Providing reliability in recommender systems through Bernoulli Matrix Factorization

Ortega, Fernando; Lara-Cabrera, Raúl; González-Prieto, Ángel; Bobadilla, Jesús

doi:10.1016/j.ins.2020.12.001

Cited by 31 publications

(15 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In this section, we analyze whether this imbalance causes bias and unfair recommendations in the MovieLens dataset. In this section, we have evaluated the quality predictions performed by the most popular CF based RS: KNN based CF has been evaluated using correlation [2], cosine [2], JMSD [29], PIP [30], and singularities [31] similarity metrics; MF based CF has been tested using Probabistic Matrix Factorization (PMF) [14], Biased Matrix Factorization (BiasedMF) [15], Bernoulli Matrix Factorization (BeMF) [32], Binomial Non-negative Matrix Factorization (BNMF) [33], and Non-negative Matrix Factorization (NMF) [34]; and NN based CF has been verified using Neural Collaborative Filtering (NCF) [19]. These recommendation models have been chosen as a heterogeneous sample of the different types of CF that exist.…”

Section: Movielens Bias and Unfairnessmentioning

confidence: 99%

Bias and Unfairness of Collaborative Filtering Based Recommender Systems in MovieLens Dataset

González¹,

Ortega

Perez-Lopez³

et al. 2022

IEEE Access

Self Cite

View full text Add to dashboard Cite

Recommender Systems have become one of the most important tools for streaming and marketplace systems in recent years. Their increased use has revealed clear bias and unfairness against minorities and underrepresented groups. This paper seeks the origin of these biases and unfairness. To this end, it analyzes the demographic characteristics of a gold standard dataset and its prediction performance when used in a multitude of Recommender Systems. In addition, this paper proposes Soft Matrix Factorization (SoftMF), which tries to balance the predictions of different types of users to reduce the present inequality. The experimental results show that those biases and unfairness are not introduced by the different recommendation models and that they come from the socio-psychological and demographic characteristics of the used dataset. INDEX TERMSRecommender systems, collaborative filtering, fairness, MovieLens. SANTIAGO ALONSO received the B.S. degree in software engineering and the Ph.D. degree in computer science and artificial intelligence from the Universidad Politécnica de Madrid, in 2015. He is currently an Associate Professor with the Universidad Politécnica de Madrid, participating in master and degree subjects and doing work related with advanced databases. His main research interests include natural computing (P-systems) and did some work on genetic algorithms. His current research interests include machine learning, data analysis, and artificial intelligence.

show abstract

Section: Movielens Bias and Unfairnessmentioning

confidence: 99%

Bias and Unfairness of Collaborative Filtering Based Recommender Systems in MovieLens Dataset

González¹,

Ortega

Perez-Lopez³

et al. 2022

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

“…Accordingly, the reliability measure is employed in both the user-similarity function and rating prediction procedure simultaneously. Bernoulli distribution is employed in a matrix factorization methodology to propose an effective model-based collaborative filtering recommendation system named Bernoulli Matrix Factorization (BeMF) [53]. Unlike previous matrix factorization models, BeMF is able to generate both the prediction and reliability scores simultaneously.…”

Section: Reliability-based Recommendation Systemsmentioning

confidence: 99%

Healthy Food Recommendation Using a Time-Aware Community Detection Approach and Reliability Measurement

Ahmadian

Rostami

Jalali

et al. 2022

Int J Comput Intell Syst

View full text Add to dashboard Cite

Food recommendation systems have been increasingly developed in online food services to make recommendations to users according to their previous diets. Although unhealthy diets may cause challenging diseases such as diabetes, cancer, and premature heart diseases, most of the developed food recommendation systems neglect considering health factors in their recommendation process. This emphasizes the importance of the reliability of the recommendation from the health content perspective. This paper proposes a new food recommendation system based on health-aware reliability measurement. In particular, we develop a time-aware community detection approach that groups users into disjoint sets and utilizes the identified communities as the nearest neighbors set in rating prediction. Then, a novel reliability measurement is introduced by considering both the health and accuracy criteria of predictions to evaluate the reliability of predicted ratings. Also, the unreliable predictions are recalculated by removing ineffective users from the nearest neighbors set. Finally, the recalculated predictions are utilized to generate a list of foods as recommendations. Different experiments on a crawled dataset demonstrate that the proposed method enhances the performance around 7.63%, 6.97%, 7.37%, 15.09%, and 16.17% based on precision, recall, F1, normalized discounted cumulative gain (NDCG), and health metrics, respectively, compared to the second-best model.

show abstract

“…Thanks to the mapping function, no constraint is needed on the factors to ensure a valid Bernoulli parameter. Nonetheless, several approaches have additionally enforced the nonnegativity of one [19] or two [20] factors, or leveraged Gaussian priors [21,22]. Despite its popularity and performance, a drawback of logistic PCA stems from the fact that the link function hampers interpretability of the decomposition, which is often a desired feature, e.g., for analyzing econometric data [23].…”

Section: Logistic Pcamentioning

confidence: 99%

A majorization-minimization algorithm for nonnegative binary matrix factorization

Magron,

Févotte

2022

Preprint

View full text Add to dashboard Cite

This paper tackles the problem of decomposing binary data using matrix factorization. We consider the family of mean-parametrized Bernoulli models, a class of generative models that are well suited for modeling binary data and enables interpretability of the factors. We factorize the Bernoulli parameter and consider an additional Beta prior on one of the factors to further improve the model's expressive power. While similar models have been proposed in the literature, they only exploit the Beta prior as a proxy to ensure a valid Bernoulli parameter in a Bayesian setting; in practice it reduces to a uniform or uninformative prior. Besides, estimation in these models has focused on costly Bayesian inference. In this paper, we propose a simple yet very efficient majorization-minimization algorithm for maximum a posteriori estimation. Our approach leverages the Beta prior whose parameters can be tuned to improve performance in matrix completion tasks. Experiments conducted on three public binary datasets show that our approach offers an excellent trade-off between prediction performance, computational complexity, and interpretability.

show abstract

Providing reliability in recommender systems through Bernoulli Matrix Factorization

Cited by 31 publications

References 38 publications

Bias and Unfairness of Collaborative Filtering Based Recommender Systems in MovieLens Dataset

Bias and Unfairness of Collaborative Filtering Based Recommender Systems in MovieLens Dataset

Healthy Food Recommendation Using a Time-Aware Community Detection Approach and Reliability Measurement

A majorization-minimization algorithm for nonnegative binary matrix factorization

Contact Info

Product

Resources

About