<![CDATA[The increasing volume of conversational data on digital platforms necessitates effective automatic summarization methods. This study investigates the use of a pre-trained BART (Bidirectional and Auto-Regressive Transformers) model for abstractive dialogue summarization on the DialogSum dataset, which consists of 14,460 English dialogues. The model is fine-tuned using a sequence-to-sequence framework with systematic hyperparameter tuning, including variations in learning rate, beam size, and training epochs. The optimal configuration is achieved using a learning rate of 2e-5, a beam size of 6, and 5 training epochs. Model performance is evaluated using ROUGE and BERTScore metrics. The experimental results show that the proposed model attains an average ROUGE-L F1-score of 0.40, indicating moderate structural similarity between generated and reference summaries, and an average BERTScore F1-score of 0.69, reflecting strong semantic alignment. These findings suggest that fine-tuned BART is effective in preserving semantic relevance in abstractive dialogue summarization, although further improvements are required to enhance lexical precision and discourse-level coherence.]]>
Copyrights © 2026
