Design And Implementation Of Sentiment Analysis System Based On Multi-modal Information

In recent years,with the development of social networks,people tend to use text,pictures,videos and other multi-modal information on social platforms to express their opinions or emotions.The analysis of emotional expressions contained in social media data is of great significance for applications,such as personalized recommendation and emergency analysis.At present,most of the traditional researches focus on the single modal analysis of text or image,while the multi-modal sentiment analysis of texts and image is seldom carried out.Based on the application background of multi-modal sentiment analysis of social networks,this paper considers the effective integration of two types of information in multi-modal sentiment analysis of images and texts.The main work of this paper is summarized as follows.This paper proposes the sentiment analysis model MFC,which is based on multi-level of image-text information fusion.The model first uses multiple branches convolution neural network to extract the features of the images and text,then perform dimension compression.After that,the MFC model uses the compressed features as the input in two-way of the neural network for feature fusion.At last,the MFC model performs the sentiment classification.Furthermore,in order to improve the performance of emotion classification,this paper adds a pre-processing method for the detection of inter-modal emotion conflict.This method selects different fusion methods based on the emotional tendency distribution of images and texts.The experiment on the datasets shows that the multi-level image-text information fusion emotion analysis model based on bidirectional GRU achieves better performance.Based on the above sentiment analysis model,we then design and implement a multi-modal sentiment analysis system for social networks,which collects social network data and gives sentiment analysis results.This paper has completed the summary design,detailed design and system realization of the images and texts cross modal sentiment analysis system for social network.Finally,this paper evaluates the system based on the hot topic.And the results show that the system has an acceptable performance in cross-modal sentiment analysis of images and texts for social network with potential applications.