Research On Cross-media Topic Detection And Opinion Analysis

In recent years,with the explosion of pictures and videos in social networking sites,today's Internet presents a trend of cross media.Cross-media data refers to multiple modal data in multiple social networking sites.The underlying performance heterogeneity between them,which brings difficulties to the full mining of the relationship between cross-media data.And because social networking sites are gathering places for people to get and discuss hot topics,cross-media topic detection and opinion analysis for social networking "topics" has become an important research topic in the field of public opinion analysis technology.However,However,there is almost no research working on multiple social networking sites and multiple modal data,which fails to accurately grasp the characteristics of cross-media data,leading to the one-sidedness of the research results.Therefore,this thesis focuses on cross-media topic detection and opinion analysis.This research is based on the research project of the Beijing Education Commission's scientific research and postgraduate training—social-sensed cross-media data analysis and mining research.This thesis aims to conduct an in-depth analysis of hot topics on social media platforms through these two researches.The main research contents and innovations of this thesis include:1.Aiming at the problem that the heterogeneous of cross-media data can't directly correlate and calculation,a framework based on graph-based cross-media data fusion is proposed,and the tag information of social networking sites is used to enhance the association of data similarity.By adopting the method of graph,it is effective to integrate cross-media data into a graph,and by using tag information as a link connecting different social websites,we can eliminate the problems caused by the heterogeneous characteristics of cross-media data representation.Experiments show that this method can effectively improve the performance of cross-media topic detection.2.In view of the diversification of cross-media data information on today's social networking sites and the high noise,which leads to the one?sided problem of traditional text analysis methods,this study proposes a cross-media opinion analysis method.By using cross-media data mutual dependency,analyze the public opinion from the perspective of text and video to achieve effective identification and control of network public opinion.In addition,there is a large amount of background information on the topic-related remarks,which affects the effect of opinion clustering.A topic model(BR-LDA model)is proposed,which divides the words in the speech into background words and opinion words.This model can effective suppress the adverse effects of background words on the clustering performance.Experiments on real data sets show the effectiveness of the method.