YNU-HPCC at SemEval-2019 Task 8: Using A LSTM-Attention Model for Fact-Checking in Community Forums

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 1180–1184 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 1180 YNU-HPCC at SemEval-2019 Task 8: Using A LSTM-Attention Model for Fact-Checking in Community Forums Peng Liu, Jin Wang and Xuejie Zhang School of Information Science and Engineering Yunnan University Kunming, P.R. China Contact:xjzhang@ynu.edu.cn Abstract The objective of the task, Fact-Checking in Community Forums , is to determine whether an answer to a factual question is true, false, or whether it even constitutes a proper answer. In this paper, we propose a system that uses a long short-term memory with attention mech- anism (LSTM-Attention) model to complete the task. The LSTM-Attention model uses two LSTM(Long Short-Term Memory) to extract the features of the question and answer pair. Then, each of the features is sequentially com- posed using the Attention mechanism, con- catenating the two vectors into one. Finally, the concatenated vector is used as input for the MLP (Multi-Layer Perceptron) and the MLP’s output layer uses the softmax function to clas- sify the provided answers into three categories. This model is capable of extracting the fea- tures of the question and answer pair well. The results show that the proposed system outper- forms the baseline algorithm. 1 Introduction Many questions pertaining to various fields are posted to QA forums by users every day, where they collect answers. However, the answers do not always address the question asked. Indeed, in some cases, the answer has nothing to do with the question. There are several reasons why this is the case. For example, the responder could have mis- understood the question and so provided a wrong answer. Most QA forums have little control over the quality of the answers posted. Moreover, in our dynamic world, the true answer was true in the past, but it may be false now . Figure 1 presents an example from the Qatar Living forum. In this case, all three answers could be considered to be good since they formally answer the question. Never- theless, a1 contains false information, whereas a2 and a3 are correct, as can be established from the official government website. Figure 1: An example from the Qatar Living forum In this study, we aim to solve the problem of de- tecting true factual information in online forums. Given a question requesting factual information, the goal is to classify the provided answers into the following categories. (i) Factual - True: The answer is true and can be proved by cross referencing with an external resource. (ii) Factual - False: The answer gives a factual response, but it is either false, partially false, or the responder is uncertain about their response. (iii) Non-Factual: The answer does not provide factual information relevant to the question; it is either an opinion or an advice that cannot be veri- fied. To the best of our knowledge, various ap- proaches have been proposed for the purposes of fact-checking in community forums (Mihaylova et al., 2018), such …

Content truncated to 500 words