华中科技大学学报（自然科学版）2025，Vol.53Issue(10)：29-35,7.DOI:10.13245/j.hust.251094
基于全局特征增强的新闻推荐模型
Global feature enhancement for news recommendation model
摘要
Abstract
To address the problem that personalized news recommendation systems in recent years primarily relied on users' own browsing history,lacked a global perspective,and failed to fully consider users' potential interests and complex behavioral patterns beyond semantic information,a news recommendation model called GREEN(global feature enhancement for news recommendation)was proposed.This model could utilize other users' browsing histories to learn global news features and combine them with users' local news representations to enhance the model's personalized recommendation capability.A global news encoder was constructed,which used gated graph neural networks to learn two types of global news representations and fused various news features through a historical news aggregator.Similarly,this approach was extended to a global candidate news encoder that utilized a global entity network and candidate news aggregator to enhance candidate news features.Through evaluation on the public news dataset MIND-small,it is confirmed that the proposed model outperforms existing methods.关键词
新闻推荐模型/特征增强/全局新闻特征/图神经网络/注意力机制Key words
news recommendation model/feature enhancement/global news features/graph neural networks/attention mechanism分类
信息技术与安全科学引用本文复制引用
刘莉,王徽,杨亮,李龙杰,马笠恭..基于全局特征增强的新闻推荐模型[J].华中科技大学学报（自然科学版）,2025,53(10):29-35,7.基金项目
甘肃省科技计划资助项目(23YFGA0005,21ZD8RA008). （23YFGA0005,21ZD8RA008）
