Representing the Interaction between Users and Products via LLM-assisted Knowledge Graph Construction

To understand user behavior, representing the semantic knowledge of user-product interaction is essential. In this paper, we represent the interaction between user and product via large language model (LLM)-assisted knowledge graph construction. We capture users' behavioral actions and static properties of the products from raw text data of "user review" and "product catalog". Moreover, the information needed for updating the knowledge graph is captured by raw texts of "news related to the products". The proposed methodology integrates them as a single knowledge graph to provide causal reasoning on user-product interaction. To alleviate the situation where a small quantity of annotated text exists in these data, we use LLM as a data annotator and augmentor.