摘要
针对RippleNet模型推荐结果中存在的用户兴趣属性失真问题,提出一种利用共有属性采样改进的推荐模型Ripple-Net-CA。首先,该模型通过共有属性采样替代原模型的随机采样,来重构知识图谱上的扩展偏好集合(RippleSet),基于增加RippleSet内部节点间的共有属性频数、增强多跳扩展偏好之间的相关性,来提高用户特征包含的信息量。此外,将用户特征和物品特征送入模型,计算用户点击物品的概率,通过排序点击概率得到top-k推荐。实验结果表明,该模型在用户历史偏好多样的场景下,推荐结果更加符合用户的偏好习惯,推荐指标AUC(Area Under Curve)和多样性均获得了相应提升。
To address the distortion problem of user interest attributes in the recommendation results of the RippleNet recommendation model,an improved RippleNet recommendation model RippleNet-CA using shared attribute sampling is proposed.First of all,the model reconstructs the extended preference set(RippleSet)on the knowledge graph by replacing the random sampling of the original model with shared attribute sampling and improves the information contained in user features based on increasing the frequency of shared attributes among nodes within the RippleSet and enhancing the correlation between user historical preferences and recommendation preferences.In addition,user features and item features are fed into the model to calculate the probability of users clicking on items,and top-k recommendations are obtained by ranking the click probabilities.The experimental results show that the model's recommendation results are more consistent with users'preference habits in scenarios with diverse user history preferences,and the recommendation metrics AUC(Area Under Curve)and diversity are improved accordingly.
作者
江山
韩永华
JIANG Shan;HAN Yonghua(School of Computer Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《智能计算机与应用》
2024年第3期10-16,共7页
Intelligent Computer and Applications
