One-class Collaborative Filtering Research Articles

MSBPR: A multi-pairwise preference and similarity based Bayesian personalized ranking method for recommendation

For addressing the “One-Class Collaborative Filtering” (OCCF) problem in recommendation systems, in which the obtained user information is all single-type positive feedback, the current mainstream methods are all based on the idea of pairwise preference learning. The Bayesian Personalized Ranking (BPR) method is a classical representative of such an idea. However, the assumption in BPR, that “users tend to prefer items that they have once interacted”, may not always hold in reality. This is because for non-interacted items, a user may have different perspectives, such as potentially favorite, dislike, or something in between. For mitigating the above-mentioned issue, this paper proposes a Multi-pairwise preference and Similarity based BPR method, termed as MSBPR for brevity. MSBPR utilizes additional auxiliary feedback information to excavate and infer deeper-level user preferences. Subsequently, MSBPR borrows ideas from traditional item/user-based collaborative filtering methods to further divide non-interacted items from the angle of item/user, respectively. Afterwards, MSBPR constructs four preferences on top of the divided items, and accordingly builds up the multiple pairwise preference assumption. To optimize MSBPR, we derive an efficient learning algorithm based on the stochastic gradient descent algorithm. The computational complexity of MSBPR is also theoretically analyzed. Comprehensive experimental results demonstrate the effectiveness and efficiency of MSBPR over ten state-of-the-art methods on six benchmark real-world datasets.

Federated one-class collaborative filtering via privacy-aware non-sampling matrix factorization

In this paper, we study an emerging and important recommendation problem called federated one-class collaborative filtering (FOCCF). Specifically, we aim to build a recommendation model by exploiting each user’s one-class or implicit feedback in a distributed and privacy-aware manner rather than collecting and learning from data in a central server. For the studied problem, there are three important issues, i.e., recommendation accuracy, privacy and efficiency. As a response, we start from the state-of-the-art one-class collaborative filtering (OCCF) method, i.e., non-sampling matrix factorization (NSMF), and propose a novel federated recommendation framework called privacy-aware NSMF (P-NSMF). Our P-NSMF protects user privacy well without sacrificing recommendation accuracy and contains two variants, i.e., P-NSMF(ALS) and P-NSMF(BGD), which are based on alternating least squares (ALS) and batch gradient descent (BGD), respectively. Moreover, we design an improved strategy called group-wise concealing and adopt a secure aggregation technique in our framework for privacy and efficiency. We then analyze the security and complexity of our P-NSMF, and conduct extensive experiments on four public datasets. In particular, compared with the existing methods, our P-NSMF outperforms them in terms of recommendation accuracy, privacy and efficiency, which are important merits for real deployment.

Ranking-Based Implicit Regularization for One-Class Collaborative Filtering

One-class collaborative filtering (OCCF) problems are ubiquitous in real-world recommendation systems, such as news recommendation, but suffer from data sparsity and lack of negative items. To address the challenge, the state-of-the-art algorithm assigns uninteracted items with smaller weights of being negative and performs low-rank approximation over the user-item interaction matrix. However, the prior ratings are usually suggested to be zero but may not be well-defined. To avert the direct utilization of prior ratings for uninteracted items, we propose a novel ranking-based implicit regularizer by hypothesizing that users’ preference scores for uninteracted items should not deviate a lot from each other. The regularizer is then used in a ranking-based OCCF framework to penalize large differences of preference scores between uninteracted items. To efficiently optimize model parameters in this framework, we develop the scalable alternating least square algorithm and coordinate descent algorithm, whose time complexity is linearly proportional to the data size. Finally, we extensively evaluate the proposed algorithms on six public real-world datasets. The results show that the proposed regularizer significantly improves the recommendation quality of ranking-based OCCF algorithms, such as BPRMF and RankALS. Moreover, the ranking-based framework with the proposed regularizer outperforms the state-of-the-art recommendation algorithms for implicit feedback.

Incremental one-class collaborative filtering with co-evolving side networks

One-class collaborative filtering (OCCF) is a fundamental research problem in a myriad of applications where the preferences of users can only be implicitly inferred from their one-class feedback (e.g., click an ad or purchase a product). The main challenges of OCCF lie in the sparsity of user feedback and the ambiguity of unobserved preferences. To effectively address the above two challenges, side networks from users and items are extensively exploited by state-of-the-art methods, which are predominantly focused on static settings. However, as real-world recommender systems evolve over time (where both the user–item ratings and user–user/item–item side networks will change), it is necessary to update OCCF results (e.g., the latent features of users and items) accordingly. The main obstacle for OCCF online update with co-evolving side networks lies in the fact that the coupled system is highly sensitive to local changes, which may cause massive perturbation on the latent features of a large number of users and items. In this paper, we propose a novel incremental one-class collaborative filtering (OCCF) method that can cope with co-evolving side networks efficiently. In particular, we model the evolution of latent features as a linear transformation process, which enables fast update of the latent features on the fly. Empirical experiments demonstrate that our method can provide high-quality recommendation results on real-world datasets.

