A Lightweight AI Framework for Personalized Product Recommendation Using Small Language Models and Behavioral Intelligence

Abstract

Personalized recommendation has emerged as an essential component of modern e-commerce platforms for improving customer engagement, satisfaction, and purchasing decisions. However, conventional recommendation techniques are often constrained by sparse user interactions, cold-start problems, limited contextual understanding, and the substantial computational cost of Large Language Models (LLMs). To address these challenges, this study proposes an AI-Driven Personalized Shopping Recommendation Framework that integrates user behavior analytics with Small Language Models (SLMs) for efficient and context-aware product recommendation. The proposed framework analyzes purchase frequency, browsing patterns, search history, dwell time, category preferences, brand affinity, and price sensitivity to construct dynamic user preference profiles. SLMs facilitate contextual preference summarization, semantic product understanding, and personalized recommendation generation. A hybrid ranking mechanism combines behavioral relevance, semantic similarity, SLM-derived preference scores, and product popularity, while an explainability layer provides recommendation rationales and confidence scores. Experimental evaluation demonstrates that the proposed framework can improve recommendation quality while reducing inference latency and computational resource requirements, highlighting its potential for scalable and practical deployment in real-world e-commerce environments.