Unveiling Human Values: The Heart of Arguments

An experimental study is presented in which human values are classified based on arguments. The identification of human values behind arguments involves the understanding of the underlying principles or beliefs that are held and utilized by people to support their opinions. This process encompasses common abstractions, whether in debate, discussion, or everyday conversation. The aim of this task is to achieve an efficient solution for the detection of human values behind arguments by implementing several models and determining the best approach. The full list of models used in experimentation includes machine learning models namely SVM, Naive Bayes, Random Forest, and transformer models such as BERT-base, BERT-large, and roBERTa-base, as well as the utilization of the ELMo model for classification. Among these models BERT-large gave best result of 0.45 as macro average F1 score with 0.46 as precision and recall followed by BERT-base and RoBERTa-base with 0.44 as macro average F1 scores both. 0.35 was the macro average F1 score obtained by SVM model with which we started as our base model. The results were obtained from these models according to their nature, strengths, and weaknesses. The best result was attained through the utilization of the BERT-large model. Insights gained from this study can potentially be used for further advancements in the field of human language analysis.