Abstract
Online shopping websites can generate a large amount of text and image data. By mining multimodal data from online shopping platforms, not only can users' evaluations or intentions be well understood, but also plays a very important role in semantic expression and fusion of multimodal data, which is also the focus of this study. Based on the close semantic association between multimodal data, this paper focuses on the temporal heterogeneity of online user comments based on multimodal data fusion. By analyzing user text and image data, and combining in-depth learning methods, the semantic features of multimodal data are fused and expressed, and in-depth research on multimodal data fusion is conducted.
The main research work of this paper is as follows: First, the research on the correlation characteristics of user comment behavior time series. This chapter takes the user comment data as the index of the technology acceptance model, analyzes the correlation characteristic index of the user comment time series based on the behavior dynamics, innovatively proposes the method of dividing the comment time interval into time series, and carries out the correlation characteristic law analysis and knowledge discovery research of the time series for the comment metadata content. Secondly, based on the analysis of the relevance characteristics of the time series of user comments based on the usefulness of comments, the usability and usability indicators of the technology acceptance model are converted into online user comments quality factors and applied to this study. Through text mining methods, combined with the fuzzy comprehensive evaluation method, the quality of comments is sorted and classified, the quality distribution of comments in different time series intervals is classified, and the characteristic words of high-quality comment products The distribution of emotion words and the correlation degree of "feature words - emotion" are analyzed to explore the knowledge discovery of the correlation characteristics of online comment quality time series. Finally, based on the rules of the correlation characteristics of the time series of online user comment behavior, the user comment emotion is regarded as the use attitude indicator of the technology acceptance model. Through the text semantic emotion mining method, the emotional polarity, emotional intensity and emotional semantic relevance of different time series intervals are analyzed. Through the word frequency statistics of emotional words at different time intervals, Support vector machine classification method is used to effectively classify emotional polarity, summarize the distribution characteristics of time series of different emotional polarity, and find the correlation characteristics of time series of comments on emotion.
Key words: multimodal data fusion; Online users; Comment behavior; Law of time heterogeneity