多粒度关系融合的微博信念网络检索模型  

A Microblog Belief Network Retrieval Model Integrating Multi-granularity Relationships

作　　者：张雄涛 甘明鑫[1] 李硕 ZHANG Xiongtao;GAN Mingxin;LI Shuo(School of Economics and Management,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]北京科技大学经济管理学院,北京100083

出　　处：《管理科学》2022年第5期67-79,共13页Journal of Management Science

基　　金：国家自然科学基金(72271024,71871019)。

摘　　要：微博检索系统是微博平台实现个性化信息过滤的重要工具,建立合理的微博检索模型不仅有利于满足用户的个性化信息需求,而且有利于提升微博平台的信息服务水平。与传统文本检索相比,微博检索面临两方面挑战,一是较短的微博查询语句难以准确表达用户的检索需求,二是较短的微博文档难以充分表现微博的语义特征。这使查询语句与文档之间难以准确进行匹配计算,致使微博检索性能受限。结合微博术语和文档的特点,将术语关系和文档关系融入信念网络检索模型,提出多粒度关系融合的微博信念网络检索模型。首先,基于混合语义信息和时间信息对微博术语关系进行量化,以更准确地建立微博术语之间的相关性;其次,基于混合语义信息和作者信息对微博文档关系进行量化,以更准确地建立微博文档之间的相关性;最后,结合量化的术语关系和文档关系,在基本信念网络检索模型的基础上,给出微博信念网络检索模型的概率推导过程。采用网络爬虫工具,从新浪微博平台获取真实的微博数据,对微博信念网络检索模型的有效性和合理性进行验证。研究结果表明,微博术语之间存在语义相关性和时间相关性;微博文档之间不仅存在语义相关性,还存在作者信息相关性;与主流的微博检索模型相比,微博信念网络检索模型在多项信息检索指标上体现出较优的检索性能。消融实验结果表明,微博术语之间存在语义相关性和时间相关性;微博文档之间不仅存在语义相关性,还存在作者信息相关性;与仅考虑单一粒度关系或不考虑任何关系的信念网络检索模型相比,融合多粒度关系的信念网络检索模型体现出较优性能。聚焦于微博检索情景,综合查询语句扩展和文档扩展的优势实现微博检索,有效克服了微博检索面临的挑战,显著提升了微博检索的性能。在信息过载的背The microblog retrieval system is an important tool for microblog platform to realize personalized information filtering. Establishing a reasonable microblog retrieval model is not only conducive to meet users′ personalized information demands, but also to improve the information service level of microblog platform. However, compared with the traditional text retrieval, microblog retrieval faces two challenges: on the one hand, the shorter query is difficult to accurately express the user′s retrieval demands, yet on the other hand, the shorter microblog is difficult to fully express semantic, which makes difficult to accurately match queries and documents.Combining the characteristics of microblog terms and microblog documents, the relationships between terms and documents are integrated into the belief network retrieval model, and a microblog belief network retrieval model integrating multigranularity relationships is proposed. Firstly, the relationship between microblog terms is quantified by mixing semantic information and temporal information, so as to model the relevance between microblog terms more accurately. Secondly, the relationship between microblog documents is quantified by mixing semantic information and author information, so as to model the relevance between microblog documents more accurately. Finally, based on the basic belief network retrieval model, the probabilistic derivation process of microblog belief network retrieval model is given by combining the quantitative term relationship and document relationship. The study uses web crawler to obtain real microblog data from Sina Weibo to verify the validity and rationality of microblog belief network retrieval model.The results show that there are semantic relevance and temporal relevance between microblog terms, while there are semantic relevance and author relevance between microblog documents. In addition, the results also show that compared with the mainstream microblog retrieval models, the microblog belief network retrieval model has b

关 键 词：多粒度关系融合 微博检索 信念网络检索模型 术语关系 文档关系 

分 类 号：G203[文化科学—传播学]