PENDEKATAN HIBRIDA PERINGKASAN TRANSKRIPSI VIDEO DENGAN KLASTERISASI HIERARCHICAL BERBASIS BERT DAN BART

The use of video as a medium for conveying information and education continues to grow, but long durations and varied conversational structures often make it difficult to grasp the core content. This poses a challenge in the development of automatic summarization methods, particularly for monologues, interviews, podcasts, and short interviews. Extractive methods generally
retain important information but lack coherence, while abstractive methods tend to be more concise but risk losing context. To address these issues, this study proposes a hybrid approach that
combines extractive summarization based on sentence representation using BERT and hierarchical clustering, with abstractive summarization using the BART. Evaluation results show that the hybrid approach delivers competitive performance and tends to excel on content with complex conversational structures, such as interviews and podcasts, with average ROUGE-1 scores of 0.30 and 0.32, respectively. This approach is also capable of producing summaries that are more concise and informative compared to single-method approaches, although for shorter and more structured content such as monologues and short interviews, the difference is not always significant. These findings confirm that the hybrid approach has strong potential as an adaptive video transcription summarization solution for various content characteristics.