Explainable Recommendation With Elastic Serendipity

Recommendation systems provide good guidance for users to find their favorite items from an overwhelming amount of options.However,most systems excessively pursue the recommendation accuracy and give rise to information cocoons and longtail effects,which triggers the emergence of serendipity.Serendipity recommendation attempts to offer items that interest users but are beyond their disc overy.There are three limitations in the existing methods: the lack of objectivity in existing definitions,the unrelated recommendation problem and the lack of explainability.To address above three issues,we propose an explainable recommendation method with elastic serendipity,trying to trade off the recommendation accuracy and difference as well as improve the explainability.Specifically,in this paper,our main research contents,contributions and innovations are as follows:(1)Proposing more objective definition of recommendation serendipity.Different from existing definitions basing on subjective id eas,we analyze a real-world dataset with user ratings on serendipity and related attributes.According to the results,we propose the definition of serendipity,balance between accuracy and difference.(2)Proposing the concept of elasticity in recommendation and a serendipity recommendation method with elasticity.Users with extensive preferences tend to accept items which are different from their histories,and vice versa.Based on the phenomenon,we propose the concept of elasticity and a serendipity recommendation approach with elasticity.It attempts to flexibly adjust the level of recommendation serendipity for different users to balance the accuracy and difference.(3)Proposing the concept of serendipity vector and a directional recommendation approach.Existing approaches usually measure user-item relevance with a scalar instead of a vector,ignoring user preference direction.To deal with the limitation,we build the serendipity vector with users' long-term preferences and short-term demands,and then generate recommendations and the corresponding explanations.It enhances the guiding role of users' long-term preferences in the recommendation process,as well as improves the recommendation explainability.