微博用户兴趣建模及推荐方法研究

发布时间：2018-03-09 07:05

本文选题：微博　切入点：用户兴趣获取　出处：《西北师范大学》2014年硕士论文　论文类型：学位论文

【摘要】：微博简短、即时、便捷的特性使其很快吸引了大量的用户群，并通过大量的转发及评论裂变式的快速传播与扩散,产生大量信息流。为了从海量信息中找出用户所感兴趣的微博话题，就需要设计合理的兴趣表示模型和准确高效的微博用户兴趣推荐方法。其中，微博用户兴趣模型是实现微博用户兴趣推荐的基础,微博用户兴趣推荐是微博用户兴趣模型的应用。因而，针对微博用户兴趣建模与推荐研究对于微博网站的发展十分重要。本文以构建微博用户兴趣模型及推荐方法作为研究背景，针对微博长度短、信息量少，高维稀疏等特点，研究用户兴趣信息获取、微博用户兴趣模型构建及推荐方法，旨在构建合理的微博用户兴趣模型及推荐方法，从而实现对微博用户准确高效的微博话题推荐。本文以新浪微博作为数据来源,主要做了如下工作： (1)分析微博中所包含的各种信息、微博用户行为及其和微博用户兴趣间的关系,选择合适的信息作为微博用户兴趣信息的来源，从而得到能够准确有效表示微博用户兴趣的信息。 (2)提出了一种基于词项关联关系与归一化割加权非负矩阵分解的微博用户兴趣模型构建方法。该方法首先基于词分布上下文语义相关性来建立词项关联关系矩阵刻画词项间相似度，接着应用归一化割加权非负矩阵分解算法获取用户-主题矩阵，产生用户感兴趣的微博主题聚类结果。实验表明，此方法能有效地进行微博主题聚类，并支持微博用户兴趣模型构建。 (3)提出了一种基于用户兴趣模型与会话抽取算法的微博推荐方法。该方法首先应用基于归一化割加权非负矩阵分解的微博用户兴趣模型获取用户-主题矩阵，产生用户感兴趣的微博主题，然后结合基于Single-Pass聚类模型的会话在线抽取算法SPFC(single-pass based on frequency and correlation)获取微博的会话队列并与用户感兴趣的微博主题进行相似计算，最后等到实时的微博推荐结果。实验表明，此方法能有效地进行微博推荐。实验表明：本文提出的微博用户兴趣模型及推荐方法能够有效地表征微博用户的兴趣并给出了相对准确的推荐结果。
[Abstract]:Weibo's short, immediate and convenient features quickly attracted a large number of users, and spread and spread rapidly through a lot of retweeting and commenting on fission. In order to find out the topics of interest to users from the mass of information, we need to design a reasonable model of interest representation and an accurate and efficient recommended method of user interest. Weibo's user interest model is the basis of user interest recommendation for Weibo, and Weibo user interest recommendation is the application of user interest model. Therefore, the research on user interest modeling and recommendation is very important for the development of Weibo website. Based on the research background of constructing Weibo user interest model and recommending method, aiming at the characteristics of Weibo, such as short length, little information, high dimension sparse and so on, this paper studies the acquisition of user interest information and the building and recommendation method of user interest model. The purpose of this paper is to construct a reasonable user interest model and recommendation method for Weibo users, so as to realize the accurate and efficient topic recommendation for Weibo users. 1) analyzing the various information contained in Weibo, the user behavior of Weibo and the relationship between the user's interest and the user's interest, and selecting the appropriate information as the source of the user's interest information. Thus can accurately and effectively express Weibo user interest information. (2) A method of constructing Weibo user interest model based on word item association relation and normalized cut weighted nonnegative matrix decomposition is proposed. The method is based on the semantic relevance of word distribution context to establish the word item correlation matrix. Describe the similarity between words, Then the normalized cut weighted non-negative matrix decomposition algorithm is used to obtain the user-topic matrix, and the result of Weibo topic clustering of interest to the user is obtained. The experiment shows that this method can effectively carry out Weibo thematic clustering. And support Weibo user interest model construction. In this paper, we propose a Weibo recommendation method based on user interest model and session extraction algorithm. In this method, we first obtain the user-topic matrix by the Weibo user interest model, which is based on normalized cut weighted nonnegative matrix decomposition. The Weibo topic of interest to users is generated, and then the online session extraction algorithm based on Single-Pass clustering model, SPFC(single-pass based on frequency and correlation, is used to obtain the session queue of Weibo, and the similar computation is carried out with the topic Weibo of interest to the user. Finally, the real-time Weibo recommendation results. Experimental results show that this method can effectively recommend Weibo. The experimental results show that the proposed Weibo user interest model and the recommended method can effectively represent the user's interest and give a relatively accurate recommendation result.
【学位授予单位】：西北师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092

【参考文献】