A Study on Item Recommendation Model by Evaluating the Effect of Individual Intervention

Abstract

Collaborative filtering is one of the item recommendation models which recommends items with high estimated purchase probability to each user. However, among the items with high estimated purchase probability, there are some items that are purchased regularly and thus have little need of recommendation. Therefore, one of the challenges for recommender systems is to identify items with high recommendation effect. CounterFactual Regression (CFR) is known as a model to estimate the effect of individual intervention such as recommendation effect. However, since this model can only handle one type of intervention, it is difficult to apply it to a recommendation system where there are many recommended items as interventions. In this study, we extend the model to estimate recommendation effect with a single CFR by combining user and item features to form covariates of user-item pairs. Finally, we demonstrate the effectiveness of the proposed method through experiments using artificial data.