, personalized recommendations are ubiquitous and research in this highly successful application area of AI is flourishing more than ever. Much of the research in the last decades was fueled by advances in machine learning technology. However, building a successful recommender system requires more than a clever general‐purpose algorithm. It requires an in‐depth understanding of the specifics of the application environment and the expected effects of the system on its users. Ultimately, making recommendations is a human‐computer interaction problem, where a computerized system supports users in information search or decision‐making contexts. This special issue contains a selection of papers reflecting this multi‐faceted nature of the problem and puts open research challenges in recommender systems to the forefront. It features articles on the latest learning technology, reflects on the human‐computer interaction aspects, reports on the use of recommender systems in practice, and it finally critically discusses our research methodology.

Research on Cosine Similarity and Pearson Correlation Based Recommendation Models

Recommendation models are based on data mining and machine learning techniques to suggest product and services to users. Advancement in the machine learning field results in the growth of more efficient recommendation models. Recommendation systems can contribute a lot to increase the business of internet based companies. Techniques for making recommendations generally fall into two categories namely collaborative filtering based models and content based recommendation models. The first techniques explores similarity based approaches like cosine similarity, Pearson correlation similarity etc. The second approaches uses predictions based on explicit and implicit ratings given by users. Recommendation systems try to suggest products to the users based on their previous purchases or general nature. A successful recommendation system is capable of improving the product sales to a large extend. This paper reviews about various recommendation models and performs a comparison of these models.

Multi-criteria collaborative filtering using rough sets theory

Recommender systems have recently become a significant part of e-commerce applications. Through the different types of recommender systems, collaborative filtering is the most popular and successful recommender system for providing recommendations. Recent studies have shown that using multi-criteria ratings helps the system to know the customers better. However, bringing multi aspects to collaborative filtering causes new challenges such as scalability and sparsity. Additionally, revealing the relation between criteria is yet another optimization problem. Hence, increasing the accuracy in prediction is a challenge. In this paper, an aggregation-function based multi-criteria collaborative filtering system using Rough Sets Theory is proposed as a novel approach. Rough Sets Theory is used to uncover the relationship between the overall criterion and the individual criteria. Experimental results show that the proposed model (RoughMCCF) successfully improves the predictive accuracy without compromising on online performance.

Research on Rating Error and Quality Metrics for Collaborative Filtering Recommendation Methods

The collaborative filtering recommendation system has been widely used in E-commerce as a relatively successful recommendation system. At present, the focus of collaborative filtering recommendation research is mainly on how to improve the accuracy of recommendation by improving the recommendation algorithm. However, in real world, the user’s rating behaviour is not perfectly rational. It is no odd that there is deviation of rating to any a given item for a user in real evaluation. In this case, what it means for the improvement of collaborative filtering recommendation methods, and how they performed when we use the commonly used quality metrics to evaluate the collaborative filtering recommendation methods? In view of these problems, this paper presupposes that the user’s rating behaviour is a bounded rational behaviour, and based on the normality hypothesis of rating error, introduces a simulated rating experiment innovatively to analyse the effect that rating prediction can achieve in sense of the commonly used quality metrics. This study is of significance for the research and application of collaborative filtering recommendation technology.

Bayesian dual neural networks for recommendation

Most traditional collaborative filtering (CF) methods only use the user-item rating matrix to make recommendations, which usually suffer from cold-start and sparsity problems. To address these problems, on the one hand, some CF methods are proposed to incorporate auxiliary information such as user/item profiles; on the other hand, deep neural networks, which have powerful ability in learning effective representations, have achieved great success in recommender systems. However, these neural network based recommendation methods rarely consider the uncertainty of weights in the network and only obtain point estimates of the weights. Therefore, they maybe lack of calibrated probabilistic predictions and make overly confident decisions. To this end, we propose a new Bayesian dual neural network framework, named BDNet, to incorporate auxiliary information for recommendation. Specifically, we design two neural networks, one is to learn a common low dimensional space for users and items from the rating matrix, and another one is to project the attributes of users and items into another shared latent space. After that, the outputs of these two neural networks are combined to produce the final prediction. Furthermore, we introduce the uncertainty to all weights which are represented by probability distributions in our neural networks to make calibrated probabilistic predictions. Extensive experiments on real-world data sets are conducted to demonstrate the superiority of our model over various kinds of competitors.