M-BPR: A novel approach to improving BPR for recommendation with multi-type pair-wise preferences

In this paper, we examine the two assumptions of the Bayesian personalized ranking (BPR), a well-known pair-wise method for one-class collaborative filtering (OCCF): (1) a user with the same degree of negative preferences for all her unrated items; and (2) a user always preferring her rated items to all her unrated items. We claim that (A1) and (A2) cause recommendation errors because they do not always hold in practice. To address these problems, we first define fine-grained multi-type pair-wise preferences (PPs), which are more sophisticated than the single-type PP used in BPR. Then, we propose a novel pair-wise approach called M-BPR, which exploits multi-type PPs together in learning users’ more detailed preferences. Furthermore, we refine M-BPR by employing the concept of item groups to reduce the uncertainty of a user’s a single item-level preference. Through extensive experiments using four real-life datasets, we demonstrate that our approach addresses the problems of the original BPR effectively and also outperforms seven state-of-the-art OCCF (i.e., four pair-wise and three point-wise) methods significantly.

CoFi-points

With the explosive growth of web resources, an increasingly important task in recommender systems is to provide high-quality personalized services by learning users’ preferences from historically observed information. As an effective preference learning technology, collaborative filtering has been widely extended to model the one-class or implicit feedback data, which is known as one-class collaborative filtering (OCCF). For a long time, pairwise ranking-oriented learning scheme has been viewed as a superior solution than the pointwise scheme for OCCF due to its higher accuracy in most cases. However, we argue that with appropriate model design, pointwise preference learning can achieve comparable or even better performance than the counterpart, i.e., pairwise preference learning. In particular, we propose a new preference assumption, i.e., pointwise preference on user/item-set. Based on this new assumption, we develop a novel, simple, and flexible solution called collaborative filtering via pointwise preference learning on user/item-set (CoFi-points). Furthermore, we derive two specific algorithms of CoFi-points with respect to the involved user-set and item-set, i.e., CoFi-points(u) and CoFi-points(i), referring to preference assumptions defined on user-set and item-set, respectively. Finally, we conduct extensive empirical studies on four real-world datasets with the state-of-the-art methods, and find that our solution can achieve very promising performance with respect to several ranking-oriented evaluation metrics.

K-Reciprocal nearest neighbors algorithm for one-class collaborative filtering

In this paper, we study an important recommendation problem of exploiting users’ one-class positive feedback such as “likes”, which is usually termed as one-class collaborative filtering (OCCF). For modeling users’ preferences beneath the observed one-class positive feedback, the similarity between two users is a key concept of constructing a neighborhood structure with like-minded users. With a well-constructed neighborhood, we can make a prediction of a user’s preference to an un-interacted item by aggregating his or her neighboring users’ tastes. However, the neighborhood constructed by a typical traditional method is usually asymmetric, meaning that a certain user may belong to the neighborhood of another user but the inverse is not necessarily true. Such an asymmetric structure may result in a less strong neighborhood, which is our major finding in this paper. As a response, we exploit the reciprocal neighborhood among users in order to construct a better neighborhood structure with more high-value users, which is expected to be less influenced by the active users. We then design a corresponding recommendation algorithm called k-reciprocal nearest neighbors algorithm (k-RNN). Extensive empirical studies on two large and public datasets show that our k-RNN performs significantly better than a closely related algorithm with the traditional asymmetric neighborhood and some competitive model-based recommendation methods.

Bayesian pairwise learning to rank via one-class collaborative filtering

With the ever-growing scale of social websites and online transactions, in past decade, Recommender System (RS) has become a crucial tool to overcome information overload, due to its powerful capability in information filtering and retrieval. Traditional rating prediction based RS could learn user’s preference according to the explicit feedback, however, such numerical user-item ratings are always unavailable in real life. By contrast, pairwise learning algorithms could directly optimize for ranking and provide personalized recommendation from implicit feedback, although suffering from such data sparsity and slow convergence problems. Motivated by these, in this article, a novel collaborative pairwise learning to rank method referred to as BPLR is proposed, which aims to improve the performance of personalized ranking from implicit feedback. To this end, BPLR tries to partition items into positive feedback, potential feedback and negative feedback, and takes account of the neighborhood relationship between users as well as the item similarity while deriving the potential candidates, moreover, a dynamic sampling strategy is designed to reduce the computational complexity and speed up model training. Empirical experiments over four real world datasets certificate the effectiveness and efficiency of BPLR, which could speed up convergence, and outperform state-of-the-art algorithms significantly in personalized top-N recommendation.

