Research On Document Emotion Summary Based On LDA Model

With the rapid development of the network technology, the comment text on the networkincreased sharply. The comment text usually contains lots of sentiment information, but thisinformation is disordered and indistinct. In order to remove redundant information and getoverall sentiment orientation of the comment text, in this paper, sentiment summarizationtechnology was used to analyze, process and integrate the comment text information on thenetwork. Finally, the users could get crystal clear sentiment summarization.In this paper, first of all, some of the key technologies are discussed, which are used inhandling the comment text information. Then, based on the tagging of Hidden Markov Model,the subjective sentences in comment text are recognized. Finally, the model Latent DirichletAllocation (LDA) is applied in the sentence level and sentiment summarization is generated.The main work of this paper includes the following three aspects:1) Researched s ubjective sentence recognition in sentiment analysis, and then themethod of subjective sentence recognition based on Hidden Markov Model was proposedThe subjective text was usually used in text sentiment analysis technology. Therefore, thesubjective sentence recognition was needed before sentiment analysis. On the basis of thesyntax and semantic relationship and context-dependent between the feature of subjective andobjective text were adequate took into account, then the method of subjective sentencerecognition based on Hidden Markov Model was proposed.In this method, the information gain and chi-square statistics were hierarchical acted onthe subjective and objective text. On this basis, subjective and objective feature collectionswere extracted, which not only held good discriminative validity, but could on behalf of thetype of subjective and objective sentence. Then, using the application of hidden markov modelin part-of-speech tagging, each sentence was tagged by HMM and subjective sentences wererecognized based on the weight of sentences. The experimental results show that this methodcan effectively identify the subjective sentences in comment text.2) Researched sentiment text modeling and the docume nt sentime nt summarizationtechnology, then the method of document sentime nt summarization based on LDA modelwas proposedBefore generating the sentiment summarization, the document modeling for subjective textwas needed. Due to the LDA topic model could avoid the problem of high dimension andsparse characteristics in the traditional vector space model, hence, LDA model was used tosentence level and the method of document sentiment summarization based on LDA mode l wasproposed in this paper. In this method, all the subjective sentences was described by model LDA and the potentialtopic in the documents collection was mined; then, the parameters of LDA model was estimatedby Gibbs sampler; finally a summarization was generated based on the weight of sentences,inwhich maximal marginal relevance was used. The experimental results show that the documentsentiment summarization which was generated by this method is closer to the specialistsummarization.3) Designed and implemented the document sentime nt summarization prototypesystem based on LDA model.On the basis of analyzing the generation process of network comment text sentimentsummarization, the corresponding function module was designed for each process. And finallythe document sentiment summarization prototype system based on the LDA model wasimplemented.This prototype system could effectively extract, analyze and process the comment text datathat exists in the network and provide the distinct and integrated sentiment summarization forthe user. This sentiment summarization not only holds overall sentiment orientation of thecomment text, but expresses the dominating content.