NLUBot101 at SemEval-2023 Task 3: An Augmented Multilingual NLI Approach Towards Online News Persuasion Techniques Detection

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1636–1643 July 13-14, 2023 ©2023 Association for Computational Linguistics NLUBot101 at SemEval-2023 Task 3: An Augmented Multilingual NLI Approach Towards Online News Persuasion Techniques Detection Genglin Liu, Yi R. Fung, Heng Ji University of Illinois Urbana-Champaign {genglin2, yifung2, hengji}@illinois.edu Abstract We describe our submission to SemEval 2023 Task 3, specifically the subtask on persuasion technique detection. In this work, our team tackled a novel task of classifying persuasion techniques in online news articles at a para- graph level. The low-resource multilingual datasets, along with the imbalanced label dis- tribution, make this task challenging. Our team presented a cross-lingual data augmentation ap- proach and leveraged a recently proposed mul- tilingual natural language inference model to address these challenges. Our solution achieves the highest macro-F1 score for the English task, and top 5 micro-F1 scores on both the English and Russian leaderboards. We have made the source code of our models and experiments publically available at 1. 1 Introduction We describe UIUC BLENDER Lab’s participation in SemEval 2023 Task 3. Detecting persuasion techniques in multilingual online news articles is valuable for several reasons. The rise of digital media and social networks has led to an increase in the amount of news and information available, making it difficult for individuals to navigate the vast amount of information and identify credible sources. By detecting persuasion techniques in news articles, NLP can help individuals distinguish between objective reporting and biased reporting, become more aware of these techniques and make more informed decisions. From a journalist’s per- spective, detecting persuasion techniques in online news articles using an automated system can help them and editors identify and remove any biased language, leading to more objective reporting and higher-quality journalism. Regarding detecting per- suasion techniques in online news articles, it is useful in a multilingual setting for several reasons. 1https://github.com/yrF1/ SemEval23-Task-3-UIUC-Team/tree/main Types of Persuasion Techniques 0.00 0.25 0.50 0.75 1.00 Whataboutism' Guilt_by_Association' Appeal_to_Hypocrisy' Straw_Man' Obfuscation- Red_Herring' Causal_Oversimplifica Conversation_Killer' Appeal_to_Authority' Slogans' Appeal_to_Popularity' False_Dilemma- Flag_Waving' Exaggeration- Appeal_to_Fear- Repetition' Doubt' Name_Calling- Loaded_Language' f1-score normalized-frequency Figure 1: F1-score on English Dev Set and Normalized Frequency for Each Label In today’s interconnected world, multilingual NLP can help identify the persuasion techniques used in news articles in different languages to understand the global media landscape better. Multilingual NLP can help identify patterns and common tech- niques used to spread propaganda and misinforma- tion in different languages and help combat these issues. SemEval 2023 released a new dataset covering several aspects of what makes a text passage per- suasive. Our training data comes from articles in six languages: English, French, German, Italian, Polish, and Russian. They were collected from 2020 to the middle of 2022, focusing on a set range of topics. The details are presented by the task organizers (Piskorski et al., 2023). The task pro- vides news articles in six languages, and our team decided to focus on the persuasion technique de- tection subtask since it is unique, challenging, and aligns …

Content truncated to 500 words