Personalised novel and explainable matrix factorisation

Abstract

Recommendation systems personalise suggestions to individuals to help them in their decision making and exploration tasks. In the ideal case, these recommendations, besides of being accurate, should also be novel and explainable. However, up to now most platforms fail to provide both, novel recommendations that advance users' exploration along with explanations to make their reasoning more transparent to them. For instance, a well-known recommendation algorithm, such as matrix factorisation (MF), optimises only the accuracy criterion, while disregarding other quality criteria such as the explainability or the novelty, of recommended items. In this paper, to the best of our knowledge, we propose a new model, denoted as NEMF, that allows to trade-off the MF performance with respect to the criteria of novelty and explainability, while only minimally compromising on accuracy. In addition, we recommend a new explainability metric based on nDCG, which distinguishes a more explainable item from a less explainable item. An initial user study indicates how users perceive the different attributes of these "user" style explanations and our extensive experimental results demonstrate that we attain high accuracy by recommending also novel and explainable items.

A framework for unbiased explainable pairwise ranking for recommendation

“…• Explainable Bayesian Personalized Ranking (EBPR): This is our proposed explainable BPR loss function [6]. This loss function is based on BPR and relies on neighborhood-based explainability [5,10,11] to rank relevant and explainable recommendations at the top of the recommendation list for a user. The explanations in this case are in the form ''This item was recommended because you also liked these similar items.…”

Parameter Tuning of Reranking-based Diversification Algorithms using Total Curvature Analysis

“…The authors in [10] considered the diversified recommendation as a structured supervised learning problem and proposed a structured Support Vector Machine-based objective function. The work in [11] proposed an explainable matrix factorization by incorporating a novelty term. Such approaches always involve solving complex optimization problems, often non-convex and, thus, resulting in extensively high computational cost while lacking of scalability.…”

Debiased Explainable Pairwise Ranking from Implicit Feedback

“…It also has the advantage of not being a post-hoc approach, and hence not incurring the cost of learning two separate models, nor risking lack of fidelity from deviations between the explaining model and the predictive model. For all these reasons, EMF was later adopted in several works, such as [12] which extended it and tried to improve the novelty of the recommendations; and in [45] which modified the calculation of the explainability matrix by integrating the neighbors' weights to improve performance. Other works used influence functions to generate neighborhood-based explanations.…”

“…Item-based explanations use itemsimilarities and generate explanations in the form "the item was recommended because you liked similar items". We extend the idea of neighborhood-based explainability from [2] because it has shown success as an intuitive method for modifying loss-based recommendation models [12,45]. Both item-based and user-based measures of explainability can be defined by relying solely on the interaction matrix (or rating matrix, depending on the type of feedback).…”