| With the development of the Internet and the maturity of social networking platforms,more and more information is emerging on social platforms.Among them,the network user community has a large base,and the domestic network social platform that involves a wide range of people in the community belongs to Sina Weibo.How we can effectively extract high-quality,emotionally-oriented texts from these cluttered,short,disorderly,and flooded blog posts has become an important topic in natural language processing.This paper mainly constructs two methods of textual sentiment classification,namely the text sentiment classification method based on LS-SO algorithm and the text sentiment classification method based on the Attention mechanism Bi-LSTM model.Compared with the previous text sentiment classification methods are optimized and improved.The research work of this paper is as follows:1.Construct a sentiment dictionary.Li Tung University of China Tsinghua University's Chinese decency dictionary and the emotional vocabulary ontology database of Dalian University of Technology are integrated into seven categories of emotions.At the same time,the HowNet sentiment dictionary of HowNet and the NTUSD Simplified Chinese Emotional Dictionary of Taiwan University are integrated into two categories of emotions.A total of eight dictionaries are marked: basic emotion dictionary,target emotion dictionary,network language emotion dictionary,emoji emotion dictionary,negative word dictionary,question word dictionary,degree adverb dictionary,and conjunction dictionary to provide protection for text sentiment classification.2.Automatically expand the basic emotional dictionary.The PMI-IR algorithm is used to synonymously expand the positive and negative sentiment words under the 7 major emotion categories of the basic sentiment lexicon to form a standard basic emotional word dictionary.3.Automatically expand the domain emotional dictionary,including the expansion of emotional dictionaries and the expansion of emotional expressions.Based on the PMI-IR algorithm,a LA-SO algorithm is proposed to automatically expand the emotional sub-categorization of micro-blog-related domain sentiment lexicon.4.Formulate candidate word extraction rules and microblog text semantic analysis rules.According to the extraction rules we make,the candidate emotional words in the text are extracted to better identify the unregistered words.Calculate the sentiment value of the text according to the semantic analysis rules.At the same time,the emotional extreme value of the emoticon is calculated and it is merged with the extreme value of the emotional word to further modify the emotional weight value of the microblog text.5.build a deep learning model.On the basis of constructing Bi-LSTM,attention mechanism Attention is introduced.The model structure is divided into four levels: word vector representation layer,semantic information coding layer,global feature extraction layer and emotional text classification layer.Among them,the word vector presentation layer uses the Word Embedding mechanism to map text data into a low-dimensional real number vector.The semantic information encoding layer calculates the value of each word's contribution to the sentence,and saves the context information of each word.At the global feature extraction layer,the CNN idea is used to fuse the feature values of the forward and backward output,so as to improve the accuracy of model emotion classification.The emotional text classifier uses the Softmax classifier to classify emotional texts. |
