Graph-Based Dialogue Modeling in Dialog Systems

In an age of rapid technological advancement and digital evolution, dialogue systems have become an integral part of everyday life. Modern conversational assistants are complex software entities capable of analysing human speech, understanding the context of the dialogue, and providing contextually appropriate responses. However, accurately predicting the intentions of dialogue participants remains a key challenge in the development of dialogue systems. This work presents an approach to conversational system design based on scenario-based dialogue graphs. Taking into account the nuances of domain-specific conversational systems, this methodology provides an automated framework for creating scenario-based dialogue graphs from unlabelled dialogue data. Thus, the decision-making process is streamlined and made more understandable, thereby increasing the transparency of predictions. Furthermore, the study explores the specifics of Russian-language dialogue data and develops a methodology that integrates text encoders and graph modelling techniques to effectively address the task of predicting next intentions in dialogue systems.