基于用户聚类的微博话题推荐方法研究

发布时间：2017-12-31 08:35

本文关键词：基于用户聚类的微博话题推荐方法研究　出处：《安徽理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,微博的规模不断扩大。微博规模扩大的同时也带来了微博平台上话题的爆炸式增长,而微博平台中大规模的话题资源和用户的辨识能力之间存在着一定的矛盾。面对用户浏览和关注话题的需求,微博平台很难在大量的话题中准确的提供满足用户偏好的话题。针对以上研究背景,本文提出了面向微博话题的推荐系统,该系统能够有效的在微博巨大的话题资源中找出满足用户偏好的话题集合,减少了用户浏览微博话题的时间成本,提高了用户效率。具体的研究内容如下:首先,微博用户影响力的计算。具有高影响力的用户关注的微博话题往往更容易让其他人关注(例如一个明星关注的话题很容易被其粉丝关注)。在微博中,用户和用户之间存在着关注和被关注的关系,在本论文中我们利用用户网络图来分析和保存用户关系,并在此基础上,用一个算法(类似PageRank)来计算每个用户的影响力。其次,微博用户偏好的表达形式。如何表达一个用户的话题偏好是微博话题推荐的一个难点。在本论文中,我们根据中文分词,特征词提取和关联规则挖掘,基于同一话题下的微博和用户历史微博,提出了话题特征图(TFG)和用户微博特征图(UTFG)。利用TFG和UMFG,进一步抽象出了用户话题特征矩阵(UTFM),用来表示用户的偏好。最后,如何准确高效的进行推荐。同一偏好的用户往往喜爱的话题也类似。在本论文中,我们基于UTFM,利用余弦距离通过凝聚层次算法进行用户聚类,在同一聚类簇内,结合话题热度和用户影响力,对用户进行推荐。通过对该推荐系统的分析和验证,该推荐系统在计算用户影响力、定义用户话题偏好表达形式和推荐微博话题算法上满足一定的合理性,能够准确高效的把满足用户偏好的话题推荐给用户。在微博话题数量巨大的情况下,能够减少用户的认知负担,大大缩短用户浏览微博所需要的时间,提高了用户效率。
[Abstract]:In recent years, Weibo's scale has been expanding. Weibo's scale has also brought about the explosive growth of the topic on Weibo platform. However, there is a certain contradiction between the large-scale topic resources and the identification ability of users in Weibo platform. Weibo platform is difficult to accurately provide topics that satisfy user preferences in a large number of topics. In view of the above research background, this paper proposes a recommendation system for Weibo topics. This system can effectively find out the topic set that satisfies the user's preference in Weibo's huge topic resources, and reduces the time cost for users to browse Weibo topic. Improve the efficiency of users. The specific content of the study is as follows: first. The calculation of Weibo's user influence. Weibo topics that have high impact users tend to get other people's attention (for example, a star's topic is easily followed by his fans.) in Weibo. There is a relationship of concern and attention between users and users. In this paper, we use user network graph to analyze and save user relationships, and on this basis. Use an algorithm (similar to PageRank-like) to calculate the impact of each user. Second. The expression of Weibo user preference. How to express a user's topic preference is a difficulty of Weibo topic recommendation. In this paper, we use Chinese word segmentation, feature word extraction and association rules mining. Based on Weibo and user history Weibo under the same topic, a topic feature map (TFG) and a user Weibo feature map (UF GG) are proposed. TFG and UMFG are used. Furthermore, the user topic feature matrix is abstracted to represent the user preference. In this paper, we use cosine distance to cluster users based on UTFM. In the same cluster, the user is recommended by combining the topic heat and user influence. Through the analysis and verification of the recommendation system, the recommendation system calculates the user influence. It is reasonable to define the expression form of user topic preference and recommend Weibo topic algorithm, which can accurately and efficiently recommend the topic satisfying user preference to the user. In the case of a large number of Weibo topics. It can reduce the cognitive burden of users, greatly shorten the time that users need to browse Weibo, and improve the efficiency of users.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】