Research And Application Of Sentiment Mining Model Based On Text Analysis

Recently,with the frequent updates of Internet technology,online social media is showing a vigorous development trend,and people's channel of expressing views has been greatly expanded in Internet,resulting in rapid growth of network information resources.Network information resources are usually unstructured data,and text information as the most typical unstructured data occupy a large proportion.However,these text information often contains users' sentiment orientation of hot events or brand reputation.It is challenging to obtain useful emotional information in these massive texts.Therefore,sentiment mining technology come into being,it extractes emotional information and classify the sentiment of text mainly through the text analysis.Sentence-level sentiment mining technology is the most widely used,often applied in brand reputation evaluation and public opinion monitoring.This thesis mainly studies the sentence-level sentiment mining methods,the work can be summaried as:1)The research of sentiment classification based on syntactic dependence analysis.Traditional method of sentiment mining focuses on the word itself generally,ignoring the syntactic structure of sentences and the semantic relevance between words.But these messages play an important role in text sentiment mining.For this reason,this thesis proposes a sentiment classification model based on syntactic dependency analysis,mainly including the extraction strategy of emotional information on sentence level and the sentiment judgment of dependency relationship cluster based on KNN algorithm.We used this model in NLP&CC 2013 data sets for testing,compared with the traditional method,the precision improved remarkably,recall rate was basically flat,F1-Score slightly improved.In addition,this model is more susceptible to the scale of the training data sets.2)The research of sentences' sentiment orientation based on Chinese emotional words.PMI-IR algorithm as a typical algorithm in the field of English sentiment orientation,can not be fully applied to Chinese.In addition,this algorithm uses a single standard word,the granularity of sentiment classification is not enough.Because of sending HTTP requests for information retrieval makes the calculation more time-consuming.Therefore,this thesis puts forward corresponding improvement strategy for the above defects,including standard words expansion,fine-grained sentiment orientation analysis,the definiton of extraction patterns in Chinese emotional phrases and adding a cache layer to reserve the semantic orientation of emotional phrases.Finally,we have done some experiments on the improved PMI-IR algorithm.The results show that improved PMI-IR algorithm has better and stable performance on sentiment classification.3)The design and implementation of sentiment classification system based on text analysis.Sentiment classification model this thesis proposed and improved PMI-IR algorithm have been applied in this system.Users can select the function mode individually,configure some parameters before text sentiment classification.