Application On Semantics Emotion Enhancement In Short Text Sentiment Analysis

The existing research on short text sentiment classification mainly considers the textual features of emoticons in sentences,ignoring the semantic association between emoticons and sentence emotions.And emoticons are independent of text grammar.Considering the different emotional tendencies of emoticons in different contexts,this paper mainly studies the semantics enhancement of emoticons to text emotions.Through in-depth research on feature extraction and emotional reasoning in short texts,a low-dimensional mixed text feature model is proposed for sentence representation,and an emotion-enhanced reasoning model based on fuzzy reasoning is proposed for sentiment analysis.On this basis,the semantic,affective affiliation information of emoticons and sentence grammar information are integrated,and a tree-based fuzzy pattern recognition emotion enhancement model is proposed.Experimental results have verified the effectiveness.Finally,a semantic emotion enhancement annotation system is designed and implemented based on the proposed models.The main works are as follows.1.Combining with the advantages of the small calculation amount of the emotion analysis method based on the emotion dictionary,we construct the emoji dictionary by calculating the emotional membership information of the emoji,and add the network emotion vocabularies in the Tsinghua University Chinese commendatory and derogatory dictionary.And then a low-dimensional mixed feature model is proposed for text representation.Experiments have verified that low-dimensional mixed text feature input is faster in training and better in classification performance in neural networks than traditional word sequence methods.To solve the uncertainty of text emotion,an emotion-enhanced reasoning model based on fuzzy reasoning and emotion dictionary is established.Moreover,comparing with traditional deep learning methods in the data set,it is found that the provided model can effectively improve the accuracy of emotion classification and it can be used as an optimization model to combine with various deep learning models.2.Based on the information of modifier relation,grammatical structure and word dependency relation,the emotion enhancement reasoning model based on grammar tree is further established.The membership degree of different emotional tendencies of emoticons not only reflects the emotional distribution in the corpus,but also reflects the different emotional expressions related to the context.Combining the emotional distribution of emotional symbols related to the context and the fuzzy emotional recognition of emoticons,a tree structure-based F pattern recognition emotional enhancement model is established.Experiments show that this model can further to improve the results of short text sentiment classification.3.According to the low-dimensional mixed text feature model and semantic emotion enhancement reasoning model proposed in this paper,a short text emotion labeling system is designed and developed,which can quickly perform sentiment labeling with short texts,indicating that the proposed model is practical.