考虑用户—发布者关系建模的微博个性化搜索方法研究

发布时间：2018-10-29 09:43

【摘要】：随着WEB 2.0的迅速发展,微博、微信及各类社交网络的应用也越来越广泛,社交网络中海量的实时的信息在给用户带来更多乐趣的同时,也使得用户搜索自己感兴趣的内容越来越难。比起传统的Web搜索,微博具有很强的社会性、主题更为丰富、文本更加短小,这使得Web的个性化搜索方法并不适用。另外,每一条微博都有其明确的发布者,用户在社交网络进行搜索时,除了关注微博的内容,往往还关注该内容的发布者。因此,如何综合用户兴趣、微博内容和微博发布者的相关信息,提高微博信息的个性化搜索效果需要进行深入的理论研究。本文提出了一种考虑用户与发布者关系建模的微博个性化搜索模型,运用主题模型与语言模型来为用户在主题上的兴趣进行建模的同时,融合了用户对信息提供者的兴趣,包括在兴趣主题上的相似度、共同关注的人,微博信息,包括微博长度、链接、标签和转发,以及影响力、传播能力、认证名人特征、用户关联度等用户社交网络的诸多特征,从用户与发布者关联、发布者特性以及微博特征三个角度进行了建模。在建模的基础上,本文设计了一套个性化排序机制。以语言模型的检索方法为基础,排序融合了用户与发布者关联、用户特性以及微博特征三个角度获取用户对微博的最终得分,根据得分对微博进行重排序。最后采用网络爬虫抓取新浪微博数据,使用了真实的用户搜索反馈进行了实验验证。本文在上述研究的基础之上设计了一套考虑用户发布者关系建模的个性化搜索框架。结果表明,考虑发布者信息可有效提高搜索效果,为用户提供个性化的微博搜索服务。
[Abstract]:With the rapid development of WEB 2.0, Weibo, WeChat and all kinds of social networks are more and more widely used. It also makes it harder and harder for users to search for content they are interested in. Compared with traditional Web search, Weibo has strong sociality, rich theme and shorter text, which makes the personalized search method of Web not suitable. In addition, every Weibo has its own specific publisher, users in the social network search, in addition to paying attention to Weibo content, often also pay attention to the publisher of the content. Therefore, how to synthesize the user's interest, Weibo's content and the relevant information of Weibo's publisher, and improve the personalized search effect of Weibo's information need to carry on the thorough theory research. This paper presents a personalized search model based on Weibo, which considers the relationship between users and publishers. It uses topic model and language model to model the interest of users on the topic and combines the interest of users in information providers. Including similarities on topics of interest, people of common concern, Weibo information, including the length, links, tags and retweets, and many of the features of user social networks, such as the length, links, tags, and retweets, as well as the power of influence, the ability to spread, the features of authenticating celebrities, the user association degree, and so on. The model is built from three aspects: user-publisher association, publisher characteristics and Weibo features. On the basis of modeling, this paper designs a set of personalized sorting mechanism. Based on the retrieval method of language model, the sorting combines user and publisher association, user characteristics and Weibo features to obtain the final score of user to Weibo, and reorder Weibo according to the score. Finally, using web crawler to grab Sina Weibo data, using real user search feedback for experimental verification. Based on the above research, this paper designs a personalized search framework which considers the modeling of user publisher relationship. The results show that considering publisher information can effectively improve the search effect and provide users with personalized Weibo search service.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.092;TP391.3

【相似文献】