Design and Experiment of Multi-Modal Sentiment Analysis Model by Fusing Multi-scale Features
The research of sentiment feature extraction exists some challenges,such as poor characterization of extracted feature and low efficiency of sentiment analysis.In view of this,this paper aims to design the multi-modal sentiment analysis model by fusing multi-scale features(MSAM),to efficiently fulfill the multi-modal sentiment analysis task.In this model,the multi-modal data are represented as vector.Then,multi-scale convolutional neural network(MSCNN)is used to feature extraction.The average,maximum and minimum of the extracted feature can be obtained after statistical pooling.The features are fused with the help of the attention mechanism.Finally,it can be predicted the sentiment tendency based on the multi-modal data and the extracted features.The comparative experimental results show that the model proposed in this paper performs better than several traditional models in the mono-modal and multi-modal situations.Specially,the statistical feature and the weighted statistical feature play an important role in multi-modal sentiment analysis.
NETL
NSTL
万方数据
