Application of NLP Methods for Analyzing the Mental State of a Conversational Partner in Text–Based Communication

The paper considers the application of modern methods of natural language processing (NLP), in particular large language models (LLM), for the analysis of a person’s psycho-emotional state based on text messages. The purpose of the study is to apply, compare and evaluate the practical applicability of NLP methods for the analysis of the interlocutor’s psycho-emotional state in text communication. Open datasets with expert markup were used for the experiments. Particular attention was paid to the issues of interpretability of model responses using LLM, therefore, a dialogue agent based on ReAct-prompting and using pre-trained models solving the above-described problems was implemented to test the system in various scenarios and form the final interpretation. A web service was developed for the agent to operate. The results of the experiments show that the adapted LLM models demonstrate high quality in the tasks of detecting depression and suicidal tendencies, and the proposed agent can be used as a tool for primary psychological assessment and identification of risk factors.