A User-Centric Evaluation to Generate Case-Based Explanations Using Formal Concept Analysis

摘要

Recommender systems are useful to find relevant products for a certain user. Some recommender techniques based on models, for example, Matrix Factorization, act as a black box for users. Explanations for recommender systems are useful to make recommendations more effective and help the users to trust the system and understand why certain items have been recommended. In this paper, we propose a post-hoc model-agnostic explanation system for MF recommendations based on Case-Based Reasoning and Formal Concept Analysis. We have conducted an experimental evaluation with real users to define what are the most useful explanation features that allow users a better understanding of the system recommendation.