支持联机分析处理的推特用户兴趣维层次提取方法

摘要

从海量推特数据中探索用户兴趣的分布规律和相关性有利于实现精确的个性化推荐.联机分析处理(On-Line Analytical Processing,OLAP)提供了一种适合人们探究数据的直观形式.将OLAP技术应用于推特数据的关键是如何挖掘和构建推特用户的兴趣维层次.针对现有方法只能提取单一层次兴趣的不足,该文提出一种支持联机分析处理的推特用户兴趣维层次提取方法.该方法首先通过RestAPI获取推特数据,然后通过改进的LDA(Latent Dirichlet Allocation)模型挖掘用户的兴趣和子兴趣,最后在此基础上构建兴趣维层次结构.实验评估了该方法的模型效果和可扩展性,并证实与LDA和hLDA相比可以更有效地提取出推特用户的兴趣维层次并应用于联机分析处理.%To explore the distribution and correlation from massive Twitter data helps the accurate personalized recommendation. On-Line Analytical Processing (OLAP) provides an intuitive form that is suitable for people to explore the Twitter data. The key of applying OLAP to Twitter data is how to mine and build dimension hierarchy of tweeter interests. Different from the existing approaches that can extract interests of tweeters with only one level, an approach to the extraction of dimension hierarchy of interests for OLAP is proposed. Firstly, it retrieves the Twitter data through RestAPI. Afterwards, it detects the interests and sub-interests using an improved (Latent Dirichlet Allocation, LDA) model. Based on the extracted interests and sub-interests it finally constructs the dimension hierarchy of interests. The experiment verifies its effectiveness and scalability, and demonstrates it can extract dimension hierarchy of tweeters' interests for OLAP more effectively than LDA and hLDA.