基于主题偏好的在线健康社区用户兴趣群体识别研究——以医享网为例

Research on the Identification of Users’ Interest Groups in Online Health Community Based on Topic Preference :an Investigation of yx129.com

【目的/意义】用户对健康信息的需求存在一定的差异,并形成不同的兴趣群体,对用户兴趣群体的发现以及组成机构的分析,有助于探索在线健康社区中知识转移与扩散的相关规律。【方法/过程】本研究借助ATM模型,通过文本聚类识别用户偏好主题,引入用户映射表和海林格距离算法,对用户兴趣群进行识别,并其分布结构和偏好主题内容进行了分析;同时,将医享网作为案例进行研究,进一步识别和分析健康社区的用户兴趣群及其相关特征。【结果/结论】采用本研究框架识别出来医享网痛风社区可分为保健、症状与治疗三个偏好主题;在此基础上结合用户兴趣群体结构分析和偏好主题具体内容,发现不同兴趣群体规模有所差别,组成结构具有一定的交叉性,并在一定程度上根据疾病发展的阶段性特征,存在动态迁移性。【创新/局限】基于主题-用户映射关系,对在线健康社区中偏好相似用户群和主题偏好分布规律进行探索,今后可进一步完善模型对不均衡文本的处理能力。

【Purpose/significance】Users have different needs for health information and form different interest groups. The discovery of user interest groups and the analysis of their constituent organizations will help to explore the relevant laws of knowledge transfer and diffusion in online health communities.【Method/process】This study uses ATM model to identify user preference topics through text clustering, introduces user mapping table and Hellinger distance algorithm, identifies user interest groups, and analyzes the structure and preference content of user interest groups. At the same time, the yx129.com website was used as a case to further identify and analyze the user interest groups of the healthy community and their related characteristics.【Result/conclusion】the gout community of yx129.com website can be divided into three preference themes: health care, symptoms and treatment. On the basis of this, combined with the user interest group structure analysis and the specific content of preference theme. It is found that the scale of different interest groups is different, the composition structure has certain cross-cutting, and to some extent, according to the stage characteristics of disease development, there is dynamic mobility.【Innovation/limitation】Based on the topic user mapping relationship, this paper explores the distribution of user groups with similar preferences and topic preferences in online health community, In the future, we need to further improve the ability of the model to deal with unbalanced text in the future.