基于节点用户和概率矩阵分解模型的推荐算法

发布时间：2018-04-22 05:14

本文选题：节点用户 + 概率矩阵分解　；参考：《天津大学》2016年硕士论文

【摘要】：基于矩阵分解的推荐算法在推荐系统领域已得到长足发展,同时随着社交网络等的兴起,用户社交关系在推荐算法中也越来越扮演着举足轻重的地位。社交网络较传统社交的最大区别在于社交网络中信息流动会显著的受一些影响力很大的用户影响,这些用户被称为“节点用户”。单纯的矩阵分解模型及将社交网络引入矩阵分解的模型在近些年被大量提出,但对于社交网络中的节点用户以及节点用户影响力对模型和推荐效果的影响鲜有研究。论文主要研究内容包括:1)分析社交网络中用户影响力指标,定义节点用户,并利用斯皮尔曼等级系数分析各指标之间关系;2)将节点用户影响力引入矩阵分解模型,提出基于节点用户的概率矩阵分解模型;3)提出二度拓展和基于影响力的限制性拓展两种社交关系拓展方法,通过缓解模型中社交关系数据的稀疏性,来提高预测准确度。论文首先研究社交网络中的用户影响力指标并对节点用户进行定义,提出基于节点用户的概率矩阵分解模型,并针对该模型采用豆瓣电影和Yelp数据集进行验证。实验证明了该模型的有效性,并且在社交关系数据质量较高时较其他模型优势明显,同时结果也展示出在不同特点的环境和数据下应选用不同影响力指标;其次本文提出了两种缓解社交关系稀疏性的方法,同样通过上述数据集进行了验证。实验结果显示了两种方法分别在不同情况下的有效性,且可以进一步提高推荐算法的预测准确性。
[Abstract]:The recommendation algorithm based on matrix decomposition has made great progress in the field of recommendation system. With the rise of social network, user social relationship plays a more and more important role in recommendation algorithm. The biggest difference between social networks and traditional social networks is that the flow of information in social networks is significantly influenced by a number of influential users, known as "node users". The simple matrix decomposition model and the model of introducing social network into matrix decomposition have been put forward in recent years, but few researches have been done on the influence of node users and their influence on the model and recommendation effect. The main contents of this paper include: (1) analyzing the index of user influence in social network, defining node user, and using Spelman coefficient to analyze the relationship between the indexes. (2) introducing the influence of node user into matrix decomposition model. A probabilistic matrix decomposition model based on nodal users is proposed. Two methods of extending social relations based on quadratic extension and restrictive extension based on influence are proposed. The prediction accuracy is improved by reducing the sparsity of social relationship data in the model. In this paper, the user influence index in social network is studied and the node user is defined. A probability matrix decomposition model based on node user is proposed, and the model is verified by using Douban film and Yelp data set. The experimental results show that the model is effective and has obvious advantages over other models when the quality of social relationship data is high. At the same time, the results also show that different influence indicators should be selected under different environment and data. Secondly, we propose two methods to reduce the sparsity of social relationships, which are also verified by the above data sets. The experimental results show that the two methods are effective in different cases and can further improve the prediction accuracy of the recommendation algorithm.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】