ILP Recommender System: Explainable and More

In this paper, we explore the use of ILP thoroughly in generating explainable, negative, group and context-aware recommendation. ILP provides recommendation rules in if-then logical format that allows us to form a clear and concise explanation to accompany the suggested items. It can indirectly derive negative rules which tell us not to recommend certain products to users. It also emphasizes the use of universal representations which enables us to suggest the items to a group of users who share the same interest. Additionally, ILP requires no re-training if new contexts (e.g., location, time and mood) are added to the system to generate context-aware recommendations (CARS), only predicates and settings are simply specified. In this paper, we also propose the explainability evaluation in terms of transparency by comparing the items/features appearing in the explanation with the features presented in the user's review. The negative, group and dynamic recommendations can be evaluated using the standard measurement.