Research On Text Sentiment Analysis Of Weibo Hot Topics Based On Deep Learning

The rapid development of Internet technology brought a large amount of network information platforms.It also reduced greatly the threshold for people to publish and receive information.Sina Weibo(hereinafter referred to as Weibo)is one of the main public opinion platforms in China,carrying huge information released by the public,most of which have emotional tendencies.It is of social and commercial significance for government organizations or enterprises to understand the public opinion trend by processing the Weibo hot topic texts and analyzing or mining the emotional tendencies.In the era of big data,text sentiment analysis methods based on emotion dictionary and traditional machine learning are inefficient while text sentiment analysis methods based on deep learning are beginning to emerge,among which convolutional neural network(CNN)and long-term memory neural network(LSTM)have become the mainstream methods in the field of text sentiment analysis with their respective advantages.However,there are still some problems in the current general deep learning methods in dealing with text tasks.For example,the single quantitative representation of text words and the inability of connection to the importance of words in sentences.In view of these problems,the research work of this thesis is as follows:1.The ConvBiLSTM model is constructed through taking advantages of CNN and BiLSTM in text processing.In the convolution layer and pooling layer,the model can exploit the CNN's advantages in extracting the deep abstract features of text sentences to the full.In the bidirectional long-term and short-term memory network layer,BiLSTM can better give consideration the previous and subsequent information.Experiments show that the accuracy and Fl score of this model are improved compared with CNN,LSTM,BiLSTM and other models in the data set of this thesis.2.To solve the problem of the unitary expression of the vectorization of the words in the text,this thesis proposes a word vector which integrates the part of speech features of each word on the basis of the original word vector,and takes it as the input of the text emotion analysis model.The experimental results show that compared with the original word vector,the word vector with part of speech features has a significant improvement on the classification effect of emotion analysis model.3.For the problem that ordinary deep learning methods failing in connection to the importance of words in sentences,this thesis constructs the ConvBiLSTM-SA model by bring in the Self-Attention mechanism.It can effectively be related to the importance of words in sentences to the whole sentence and give different weights.In addition,compared with the traditional Attention mechanism,Self-Attention mechanism has the advantage of independency of external information.The comparison experiment also proves that the model with Self-Attention mechanism is better than the model without Attention mechanism and traditional Attention mechanism in classification accuracy.