Proposal of Extending Logical Formulae for Recommendation Based on Logic-Integrated Neural Network

Abstract

This paper proposes several logical formulae for recommendation based on Logic-Integrated Neural Network (LINN). Deep neural networks have attracted interest in many domains due to its superior performance. However, it does not explicitly handle logical reasoning. To integrate logical expression into neural network-based reasoning, LINN has been proposed and applied to a recommendation. While the existing study represents the user's interaction history as a logical formula, its expression power was limited. This paper extends such a simple formula by introducing the item's genre information and complicated interaction between items. The experimental results show the effectiveness of the proposed logical formulae.