基于协同过滤和行为分析的微博推荐系统

发布时间：2018-01-26 08:38

本文关键词： 推荐系统个性化社交网络协同过滤行为分析　出处：《南京理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：推荐系统是帮助用户发现内容、克服信息过载问题的有力工具,社交网络是发布、传播、接受信息的重要平台,作为当前学术界和工业界的研究热点,二者的结合是大势所趋。本文以社交网络中的关注对象推荐为任务,对推荐系统和社交网络的特性进行了深入研究,主要工作如下： 1推荐系统的方法论综述：对推荐系统研究中现有的理论和方法进行了分类总结,包括内容过滤、协同过滤和社会化过滤等,并着重分析了评分预测问题中的主流算法。 2推荐算法结合排序学习：评分预测是推荐系统的一个基本问题,但实际应用场景中更普遍的形式是TopN推荐。基于评分预测问题中的多种建模方法,本文尝试将排序学习应用于这些模型,实验结果表明推荐性能显著优于传统评分模型。 3社交网络中的用户兴趣建模：社交网络中的用户拥有众多的异质信息,如内容信息、社交信息等,本文尝试利用各种信息对用户兴趣进行建模,并通过实验分析了不同信息对反映用户兴趣的贡献度。 4社交网络中的用户行为模式分析与建模：推荐系统是通过分析用户的历史行为推测用户兴趣,因此大部分推荐系统的研究都集中于对用户兴趣进行建模,针对用户本身行为模式的建模研究相对较少。针对这一问题,本文提出了3类从不同角度对用户行为进行分析建模的方式,并尝试与推荐模型结合,实验结果表明融合行为模式可以有效地提高推荐系统的精度。
[Abstract]:Recommendation system is a powerful tool to help users find content and overcome the problem of information overload. Social network is an important platform for publishing, disseminating and receiving information. The combination of the two is the trend of the times. In this paper, the characteristics of recommendation system and social network are studied. The main work is as follows: 1 Summary of the methodology of recommendation system: the existing theories and methods in the research of recommendation system are classified and summarized, including content filtering, collaborative filtering and socialized filtering. The main algorithms of scoring prediction are analyzed. 2 recommendation algorithm combined with ranking learning: score prediction is a basic problem in recommendation system, but the more common form in practical application scenario is TopN recommendation. This paper attempts to apply ranking learning to these models. The experimental results show that the performance of recommendation is significantly better than that of traditional scoring model. 3 user interest modeling in social network: users in social network have a lot of heterogeneous information, such as content information, social information, etc. This paper attempts to use all kinds of information to model user interest. The contribution of different information to reflect user's interest is analyzed through experiments. 4user behavior pattern analysis and modeling in social network: recommendation system infer user's interest by analyzing user's historical behavior, so most of the research of recommendation system focuses on modeling user's interest. To solve this problem, this paper proposes three kinds of modeling methods to analyze user behavior from different angles, and try to combine with recommendation model. Experimental results show that the fusion behavior model can effectively improve the accuracy of the recommendation system.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.3;TP393.092

【相似文献】