Improving One-Class Collaborative Filtering via Ranking-Based Implicit Regularizer

One-class collaborative filtering (OCCF) problems are vital in many applications of recommender systems, such as news and music recommendation, but suffers from sparsity issues and lacks negative examples. To address this problem, the state-of-the-arts assigned smaller weights to unobserved samples and performed low-rank approximation. However, the ground-truth ratings of unobserved samples are usually set to zero but ill-defined. In this paper, we propose a ranking-based implicit regularizer and provide a new general framework for OCCF, to avert the ground-truth ratings of unobserved samples. We then exploit it to regularize a ranking-based loss function and design efficient optimization algorithms to learn model parameters. Finally, we evaluate them on three realworld datasets. The results show that the proposed regularizer significantly improves ranking-based algorithms and that the proposed framework outperforms the state-of-the-art OCCF algorithms.

Neighborhood-enhanced transfer learning for one-class collaborative filtering

Recommender systems have become more prevalent in recent years for providing users with personalized services such as movie recommendation and news recommendation. In real-world scenarios, they are naturally thought of as one-class collaborative filtering (OCCF) problems because most behavioral data are users’ interaction records, e.g., browses or clicks, which are referred to as one-class feedback or implicit feedback. In these problems, the sparsity of observed feedback and the ambiguity of unobserved feedback make it difficult to capture users’ true preferences. In order to alleviate that, two well-known approaches have been proposed, including factorization-based methods aiming to learn the relationships between users and items via latent factors, and neighborhood-based methods focusing on similarities between users or items. However, these two types of approaches are rarely studied in one single framework or solution for OCCF. In this paper, we propose a novel transfer learning solution, i.e., transfer by neighborhood-enhanced factorization (TNF), which combines these two approaches in a unified framework. Specifically, we extract the local knowledge of the neighborhood information among users, and then transfer it to a global preference learning task in an enhanced factorization-based framework. Our TNF is expected to exploit the local knowledge in a global learning manner well. Extensive empirical studies on five real-world datasets show that our proposed solution can perform significantly more accurate than the state-of-the-art methods.

SDBPR: Social distance-aware Bayesian personalized ranking for recommendation

Recommendation systems recommend new items to users. Because training data contain only binary forms of implicit feedback in many cases, such as in IoT and IoV, one-class collaborative filtering, which can be solved by using rating-based methods to estimate the numeric scores of items or ranking-based methods based on the preferences of each user for items, must be addressed. In addition, because of the sparsity of such data, ranking-based methods are often preferred over rating-based methods when only implicit feedback is available. Social information has recently been used to improve the accuracy of rankings. Traditional approaches simply consider the direct friends of users in a social network, but this process fails to consider the propagation of influence along connections in the social network and cannot reveal the complex graph structure of the social network. In this paper, a novel social distance-aware Bayesian personalized ranking model, called SDBPR, is proposed to generate more accurate recommendations. SDBPR uses a random walk to travel the social network and then makes pairwise assumptions about the ranking order based on the distance between users along the random walk. The experimental results on two real datasets show that the proposed approaches significantly outperform the baseline approaches in terms of ranking prediction.

Exploit latent Dirichlet allocation for collaborative filtering

Previous work on the one-class collaborative filtering (OCCF) problem can be roughly categorized into pointwise methods, pairwise methods, and content-based methods. A fundamental assumption of these approaches is that all missing values in the user-item rating matrix are considered negative. However, this assumption may not hold because the missing values may contain negative and positive examples. For example, a user who fails to give positive feedback about an item may not necessarily dislike it; he may simply be unfamiliar with it. Meanwhile, content-based methods, e.g. collaborative topic regression (CTR), usually require textual content information of the items, and thus their applicability is largely limited when the text information is not available. In this paper, we propose to apply the latent Dirichlet allocation (LDA) model on OCCF to address the above-mentioned problems. The basic idea of this approach is that items are regarded as words, users are considered as documents, and the user-item feedback matrix constitutes the corpus. Our model drops the strong assumption that missing values are all negative and only utilizes the observed data to predict a user’s interest. Additionally, the proposed model does not need content information of the items. Experimental results indicate that the proposed method outperforms previous methods on various ranking-oriented evaluation metrics. We further combine this method with a matrix factorization-based method to tackle the multi-class collaborative filtering (MCCF) problem, which also achieves better performance on predicting user ratings.

