List your company

Topic Segmentation on Conc-Wiki 1.0 (test)

46Boundary Similarity (n=2)BERT
17.9225.2132.539.79Dec 27, 2025

Evaluation results

MethodLinks
BERTthreshold=0.3, embedding=SentenceBERT2025.12469150
BERTGraphclustering=graph-based, embedding=SentenceBERT2025.12457954
FlanT5backbone=Flan-T5 large (780M), topk=1, training=LoRA fine-tuned2025.12416563
GPT3.5model=gpt-3.5-turbo-16k, topk=1, prompting=overlapping and recursive2025.12294263
Split5rule=boundary every 5 sentences2025.12192552