Research On Public Opinion Analysis Methods Based On Text And Image

With the development and popularization of Internet and mobile technology,social media has become an indispensable part of people's daily life.Nowadays,people are increasingly interested in publishing their own opinions,comments,emotional states and other information on these social media.The correct analysis of these data can not only help the enterprises to improve their products but also help government organs to carry out correct guidance of public opinion and actively respond to all kinds of crises.At present,most of the known public opinion analysis systems are focus on the sentiment analysis of the text data,and these methods do not consider or highlight the influence of emotional information on the sentiment analysis of text public opinion.How to mine enough effective emotional features by using limited text data is still a key challenge.In addition,the image has a complementary effect on the emotion of the text.However,there are few studies on the public opinion analysis of the fusion of image and text.And it's difficulty to handle the emotional mutual exclusion problem between the text and image due to the randomness and subjectivity of social media in practical applications.In view of the above two challenges,this paper proposes an emotional saliency features extraction method based on the public opinion of text and a analysis method based on cross-modality public opinion regression of the fusion of image and text.The main contents and innovations of the article are as follows:(1)The emotional saliency features extraction method based on the public opinion of text is proposed.The main idea of this method is to enrich the emotional information of the text by combining the emotional punctuation and adjacent words as a new emotional word and expand the emotional words by calculating the emotional relevance based on the emotion dictionary in the preprocessing stage.And then to study textual emotional saliency features which contains both the text semantic information and the emotional information by improving the CBOW model(Continuous Bag of Words Model).Finally,in order to verify the effectiveness of the proposed method,the textual emotional saliency features are input into CNN and LSTM classifiers for sentiment classification.(2)The analysis method based on cross-modality public opinion regression of the fusion of image and text is proposed.The main idea of this method is to introduce the modal contribution calculation on the basis of the cross modal regression model.And then to choose the appropriate fusion strategy by calculating and analyzing the contribution of each mode to the whole.The method mainly consists of three stages.The first stage is to extract the features of texts and images by the emotional saliency features extraction method and CNN model.The second stage is to learn the emotional correlation weights between the fusion features and each modal feature by crossmodality public opinion regression model.The third stage is to predict the sentiment tendency of public opinion with the trained classifier.(3)We design and implement the prototype system of public opinion analysis based on text and image.The Python and OpenCV are used to realize the prototype system of public opinion analysis based on text and image.The system includes four modules: data collection,data preprocessing,public opinion tendentiousness analysis and report display.In the paper,the emotional saliency features extraction method based on the public opinion of text module and the analysis method based on cross-modality public opinion regression of the fusion of image and text module are implemented in the prototype system.The feasibility and practicability of the proposed methods are verified by the prototype system.