Knowledge Augmented Expert finding framework via knowledge graph embedding for Community Question Answering

作     者：Liu, Yue Liu, Zitu Yu, Zhenyao Fu, Qingshan Tang, Weize Yao, Wenxuan Sun, Zhibin 

作者机构：Shanghai Univ Sch Comp Engn & Sci Shanghai 200444 Peoples R China Shanghai Engn Res Ctr Intelligent Comp Syst Shanghai 200444 Peoples R China Nanjing Normal Univ Sch Math Sci Jiangsu Key Lab Numer Simulat Large Scale Complex Nanjing 210023 Peoples R China 

出 版 物：《ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE》 (Eng Appl Artif Intell)

年 卷 期：2025年第142卷

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China [92270124  52073169  61936001] 

主　　题：Expert finding Community question answering Knowledge graph Knowledge graph embedding 

摘      要：Expert Finding in Community Question Answering aims to recommend appropriate experts to answer posted questions. However, existing approaches focus on exploiting semantic extraction and authority analysis techniques, failing to realize the latent knowledge-aspect connections between questions and experts. Therefore, the experts recommended for posted questions are limited to text-to-text approximation or domain-independent authority. In this study, we propose a Knowledge Augmented Expert finding framework (KAExpert) that introduces knowledge level information into semantic-based and authority-based expert finding method. A community knowledge graph is firstly constructed by acquiring ternary relations from public databases based on question tags and knowledge-related phrases. Then a flexible knowledge graph embedding is designed to extract the matching relationship between questions and experts at the knowledge level. Along this line, the knowledge- level authority is calculated based on the knowledge graph embedding to optimize the results of domain matching. According to knowledge graph embedding and knowledge-level authority, KAExpert is constructed to optimize the results of domain matching so that the found experts set consists of high-level expert with semantic matching and knowledge matching. Finally, experimental results on two real-world datasets collected from two major commercial question answering web sites show that KAExpert can outperform baseline methods with a significant margin.