基于异构信息网络聚类的APP推荐算法研究

发布时间：2018-04-14 19:31

本文选题：异构信息网络 + 排序　；参考：《浙江大学》2016年硕士论文

【摘要】：移动应用(Mobile Application, APP)市场采用推荐技术将APP推荐给不同的用户,以此帮助用户从海量的APP中发现感兴趣的内容。但是由于APP所处领域的限制,APP推荐存在一系列的问题,例如APP使用分布容易出现重头和长尾现象、数据稀疏问题、冷启动问题等。其中,重头和长尾现象的出现不利于APP领域的发展,数据稀疏问题则限制了推荐算法的效率,而且随着APP数量的增长,这些问题将日益突出。Netflix竞赛的出现在很大程度上推动了推荐领域的发展,但是APP推荐算法研究还不成熟。目前移动应用市场APP推荐算法主要集中在关联推荐、热门推荐以及新品推荐等,这些传统的推荐方法没有从根本上解决APP推荐面临的问题。随着APP市场的日益完善,APP市场将拥有更加完备的APP信息以及用户信息,如何利用这些信息帮助解决APP推荐面临的一系列问题变得十分重要。针对APP数据集特点,本文提出将排序方法、聚类技术与推荐算法相结合,共同挖掘APP多维度文本数据中蕴藏的关系信息,在此基础上开展基于异构信息网络聚类的APP推荐算法研究。第一,构造多维度文本组成的APP信息网络模型,包括用户、APP、描述文本、发布者信息以及分类信息等。第二,通过两种排序算法获取附属类型对象的排序分布。第三,在排序分布的基础上建立一个针对中心类型的混合概率生成模型,使用EM算法估计参数的最优值,然后依据贝叶斯理论获得对象的后验概率,根据对象的聚类分布重新划分类簇。第四,根据APP以及用户聚类结果开展两种不同的协同过滤算法,即基于伪评分的IBCF (Iterm-Based Collaborative Filtering)算法以及基于时间衰退的UBCF(User-Based Collaborative Filtering)算法。本文采用360手机助手应用市场中的数据集进行实验分析,实验结果表明本文提出的APP推荐算法能够增强APP推荐的实际效果。
[Abstract]:Mobile application (Mobile Application APP) market recommended by APP technology will be recommended to different users, to help users find content of interest from the mass of APP. But because of the limitation of APP, APP recommended a series of problems, such as the use of APP distribution to head and tail phenomenon, data the sparsity, cold start problem. The emergence of heavy head and long tail phenomenon is not conducive to the development of APP, the data sparseness problem is limited by the efficiency of the algorithm, and with the increase in the number of APP, these problems will appear increasingly prominent.Netflix competition to promote development of the recommended field to a great extent, but APP research on recommendation algorithm is not mature. At present, the mobile application market APP recommendation algorithm mainly focus on the related recommendations, recommendations and new products, recommend these traditional methods not fundamentally To solve the problem. With the APP recommended APP market is improving, the APP market will have a more complete APP information and user information, how to use these information to help solve a series of problems facing APP recommendation becomes very important. According to the APP data set, this paper presents the ranking method, combining clustering technique and recommendation the common information mining algorithm, the relationship is APP multi dimension text data, on the basis of research on clustering heterogeneous information network APP recommendation algorithm based on APP. First, the information network model, construct multi dimension text including user, APP, description of the text, the publisher information and classification information. Second, get sort of distribution the affiliated type object through two kinds of sorting algorithms. Third, based on the distribution of the order for the establishment of a generation model of mixed probability center type, using the EM algorithm. Optimal values of the parameters, then the probability on the basis of Bias theory to obtain the object, based on the object clustering distribution re divided clusters. Fourth, according to APP and user clustering results carried out two kinds of collaborative filtering algorithm, which is based on the pseudo score (Iterm-Based Collaborative IBCF Filtering) algorithm and UBCF based on User-Based (time of recession Collaborative Filtering) algorithm. This paper uses the 360 mobile phone assistant application market data sets were analyzed, the experimental results show that the proposed APP recommendation algorithm can enhance the actual effect of the APP recommendation.

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】