Research On Product Feature Extraction And Sentiment Analysis Of Online Review

With blowout development of electronic commerce, the network shopping is beingaccepted and used by more and more people, and then explosive growth starts.Study showsthat64.4%of e-commerce websites consumers make a buying decision with information fromproduct comments of already buyers. Opinion mining technology can help potential buyers、dealers and producers to get effective information from mass product reviews which spread inweb pages.This thesis focuses on feature extraction and application of opinion mining ine-commerce based on opinion mining framework.Firstly, we add time dimension extraction inopinion element extraction, and establish a hierarchical product attributes set. What’s more,we propose pos tag algorithm and sentiment set extension algorithm based on JWNL.Secondly, we add grammar elements into the maximum entropy model such as words aroundcandidate product feature, its pos tag, and positive relation between candidate feature andopinion word, which make the model to achieve good result. Besides, we propose implicitfeature extraction algorithm based on PMI-IR algorithm, and apply SO-PMI algorithm tosentiment oritation recognition. Finally, we conduct experiment with online reviews of fourtypes digital cameras in Amazon e-commerce website.After collect computation of sentimentscore, we put forward opinion mining application in e-commerce recommendation systembased on hierarchical product feature set. In comparison with sentiment score between fourcameras, we can see clearly the advantages or disadvantages of four cameras.We display thecomparision result with PowerBuilder programming tools.Then, we analyze regular pattern ofsentiment score time series and product feauture heat with consumer psychologyknowledge,product life cycle theroy and product market diffusion theory. The innovations ofour research are as follows: propose sentiment set extension algorithm based on JWNL; takecontext characteristics of candidate product feature into account in feature extraction model.We add time dimension to feature heat and sentiment score to explore the tendency of changes,and ananlyze the result combined with knowledge of social science.Through our research,cosumers can get information which they are interested from huge product reviews, andenterprise can also obtain data to support their management decision.