Research On Text Sentiment Analysis Combining Sentiment Dictionary And Neural Network

The traditional sentiment analysis methods mainly include sentiment dictionary and machine learning.The idea of the sentiment dictionary is to match words in the corpus with words in the sentiment dictionary to obtain sentiment polarity,but due to the phenomenon of polysemy,it is difficult to construct A complete dictionary,so this method has obvious flaws.Machine learning use Naive Bayesian,maximum entropy,and SVM(Support Vector Machine)and other machine learning algorithms to implement sentiment analysis,but they lack generalization ability and cannot be used in a variety of scenarios.The deep learning method can automatically optimize the parameters and continuously optimize the model,which can not only obtain a good classification effect but also have a strong generalization ability,and can be applied in a variety of scenarios.CNN(Convolutional Neural Networks)uses convolution operations and down sampling to obtain deep semantic information of text and main features of text capture,reducing network dimensions and improving training speed.LSTM(Long Short-Term Memory)can obtain long-distance semantic information through the memory cell mechanism,learn the dependency relationship between sentences,and reduce the loss of emotional semantics.The attention mechanism assigns different weights to each word.The stronger the emotional polarity is,the greater the weight is.In terms of feature acquisition,it is better than only acquiring a few words with a high emotional polarity or obtaining information on each word on average.Based on the above ideas,this paper designs and optimizes models through the fusion of sentiment dictionary,deep neural network and attention mechanism,and proposes two mixed models for sentiment analysis to improve the effect of sentiment classification.The main work of this paper is as follows:1.Aiming at the problem of single traditional neural network model and insufficient input data processing,the sentiment dictionary is used to filter the polarity of words in the text data,based on CNN,BiLSTM(Bi-directional Long Short-Term Memory)and attention mechanism,a text sentiment analysis model(SDPCNN-BiLSTM)combined with sentiment dictionary and parallel neural network is proposed to deal with the binary classification problem.In the English data set IMDB(Internet Movie Database,a large foreign movie review data set)and SST(Stanford sentiment treebank,a sentiment tree bank developed by Stanford University on movie review data).The experimental results show that the SDPCNN-BiLSTM model improves the precision and recall.2.Aiming at the problem of traditional deep learning algorithms for sentiment analysis that does not fully consider the text features and input optimization,the sentiment dictionary is used to rank the sentences in the text data,based on CNN,BiLSTM and attention mechanism,the two-layer CNN-BiLSTM model combined with attention mechanism and sentence ordering(DASSCNN-BiLSTM)is proposed to deal with the multi-classification problem.In the English dataset IMDB,Yelp2014(the restaurant dataset of Yelp in 2014)and Yelp2015(the restaurant dataset of Yelp in 2015).The experimental results show that the DASSCNN-BiLSTM model improves the accuracy and reduces the MSE value.