HITSZ-ICRC: Exploiting Classification Approach for Answer Selection in Community Question Answering

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 196–202, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics HITSZ-ICRC: Exploiting Classification Approach for Answer Selection in Community Question Answering Yongshuai Hou, Cong Tan, Xiaolong Wang Yaoyun Zhang, Jun Xu and Qingcai Chen Key Laboratory of Network Oriented Intelligent Computation Department of Computer Science and Technology Harbin Institute of Technology Shenzhen Graduate School HIT Campus, The University Town of Shenzhen, Shenzhen, 518055, China {yongshuai.hou, viptancong}@gmail.com, wangxl@insun.hit.edu.cn {xiaoni5122, hit.xujun,qingcai.chen}@gmail.com Abstract This paper describes the participation of the HITSZ-ICRC team on the Answer Selection Challenge in SemEval-2015. Our team parti- cipated in English subtask A, English subtask B and Arabic task. Two approaches, ensemble learning and hierarchical classification were proposed for answer selection in each task. Bag-of-words features, lexical features and non-textual features were employed. For the Arabic task, features were extracted from both Arabic data and English data that translated from the Arabic data. Evaluation demonstrat- ed that the proposed methods were effective, achieving a macro-averaged F1 of 56.41% (rank 2nd) in English subtask A, 53.60 % (rank 3rd) in English subtask B and 67.70% (rank 3rd) in Arabic task, respectively. 1 Introduction In recent years, community question answering (CQA) systems are becoming more and more pop- ular on the Internet. By using CQA system, a user can post his/her question on CQA portal and re- ceive answers from other users. All users can post questions and answers on CQA portal freely. Al- though it makes CQA users to get answers easily, the answer quality evaluation becomes a challenge for questions with multiple answers. To reduce the inconvenient in going through plenty of candidate answers, it makes sense to evaluate the quality of answers and select high-quality answers automati- cally for CQA systems. As a consequently, the task of answer quality evaluation and answer selection in CQA have attracted more and more attention in recent years (Arai and Handayani, 2013; Shah and Pomerantz, 2010; Agichtein et al., 2008). The Answer Selection in CQA challenge was opened as one new task in SemEval-2015: SemEv- al-2015 Task 3 (Màrquez et al., 2015). It created a venue and provided annotated datasets for re- searchers to compare their methods for answer se- lection in CQA. This challenge consisted of Subtask A and Subtask B. Subtask A required par- ticipant system to classify answers as relevant, po- tentially useful and bad for each question. Subtask B required participant system to decide whether the answer to a YES_NO question should be Yes, No or Unsure based on the answer list. Subtask A was offered for two languages: English and Arabic. Data for the two languages was in different data set format. In remainder of this paper, Subtask A in English is abbreviated to English subtask A, Sub- task A in Arabic is abbreviated to Arabic task and Subtask B in English is abbreviated to English sub- task B. HITSZ-ICRC team participated in English sub- task A, English subtask B and Arabic task. This …

Content truncated to 500 words