GOCCF: Graph-Theoretic One-Class Collaborative Filtering Based on Uninteresting Items

We investigate how to address the shortcomings of the popular One-Class Collaborative Filtering (OCCF) methods in handling challenging “sparse” dataset in one-class setting (e.g., clicked or bookmarked), and propose a novel graph-theoretic OCCF approach, named as gOCCF, by exploiting both positive preferences (derived from rated items) as well as negative preferences (derived from unrated items). In capturing both positive and negative preferences as a bipartite graph, further, we apply the graph shattering theory to determine the right amount of negative preferences to use. Then, we develop a suite of novel graph-based OCCF methods based on the random walk with restart and belief propagation methods. Through extensive experiments using 3 real-life datasets, we show that our gOCCF effectively addresses the sparsity challenge and significantly outperforms all of 8 competing methods in accuracy on very sparse datasets while providing comparable accuracy to the best performing OCCF methods on less sparse datasets. The datasets and implementations used in the empirical validation are available for access: https://goo.gl/sfiawn.

ANTENNA, a Multi-Rank, Multi-Layered Recommender System for Inferring Reliable Drug-Gene-Disease Associations: Repurposing Diazoxide as a Targeted Anti-Cancer Therapy.

Existing drug discovery processes follow a reductionist model of "one-drug-one-gene-one-disease," which is inadequate to tackle complex diseases involving multiple malfunctioned genes. The availability of big omics data offers opportunities to transform drug discovery process into a new paradigm of systems pharmacology that focuses on designing drugs to target molecular interaction networks instead of a single gene. Here, we develop a reliable multi-rank, multi-layered recommender system, ANTENNA, to mine large-scale chemical genomics and disease association data for prediction of novel drug-gene-disease associations. ANTENNA integrates a novel tri-factorization based dual-regularized weighted and imputed One Class Collaborative Filtering (OCCF) algorithm, tREMAP, with a statistical framework based on Random Walk with Restart and assess the reliability of specific predictions. In the benchmark, tREMAP clearly outperforms the single-rank OCCF. We apply ANTENNA to a real-world problem: repurposing old drugs for new clinical indications without effective treatments. We discover that FDA-approved drug diazoxide can inhibit multiple kinase genes responsible for many diseases including cancer and kill triple negative breast cancer (TNBC) cells efficiently [Formula: see text]. TNBC is a deadly disease without effective targeted therapies. Our finding demonstrates the power of big data analytics in drug discovery and developing a targeted therapy for TNBC.

Incorporating temporal dynamics into LDA for one-class collaborative filtering

In the one-class collaborative filtering (OCCF) scenario, the elements of the user-item rating matrix consist each take one of only two values: either “like” or unknown. Previous methods for solving the OCCF problem can be roughly categorized into content-based methods, pointwise methods, and pairwise methods. A fundamental assumption of these approaches is that all missing values in a rating matrix can be treated as “dislike”. However, this assumption may not hold because the missing values are not always negative. Sometimes users do not give positive feedback on an item simply because they are not familiar with it rather than because they dislike it. In addition, content-based methods usually require textual information on the items. In many cases, however, sufficient textual information is not available; therefore, content-based methods are not applicable. Moreover, a user’s preference for items usually drifts over time, but the previous methods cannot capture the temporal dynamics of this drift. In this paper, we propose to modify the latent Dirichlet allocation (LDA) model to address the above-mentioned problems. Our method uses only observed rating data to predict users’ interests and effectively avoids the issue of data skew. Furthermore, to address the issue that users’ preferences for items usually drift over time, we assign a different weight to each rating according to its timestamp when using Gibbs sampling to estimate the parameters of the LDA model. In this way, the temporal dynamics of the user preferences can be captured. We report experiments conducted to evaluate our model. The results show that the proposed model outperforms state-of-the-art approaches for the OCCF problem.

Boolean kernels for collaborative filtering in top-N item recommendation

In many personalized recommendation problems available data consists only of positive interactions (implicit feedback) between users and items. This problem is also known as One-Class Collaborative Filtering (OC-CF). Linear models usually achieve state-of-the-art performances on OC-CF problems and many efforts have been devoted to build more expressive and complex representations able to improve the recommendations. Recent analysis show that collaborative filtering (CF) datasets have peculiar characteristics such as high sparsity and a long tailed distribution of the ratings. In this paper we propose a boolean kernel, called Disjunctive kernel, which is less expressive than the linear one but it is able to alleviate the sparsity issue in CF contexts. The embedding of this kernel is composed by all the combinations of a certain arity d of the input variables, and these combined features are semantically interpreted as disjunctions of the input variables. Experiments on several CF datasets show the effectiveness and the efficiency of the proposed kernel.

