大语言模型增强的知识图谱推荐算法在电子商务中的应用与推广

Applications and Deployment of Large Language Model-Enhanced Knowledge Graph Recommender Algorithm in E‑Commerce

摘要: 在电子商务快速发展的背景下，推荐系统已成为提升用户体验与促进平台增长的关键技术。传统方法面临交互稀疏、冷启动与长尾问题，限制了推荐性能与商业价值。知识图谱(KG)增强推荐通过建模结构化语义关系，缓解了部分挑战，但现有图神经网络方法仍受限于知识缺失、语义不足与计算冗余。本文提出一种融合大语言模型(LLM)与知识图谱的推荐框架，利用LLM对物品的本地知识子图进行自然语言理解与语义建模，生成高质量表示，避免多跳传播带来的效率与泛化瓶颈。该算法在电商场景中可有效捕捉商品属性与用户兴趣之间的深层关系，提升冷启动与长尾推荐能力，助力平台实现智能推荐优化与数字经济价值增长。

Abstract: Against the backdrop of rapid e-commerce development, recommender systems have become a key technology for enhancing user experience and driving platform growth. Traditional methods suffer from sparse user-item interactions, cold-start issues, and insufficient coverage of long-tail items, thereby constraining recommendation performance and commercial value. Knowledge graph (KG)-enhanced recommendation alleviates some of these challenges by modeling structured semantic relations; however, existing graph neural network approaches remain limited by knowledge incompleteness, coarse-grained semantics, and computational redundancy. This paper proposes a novel recommendation framework that integrates large language models (LLMs) with knowledge graphs. By applying LLMs to the natural-language understanding and semantic modeling of local knowledge subgraphs of items, the framework produces high-quality representations and avoids the efficiency and generalization bottlenecks caused by multi-hop propagation. In e-commerce scenarios, the proposed algorithm effectively captures deep relationships between product attributes and user interests, enhances recommendations for cold-start and long-tail items, and thereby supports intelligent recommendation optimization and the growth of digital economic value for online platforms.