Text Segmentation in Multi-Party Dialogues

The article discusses a full-fledged open tool that allows segmentation of multi-participant dialogues with minimal or no training texts; It is proven that the main problem of existing solutions is the inferiority of their tools, which serve only as a demonstration of possible solutions; a hybrid segmentation model is proposed, based on the use of BERT embeddings and some linguistic features unique to multi-participant dialogues; good model results are presented, which can be compared with the leading results described in works on this topic; the result of the work was the creation of a full-fledged tool that allows automatic segmentation reproduction.