Design And Implementation Of A Fine-grained Sentiment Analysis System For User Reviews

With the rapid development of the Internet,various e-commerce websites and shopping websites have also shown an unprecedented growth rate,which has led to the emergence of a large number of short text online reviews.These short comments contain a lot of useful user evaluation information.By analyzing these short comments effectively,we can not only get important information,but also promote the development and prosperity of e-commerce.Therefore,collecting,sorting out and researching a large number of data on the network has become the focus of current network public opinion research.In view of this background,this paper designs and implements a fine-grained emotional analysis system,which solves the problems of time-consuming,laborious,incomplete and inefficient manual analysis of Web comment text.Firstly,this paper collects,analyses and processes the evaluation information in the network,and finally realizes fine-grained emotional analysis.The main work of this paper includes information acquisition,data processing and emotional orientation analysis.The main contents of this paper are as follows:(1)An unsupervised spam review detection method based on clustering is proposed.Firstly,this paper preprocesses the collected text,transforms the semi-structured web pages into structured data forms,and then finds that the excessive amount of information will bring great trouble to our emotional orientation analysis.Therefore,considering the information filtering before emotional analysis,this paper proposes an unsupervised garbage filtering method based on clustering,and evaluates these results.After a series of processing,information is clustered according to its similarity,and then processed on the basis of this clustering.Experiments show that this method has certain practicability and effectiveness.(2)Using CRF algorithm to extract evaluation words iteratively to improve feature sparsityIn this paper,CRF algorithm is used to extract emotional objects,emotional words and emotional modifiers.The results of unsupervised extraction are used as input of CRF,and feature words are extracted iteratively by combining word,part of speech and location features.Since most of the evaluation objects are nouns and noun phrases,in order to improve the efficiency of mining transaction sets,only the part of part of speech as nouns is used to construct the initial transaction set,and the candidate evaluation objects are used as input of CRF to extract the evaluation object set.Because CRF seek the global optimal solution,it effectively avoids the limitation of context feature selection caused by the output independence hypothesis of traditional implicit Markov model and the problem of label bias in maximum entropy model.(3)Design and implementation of fine-grained affective analysis systemThe system achieves the collection and analysis of network evaluation information through the related technologies and methods of network crawler,data processing and fine-grained emotional analysis,and can be displayed visually.Users can query through the query interface to obtain the relevant information they need.The system can quickly and effectively collect the relevant information in the network,integrate it,analyze and evaluate it,and finally show it in the form of graphics.It has the advantages of practicability,operability and intuition.Through the design and implementation of this system,we can automatically process the comment text,and display the emotional analysis results in front of users,so as to facilitate users to obtain information.