Enhancing Idiom Classification on Text-based Conversational Data: Evaluating Transformer Model Performance through k-fold Cross-Validation

This study focuses on enhancing the classification of idioms within conversational data by leveraging transformer models. Idioms pose a challenges in natural language understanding due to their figurative nature and context-dependent meaning. Transformer models, such as BERT and T5, which are well-known for their ability to capture complex linguistic patterns and have been extensively explored for this task. The study incorporates embedding methods with these models and observes the results: a rigorous evaluation methodology, k-fold cross-validation, assesses of their performance. The dataset is divided into multiple folds, allowing for comprehensive training and evaluation of the models. This evaluation approach provides reliable and robust measures of the transformer models' effectiveness in accurately classifying idioms in conversational contexts. The findings from this study contribute to advancing the understanding of idiomatic expressions and demonstrate the potential of transformer models for improving idiom classification in conversational data. The transformer model BERT with ELMo embedding and the T5 model achieved better performance than the BERT base model.