| Into the age of the web2.0, the population of the Chinese netizen is growing,whichmakes the emerging media represented by Internet become popular expression desire and theimportant tool of communication. User’s identity information is also turned to the publisher forinformation from past recipients, which makes a lot of user information accumulated on theInternet,the information contains the emotional attitude of the users of these information andpolitical tendency. Mining the emotional information from the information that user generatedand analyzing of the user’s emotional tendency has great significance to the understanding ofpublic opinion, etc.Comparative sentences is a common sentence patterns in user generated information, bycomparing two things, From that,it is easy to judge two things of the same kind can determinethe similarities and differences and the advantages and disadvantages, By the study of this kindof problem,we can solve in product reviews, customer information management, publicopinion, and other fields to provide information support, collection of entity recognition,relationship extraction and category balance, hot issues of study have certain reference value.The orientation study of comparative sentences is integrated in the natural language processingmultiple key technology in the integration of research, including the text categorization, entityextraction, the emotional analysis, etc.Studying the contents of this paper is on the fourth and the fifth Chinese bias evaluation task,on the basis of further research on Chinese comparative sentences, including comparativesentences, comparative identification of extraction and emotional orientation of comparativesentences, the main research work includes the following aspects:(1) Use the correlation of key table is used to identifying the comparative sentences.Association rules is the containing type shape such as X and Y. Among them, the X and Y,respectively, referred to as the leading and subsequent association rule mining of associationrules is to find in the transaction database D with users of the given minimum support minsupand minimum confidence minconf association rules. In this paper, based on the basic principle ofthe algorithm to construct correlation characteristic dictionary, to summarize the characteristicsof comparative sentences is stored in the form of rules to the correlation characteristic in thedictionary is used for the identification of Chinese comparative sentences.(2)The use of conditional random field model to compare the relation extraction.Comparative relation extraction is refers to the key element to extract the comparativesentences. Comparative sentence elements including comparative subject and comparative object,comparing attributes and compare the results. Conditional random field theory is been putforward for the first time in2001, is a combination of the maximum entropy model and thecharacteristics of the hidden markov model, based on the analysis in recent years, and part ofspeech tagging, named entity recognition sequence annotation tasks such as obtain good effect, iswidely used in the field of information extraction. Respectively in this paper by using theconditions with the airport learning method and features of comparative sentences are extracted forecasting method and model training, extract compared other main component, and thesentence features of comparative sentences analysis comparative subject and comparative object,comparing attributes and compare the results.(3) Compare entity to make use of emotional dictionary tendentiousnessJudging comparative sentences tendency analysis is mainly reviews the comparative subjectand comparative object in comparative sentences emotional tendencies. In this paper,we are onthe basis of the former two parts of research work by building more emotional dictionary entityin the methods to calculate the comparative sentences. |
