Chinese Comments Sentiment Analysis Based On Deep Neural Networks

In recent years,the development of the Internet has given a large number of users the opportunity to express their opinions on the Internet,and Weibo.com,which provides a new space for individuals to talk and communicate in a fragmented way,has a great influence on all kinds of people in the society.Weibo users are free to post their comments on various things and events and share their experiences.These comments or shares contain the views and feelings of the publishers,and the emotional information extracted from the text can reflect the emotional state of the publishers at that time,which has great application value in personalized recommendation,public opinion control,policy-making based on social investigation and other aspects.Firstly,this paper chooses Weibo comment data for research and uses Python language to conduct network data crawling.Aiming at the shortage of Weibo data set and the confusion of labeling available data,this paper collects and sorts available Weibo comment data set publicly available on the Internet.At the same time,the hidden markov model and viterbi algorithm are used to design the word segmentation technology.Using the transition probability between the comprehensive states and the probability of the previous state,the state transition path with the highest probability is calculated,and the maximum probability path is backtracked and recorded to find the most likely correct word segmentation scheme.Secondly,in view of the traditional word to quantify the sole hot coding method of data sparseness of the existence of problems,this paper uses a word vector text feature extraction tools improvement method,using the Sub-word child word Fast Text embedded technology to obtain the word vector is more semantic and syntactic relations,and its tree structure characterization of category,on its output using H-softmax multilayer classifier method,at the same time with traditional Word2 Vec jumps model of text feature extraction method of comparison,proved the effectiveness of the method,improve the computational efficiency of the model and the training speed.Finally,an Attention mechanism is introduced into the model,and the background variables are obtained by weighted averaging the hidden states of all time steps of the encoder.It can intuitively explain the contribution and importance of each sentence and word to the classification.The attention mechanism is connected between the LSTM model and the output layer,which shows the corresponding relationship between a word in the output targetsentence and each word in the input sentence.After adding the attention mechanism,the effect of the model on the emotional classification of Weibo comments is improved.