UM6P at SemEval-2023 Task 3: News genre classification based on transformers, graph convolution networks and number of sentences

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 856–861 July 13-14, 2023 ©2023 Association for Computational Linguistics UM6P at SemEval-2023 Task 3: News genre classification based on transformers, graph convolution networks and number of sentences Hamza Alami1, Abdessamad Benlahbib2, Abdelkader El Mahdaouy3, Ismail Berrada1 1School of Computer Science, Mohammed VI Polytechnic University, Morocco 2Laboratory of Informatics, Signals, Automatics, and Cognitivism (LISAC), Faculty of Sciences Dhar El Mahraz, Sidi Mohammed Ben Abdellah University, Morocco 3Modeling, Simulation and Data Analysis, Mohammed VI Polytechnic University, Morocco {firstname.lastname}@{um6p.ma1,3, usmba.ac.ma2} Abstract This paper presents our proposed approach for English document genre classification in the context of SemEval-2023 Task 3, Subtask 1. Our method uses an ensemble technique to combine four distinct model predictions: Longformer, RoBERTa, GCN, and a sentence number-based model. Each model is optimized on simple and easy-to-understand objectives. We provide snippets of code that define each model to make the reading experience better. Our method ranked 12th in document genre (subtask 1) classification for English texts. 1 Introduction Detecting the genre, framing, and persuasion tech- niques in online news articles is a crucial task in media analysis, with applications ranging from con- tent curation to political campaign analysis. With the increasing diversity of online news sources and the proliferation of multiple languages, the ability to perform such analysis in a multilingual setup is becoming ever more important. In this context, the authors (Piskorski et al., 2023) proposed a new competition in the context of the SemEval 2023 shared task, which aims to encourage the use of artificial intelligence (AI) in media analysis. The competition’s dataset is not publicly available and covers various aspects of what makes a text persua- sive, such as the genre (opinion, report, or satire), framing (which key aspects are highlighted), and rhetoric (persuasion techniques used to influence the reader). The shared task includes three subtasks: news genre categorization, framing detection, and persuasion techniques detection. The participants can choose to take part in any number of subtask- language pairs. The dataset is available in six lan- guages, and participants can train their systems using the data for all languages in a multilingual setup. As part of our participation in the SemEval 2023 task 3, we will be focusing on the News Genre Categorisation (subtask 1), working specifically in a monolingual setting with English language texts. Our aim is to develop and test systems that can accurately classify news articles according to their genre. Through our participation in this shared task, we hope to contribute to the growing field of AI-based media analytics. In this paper, we present our method for En- glish document genre classification, which com- bines four different models to achieve promising results. Our method integrates 1) Longformer and RoBERTa, two deep learning models based on Transformer architecture that have demonstrated strong performance in various NLP tasks (Vaswani et al., 2017; El Mekki et al., 2022, 2020), with 2) a Graph Convolutional Network (GCN) that cap- tures the relationships between different parts of the …

Content truncated to 500 words