Data-driven Analysis of Domain Specificity for Explainable Session-based Recommendation System

抄録

When utilizing sophisticated AI platforms for managerial decisionmaking based on observed user logs, the explainability of predictive models becomes key to trust. This explainability is significantly enhanced by understanding the domain-specificity within the predictive space. Therefore, attempts to capture domain-specificity from the internal states of large-scale language models applied in real business contexts can aid in model improvement and in formulating problems based on contextual information. In this paper, we propose an algorithmic framework that uses methods of inductive logic programming to extract logic rules directly from the sequence data logs and learned internal states of various session-based recommendation systems.