Alleviating sparsity by specificity‐aware ontology‐based clustering for improving web service recommendation

The rapid development of the Internet in recent years has led to a vast increase in the numbers of Web services, which challenge users' capability to find their favorite services quickly and accurately. There is thus an urgent demand for service recommendations that help discover applicable services. Although the collaborative filtering technique is one of the most successful recommendation system technologies, it suffers from data sparsity and cold‐start problems, which in turn lead to inaccurate results. In this article, we deal with these issues by applying a novel ontology‐based clustering approach that uses domain specificity and service similarity for ontology generation. This clustering approach can easily and effectively increase the data density of the user‐service dataset by assuming blank user preferences according to the history of user‐favored domain(s). Then, user ratings are predicted by calculating the trust value between users. The experimental results indicate that the proposed approach can effectively alleviate the data sparsity and cold‐start problems with lower prediction error with the best recommendation performance, and our method performed better than baseline approaches in terms of accuracy and novelty. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Stability Properties of Graph Neural Networks

Graph neural networks (GNNs) have emerged as a powerful tool for nonlinear processing of graph signals, exhibiting success in recommender systems, power outage prediction, and motion planning, among others. GNNs consist of a cascade of layers, each of which applies a graph convolution, followed by a pointwise nonlinearity. In this work, we study the impact that changes in the underlying topology have on the output of the GNN. First, we show that GNNs are permutation equivariant, which implies that they effectively exploit internal symmetries of the underlying topology. Then, we prove that graph convolutions with integral Lipschitz filters, in combination with the frequency mixing effect of the corresponding nonlinearities, yields an architecture that is both stable to small changes in the underlying topology, and discriminative of information located at high frequencies. These are two properties that cannot simultaneously hold when using only linear graph filters, which are either discriminative or stable, thus explaining the superior performance of GNNs.

A Stable Collaborative Filtering Algorithm for Long Tail Recommendation

As a relatively successful recommendation system, the collaborative filtering recommendation system (CFRS) has been widely used in e-commerce. However, the current CFRS is mainly based on mainstream or popular products to recommending similar items for users and is less efficiency in recommend the so called “Long Tail” products to meet the individual needs of users. Based on the Item-based system filtering recommendation algorithm, this paper proposes a collaborative filtering recommendation algorithm that implements long tail recommendation by using the item rating probability matrix and item rating reliability. Compared with the traditional collaborative filtering algorithm, the experimental result based on MovieLens 1M dataset shows that the proposed algorithm can deal with the data sparsity problem better, and is better for producing recommendation for the long tail products effect, and furthermore, it shows stability to a certain extent in producing recommendations under different situations of data sparseness.

A scalable and robust trust-based nonnegative matrix factorization recommender using the alternating direction method

Matrix Factorization (MF) has been proven to be an effective approach for the generation of a successful recommender system. However, most current MF-based recommenders cannot obtain high prediction accuracy due to the sparseness of the user–item matrix in collaborative filtering models. Moreover, they suffer from scalability issues when applied to large-scale real-world tasks. To tackle these issues, a social regularization method, called TrustANLF, is proposed, which incorporates users’ social trust information in a nonnegative matrix factorization framework. The proposed method integrates trust statements as an additional information source along with rating values into the recommendation model to deal with the data sparsity and cold-start issues. Moreover, the alternating direction optimization method is used for solving the trust-based nonnegative MF model in order to improve convergence speed as well as reduce computational and memory costs. To evaluate the effectiveness of the proposed method, several experiments are performed on three real-world datasets. The obtained results demonstrate the significant improvements of the proposed method over several state-of-the-art methods.

APRA: An approximate parallel recommendation algorithm for Big Data

Finding relevant and interesting items according to the preferences of each user has become an important challenge in the era of Big Data. Recommender systems have emerged in response to this problem. Collaborative Filtering (CF) is one of the most successful recommender systems used by several big online shopping companies. However, CF is computationally demanding, especially in Big Data context, where the number of users and items are too big to be effectively processed by traditional approaches.In this paper, we propose a new solution based on Spark, which is tailored to handle large-scale data and provide better results. Particularly, we take advantage of the in-memory operations available through Spark, to improve the performance of recommendation systems in context of Big Data. Experimental results on two real-world data sets confirm the claim.

Learning social regularized user representation in recommender system

Appropriate user and item representation learning is critical to a successful recommender system. A set of models from natural language processing domain, namely neural language models, have recently been utilized to user and item representation learning in standard recommendation tasks. This paper proposes to exploit neural language models in the social recommendation task. Unlike previous studies that focus on modeling the user-item interaction matrix and only consider the item-level context, this paper models user social relationship information and adds an additional layer to incorporate user-level context. The derived representation is very like the social regularization imposed in matrix factorization-based recommendation, but with more flexible context. Experiments on a Douban movie dataset validate the advantage of the proposed model.

