Lee at SemEval-2020 Task 5: ALBERT Model Based on the Maximum Ensemble Strategy and Different Data Sampling Methods for Detecting Counterfactual Statements

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 664–669 Barcelona, Spain (Online), December 12, 2020. 664 Lee at SemEval-2020 Task 5: ALBERT model based on the maximum ensemble strategy and different data sampling methods for detecting counterfactual statements Junyi Li, Yuhang Wu, Bin Wang, Haiyan Ding∗ School of Information Science and Engineering Yunnan University, Kunming 65091, P.R. China ∗Corresponding author, hyding@ynu.edu.cn Abstract This article describes the system submitted to SemEval 2020 Task 5: Modelling Causal Reasoning in Language: Detecting Counterfactuals. In this task, we only participate in the subtask A which is detecting counterfactual statements. In order to solve this sub-task, first of all, because of the problem of data balance, we use the undersampling and oversampling methods to process the data set. Second, we used the ALBERT model and the maximum ensemble method based on the ALBERT model. Our methods achieved a F1 score of 0.85 in subtask A. 1 Introduction Now, with the rapid development of Internet technology, our society has entered a highly information era. Life is full of different kinds of information. However, the messy information is likely to change or even become the wrong information during the non-stop transmission process. This will have some bad effects. In order to solve this problem, we have created some positions where the verification information is right or wrong. However, this will consume a lot of labor costs and cause a waste of resources. At the same time, there are usually some misjudgments in human detection. Therefore, we need some new methods to deal. In the past two years, with the rapid development of artificial intelligence, some related problems have emerged in the field of deep learning. For example, common sense discrimination and natural language inference. The emergence of these problems provides some mechanical methods to solve the problem of error messages. Our research is also carried out. Semeval 2020 task 5 (Yang et al., 2020) is to model causal inference in language: detect counterfactuals. This task is through the establishment of a model, and at the same time to detect whether it is a factual statement based on a large amount of text data. The generation of this task can help us to make a machine judgment of information correctness to a great extent. Subtask A is to detect counterfactual statements. This subtask requires us to determine whether a given statement is counterfactual. A counterfactual statement describes an event that has not actually occurred or cannot occur, and the possible consequences of the event. More specifically, counterfactuals describe events that are contrary to facts, and therefore naturally contain common sense, knowledge, and reasoning. Solving this problem is the basis of using natural language for all downstream counterfactual causal reasoning analysis. Subtask B is to detect precedents and consequences. This subtask states that causal insights are inherent features of counterfactuals. For this task, we mainly participated in subtask A. In this task, we only participate in subtask A: Detect counterfactual statements. For this task, we use a …

Content truncated to 500 words