PREVENTING DESTRUCTIVE ACTIONS IN USERS TEXTS WITH PTSD USING NATURAL LANGUAGE PROCESSING METHODS

Abstract

The study addresses the problem of automated detection of destructive actions in texts produced by users with PTSD using natural language processing methods. The article proposes a method for automated detection and prevention of destructive actions among users with PTSD based on a multi-step analysis of their textual messages. The architecture of the method includes sequential identification of PTSD indicators, detection of comorbid mental states (depression, anxiety, panic, narcissism, anger), determination of emotional characteristics of the content, and calculation of integral indicators of susceptibility to destructive actions. An important component of the approach is the construction of temporal visualizations that enable tracking of risk dynamics and identification of critical periods of elevated threat. The method was tested on the dataset “Depression: Twitter Dataset + Feature Extraction”, which contains 20,000 labeled English-language tweets suitable for temporal analysis. The dataset’s existing annotations are consistent with the results obtained by the proposed method.

The scientific novelty of the work lies in the comprehensive integration of NLP analysis, neural network models, and time series visualization for generating an integral risk assessment of destructive actions. The practical significance is determined by the potential application of the proposed method in psychodiagnostic systems, early risk detection tools, and preventive mental health support solutions. The obtained results demonstrate the feasibility and effectiveness of the approach and highlight future directions for development, including the integration of multimodal data and adaptation of models to the Ukrainian language environment. The study is also aligned with the United Nations Sustainable Development Goals, contributing to the promotion of human well-being, health, and safety in the context of contemporary social challenges.