Rating-based Recommender Systems Research Articles

Fairness metrics and bias mitigation strategies for rating predictions

Algorithm fairness is an established line of research in the machine learning domain with substantial work while the equivalent in the recommender system domain is relatively new. In this article, we consider rating-based recommender systems which model the recommendation process as a prediction problem. We consider different types of biases that can occur in this setting, discuss various fairness definitions, and also propose a novel bias mitigation strategy to address potential unfairness in a rating-based recommender system. Based on an analysis of fairness metrics used in machine learning and a discussion of their applicability in the recommender system domain, we map the proposed metrics from the two domains and identify commonly used concepts and definitions of fairness. Finally, to address unfairness and potential bias against certain groups in a recommender system, we develop a bias mitigation algorithm and conduct case studies on one synthetic and one empirical dataset to show its effectiveness. Our results show that unfairness can be significantly lowered through our approach and that bias mitigation is a fruitful area of research for recommender systems.

Hypothetical tensor-based multi-criteria recommender system for new users with partial preferences

Multi-Criteria Recommender Systems (MCRSs) have been developed to improve the accuracy of single-criterion rating-based recommender systems that could not express and reflect users? fine-grained rating behaviors. In most MCRSs, new users are asked to express their preferences on multi-criteria of items, to address the cold-start problem. However, some of the users? preferences collected are usually not complete due to users? cognitive limitation and/or unfamiliarity on item domains, which is called ?partial preferences?. The fundamental challenge and then negatively affects to accurately recommend items according to users? preferences through MCRSs. In this paper, we propose a Hypothetical Tensor Model (HTM) to leverage auxiliary data complemented through three intuitive rules dealing with user?s unfamiliarity. First, we find four patterns of partial preferences that are caused by users? unfamiliarity. And then the rules are defined by considering relationships between multi-criteria. Lastly, complemented preferences are modeled by a tensor to maintain an inherent structure of and correlations between the multi-criteria. Experiments on a TripAdvisor dataset showed that HTM improves MSE performances from 40 to 47% by comparing with other baseline methods. In particular, effectivenesses of each rule regarding multi-criteria on HTM are clearly revealed.

Leveraging multi-criteria customer feedback for satisfaction analysis and improved recommendations

Travel websites and online booking platforms represent today’s major sources for customers when gathering information before a trip. In particular, community-provided customer reviews and ratings of various tourism services represent a valuable source of information for trip planning. With respect to customer ratings, many modern travel and tourism platforms—in contrast to several other e-commerce domains—allow customers to rate objects along multiple dimensions and thus to provide more fine-granular post-trip feedback on the booked accommodation or travel package. In this paper, we first show how this multi-criteria rating information can help to obtain a better understanding of factors driving customer satisfaction for different segments. For this purpose, we performed a Penalty-Reward contrast analysis on a data set from a major tourism platform, which reveals that customer segments significantly differ in the way the formation of overall satisfaction can be explained. Beyond the pure identification of segment-specific satisfaction factors, we furthermore show how this fine-granular rating information can be exploited to improve the accuracy of rating-based recommender systems. In particular, we propose to utilize user- and object-specific factor relevance weights which can be learned through linear regression. An empirical evaluation on datasets from different domains finally shows that our method helps us to predict the customer preferences more accurately and thus to develop better online recommendation services.