面向用户需求主题的在线问答社区信息多层级分类研究  

Information Multi-hierarchical Classification of Online Q&A Community Oriented to Users’ Needs Topics

作　　者：成全 张燕刚 Cheng Quan;Zhang Yangang(School of Economics and Management,Fuzhou University,Fuzhou 350116)

机构地区：[1]福州大学经济与管理学院,福州350116

出　　处：《情报学报》2022年第8期860-871,共12页Journal of the China Society for Scientific and Technical Information

基　　金：国家社会科学基金一般项目“多源在线医疗健康信息的语义融合及精准推荐服务研究”(19BTQ072)。

摘　　要：互联网在线问答社区服务的广泛性、便捷性、互动性与个性化特征促进了该模式的飞速发展,在线问答社区已逐渐成为人们获取各类生活信息的重要平台。然而,社区内信息资源存在缺乏有效组织与语义关联缺失等现实瓶颈,以及生活类信息的高复杂、多层级等特性,导致用户的在线信息需求服务体验效果不尽如人意。为实现对在线问答社区内各类信息资源的细粒度组织与语义关系揭示,进而达到面向用户需求主题实施信息精准分类的目标,本研究利用在线母婴社区内用户提问数据构建用户需求的多层级架构体系,进而生成经过验证的、具有多层级需求主题的标签化数据实验样本。最后,通过比对本研究所构建的面向用户需求主题的信息多层级分类模型(users’needs topicshierarchical classification,UNT-HC)与TextAttBiRNN (text attention bi-directional recurrent neural network)单层级分类模型及HFT-CNN (hierarchical fine-tuning conventional neural network)、HCCNN (hierarchical classification conventional neural network)等多层级分类模型的分类效果,验证了UNT-HC模型在实现在线问答社区中多层级单标签、超细粒度文本信息分类应用中性能的优越性。The extensiveness, convenience, interactivity and individual characteristics of the online question and answer(Q&A) community service has promoted the rapid development of the proposed model. The online Q&A community has gradually become an important platform for people to obtain various types of life-related information. However, due to the lack of effective organization of the information resources in the community and the lack of semantic relevance and other practical bottlenecks, as well as the high complexity and multi-hierarchical characteristics of such information, users’ experience in this regard is not typically satisfactory. To achieve the fine-grained organization and the semantic relationship disclosure of various information resources in the online Q&A community and subsequently achieve the goal of accurate classification of information based on user needs, a multi-hierarchical architecture system of user needs is built using the user search data in the online mother-infant community. Further, verified experimental samples of the labeled data are generated with multi-hierarchical needs themes. Finally, the classification effect of the UNT-HC(users ’ needs topics-hierarchical classification) is verified by comparing the information multi-hierarchical classification model(UNT-HC) constructed by this research institute for user-oriented topics with the TextAttBiRNN(text attention bi-directional recurrent neural network) single-hierarchical classification model and multi-hierarchical classification models, such as HFT-CNN(hierarchical fine-tuning conventional neural network) and HCCNN(hierarchical classification conventional neural network). The proposed model showed superior performance in realizing the multi-hierarchical single-label and ultra-fine-grained text information classification applications in the online Q&A communities.

关 键 词：用户信息需求 需求主题 在线问答社区 多层级分类 机器学习 

分 类 号：G252.61[文化科学—图书馆学]