Method for analyzing semantic dynamics of dialogues with result explanation in digital psychological support services

Abstract

The article addresses the problem of analyzing semantic dynamics of dialogues in digital psychological support services. The relevance of the study is determined by the fact that the meaning of a dialogue is formed not only by individual utterances, but also by a sequence of states, thematic transitions, emotional changes, communicative intentions, and contextual links between messages. The aim of the study is to develop a method for analyzing semantic dynamics of dialogues with result explanation based on a graph-context representation. The proposed method represents a dialogue as a trajectory of semantic states constructed from semantic utterance profiles, typed graph relations, local dialogue fragments, and a transition matrix between states. The method formalizes indicators of thematic and emotional change, transition coherence, state trajectory similarity, and rules for identifying dynamic situations, including coherent development, topic shift, state repetition, stagnation, and escalation fragments. To interpret the result, an explanatory subgraph is formed that contains the most significant utterances, topics, emotional markers, communicative intentions, states, and relationships between them. The experimental study was performed using the open MentalChat16K dataset. An experimental sample of 720 dialogues and 4800 utterances was formed. Comparison with textual and linear-context approaches showed that the graph-context model achieved a TransitionAcc value of 0.78 and a SeqSim value of 0.74. Adding the explanation mechanism increased Fidelity from 0.16 to 0.24 and reduced Compactness to 0.18. The obtained results confirm the feasibility of using the proposed method for automated analysis of dialogue development and generation of explainable results in digital psychological support services.