ECNU at SemEval-2017 Task 3: Using Traditional and Deep Learning Methods to Address Community Question Answering Task

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 365–369, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics ECNU at SemEval-2017 Task 3: Using Traditional and Deep Learning Methods to Address Community Question Answering Task Guoshun Wu1, Yixuan Sheng1, Man Lan1,2∗, Yuanbin Wu1,2 1Department of Computer Science and Technology, East China Normal University, Shanghai, P.R.China 2Shanghai Key Laboratory of Multidimensional Information Processing 51141201064,51164500026@stu.ecnu.edu.cn mlan,ybwu@cs.ecnu.edu.cn Abstract This paper describes the systems we sub- mitted to the task 3 (Community Ques- tion Answering) in SemEval 2017 which contains three subtasks on english corpora, i.e., subtask A: Question-Comment Simi- larity, subtask B: Question-Question Sim- ilarity, and subtask C: Question-External Comment Similarity. For subtask A, we combined two different methods to rep- resent question-comment pair, i.e., super- vised model using traditional features and Convolutional Neural Network. For sub- task B, we utilized the information of s- nippets returned from Search Engine with question subject as query. For subtask C, we ranked the comments by multiplying the probability of the pair ”related ques- tion õ comment” being Good by the re- ciprocal rank of the related question. 1 Introduction The purpose of Community Question Answering task in SemEval 2017 (Nakov et al., 2017) is to provide a platform for finding good answers to new questions in a community-created discus- sion forum, where the main task (subtask C) is defined as follows: given a new question and a large collection of question-comment threads cre- ated by a user community, participants are re- quired to rank the comments that are most use- ful for answering the new question. Obvious- ly, this main task consists of two optional sub- tasks, i.e., Question-Comment Similarity (subtask A, also known as answer ranking), which is to re-rank comments/answers according to their rele- vance with respect to the question, and Question- Question Similarity (i.e., subtask B, also known as question retrieval), which is to retrieve the simi- lar questions according to their semantic similarity with respect to the original question. More, a new subtask: Multi-Domain Duplicate Detection Sub- task (i.e., subtask E) which is to identify duplicate questions in StackExchange has been added to Se- mEval 2017 task 3. To address subtask A, we explored a tradi- tional machine learning method which uses mul- tiple types of features, e.g., Word Match Fea- tures, Topic Model-based Features, and Lexical Semantic Similarity Features. Additionally, for subtask A, we also built a Convolutional Neural Network (CNN) model to learn joint representa- tion for question-comment (Q-C) pair. For sub- task B, we utilized the information of snippets re- turned from Search Engine with question subject as query, e.g., we counted the frequency of each word in each snippets list and added the words which appear in the subject of original question and the frequency is more than 1 to the subject of related question. Since subtask C can be re- garded as a joint work of the two above-mentioned subtasks, we ranked the comments by multiplying the probability of the …

Content truncated to 500 words