Transformer-based Text Segmentation for Indonesian: Benchmarking and Application to Abstractive Summarization

Abstrak

This study benchmarks transformer-based, topic-modeling, and classical methods for text segmentation in Indonesian. Experiments on the Indonesian Wikipedia dataset demonstrate that transformer-based embeddings outperform baselines in segmentation accuracy. A case study on abstractive summarization demonstrates that coherent segmentation improves human-judged summary quality. This work establishes the first benchmark for Indonesian text segmentation and highlights its benefits for downstream NLP tasks. The source code and datasets used in this study are publicly available on https://github.com/blacfli/indo_text_segmentation

Identifier

• OpenAlex: https://openalex.org/W7155636456
• DOI: https://doi.org/10.1145/3789418.3789430