I2C Huelva at SemEval-2023 Task 4: A Resampling and Transformers Approach to Identify Human Values behind Arguments

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1382–1387 July 13-14, 2023 ©2023 Association for Computational Linguistics I2C Huelva at SemEval-2023 Task 4: A Resampling and Transformers Approach to Identify Human Values behind Arguments Nordin El Balima Cordero, Jacinto Mata Vázquez, Victoria Pachón Álvarez, Abel Pichardo Estévez Escuela Técnica Superior de Ingeniería Universidad de Huelva (Spain) nordin.elbalima531@alu.uhu.es, mata@dti.uhu.es, vpachon@dti.uhu.es, abel.pichardo107@alu.uhu.es Abstract This paper presents the approaches proposed for I2C Group to address the SemEval-2023 Task 4: Identification of Human Values behind Arguments (ValueEval)" (Kiesel et al., 2023), whose goal is to classify 20 different categories of human values given a textual argument. The dataset of this task consists of one argument per line, including its unique argument ID, conclu- sion, stance of the premise towards the conclu- sion and the premise text. To indicate whether the argument draws or not on that category a bi- nary indication (1 or 0) is included. Participants can submit approaches that detect one, multi- ple, or all of these values in arguments. The task provides an opportunity for researchers to explore the use of automated techniques to identify human values in text and has potential applications in various domains such as social science, politics, and marketing. To deal with the imbalanced class distribution given, our ap- proach undersamples the data. Additionally, the three components of the argument (conclu- sion, stance and premise) are used for training. The system outperformed the BERT baseline according to official evaluation metrics, achiev- ing a f1 score of 0.46. 1 Introduction Human values refers to the beliefs, principles and standards that individuals or groups hold to be im- portant and worthwhile. These values guide peo- ple´s attitudes and behaviors; they can vary across cultures (Civitillo et al., 2019), communities, and individuals. According to a study of Global Values Survey (White et al., 2020), the most widely held values across the world are a sense of community, a sense of national pride, and a desire for social order. Other commonly held values include equal- ity, respect for others, and a desire for a peaceful world. In this context, the classification of human values in textual arguments is an important task in the field of Natural Language Processing. Understanding the values that underlie an argument could provide valuable insights into people´s beliefs, attitudes, and motivations. It could also be useful in various applications such as opinion analysis (Hemmatian and Sohrabi, 2019), argumentation mining, emo- tion recognition, and persuasive technology, among others. Despite its importance, the automatic classifi- cation of human values in arguments remains a challenging problem. The task requires the ability to identify values in a text, understand the argument structure, and make a binary judgment about the presence of a value in the argument. It aims to ad- vance the state of the art in human value classifica- tion in textual arguments , these textual arguments are compiled from the social science literature and described in detail in the accompanying ACL paper (Kiesel et al., …

Content truncated to 500 words