Research On Sentiment Analysis Of Chinese Short Text Based On Deep Learning

With the rapid development of the Internet,a large number of product reviews have been generated on the Internet.These product reviews often contain a lot of valuable information.By analyzing the sentiment tendency of online product reviews,it can provide support for users and businesses to make decisions.At present,the sentiment analysis of Internet text comments has become a popular field of text mining.Although the sentiment analysis method based on neural network overcomes the difficulty of feature extraction in the machine learning method,the neural network still cannot perceive the importance of different words.Unable to learn the internal structure of sentences and unable to use the position information of words.At the same time,product reviews are short texts,which have the characteristics of less vocabulary and more noise.When using word vectors to represent them,there is a problem of sparse features,resulting in insufficient short text representation.In response to the above-mentioned problems,this article mainly conducts the following research:1.Aiming at the problem that neural network can't distinguish the importance of features and the lack of short text representation,a MCNN-BLSTM sentiment analysis model integrating part of speech and attention is proposed.Firstly,the word vector and part of speech vector are fused as the feature representation of the text,and then the local features of different granularity are extracted by multi-channel convolution neural network.At the same time,the semantic features of text are extracted by bidirectional long short-term memory network,and the extracted features are fused as the deep semantic expression of the text.Finally,different weights are assigned to the fused features by attention mechanism to distinguish the importance of different features.According to the experimental results,the accuracy of the model is92.65%,93.03%,93.27% and 93.77% on htl-2000,htl-4000,htl-6000 and htl-10000 hotel review data sets,respectively.On htl-6000 balanced data set,the accuracy of this model is improved by 3.09% compared with BGRU model and self-attention mechanism.On htl-10000 unbalanced data set,the accuracy of this model is improved by 2.07% compared with CNN model based on character vector and word vector.2.Aiming at the problem that the neural network cannot pay attention to the local key features of the text sequence and can not learn the hidden feature information of the text sequence,a sentiment analysis model based on self-attention and capsule network is proposed.The model first encodes the text sequence through a bidirectional long short-term memory network,learns the dependencies between the text sequences,then captures the internal structure of the sentence through the self-attention mechanism,pays attention to the local key features of the text sequence,and finally mines the text through the capsule network hidden feature information such as the position information of words in the sequence and the syntax structure of the text.According to the experimental results,the model achieved an accuracy of94.27% on the shop comment data set.Compared with the model based on CNN-BGRU and attention mechanism,the accuracy of the model is improved by 2.24%.