Research And Application Of Sentiment Analysis Based On Image-text Fusion

The advancement of social media technology and the popularity of applications have caused uncountable social data generated every day.How to obtain a higher accuracy rate of sentiment analysis results has always been a hot issue in the field of public opinion analysis.These hot issues are mainly divided into two categories.One is that as the age of user changes,social data gradually shifts from standard text to weak-rule text.The other is that the advancement of technology has made it easy to upload pictures.Thus,social media data combined with images and text began to replace the original plain text data.At present,most of the known sentiment analysis methods perform sentiment analysis on standard text data.These methods use grammatical rules to judge sentiment words.However,the images appearing with the text often also contain the user’s emotional information,and they are complementary to the emotion expressed by the text.It becomes important to analyze the image-text fusion data.At present,there are still few methods for sentiment analysis on image-text fusion data.How to fuse the features of text and image is still a research hotspot.Aiming at the above two problems,this thesis respectively proposes a text sentiment analysis method based on the combination of sentiment lexicons and neural network and a sentiment analysis method of image-text fusion,then designs and implements a sentiment analysis system.The main work of this thesis is as follows:1.This thesis proposes a text sentiment analysis method based on the combination of sentiment lexicons and neural network,the purpose of which is to realize sentiment analysis of text with weak rules.We expand the sentiment lexicons and propose the sentiment score vector representation method.Then,the sentiment analysis of the weakrule text is realized by combining the sentiment score vector with neural network.Experiments have proved that the method proposed in this thesis has better accuracy on weak-rule texts represented by Weibo.2.This thesis proposes a sentiment analysis method based on image and text fusion,the purpose of which is to realize sentiment analysis of image and text fusion data.We use dynamic weight distribution function to dynamically assign the weights of the fusion of text sentiment analysis scores and image sentiment analysis scores at the decisionmaking layer,which effectively improves the accuracy of sentiment analysis on imagetext fusion data when the text length is too short or too long.At the same time,combined with the text sentiment analysis method proposed in this thesis,the text sentiment analysis part of the model has a better sentiment analysis effect for weak-rule texts.Experiments show that the method proposed in this thesis has a higher accuracy rate than the imagetext fusion sentiment analysis method with fixed weight distribution.3.In order to solve the problem of sentiment analysis of weakly ruled text and imagetext fusion data,we combined with the text sentiment analysis method based on the combination of sentiment dictionary and neural network and the sentiment analysis method of image-text fusion proposed in this thesis to design and implement the sentiment analysis system.The system realizes the upload of sentiment lexicons and sentiment analysis of text,image-text fusion data.The user can complete the above functions on the visual interface after logging in.