面向微博系统的实时个性化推荐研究

发布时间：2018-01-14 22:26

本文关键词：面向微博系统的实时个性化推荐研究　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来社交网络服务(Social Networks Services,SNS)发展迅猛,尤其是微博系统凭借其社交特性、媒体特性、终端可扩展性,逐渐发展成为主流社交媒体,是目前人们实现社会交往、获取和分享实时信息的一种重要方式,大量用户频繁的交互以及实时更新的内容,产生了海量的数据流,导致信息超载,阻碍了用户从大量数据中高效检索出感兴趣的信息,限制了社交网络服务的发展,也为现有的推荐系统研究带来了巨大的挑战。一方面,巨大的数据量,使得传统的推荐方法不能及时处理,同时还面临着候选集过大的问题。另一方面,社交网络中话题的不断更新和用户兴趣的不断变化,使得推荐的时效性成为研究难点。针对这些问题,本文深入分析微博的自身特征,在构建用户兴趣向量模型时,结合微博内容信息和用户间的信任度得到用户的兴趣向量,通过主题模型计算微博主题特征词的分布概率;提出微博局部索引机制(Partial Index Mechanism),改进倒排索引结构,对海量且持续更新的微博数据进行有效组织,减少索引成本,通过微博评分函数对新发布微博进行排序,索引列表中维持最受欢迎的微博;在微博实时个性化推荐部分,通过局部索引机制来维护最新发布或更新的微博,计算索引中的微博和用户之间的兴趣相似度,对结果进行排序。从索引列表中选取相应受欢迎度高的微博,构建用户个性化推荐列表。为验证微博实时个性化推荐方法的实时性及有效性,选取真实的微博平台用户数据作为实验对象进行多组实验,并对实验结果进行分析。实验结果表明,本文提出的算法能够有效的从海量微博信息中实时推荐用来感兴趣的微博,在准确性、实时性方面要优于传统方法。
[Abstract]:In recent years, social networking service (Social Networks, Services, SNS) is developing rapidly, especially the micro-blog system with its social characteristics, media characteristics, terminal scalability, gradually become the mainstream of social media, is the realization of social communication, an important way to access and share information in real time, a large number of frequent and real-time user interaction the updated content, produced a huge amount of data flow, resulting in information overload, block users from large amounts of data, retrieve information of interest, limiting the development of social networking services, it also brings great challenges to the existing recommendation system research. On the one hand, the huge amount of data, the traditional method cannot be recommended the timely treatment, but also facing the candidate set is too large. On the other hand, changing constantly updates and user interest topics in social networks, which makes the recommended time Becomes difficult. To solve these problems, this paper analyzed the characteristics of micro-blog, in the construction of user interest vector model, combined with micro-blog content information between the user and the user's trust to get interest vector, the distribution probability calculation of feature words by micro-blog theme topic model; put forward micro-blog local index mechanism (Partial Index Mechanism) improved, the inverted index structure, micro-blog data on the massive and continuous updating of the effective organization, reduce the cost of the index, by sorting the newly released micro-blog micro-blog scoring function, maintain the most popular micro-blog index list; recommendation part in micro-blog real-time personalization, through local index mechanism to maintain the latest release or update micro-blog and interest in computing the similarity between the index and micro-blog users, to sort the results from the index list. Select the appropriate high popularity of micro-blog, The construction of personalized recommendation list. The real-time and effectiveness of the recommended method for verification of micro-blog real-time personalization, user selection of micro-blog platform real data as experimental subjects for many experiments, and the experimental results are analyzed. The experimental results show that the proposed algorithm can effectively real-time information from the massive micro-blog recommended for interest micro-blog, in accuracy, real-time performance is superior to the traditional method.

【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】