面向微博用户的推荐多样性研究

发布时间：2018-04-28 10:27

本文选题：社交网络 + 推荐系统　；参考：《哈尔滨工业大学》2013年硕士论文

【摘要】：微博作为一种社交媒体正在越来越深刻的改变人们获取信息和交流的方式。越来越多的人愿意在微博中发表自己的观点，表达自己的心情，发布最新的消息。然而随着用户在微博中关注了更多的人，用户被大量的微博信息包围，其中很大一部分是用户不感兴趣的。这给用户在浏览微博时造成了一定的困扰，降低了用户的体验。用户浏览的微博默认按照时间的先后顺序排列，本文旨在使用推荐系统将用户感兴趣的微博靠前放置，帮助用户更快更好的获得信息。本文首先使用了基于特征矩阵分解的方法做微博推荐。矩阵分解是推荐系统中效果最好的方法之一，我们将其应用在微博推荐场景中。同时针对微博的特点，在做推荐时，我们同时考虑了微博的内容和社交特征以进行更好个性化的推荐。本文将微博推荐定义为列表的排序问题，采用了搜索引擎评价中流行的MAP值作为评价指标，实验结果证明了我们的方法较默认的时间排序能够更好的将用户感兴趣的微排在列表的前面，同时验证了微博推荐可以提升用户体验。在使用推荐系统进行微博推荐时会出现将过多的同类内容推荐给用户的问题。虽然这些同类微博可能都是用户感兴趣的，，但是这样的结果并不利于用户获取多样的信息，长期使用会导致用户体验的降低。所以本文在普通推荐的基础上引入列表多样性的度量。我们希望推荐列表中微博间的相似度要尽量的低。本文针对微博的特点提出了一系列的特征来度量微博相似性，使用增量Single-Pass聚类算法和聚类集成的框架将微博聚类，将该聚类结果与人工标注的结果进行了对比。使用了聚类评价中常用的NMI和B-cubed评价指标。实验结果证明了我们的提出的特征在计算微博相似性时是比较有效的。根据微博间的相似度，我们使用贪心的算法将推荐列表多样化，并采用人工在线评价的方式验证了多样化的推荐列表在用户体验上优于普通的推荐列表。
[Abstract]:Weibo, as a kind of social media, is changing the way people get information and communication. More and more people are willing to express their views, express their feelings and release the latest news in Weibo. However, as users pay attention to more people in Weibo, users are surrounded by a large amount of Weibo information, a large part of which is not of interest to users. This gives users a certain amount of trouble when browsing Weibo, reducing the user's experience. Weibo, who browsed by users, arranged according to the order of time by default. The purpose of this paper is to use the recommendation system to place the user's interesting Weibo forward, so as to help the user to obtain information faster and better. This paper first uses the method based on eigenmatrix decomposition to recommend Weibo. Matrix decomposition is one of the most effective methods in recommendation system. At the same time, according to Weibo's characteristics, we also consider the content and social features of Weibo in order to make a better personalized recommendation. In this paper, Weibo's recommendation is defined as the ranking problem of list, and the popular MAP value in search engine evaluation is used as the evaluation index. The experimental results show that our method is better than the default time ranking method to put the users of interest in the front of the list, and it also proves that Weibo recommendation can improve the user experience. In the use of recommendation system Weibo recommendation will appear too much of the same kind of content recommended to users. Although these similar Weibo may be of interest to users, such results are not conducive to access to a variety of information, long-term use will lead to a decline in the user experience. So this paper introduces the measurement of list diversity on the basis of general recommendation. We hope the similarity between Weibo in the recommended list should be as low as possible. In this paper, according to Weibo's characteristics, we propose a series of features to measure Weibo similarity. We use incremental Single-Pass clustering algorithm and clustering ensemble framework to cluster Weibo, and compare the clustering results with those of manual tagging. The NMI and B-cubed evaluation indexes which are commonly used in clustering evaluation are used. The experimental results show that the proposed features are effective in calculating Weibo similarity. According to the similarity between Weibo, we use greedy algorithm to diversify recommendation list, and use artificial online evaluation to verify the diversity of recommendation list in user experience is better than that of common recommendation list.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP393.092;TP391.3

【相似文献】