The Research Of Text Sentiment Classification Based On Context Semantics

Social media,online social media,is a tool or platform for people to share their opinions,experiences and insights.Sentiment classification technology is especially important for how to classify comment texts and extract effective information.The purpose of this study is to better capture contextual semantic information,extract contextual semantic features,and further improve the effect of emotional classification.Therefore,on the basis of previous studies,extract contextual semantic features and improve feature vectors,in order to improve the effect of sentiment classification of text.Firstly,construct emotional dictionary resources.Including basic emotion dictionary,network emotion new word dictionary,negative word dictionary,conjunction dictionary,degree adverb dictionary and stop word dictionary.Predecessors have studied more on emotional words,but less on the emotional effects of punctuation in texts.This paper summarizes the emotional effects of punctuation in texts,constructs a dictionary of emotional symbols,and sets corresponding weights.The basic dictionary is extended.Through the research,a SO-PMI sentiment dictionary extension algorithm based on Simple Good-Turing smoothing is constructed to construct the extended emotional word set 1.At the same time,the semantic similarity is calculated by using Word2 Vec word vector tool to obtain extended emotion.word set 2;Finally,combined with the two methods to build an extended sentiment dictionary,after fusion,de-duplication,manual screening,build an extended sentiment dictionary.The experimental simulation shows that the sentiment dictionary constructed in this paper is effective.Then,study the sentence-level sentiment classification,extract the emotion unit,and construct the sentence emotion vector.According to the fact that sentences are constructed by a combination of different parts of speech and symbols,five emotion units are constructed,including contextual semantic information such as conjunctions,negative words,degree adverbs,emotional words and emotional symbols.In the sentence vectorization,the Word2 Vec tool is used to represent the pre-processed sentences,combined with the constructed sentiment unit,the original word vector is adjusted to obtain the sentence emotion vector.Finally,the Attention-based BiLSTM classification model is used for sentence level emotional classification.The experimental results show that the proposed method effectively improves the sentence-level sentiment classification.Finally,combined with the sentence-level sentiment classification,the chapter-level sentiment classification is carried out.Firstly,the Doc2 Vec model is used to vectorize the chapters,then the contextual semantics such as sentence sentiment orientation and sentence position are adjusted to obtain the chapter-level sentiment vector,and the Attention-based BiLSTM classification model is used for chapter-level sentiment classification.The experimental results show that the proposed method effectively improves the chapter-level sentiment classification.