Emotion Analysis Of Medical Crowdfunding Project Based On Multi-modal Fusion

With the development of multimedia technology,the types of data become more and more abundant.People prefer to express their attitudes and emotions by combining various modal information.Nowadays,multi-modal data is widely available on the Internet platform,and it is difficult for a single mode data to provide complete information about a phenomenon.Multi-modal data means that we can understand our surrounding environment by combined information from different channels,so that we can have a fuller understanding of various information content such as emotion.In recent years,medical crowdfunding platforms have developed rapidly,providing the public with a new way to raise medical funds and playing an important role in supporting national health security.Scholars have also did research on medical crowdfunding projects from various perspectives.Text and pictures in medical crowdfunding projects have rich emotional information.It is of great significance to use deep learning technology to mine emotional information from various modal data such as text and pictures.In this paper,the author firstly read papers from four research fields: crowdfunding project,text sentiment analysis,picture sentiment analysis and multimodal sentiment analysis.Based on the research needs of this paper,also introduced the theoretical basis and key technologies of text feature extraction method,image feature extraction method,deep learning method and multi-mode fusion method.This paper designed a multi-modal medical crowdfunding emotional fusion model framework based on BERT+Res Net and used the feature layer fusion method,and the accuracy rate of77.95 was obtained on the test set.On this basis,the ablation experiments were conducted to verify the effectiveness of the multi-mode fusion features.By comparing with other single-mode models and combined models,it was verified that the model selected in this paper has better accuracy and suitable training difficulty.This paper uses the model to conduct an empirical study on the data of medical crowdfunding projects on the shuidichou platform,and analyzes the results of emotion classification.The analysis results are as follows: patients in the 0-10 and80-100 age groups have stronger negative emotions;There is no significant affective gap between projects of different genders or different regions;The diseases involved in medical crowdfunding projects are mainly serious diseases,and the negative emotions towards serious diseases are strong.Items classified as negative emotions had higher retweets and received more help.