基于问答语义匹配的知识社区新问题专家推荐方法  

Question Answering Semantic Matching-Based Expert Recommendation Method for New Questions in Knowledge Community

作　　者：杜军威[1] 邹树林 李浩杰[1] 江峰[1] 于旭[1] 胡强[1] DU Jun-wei;ZOU Shu-lin;LI Hao-jie;JIANG Feng;YU Xu;HU Qiang(School of Information Science and Technology,Qingdao University of Science and Technology,Qingdao,Shandong 266061,China)

机构地区：[1]青岛科技大学信息科学技术学院,山东青岛266061

出　　处：《电子学报》2023年第7期1875-1888,共14页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61973180,No.6217072142);山东省自然科学基金(No.ZR2021MF092,No.ZR2019MF014,No.ZR2019MF033,No.ZR2022MF326)。

摘　　要：传统的知识社区专家推荐方法采用文本相似度匹配机理,并基于问题或专家描述来构建专家特征.这些方法没有利用问题与答案的语义匹配关系,因此难以充分挖掘专家回答问题的能力特征,影响推荐性能.提出一种基于综合历史和当前问答语义匹配的知识社区新问题的专家推荐方法(History-Now Semantics Expert RECommendation model,HNS-EREC).首先,采用反馈评价和负采样技术来处理数据集中的两类不平衡现象;其次,基于问答语义来提取专家回答问题能力特征;最后,提出一种基于问答语义匹配的History-Now联合专家推荐模型,该模型能够实现面向专家的历史问答和当前问答的语义联合学习.实验结果表明,相对于其他方法,本文所提出的HNS-EREC方法在新问题专家推荐方面具有显著的优势.The traditional knowledge community expert recommendation methods adopt the mechanism of text similarity matching,and construct expert features based on questions or expert descriptions.These methods do not take advantage of the semantic matching relationship between questions and answers,hence it is difficult to fully exploit the features of experts'abilities to answer questions,which will affect the recommendation performance.In this paper,we propose a question answering semantic matching-based expert recommendation method,called History-Now Semantics Expert RECommendation model(HNS-EREC),for new questions in knowledge community.First,the feedback evaluation and negative sampling techniques are used to handle two types of imbalances in data sets;second,the features of experts'abilities to answer questions are extracted based on question answering semantics;finally,a history-now joint expert recommendation model based on question answering semantic matching is proposed,which can realize the semantic joint learning of expert-oriented historical question answering and current question answering.Experimental results show that compared with other methods,the HNS-EREC method has obvious advantages in the expert recommendation for new questions.

关 键 词：专家推荐 知识社区 不平衡学习 问答语义 stack overflow 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]