DUTH at SemEval-2019 Task 8: Part-Of-Speech Features for Question Classification

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1155–1159 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1155 DUTH at SemEval-2019 Task 8: Part-Of-Speech Features for Question Classification Anastasios Bairaktaris Symeon Symeonidis Avi Arampatzis Database and Information Retrieval research unit, Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi 67100, Greece {anasbair1,ssymeoni,avi}@ee.duth.gr Abstract This report describes the methods employed by the Democritus University of Thrace (DUTH) team for participating in SemEval- 2019 Task 8: Fact Checking in Commu- nity Question Answering Forums. Our team dealt only with Subtask A: Question Classi- fication. Our approach was based on shal- low natural language processing (NLP) pre- processing techniques to reduce noise in data, feature selection methods, and supervised ma- chine learning algorithms such as NearestCen- troid, Perceptron, and LinearSVC. To deter- mine the essential features, we were aided by exploratory data analysis and visualizations. In order to improve classification accuracy, we developed a customized list of stopwords, re- taining some opinion- and fact-denoting com- mon function words which would have been removed by standard stoplisting. Furthermore, we examined the usefulness of part-of-speech (POS) categories for the task; by trying to re- move nouns and adjectives, we found some ev- idence that verbs are a valuable POS category for the opinion-oriented question class. 1 Introduction The significance of Community Question Answer- ing (CQA) forums has risen in the past years. Such forums represent a modern need for infor- mation that comes with the abundance of online sources and the needs of millions of people for answers. Popular forums like StackOverflow, Ya- hoo! Answers, and Answers.com provide plat- forms for general or specific questions in a wide range of topics by users’ and also a community- based model for user interaction. The large numbers of questions and answers lo- cated in these forums generate many opportuni- ties for information retrieval and data mining ap- plications, such as query-intent detection, opinion mining, fake news classification, etc. (Tsur et al., 2016; Jo et al., 2018; Sethi, 2017). More ad- vanced applications do not only aim at analyzing opinions but—by categorizing the feelings of the Q&As—they may be able to detect inappropriate content such as hate speech and act accordingly (Karadzhov et al., 2017; Baly et al., 2018). The SemEval Task 8, Fact Checking in Com- munity Forums, aims to determine whether the answers that are provided for a question in a fo- rum are true or false. While answers to fact- oriented questions can be deemed true or false, opinion-oriented and socializing questions evoke answers for which a true/false categorization does not make much sense. As a result, determining the question type is a necessary first step. Conse- quently, the subtask A of SemEval Task 8 has the goal of classifying questions in three categories: opinion, factual, or socializing. The rest of this report is structured as follows. Section 2 reviews some previous studies for CQA classification. Section 3 describes our system, while Section 4 presents experiments and …

Content truncated to 500 words