基于ConceptNet语义的伪相关反馈信息检索方法  

Pseudo-relevance feedback retrieval based on ConceptNet semantics

作　　者：潘敏 刘宇 裴全力 李腾 PAN Min;LIU Yu;PEI Quan-li;LI Teng(College of Computer and Information Engineering,Hubei Normal University,Huangshi 435000,China;School of Information Technology,York University,Ontario Toronto M2J4A6,Canada)

机构地区：[1]湖北师范大学计算机与信息工程学院,湖北黄石435000 [2]约克大学信息技术学院,安大略多伦多M2J4A6

出　　处：《湖北师范大学学报（自然科学版）》2023年第2期28-37,共10页Journal of Hubei Normal University：Natural Science

基　　金：2022年湖北省教育厅科研计划项目重点项目(D20222501);国家留学基金委博士后基金项目(202108420144);2023年研究生创新科研项目(20238099)。

摘　　要：伪相关性反馈技术在信息检索领域应用广泛,在考虑词频和逆文档频率等重要特征时,传统的信息检索方法容易忽略查询词本身的语义信息。提出了一种基于语义的伪相关性反馈信息检索方法SPRF(Semantic Pseudo-Relevance Feedback),充分利用ConceptNet获取语义信息,不仅考虑了查询词在文档中的词频重要性,还将查询词的语义信息整合到伪相关反馈框架中,以改善查询扩展词的选择。在6个TREC数据集上实验结果表明:SPRF方法对比较强基线模型和几种基于神经网络的方法在P@10和MAP两个指标上具有显著提升。Pseudo-relevance feedback is widely used in information retrieval.Traditional information retrieval methods have neglected the query terms semantic information when retrieving important features such as word frequency and inverse document frequency.This paper proposed a semantic-based pseudo-relevance feedback information retrieval method SPRF(Semantic Pseudo-Relevance Feedback)and took full advantage of ConceptNet to capture semantic information,which took importance of the query into consideration and integrated the semantic information into the pseudo-relevance feedback framework to enrich the query selection of expansion terms.Experimental results on six TREC datasets indicates that compare with stronger baseline models and several neural network-based approaches,the SPRF has significant improvement in terms of precision at 10(P@10)and mean average precision(MAP).

关 键 词：ConceptNet 伪相关反馈 查询扩展 信息检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]