Get the latest tech news
TreeSeg: Hierarchical Topic Segmentation of Large Transcripts
The Augmend blog covers information on our products as well as related areas in the AI + Collaborative developer tools space. Feel free to reach us on twitter if you have comments or thoughts on any of our posts. We love to hear from others passionate about this stuff!
The BERT-based approach of Solbiati et al. (2021)(henceforth referred to as BertSeg) is geared towards segmenting meeting transcripts and can be regarded as a modern version of TextTiling. TreeSeg is motivated by an idealized UI that provides such an affordance to the user ; the ability to transition from coarse to fine chapter segmentations seamlessly and to choose the desired level of granularity. In a future post I will describe how we use the resulting segment tree to partition large transcripts for downstream tasks, providing a scaffold to move between local and global context.
Or read this on Hacker News