Research On Multi-features Hierarchical Answer Quality Evaluation Method

With the rapid development of web2.0, the online community-based Question Answering(CQA) has been widely used in the form of a user according to their needs to ask questions, the answers are given by other users, so that users can get a wider range of what they need. Thus the question and answer system are produced such as Answer Twitter, Yahoo, Facebook, Baidu know and other communities. However, in the community question and answer, the user can freely speech, will inevitably produce some low-quality information, these low-quality answers that appear in the QA system, seriously affecting the quality of QA system, therefore, how to evaluate the quality of the answer becomes particularly important.This thesis focuses on automatically evaluating the quality of user-generated answer-s in CQA portals. Specifically, this thesis mainly consists of the following parts.(1)Hierarchical classification modelThis thesis analyzes the characteristics and research status of community question answering system, according to previous studies, most of the existing methods based on text or non-text feature to evaluate the quality of the answers, were not noticed the different characteristics of the evaluation of the quality of the answers with the influence of the problem category vary this phenomenon. For example, for a yes-no class question, the answer mostly will be “yes” or “no”, for the fact that the class question, the answer will always be a number of terms, for the recommendations class question, the answer generally will be “I think”, “I feel” and a series of words. So you can take advantage of the question category analyze the quality of the answers. Based on that, this thesis presents a hierarchical classification model, the first analysis of the question type, according to the feature of text data, the question is classified according to the feature of syntactic structure. Experimental results show that this method can remove noise of question classification, feature extraction of a specific category focus and improve problem classification accuracy. Then extract the text, non-text, Language translation features, the number of links in answer feature of four categories, based on the influence of feature classification problem with this type differ objective phenomenon, using logistic regression algorithm to evaluate the quality of the answers of each type of question, achieve better test results. Finally, analyzes the main characteristics of the quality of the answers to all kinds of question.(2)Combined with sentiment analysis answers Quality Evaluation MethodThrough the analysis of existing research found that sentiment analysis played a key role in the community question answering system. However, the existing research can’t effectively analyze the emotion of the question and answer in the question and answer community system. In view of the deficiency of the prior research on the quality evaluation method, this paper proposes a method of evaluating the quality of the answer based on the emotion analysis. This method combines the machine learning and the emotion analysis method based on emotion dictionary. Through construction of emotional words, degree adverbs, negative words table, punctuation, and other features the user point of view, combined with emotional characteristics proposed new emotional value calculation formula, combining classifiers, the quality of the answer is evaluated by the supervised method. Experimental results show that better performance combined with sentiment analysis answers evaluation.