Research On Sentiment Analysis Of Weibo Text Based On Hybrid Neural Network

With the rapid development of information technology,various social media tools are also emerging in endlessly.Weibo has become an important platform for the public to understand current affairs and express emotions due to its wide spread range and large user traffic.This paper focuses on the sentiment binary classification task of Weibo text comments.It mainly aims at the problem of incomplete feature extraction of Weibo text and ignoring the weight of sentiment information,and proposes a microblog text sentiment analysis model based on hybrid neural network.The model adopts a dual-channel computing structure,which can comprehensively analyze the local features and timing features in the microblog text.The research work and innovations of this article are as follows:Aiming at the problem that the word vector representation generated by word2 vec does not contain emotional information and word weights,this paper proposes an emotional word vector representation method that combines word2 vec,Weibo sentiment dictionary,and TFIDF weight information.First,count the word frequency,use the TF-IDF algorithm to filter some low-frequency words and some high-frequency but meaningless words.Aiming at the problem that the probability of candidate words and emotional benchmark words appearing at the same time in the text is very low,based on the Semantic Orientation Pointwise Mutual Information(SO-PMI)algorithm,the correlation between the candidate words and the emotional polarity of the corresponding text is calculated,and the emotional tendency of the words is determined through the calculation results,and the microblog is completed.The establishment of the domain sentiment dictionary.The SO-PMI value and word frequency of the words are merged with the word vectors.Complete an emotional word vector representation that integrates Weibo context semantics,emotional information,and word weights.Design a deep learning model to verify the effectiveness of sentiment word vectors in sentiment analysis tasks of Weibo texts.Aiming at the problem of incomplete feature extraction of a single deep neural network model and ignoring the weight of emotional words,this paper designs and implements a Hybrid-Att hybrid neural network model.The model uses parallel computing to extract the local emotional features of the text from the convolutional neural network layer,extract the time series features of the text from the bidirectional long and short-term memory network channel,and input the output time series feature vector to the attention layer.The force mechanism increases the focus on key words and optimizes the weights of emotional features to extract emotional information in a deeper level.Mixing the CNN model with the Bi-LSTM＿Attention model can not only extract local emotional semantic features from the microblog text,but also consider the temporal features of the global text.Then connect the local feature vector with the time series feature vector,and finally output the result through the sentiment classification layer.In order to verify the effectiveness and feasibility of the hybrid neural network model Hybrid-Att in the field of Weibo text sentiment analysis,multiple sets of network models are designed for comparison experiments.The results showed that the accuracy rate reached 95.28%on the weibo＿senti＿100k public data set,and the accuracy rate,recall rate and F1 value were also slightly improved,and it also performed well on the data set crawled by the web.It is proved that the model can be well qualified for the sentiment analysis task of Weibo text.