Selecting Sentences versus Selecting Tree Constituents for Automatic Question Ranking

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2515–2525, Osaka, Japan, December 11-17 2016. Selecting Sentences versus Selecting Tree Constituents for Automatic Question Ranking Alberto Barr´on-Cede˜no and Giovanni Da San Martino and Salvatore Romeo and Alessandro Moschitti Qatar Computing Research Institute, Hamad bin Khalifa University Doha, Qatar {albarron, gmartino, sromeo, amoschitti}@qf.org.qa Abstract Community question answering (cQA) websites are focused on users who query questions onto an online forum, expecting for other users to provide them answers or suggestions. Unlike other social media, the length of the posted queries has no limits and queries tend to be multi-sentence elaborations combining context, actual questions, and irrelevant information. We approach the problem of question ranking: given a user’s new question, to retrieve those previously-posted questions which could be equivalent, or highly relevant. This could prevent the posting of nearly- duplicate questions and provide the user with instantaneous answers. For the first time in cQA, we address the selection of relevant text —both at sentence- and at constituent-level— for parse- tree-based representations. Our supervised models for text selection boost the performance of a tree kernel-based machine learning model, allowing it to overtake the current state of the art on a recently released cQA evaluation framework. 1 Introduction Community-driven question-and-answering (cQA) sites are popular forums in which users ask and an- swer questions on diverse topics. The freedom in these websites, and the diversity of their users, promote massive participation, resulting in large amounts of texts in the form of both questions and answers. All the steps in the assembly line of these highly-collaborative sites —from the question posting up to the seek for sensitive answers among those triggered by the question—, pose interesting natural language processing (NLP) challenges. When a user posts a query question, a model can search for previously- posted equivalent or relevant questions which might address the user’s information need at once. Once a number of comments have been posted intending to answer a question, another model can select the most appropriate one, or at least rank them on the basis of their quality. The same technology can be applied to discard inappropriate or diverting answers. When a website has accumulated a significant amount of posts, a model can be implemented to look for near-duplicates or related questions and answers. Different approaches have been proposed to address these tasks. Here we focus on the first of them: question re-ranking in cQA. That is, given a query question q and a pool of previously-posted questions D, rank the questions in D according to their relevance against q. This problem has attracted the attention of a manifold of research works which rely on the use of standard lexical similarity metrics, semantic representations, machine-translation models, tree kernels, or neural networks, to mention just some ex- amples of representations and models. Nevertheless, those approaches neglect one of the inherent facets of cQA sites: opposed to other social media, users are free to post potentially ill-formed texts of anarchic lengths. …

Content truncated to 500 words