Методи XAI в мультимодальному аналізі Instagram-контенту для малого ритейлу

Abstract

This research proposes a hybrid Instagram data collection pipeline for small retail trend analysis, integrating official API metrics with web scraping. By utilizing a multimodal AI framework – BERT for semantic text analysis and CLIP for visual feature extraction – the system transforms unstructured social media streams into structured style embeddings and time-series proxy metrics. These outputs provide a robust mathematical foundation for intelligent decision support, enabling small enterprises to identify latent demand vectors and adapt assortment strategies to volatile market environments.