ICL00 at SemEval-2016 Task 3: Translation-Based Method for CQA System

Proceedings of SemEval-2016, pages 857–860, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics ICL00 at SemEval-2016 Task 3: Translation-Based Method for CQA System Minghua Zhang Yunfang Wu Institute of Computational Linguistics, Peking University Abstract We participate in the English subtask B and C at SemEval-2016 Task 3 “Community Ques- tion Answering”. This paper is concerned with the description of our participating system. We propose a ranking model that combines a trans- lation model with the cosine similarity-based method. Compared to the traditional bag of words method, the proposed model is more ef- fective because the relationships between words can be explicitly modeled through word- to-word translation probabilities. Experiments conducted on the official test data demonstrate that our proposed ranking method obtains promising results. 1 Introduction The SemEval-2016 Task 3 (Nakov et al., 2016) Community Question Answering (CQA) covers a full task on CQA and which is, therefore, closer to a real application. To facilitate the participation of non IR/QA scholar to the task, the search engine step is already carried out which means that the task organizers explicitly provides the set of potential answers to be reranked. That is to say, given a new question (aka original question) and the set of the first 10 related questions (retrieved by a search en- gine) in subtask B, our system will focus on rerank- ing the related questions according to their similar- ity with the original question. Similar to subtask B, in subtask C that is the main English subtask, Given a new question and the set of the first 10 related questions, each associated with its first 10 com- ments appearing in its thread, we will rerank the 100 comments (10 questions * 10 comments) according to their relevance with respect to the original ques- tion. Note that the subtask B will give us enough tools to solve the main subtask. And therefore, our paper will give an overall description of the system based on subtask B. In section 2, we will briefly discuss an important difference between the subtask C and subtask B in the course of reranking. However, the major challenge for subtask B, as for most of CQA systems, is the word mismatch be- tween the original question and the related question. For example, “Where I can buy good oil for mas- sage?” and “Is there any place I can find scented massage oils in Qatar?” are two very similar ques- tions, but they only have a few words in common. To solve the word mismatch problem, we focus on translation-based approaches in this paper. The remainder of this paper is organized as fol- lows. Section 2 introduces the methods used in the ranking model clearly. Section 3 presents the trans- lation probabilities Estimation. We will talk about an overview of the system in section 4. Section 5 presents the experimental results. In Section 6, we conclude with ideas for future research. 2 Ranking Approach In SemEval-2016 Task 3, the dataset file is a se- …

Content truncated to 500 words