Design of an LLM-Driven Personalized Learning Resource Recommendation System

-- A Comparative Study

Abstract

Recently, the demand for personalized learning resources has increased substantially. Traditional recommendation algorithms, such as collaborative filtering (CF), often struggle with cold-start scenarios and data sparsity, limiting their applicability in new platforms. In order to address these challenges, this study proposes a personalized learning recommendation system based on large language models (LLMs), which use their advanced understanding capabilities to analyze user intent and generate recommendations. The system architecture employs the Spring Boot and Vue frameworks for efficient frontend-backend development, while the LangChain framework on LLM interactions. The integration of fine-tuned prompts with user-generated tags and historical learning data enables the platform to dynamically generate recommendations through APIs of LLMs, such as ChatGPT and DeepSeek. This approach effectively addresses the issue of suboptimal recommendation accuracy caused by data sparsity within the platform. However, limitations such as hallucination phenomena and API cost scalability necessitate further optimization. It is evident that the platform will require continuous optimisation in the future.