Cross Modality Sentiment Analysis Based On Weakly Supervised Learning

In the present,people are accustomed to comprehensively utilize information from plenty of modal such as text,pictures,video and others on a variety of social media platform to express their overall feelings or views.Thus,it is of great importance to explore and analyze the sentiment for monitoring aspects of social media,public relations,marketing and policy analysis and others.However,the traditional Single-modal Sentiment Analysis partly analyzes of the partial emotions in the cross-modal information.Strongly Supervised Learning,which consumes too much manpower on data annotation,is not suited to large-scale data training learning process.Therefore,Research on Cross-modal Sentiment Analysis based on Weakly Supervised Learning begins to receive attention.The focal point of cross-modal sentiment analysis based on weakly supervised learning lies in how to use this method to conduct sentiment analysis of sentence-level social media cross-modal information,and on this basis,how to conduct analysis of document-level social media cross-modal information.Aiming at the above two problems,this paper presents a set of Cross-modal Sentiment Analyses based on Weakly Supervised Learning.The main contents and innovations are as follows:1.Due to the problem of sentiment analysis of sentence-level social media cross-modal information,the paper proposes a sentence-level Cross-modal Sentiment Analysis method based on Weakly Supervised Learning.The main feature is the use of large-scale sentence-level social media cross-modal data and weak index data which could be easily and directly obtained to guide the learning process of sentiment characteristics in a single modal,and to realize the cross-modal fusion of sentiment characteristics.Experiment proves that the performance of this method is superior to the existing single-modal Sentiment Analysis,and to a certain extent it reduces the dependency on data annotation.2.Due to the problem of sentiment analysis of document-level social media cross-modal information,the paper proposes a document-level Cross-modalSentiment Analysis method based on Weakly Supervised Learning.The main feature is the construction of sentiment feature representation matrix of document-level cross-modal content,based on the sentence-level sentiment characteristics,and two modal of text and picture are used to combine the cross-modal features and study document-level sentiment expression.Experiment proves that the method is suitable for the sentiment characteristics learning process of document-level cross-modal information.With improvement of the accuracy of sentiment classification,it also reduces the dependency of the data annotation to a certain extent.Due to different forms of social media cross-modal information,this paper proposes the corresponding cross-modal Sentiment Analyses based on Weakly Supervised Learning.It verifies the correctness and validity of the proposed method by reasonable experiments and solves the above two issues.