Research On List Sorting Method Of User Tag And User Behavior In Community Q & A System

The early SE(search engine) which provides convenient ways of obtaining information for Internet users such as Baidu and Google, and the users could obtain the information quickly which is vital to them by inputting the keywords in the SE. Nevertheless, the conventional SE already can no longer meet the demands of the Internet users to get information quickly along with the information explosion on the Internet and the accelerated pace of life. However, the emergence of CQA (Community Question Answering) offset the lack of SE. The characteristics of interactivity and openness of CQA link the asker and answerer closely, which could meet the demands of Internet users to obtain the target knowledge more quickly and more directly. The answers ranking is vital in the CQA and the accuracy of answers ranking affect the quality of a CQA and the experience of users directly. In this paper, several major achievements made in the following areas for the answers ranking of the CQA:(1) Analysis the characteristics of CQA which effect the result of answers ranking The characteristics of CQA include tags and behaviors of user. The tags of user in the CQA include the level of answers, the areas of expertise of answers, the adoption rate of answers, the approved number of answers, the experience points of answers and the concerned words of answers etc. And the behaviors of user include the mark of asker, the mark of visitor and the frequently answered questions category of answers etc. The answers rank is accomplished combine above characteristics of CQA in this paper. The subsequent experiments showed that the rank methods can improve the effect of answers rank effectively when the characteristics of CQA are combined.(2) Utilise a Semi-Supervised Question Classification based on Ensemble Learning. The questions must be classified in order to match the interrelated labels such as the areas of expertise of answers and the concerned words of answers etc. The classification algorithm integrates classifiers by ensemble method, which trains a classification model combine semi-supervised learning method and training data which include small number of labeled data and a large number of unlabeled data. Then, utilizing the mode to classify new question, and the experiment indicate that the semi-supervised question classification method based on ensemble learning can improve the accuracy of question classification effectively.(3) Utilise a list sort method blend in the characteristics of tags and behaviors of user. The tags and behaviors of user are studied and analyzed in the CQA, then the characteristics of tags and behaviors of user which have utility value are selected to blend in the answers feature space. ListNet is used as the ranking method which selects Neural Networks as the model and Gradient Descent as the optimization method to structure ListNet ranking model which blends in characteristics of tags and behaviors of user. Then, the ranking mode is utilized to finish experiment combine the answers feature space, and the result of experiment show that the ListNet ranking model can improve effect of answers ranking obviously which blend in the characteristics of tags and behaviors of user.(4) Utilise a list sort method optimized based on tags and behaviors of user. The tags and behaviors of user are studied and analyzed in the CQA, then the tags and behaviors of user which have utility value are selected to optimize ListNet model. Then the ListNet ranking model is structured which is optimized based on tags and behaviors of user. Then, the ranking mode is utilized to finish experiment combine the answers feature space, and the result of experiment show that the ListNet ranking model can improve effect of answers ranking in different degree which optimized based on tags and behaviors of user.