Research On Multi-model Fusion Text Sentiment Algorithm With Deep Learning

With the rapid development of internet technology,a large amount of information has been generated in the online comment space.These comments mainly include Weibo comments,product comments,social comments and so on.These comments often have a strong sentimental color.Therefore,sentiment analysis of online comments have great social realistic sense.In this thesis,we conduct further research on the current situation of sentiment analysis and deep learning theory,and analyze some problems in this field.From the perspective of improving the accuracy of sentiment analysis and prediction,a multimodel algorithm for text sentiment fusion is proposed to address the shortcomings of a single deep learning model.Secondly,for the problem of lack of sentimental information in word vectors,a method for constructing sentimental words vectors is proposed.The specific research and innovations of this thesis are as follows:(1)In order to solve the problem that a single deep learning model is difficult to acquire the sequence features of sentences while obtaining highly abstract text features,we propose a multi-model fusion deep learning algorithm.Firstly,we use the parallel feature extraction ability of convolutional neural networks to extract the features of data.Then we combined with the advantages of the Bi-directional LSTM to extract the features of the sentence sequence.The extracted features are sent to the Bi-directional LSTM model for training,and the global average pooling method is used to reduce the complexity of the model.This thesis is carried out in the Amazon review dataset and SSTb dataset and the experiment results show that compared with the traditional algorithm,the proposed algorithm can effectively improve the accuracy of sentiment analysis.(2)In order to make the word vector have sentimental information,we construct the sentimental word vector training framework.By improving the Skip-Gram word vector training model,the sentimental information is integrated into the word vector training model to construct the sentimental word vector.For sentiment analysis tasks,the sentimental word vector can enrich the sentimental expression of the text,thereby improve the accuracy of sentiment analysis.We use the Amazon review dataset to train the sentimental word vector,and then use the different word vectors of the SSTb dataset for comparison.Firstly,the high-frequency words in the positive and negative samples in the Amazon dataset are respectively calculated,and the words with the most frequent word frequency and strong sentimental color are selected as the positive sentiment phrase and the negative sentiment phrase respectively.Then the residual word sentiment is calculated by the SO-PMI algorithm,and uses TF-IDF to weight the sentiment.Finally,the improved Skip-Gram model is used to train the word vector model,so as to obtain the expression of sentimental word vector.Experiments show that this method can effectively distinguish words with similar context structure but different sentiment,such as "good" and "bad",so as to improve the effect of sentiment analysis.