RNN-Based Approach to Cognitive Architecture of Subject Behavior in Digital Environment
Year: 2026
Type: conference-paper
Abstract: This paper proposes an approach to development of a dynamic model of subject’s behavior in digital environment based on recurrent neural networks. The study was conducted as part of a project aimed at creating a cognitive architecture that models the interaction of users in digital environments. Previously in the project a static model linking the content of text messages to 3 user behavior types via 60 behavior keys extracted from text data using a large language model was developed. The current study extends this model by shifting from analysis of individual posts to analysis of comment chains and prediction of an author's behavior keys during a reply, taking the dialogue history into account. Neural network architecture for regression based on Long Short-Term Memory with Attention mechanism is proposed to capture the context of comment chains. Two model variants were trained: one using the full set of keys and another using a reduced set comprising the most represented keys in the dataset. The results demonstrated that excluding underrepresented keys increased R2(coefficient of determination), indicating a reduction in the noise introduced by rare features. Limitations of the current dataset were identified, highlighting the need for expansion to improve the model's generalization capability. The resulting model is integrated into the project's cognitive architecture and can be applied in psychological studies of digital communication, as well as in the development of dialogue systems that simulate human behavior.
Institution: Kazan Federal University
Language: English
Topic: Mental Health via Writing
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
Funder: Russian Science Foundation
OpenAlex ID: W7224760984
