Modeling of topological relationships between trends of social platforms X and YouTube for short-term forecasting using Telegram interface

Abstract

The article presents the results of the development and research of an intelligent system designed for monitoring and forecasting the popularity dynamics of hashtags and topics across social media platforms (X/Twitter, YouTube). The relevance of this study is driven by the rapid update rate of information flows, which necessitates automated tools for identifying emerging trends in real-time. The scientific novelty of the work lies in the application of Graph Neural Networks (GNN) to analyze non-linear relationships between social media objects, allowing for the integration of both temporal dynamics and the topological structure of topic interactions. The paper provides a detailed description of the system's modular architecture. The data acquisition module is implemented through a hybrid approach: utilizing the official YouTube Data API for video content analysis and web scraping mechanisms (BeautifulSoup) to retrieve hourly Twitter trend data via the trends24.in platform. A relational SQLite database is employed to store structured information and ensure rapid access to time series data. Special attention is paid to the data preprocessing stage, which includes the normalization of popularity metrics and the formation of a graph structure where hashtags serve as nodes and edges represent their co-occurrence within a single context. The mathematical foundation of the system is based on the Graph Convolutional Network (GCN) architecture. The study justifies the selection of GCNConv layers, which implement a feature aggregation mechanism from neighboring nodes, and the Adam adaptive optimizer, ensuring efficient model training on sparse data. In the experimental section, a comparative analysis was conducted between the developed GNN model and classical approaches (Linear Regression, RNN), as well as alternative graph-based methods (GAT, Node2Vec). Testing results using MSE and RMSE metrics confirmed the superior predictive accuracy of the GNN in short-term popularity forecasting tasks. The practical significance of the work is validated by the implementation of a user interface in the form of a Telegram bot, providing analytics visualization via the Matplotlib library.