Recommending packages with validity constraints to groups of users

The success of recommender systems has made them the focus of a massive research effort in both industry and academia. Recent work has generalized recommendations to suggest packages of items to single users, or single items to groups of users. However, to the best of our knowledge, the interesting problem of recommending a package to a group of users (P2G) has not been studied to date. This is a problem with several practical applications, such as recommending vacation packages to tourist groups, entertainment packages to groups of friends or sets of courses to groups of students. In this paper, we formulate the P2G problem, and we propose probabilistic models that capture the preference of a group toward a package, incorporating factors such as user impact and package viability. We also investigate the issue of recommendation fairness. This is a novel consideration that arises in our setting, where we require that no user is consistently slighted by the item selection in the package. In addition, we study a special case of the P2G problem, where the recommended items are places and the recommendation should consider the current locations of the users in the group. We present aggregation algorithms for finding the best packages and compare our suggested models with baseline approaches stemming from previous work. The results show that our models find packages of high quality which consider all special requirements of P2G recommendation.

Bayesian Wishart matrix factorization

User tastes are constantly drifting over time as users are exposed to different types of products. The ability to model the tendency of both user preferences and product attractiveness is vital to the success of recommender systems (RSs). We propose a Bayesian Wishart matrix factorization method to model the temporal dynamics of variations among user preferences and item attractiveness in a novel algorithmic perspective. The proposed method is able to well model and properly control diverse rating behaviors across time frames and related temporal effects within time frames in the tendency of user preferences and item attractiveness. We evaluate the proposed method on two synthetic and three real-world benchmark datasets for RSs. Experimental results demonstrate that our proposed method significantly outperforms a variety of state-of-the-art methods in RSs.

Research on the Matthews Correlation Coefficients Metrics of Personalized Recommendation Algorithm Evaluation

The personalized recommendation systems could better improve the personalized service for network user and alleviate the problem of information overload in the Internet. As we all know, the key point of being a successful recommendation system is the performance of recommendation algorithm. When scholars put forward some new recommendation algorithms, they claim that the new algorithms have been improved in some respects, better than previous algorithm. So we need some evaluation metrics to evaluate the algorithm performance. Due to the scholar didn’t fully understand the evaluation mechanism of recommendation algorithms. They mainly emphasized some specific evaluation metrics like Accuracy, Diversity. What’s more, the academia did not establish a complete and unified assessment of recommendation algorithms evaluation system which is credibility to do the work of recommendation evaluation. So how to do this work objective and reasonable is still a challengeable task. In this article, we discussed the present evaluation metrics with its respective advantages and disadvantages. Then, we put forward to use the Matthews Correlation Coefficient to evaluate the recommendation algorithm’s performance. All this based on an open source projects called mahout which provides a rich set of components to construct the classic recommendation algorithm. The results of the experiments show that the applicability of Matthews correlation coefficient in the relative evaluation work of recommendation algorithm.

Exploration in Interactive Personalized Music Recommendation

Current music recommender systems typically act in a greedy manner by recommending songs with the highest user ratings. Greedy recommendation, however, is suboptimal over the long term: it does not actively gather information on user preferences and fails to recommend novel songs that are potentially interesting. A successful recommender system must balance the needs to explore user preferences and to exploit this information for recommendation. This article presents a new approach to music recommendation by formulating this exploration-exploitation trade-off as a reinforcement learning task. To learn user preferences, it uses a Bayesian model that accounts for both audio content and the novelty of recommendations. A piecewise-linear approximation to the model and a variational inference algorithm help to speed up Bayesian inference. One additional benefit of our approach is a single unified model for both music recommendation and playlist generation. We demonstrate the strong potential of the proposed approach with simulation results and a user study.

Fuzzy Computational Models of Trust and Distrust for Enhanced Recommendations

Trust models have received considerable attention in the recent past and have been employed in many of today's most successful recommender systems (RSs) for alleviating sparsity by enhancing their interuser connectivity obtained from historical preference data and to address the cold start problem. However, the incorporation of distrust in addition to trust for improved recommendations has not been analyzed fully because of the absence of publically available data sets containing gradual information about trust and distrust concepts. Our work in this paper is an attempt toward introducing recommendation strategies exploiting both the trust and the distrust in the RSs to further enhance their quality of recommendations through fuzzy computational models for trust and distrust. Since trust and distrust concepts are gradual phenomenon, therefore these can be represented more naturally by fuzzy logic using linguistic expressions. The contributions of this paper are three fold: First, we propose fuzzy computational models for both the trust and distrust concepts through similarity as well as knowledge factors based on linguistic expressions. Second, we suggest appropriate propagation and aggregation operators to deal with the data sparsity. Finally, we present a comparative analysis of proposed recommendation strategies utilizing both the trust and distrust concepts.