Social and content aware One-Class recommendation of papers in scientific social networks.

With the rapid development of information technology, scientific social networks (SSNs) have become the fastest and most convenient way for researchers to communicate with each other. Many published papers are shared via SSNs every day, resulting in the problem of information overload. How to appropriately recommend personalized and highly valuable papers for researchers is becoming more urgent. However, when recommending papers in SSNs, only a small amount of positive instances are available, leaving a vast amount of unlabelled data, in which negative instances and potential unseen positive instances are mixed together, which naturally belongs to One-Class Collaborative Filtering (OCCF) problem. Therefore, considering the extreme data imbalance and data sparsity of this OCCF problem, a hybrid approach of Social and Content aware One-class Recommendation of Papers in SSNs, termed SCORP, is proposed in this study. Unlike previous approaches recommended to address the OCCF problem, social information, which has been proved playing a significant role in performing recommendations in many domains, is applied in both the profiling of content-based filtering and the collaborative filtering to achieve superior recommendations. To verify the effectiveness of the proposed SCORP approach, a real-life dataset from CiteULike was employed. The experimental results demonstrate that the proposed approach is superior to all of the compared approaches, thus providing a more effective method for recommending papers in SSNs.

One-class collaborative filtering based on rating prediction and ranking prediction

One-Class Collaborative Filtering (OCCF) has recently received much attention in recommendation communities due to their close relationship with real industry problem settings. However, the problem with previous research studies on OCCF is that they focused on either rating prediction or ranking prediction, but no concerted research effort has been devoted to developing a recommendation approach that simultaneously optimizes both the ratings and rank of the recommended items. In order to overcome the defects of prior research, a new better unified OCCF approach (UOCCF) based on the newest Collaborative Less-is-More Filtering (CLiMF) approach and the Probabilistic Matrix Factorization (PMF) approach was proposed, which benefits from the ranking-oriented perspective and the rating-oriented perspective by sharing common latent features of users and items in CLiMF and PMF. We also provide an efficient learning algorithm to solve the optimization problem for UOCCF. Experimental results on practical datasets showed that our proposed UOCCF approach outperformed existing OCCF approaches (both ranking-oriented and rating-oriented) over different evaluation metrics, and that the UOCCF approach enjoys the advantage of low complexity and is shown to be linear with the number of observed ratings in a given user–item rating matrix. Because of its high precision and good expansibility, UOCCF is suitable for processing big data, and has wide application prospects in the field of internet information recommendation.

Group Bayesian personalized ranking with rich interactions for one-class collaborative filtering

Both researchers and practitioners in the field of collaborative filtering have shown keen interest to user behaviors of the “one-class” feedback form such as transactions in e-commerce and “likes” in social networks. This recommendation problem is termed as one-class collaborative filtering (OCCF). In most of the previous work, a pairwise preference assumption called Bayesian personalized ranking (BPR) was empirically proved to be able to exploit such one-class data well. In one of the most recent work, an upgraded model called group preference based BPR (GBPR) leverages the group preference and obtains better performance.In this paper, we go one step beyond GBPR, and propose a new and generic assumption, i.e., group Bayesian personalized ranking with rich interactions (GBPR+). In our GBPR+, we adopt a set of items instead of one single item as used in GBPR, which is expected to introduce rich interactions. GBPR is a special case of our GPBR+ when the item set contains only one single item. We study the empirical performance of our GBPR+ with several state-of-the-art methods on four real-world datasets, and find that our GPBR+ can generate more accurate recommendations.

Pairwise probabilistic matrix factorization for implicit feedback collaborative filtering

Implicit feedback collaborative filtering has attracted a lot of attention in collaborative filtering, which is called one-class collaborative filtering (OCCF). However, the low recommendation accuracy and the high cost of previous methods impede its generalization in real scenarios. In this paper, we develop a new model named pairwise probabilistic matrix factorization (PPMF) by using the advantages of RankRLS. PPMF model takes RankRLS integrated with PMF (probabilistic matrix factorization) to learn the relative preference for items. Different from previous works, PPMF minimizes the average number of inversions in ranking rather than maximize the gaps of the binary predicted values for OCCF problem. Meanwhile, we propose to optimize the PPMF model by the pointwise stochastic gradient descent algorithm based on bootstrap sampling, which is more effective for parameter learning than the original optimization method used in previous works. Experiments on two datasets show that PPMF model achieves satisfactory performance and outperforms the state-of-the-art implicit feedback collaborative ranking models by using different evaluation metrics.