基于加权动态兴趣度的微博推荐方法研究

发布时间：2019-01-28 18:30

【摘要】：微博是一种用户通过关注关系进行信息实时分享的社交网络平台,不同的用户可能会有相同的喜好,于是就会形成具有相同兴趣爱好的用户集体。这就给人们精准定位用户兴趣取向,为组织机构精准发布推荐信息提供了可能性,增加了用户获得自己感兴趣信息的概率。因此,用户的兴趣度成为微博出现以来人们研究的热点,研究产生了许多个性化的推荐方法。从现有的研究来看,对微博数据进行挖掘分析的研究有很多,其中有对微博结构的研究,也有对微博文本的研究。在这些研究模型中,对于微博用户兴趣的模型研究很少,并且没有考虑到用户的兴趣变化,由于用户的兴趣具有时间变化性,也就是用户的兴趣会因为时间的推移而产生相应的变化,可能会产生兴趣转移。基于这一特点,本文把时间作为一个影响因子引入其中,首先根据现有的潜在狄利克雷分布模型计算出微博集数据集合的主题分布,从而将用户个体的动态兴趣度计算出来;其次,由于用户之间可能形成具有相同兴趣爱好的群体,即可以通过用户之间的互动频率和相似度,计算出用户集合体之间的兴趣度,即用户兴趣的相对稳定性;再次,将用户个体的兴趣和用户兴趣集合体的兴趣进行加权,就可以获得更加准确的微博用户对于微博主题的兴趣度;最后,给出一条新的微博,根据其主题分布,以及新的微博用户对主题的兴趣度,即可计算出加权动态兴趣度。进而,逐一计算出用户的加权动态兴趣度,利用兴趣度递减的算法,对所得兴趣度进行排序,最终将TOP-N个微博推荐给用户,从而实现精准推荐。论文从模型推荐的总体精度、推荐的时间精度和不同权值对模型的影响这几个方面对提出的推荐模型进行分析,同时通过实验,将本文提出的算法与基于LDA模型的协同过滤算法和基于RT-LDA模型的协同过滤算法进行了比较。实验结果表明,本文提出的推荐模型比传统模型可以更为准确地反映用户兴趣。
[Abstract]:Weibo is a kind of social network platform where users share information in real time by paying attention to the relationship. Different users may have the same preferences, so they will form a group of users with the same interests. This gives people accurate orientation of user interest, provides a possibility for organizations to accurately publish recommendation information, and increases the probability of users getting information of their own interest. Therefore, the interest of users has become the focus of research since Weibo appeared, which has produced many personalized recommendation methods. According to the existing research, there are many researches on Weibo data mining and analysis, including the research on the structure of Weibo and the text of Weibo. Among these research models, there is little research on Weibo user interest model, and it does not take into account the change of user interest, because user interest is time-varying. That is, the interest of the user will change with the passage of time, and may generate a shift of interest. Based on this characteristic, this paper introduces time as an influence factor. Firstly, the topic distribution of Weibo set data set is calculated according to the existing potential Delikley distribution model, and the dynamic interest degree of user is calculated. Secondly, because users may form groups with the same interests, that is, through the interaction frequency and similarity between users, the interest degree between user sets can be calculated, that is, the relative stability of user interest; Thirdly, by weighting the interests of individual users and the interests of users' interest aggregates, we can obtain a more accurate degree of interest of Weibo users to the theme of Weibo; Finally, a new Weibo is given, which can calculate the weighted dynamic interest according to its theme distribution and the interest of the new Weibo user to the topic. Then, the weighted dynamic interest of the user is calculated one by one, and the interest degree is sorted by the algorithm of decreasing interest. Finally, the TOP-N Weibo is recommended to the user, so that the accurate recommendation can be realized. This paper analyzes the recommended model from the following aspects: the overall accuracy of the model, the time accuracy and the influence of different weights on the model. At the same time, through experiments, The proposed algorithm is compared with the collaborative filtering algorithm based on LDA model and the collaborative filtering algorithm based on RT-LDA model. The experimental results show that the proposed recommendation model can reflect user interest more accurately than the traditional model.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.092;G206

【参考文献】