Exploring domain specific feature weights for question classification enhancement in community-based QA services

With the rapidly developing of Web2.0, community-based question answering (cQA) services have become extremely popular knowledge and information acquisition systems. Over times, there is tremendous questions and answers with high quality devoted by human intelligence which is so called user generated content (UGC). Hence, how to correctly and effectively categorize new coming questions becomes a nontrivial task. In this paper, we propose a domain specific feature weighting scheme for the existing supervised machine learning-based classification approaches to cQA question classification. We use Baiduzhidao taxonomy as hierarchical categories which include 14 main categories and 169 sub-categories. Experimental results show that the supervised machine learning-based classifiers can be enhanced by the proposed feature weighting scheme. Further, we check the influences of different question parts on question classification task and the results show that better performance has achieved when using question title and description as features. Thus we essentially complete the feature selection progress.