计算机工程与应用2025，Vol.61Issue(24)：134-143,10.DOI:10.3778/j.issn.1002-8331.2409-0335
基于大语言模型的跨语言会话式商品推荐设计与实现
Cross-Lingual Session Recommender System Based on Large Language Models
摘要
Abstract
Using modal representations like text to build universal session recommender systems has attracted the attention of the research community.However,these researches are based on the premise that the text information obtained by the recommender system is in the same language.This paper presents a cross-lingual session recommender system,named CLSRec.By using both the pivot language and the target language,the method uses the text representation as input,does not use the item ID,performs text whitening and linear pooling,and provides it to the session recommendation backbone,and finally obtains the score for recommendation.In the pre-training phase,contrastive learning technology is used to learn general knowledge of resource-rich language data.In the fine-tuning phase,user behavior patterns on the target lan-guage data set are learned to ensure the generalization of the method and handle cold start scenarios.While learning on the recommended target,the semantic modeling information is retained as much as possible.Experiments on real-world multilingual e-commerce datasets,the method is significantly higher than the current baseline on three different language data on NDCG@10,NDCG@50,Recall@10 and Recall@50.The paper also explores the factors that affect cross-lingual session recommender systems through comprehensive comparative experiments.关键词
跨语言推荐/会话推荐/大语言模型/迁移学习/对比学习Key words
cross-lingual recommendation/session recommendation/large language models/transfer learning/contrastive learning分类
信息技术与安全科学引用本文复制引用
LI Yishan,SUN Peijie,ZHANG Min..基于大语言模型的跨语言会话式商品推荐设计与实现[J].计算机工程与应用,2025,61(24):134-143,10.基金项目
国家自然科学基金(U21B2026,62372260) （U21B2026,62372260）
泉城实验室(QCLZD202301). （QCLZD202301）
