Research On Fine-grained Sentiment Analysis Of Chinese Micro-blog Based On Key Sentence Extraction

As a new social media,microblog has become more and more popular,and people have begun to transform into information creators.By extracting users'discussions on microblog,government departments can effectively understand and control the trend of public opinion by analyzing user's sentiment attitudes towards social hot topics.For companies,product demand could fully explored by mining users' comments on related products,thereby improving marketing methods.Therefore,sentiment analysis on microblog is of great significance.Compared with ordinary text,micro-blog is random and contains many new words and emoji.With the continuous improvement of micro-blog,the original limit of 140 characters has been expanded to nearly 2000 characters,which resulting in an increase in the number of long texts.At present,most of the sentiment analysis research on microblog only consider the method of short text.Therefore,based on the current microblog environment,this article aims to improve the accuracy of sentiment analysis,launches a fine-grained sentiment analysis research on Chinese microblog based on key sentence extraction.The specific content including the following aspects:(1)Research on the influence of microblog text character length distribution and length on sentiment analysis results and conduct key sentence extraction.The microblog data set is grouped by the number of characters,the number of samples is counted,the impact of text character length on the accuracy of sentiment analysis is explored,and the threshold for dividing the text length is selected to extract key sentences from long text.In the key sentence extraction stage,the sentiment dictionary is expanded based on the SO-PMI algorithm,and the keyword dictionary is constructed based on the word graph model.Considering the sentiment words contained in the sentence,the keyword,the location of the sentence,and the similarity between the text and the microblog topic.Based on the four types of similarity characteristics,we develop a scoring strategy for sentiment key sentences,and extract key sentences according to the score.(2)Construction of microblog emoji dictionary.Considering the difference between the designers' original intention and the users' understanding,a large number of data sets are used to construct a seed dictionary of derogatory and praise words in the microblog field,and the emoji are divided based on the SO-PMI algorithm to obtain users' true corresponding sentiment to be expressed when using it,and explore the optimal weight of emoji in sentiment calculation.(3)The extraction and research of dependency relationship pairs and fine-grained sentiment calculation.Through the dependency syntactic analysis,we analyze the syntactic structure and the dependency relationship between words,refine the calculation granularity and discuss the sentiment calculation methods of the modifiers that have a dependency relationship with the sentiment word under different combinations.The corresponding phrase combination is matched to calculate the sentiment value of the phrase on the basis of calculating the sentiment value of the dependency pair.At the same time,we formulate the semantic rule set covered by Chinese,and find the optimal parameters for each rule through experiments,perform sentence-level sentiment calculation,and further calculate the text sentiment value and classification.Aiming at the research methods proposed above,corresponding experiments are designed,and the experiments prove that the algorithm proposed in this paper has good results.Figure[28]Table[15]Reference[63]...