Team JUSTR00 at SemEval-2023 Task 3: Transformers for News Articles Classification

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 121–124 July 13-14, 2023 ©2023 Association for Computational Linguistics NLP-LISAC at SemEval-2023 Task 9: Multilingual Tweet Intimacy Analysis via a Transformer-based approach and Data Augmentation Abdessamad Benlahbib1 , Hamza Alami2 , Achraf Boumhidi1, Omar Benslimane3 1 LISAC Laboratory, Faculty of Sciences Dhar EL Mehraz, USMBA, Fez, Morocco 2 School of Computer Sciences, Mohammed VI Polytechnic University, Ben Guerir, Morocco 3 LAMA Laboratory, Faculty of Sciences Dhar EL Mehraz, USMBA, Fez, Morocco abdessamad.benlahbib@usmba.ac.ma , hamza.alami@um6p.ma achraf.boumhidi@usmba.ac.ma, omar.benslimane@usmba.ac.ma Abstract This paper presents our system and findings for SemEval 2023 Task 9 Tweet Intimacy Analysis. The main objective of this task was to predict the intimacy of tweets in 10 languages. Our submitted model (ranked 28/45) consists of a transformer-based approach with data augmen- tation via machine translation. 1 Introduction According to the Cambridge Dictionary 1, intimacy is things that are said or done only by people who have a close relationship with each other. Tweet Intimacy is a term that is used to describe the re- lationship between Twitter users and the way they interact with one another through the social media platform. It refers to the level of personal connec- tion and familiarity that exists between individuals who engage with each other on Twitter. This can be determined by various factors, such as the fre- quency of interaction, the types of content shared, and the use of personal language or inside jokes. A high level of tweet intimacy suggests that the indi- viduals have a strong, personal relationship, while a low level of tweet intimacy indicates more casual or impersonal interactions. In this paper, we present our findings on Se- mEval 2023 Task 9: Multilingual Tweet Intimacy Analysis (Pei et al., 2023). Our method consists of a transformer-based approach (Vaswani et al., 2017) along with data augmentation. The rest of the paper is structured in the follow- ing manner: Section 2 provides the main objective of the Task. Section 3 describes our system. Sec- tion 4 details the experiments. And finally, Section 5 concludes this paper. 1https://dictionary.cambridge.org/ dictionary/english/intimacy 2 Task Description This task aims at using a collection of tweets in six languages (English, Spanish, Italian, Portuguese, French, and Chinese), which have been marked with scores ranging from 1 to 5 based on their level of intimacy to predict the intimacy of tweets in 10 different languages. The model’s effectiveness will be assessed on the test set consisting of the aforementioned six languages, as well as an ad- ditional test set that contains four languages not present in the training data (Hindi, Arabic, Dutch, and Korean). 3 System Description To tackle the SemEval 2023 Task 9: Multilingual Tweet Intimacy Analysis, we have fine-tuned a transformer-based model on an augmented train- ing dataset. The different steps of our system are described as follows: • We have translated the English tweets in the training dataset to Hindi, Arabic, Dutch, and Korean using the Google translate API for Python: Googletrans 2 by …

Content truncated to 500 words