CSECU-DSG at SemEval-2023 Task 4: Fine-tuning DeBERTa Transformer Model with Cross-fold Training and Multi-sample Dropout for Human Values Identification

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1988–1994 July 13-14, 2023 ©2023 Association for Computational Linguistics CSECU-DSG at SemEval-2023 Task 4: Fine-tuning DeBERTa Transformer Model with Cross-fold Training and Multi-sample Dropout for Human Values Identification Abdul Aziz, Md. Akram Hossain, and Abu Nowshed Chy Department of Computer Science and Engineering University of Chittagong, Chattogram-4331, Bangladesh {aziz.abdul.cu, akram.hossain.cse.cu}@gmail.com, and nowshed@cu.ac.bd Abstract Human values identification from a set of argu- ment is becoming a prominent area of research in argument mining. Among some options, val- ues convey what may be the most desirable and widely accepted answer. The diversity of human beliefs, random texture, and implicit meaning within the arguments make it more difficult to identify human values from the argu- ments. To address these challenges, SemEval- 2023 Task 4 introduced a shared task ValueEval focusing on identifying human value categories based on given arguments. This paper presents our participation in this task where we propose a fine-tuned DeBERTa transformers-based clas- sification approach to identify the desired hu- man value category. We utilize different train- ing strategies with the fine-tuned DeBERTa model to enhance contextual representation on this downstream task. Our proposed method achieved competitive performance among the participants’ methods. 1 Introduction In human decision-making, values play a crucial role. The meaning of values often refers to any individual’s personal beliefs and priorities about any argument which tends to go to the conclusion of that argument (Teze et al., 2019). The conclu- sion may vary from person to person though the arguments are the same. Identifying human value categories which drive a human being to conclude an argument and make perspective decisions is very crucial to build a value-driven artificial intelligence (AI) agent to mitigate future security risks from those AI agents. Nevertheless, it is beneficial for various NLP tasks also including argument mining, sentiment analysis, political discourse analysis, per- sonality analysis, and marketing analysis. However, identifying human values from the ar- guments is the most challenging and prominent **The first two authors have equal contributions. task in NLP. Because of the large variety and het- erogeneity of personal beliefs which may be am- bivalent for the same group of people regarding their personal experiences, this task becomes chal- lenging. To address the challenges of human values identification from arguments Kiesel et al. (2023) introduce a shared task ValueEval at SemEval- 2023 1. The task is composed of three categories of tracks including the argument test, Nahj al-Balagha test, and New York Times test datasets. Here, the argument test considers the main dataset for this task. To categorise the arguments, this task articu- lates 20 fixed categories of values which are shown in Figure 1. S el f- di re ct io n: P o w er : S e c ur it y: C o nf or m it y: B e n e v ol e n c e: U ni v er s al is m : thought action Stimulation Hedonism Achievement dominance resources …

Content truncated to 500 words