Optimizing model parameter for entity summarization across knowledge graphs

Abstract

Knowledge graphs, which belongs to the category of semantic networks, are considered as a new method of knowledge representation of health care data. It establishes a semantic explanation model for human perception and health care information processing. Each knowledge graph is composed of massive entities and relationships. However, it is an arduous work to search and visualize users’ interested entities and attributes since there are many attributes for an entity across different knowledge graphs. It is a natural problem how to summarize an entity based on multiple knowledge graphs. We propose a three-stage algorithm to solve the problem of entity summarization across knowledge graphs, including candidate generation, knowledge graph linkage, and entity summarization. We propose an unsupervised framework to link different knowledge graphs based on the semantic and structure information of entities. To further reduce the computational cost, we employ word embedding technique to find the similar entities in semantic, and filter some pairs of unmatched entities. Finally, we model entity summarization as personalized ranking problem in a knowledge graph. We conduct a set of experiments to evaluate our proposed method on four real datasets: historical data for user query, two English knowledge graphs (YAGO and DBpeida) and an English corpus. Experimental results demonstrate the effectiveness of our proposed method by comparing with baselines.