基于多源数据融合的微博用户兴趣挖掘方法

发布时间：2018-01-05 15:18

本文关键词：基于多源数据融合的微博用户兴趣挖掘方法　出处：《哈尔滨工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着互联网的快速发展,社交网络得到人们的广泛认可。在国内社交网络方面,越来越多的人开始通过微博发布信息,而微博网站也成为主流海量信息的发布体,对微博的研究也从显性的兴趣标签到微博本身潜在的内容进行主题挖掘。LDA(latent Dirichlet allocation)模型是近几年比较流行的一种非监督的主题模型,已经有一些研究通过在Twitter数据集上对LDA模型进行主题挖掘,但在中文微博的主题挖掘上的研究并不多。基于微博内容的用户兴趣挖掘可以获得较为精准的挖掘结果,但会面临冷启动和数据稀疏问题。基于交互关系的用户兴趣挖掘方法和基于交互信息的用户兴趣挖掘方法可以从两个不同的角度弥补基于微博内容的用户兴趣挖掘方法的缺陷。本文爬取新浪微博用户不同层次的数据用于进行多源数据融合微博用户兴趣建模研究。主要研究成果包括以下几个方面:首先,结合传统LDA模型提出一种适合中文微博的有监督的兴趣主题挖掘模型,即基于内容主题挖掘微博生成模型CTM-LDA。该模型有效的利用了先验主题信息,根据微博用户信息及用户原创内容相融合挖掘微博用户兴趣主题。其次,利用基于交互关系和交互信息的等信息源分别构建兴趣模型,利用交互关系矩阵和关注人兴趣标签以及词语间相似度生成关注人兴趣主题。最后,针对微博自定义内容,话题微博,交互信息,以及用户自定义标签等不同数据进行实验,构建微博用户兴趣模型,研究并设计了多源数据融合的用户兴趣模型,利用空间向量构建出用户最终的兴趣主题。本文提出了不同数据源的微博用户兴趣融合模型,通过研究发现可以有效利用用户之间的交互关联关系的信息对微博用户进行兴趣主题挖掘,且融合模型的效果要更优。未来通过用户的兴趣模型可有针对性的对微博用户进行个性化推荐,该模型可推广到其他社会媒体网站平台,对于企业而言具有一定商业研究价值。
[Abstract]:With the rapid development of the Internet, social networks have been widely recognized. In the domestic social networks, more and more people began to publish information through Weibo. And Weibo website also becomes the main stream massive information release body. The study of Weibo also goes from explicit interest tags to Weibo's own potential content for topic mining. LDAlatent Dirichlet location). Model is a popular unsupervised thematic model in recent years. There has been some research on topic mining for LDA models on Twitter datasets. However, there is not much research on the Chinese Weibo topic mining. The user interest mining based on Weibo content can obtain more accurate mining results. However, it will face the problem of cold startup and sparse data. The interactive method of user interest mining and the method of user interest mining based on interactive information can make up for the user interest mining based on Weibo content from two different angles. This paper crawls different levels of data from the user of Sina Weibo to carry on the multi-source data fusion Weibo user interest modeling research. The main research results include the following aspects:. First. Based on the traditional LDA model, a supervised topic mining model for Chinese Weibo is proposed. Namely based on content topic mining Weibo generation model CTM-LDA. this model effectively utilizes the prior topic information. According to Weibo user information and user-generated content fusion mining Weibo user interest topics. Secondly using interactive and interactive information sources such as information to build interest model. Using the interaction matrix, interest labels and the similarity between words and expressions to generate topics of interest. Finally, for Weibo custom content, the topic Weibo, interactive information. As well as user defined tags and other different data experiments, build Weibo user interest model, research and design multi-source data fusion user interest model. Using space vector to construct the user's final topic of interest. In this paper, Weibo user interest fusion model with different data sources is proposed. Through the research, it is found that the information of the interaction relationship between users can be used effectively to mine the topic of interest of Weibo users. And the effect of fusion model should be better. In the future, Weibo user can be personalized recommended through user interest model, this model can be extended to other social media website platform. It has certain commercial research value for enterprises.
【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】：2016
【分类号】：TP391.1;F49

【相似文献】