Studies Of Chinese Opinion Question/Answer System Based On Specific Domain

The information contained in the internet is increasing everyday, and the amount is gigantic. The comments on products appear more and more usually on BBS, web forum and Web sites. At the same time, users want to gain useful comments on products from the gigantic data and information. The traditional Question/Answer system and key word based search engine always can't deal with both the opinion questions that the users ask about and the opinion sentences on internet: first, the traditional Q/A system can only deal with normal question lacks the ability to deal with opinion question. Second, the traditional Q/A system's IR model always use key word based methods which lacks the ability to solve the opinion sentences on internet.In this paper, we design an opinioned Question/Answer system that aims to solve the problems. The system mainly contains three parts: question analyzing module, answer database generation module and answer extraction module. The system downloads opinion sentences from web forums and uses them to create the language corpus. The question analyzing module models the opinion question as . QTopic represents the topic of the question, QSentiment represents the sentiment of the question and the QType represents the type of the question. Besides, the module proposes the opinion question classifying standard for the first time.When question comes, the question analyzing module utilizes pattern matching method together with language model algorithm to analyze the opinion question and generates the classifying results; secondly the answer database generation module use conditional random field(CRF) model the identify the complex domain word and then identity the claims of the opinion sentences of the language corpus using word matrix algorithm which effectively utilizes the information contained in the dependency analyzing results while avoids the bad effects that generate from the error of the dependency analyzing results, at last the results are saved to answer database; the answer extraction module aims to generate the final result of the Q/A system, the module use concept matching algorithm to match claims in answer database and question analyzing result.