EMPLOYING COMPUTATIONAL LINGUISTIC TECHNOLOGIES AND OCULOGRAPHY TO DEVELOP DIAGNOSTIC TOOL FOR DETECTING AUTOAGGRESSIVE TENDENCIES IN YOUNG PEOPLE: A RIVETED GAZE INTO "GET RID OF THE SHACKLES OF THIS WORLD".

Abstract

Background: Early recognition of autoaggressive tendencies in young people is essential for diagnostic screening and reducing suicidality risks. This can be achieved through psycholinguistic approaches such as corpus analysis and eye-tracking studies. Corpus research helps to develop generalized speech patterns of those at risk of suicide, while oculographic methods examine perceptual cues linked to suicidal tendencies.

Methods: We formulated an algorithmic framework for constructing verbal, visual, and multimodal material to identify autoaggressive tendencies among youth. The stimuli material was created following the idiolect paradigm of forensic authorship attribution. The first stage involved analyzing corpus data including materials from social networks and social media, the Rusentiment database, and a text collection from the Privolzhsky Research Medical University. Python's NLTK and SpaCy libraries for automated text processing were used to extract corpus statistics, n-grams, keywords, and collocations for identifying linguistic markers of autoaggression. Keywords were statistically ranked using Log-likelihood, T-score, and mutual information, while collocations were derived via T-score analysis. Sentiment analysis for the Dostoevsky Python library and stylistic indices (lexical diversity, readability) were also applied. The total analyzed material comprised more than 100 million tokens. We next integrated, stimulus and filler materials into an eye-tracking application (developed by LLC Lad IT Group) using standard laptop video cameras. Oculographic data quantified gaze delay differences via a percentage excess formula to pinpoint the most diagnostically relevant stimuli. In two iterations of the pilot experiment, 66 youths from the control group and 29 from the target group participated in the oculographic experiments.

Results: In multimodal texts, most stimuli derived from corpus statistics were relevant, and all individuals in the target group showed a prolonged gaze delay; visual stimuli (pseudo-self-portraits, anime/game characters) elicited 26-36% longer gaze delay in the target group. Verbal stimuli analysis revealed prolonged gaze fixations on self-referential pronouns (12-25%) and metaphorical death expressions, although direct terms, like "suicide" showed the gaze avoidance (-11.9 to -129% deviation). We then developed a system of weighted coefficients for an automated diagnostic model. The algorithm showed 72 % accuracy in identifying autoaggression, presenting a promising tool for early diagnostic screening of this phenomenon.

Conclusions: The present methodology focuses on creating and employing a novel selective dataset consisting of visual, linguistic, and multimodal text stimuli integrated into the oculographic examination protocol. The oculographic detection of eye movement perceptual cues in response to exposure to the stimuli dataset may identify objective markers for evidence-based diagnostics of mental disorders (e.g., depression) and fundamental psychopathological phenomena (e.g., suicidality), including at-risk states (e.g., autoaggression). Furthermore, this approach may contribute to the enhancement of suicide prevention programs, particularly targeted interventions for the vulnerable population of young people who experience autoaggressive tendencies (i.e., self-aggression).