Study On The Analysis For Chinese Opinion Question And Comparative QA

Most of the studies for traditional Question Answering (QA) serve the needs of factual question. And now the study for opinion question is still very rare at home and abroad. Opinion question is a question that queries information about what people think or feel towards to some topics or events. It can express the user's thoughts and opinions more directly, so it is prevalent in everyday life. The study of Opinion QA system needs not only the traditional QA system's technology, but also the technology of Opinion Mining and comparative relation mining, and so on.In this paper, we explore and study the analysis for Chinese opinion question and Chinese comparative Opinion QA. At first, we need to identify the opinion question. We classify fact question and opinion question by the use of SVM and Na?veBayes classifier and achieve good results. Then, we analyze Chinese opinion question from syntax and semantic level, and also extract the question's other key information. In syntax level, we classify the question into four categories by C4.5 algorithm, such as yes-or-no interrogative, specially related interrogative, and so on. In semantic level, corresponding to the definition of Opinion, we define a Classification System for Chinese opinion question, including six main categories, such as Holder, Topic, Sentiment, Time, Location and Reason. And every category also includes several types. For the semantic main category classification, we take HowNet and the related attributes of interrogatives as feature, which reduces the dimension of feature space but also obtains good results.Among opinion questions, there is a kind of opinion question that asks people's opinion with comparison between different targets, and it was called comparative opinion question. Because Comparative opinion questions and their answers are often comparative sentences, in this paper we also explore and study Chinese comparative sentences, including comparative sentence identifying and comparative relation mining. We use CRF model to extract domain entities which are applied to identify comparative sentences and mine comparative relation with good results. And finally with the former study on Chinese comparative sentences, we analyze the general process for Chinese comparative Opinion QA.