基于CNN和关联规则的协同过滤混合推荐
Collaborative Filtering Hybrid Recommendation Based on CNN and Association Rules
黄甫 1李涛 1谢君臣1
作者信息
- 1. 武汉科技大学计算机科学与技术学院 武汉 430065
- 折叠
摘要
针对传统协同过滤算法在药品推荐领域中对用户评价过于依赖、数据稀疏性严重等问题,提出一种基于卷积神经网络(CNN)和关联规则的协同过滤混合推荐算法.首先利用CNN从药品文本数据中获取深层功效特征向量,然后使用Apriori算法发掘药品间的关联规则.在此基础上,从药品功效特征和关联性两个角度结合对应的相似度计算公式计算药品相似度,进而预测评分缺失值,最终对稀疏矩阵进行叠加填充实现药品推荐的优化.经试验对比,论文算法相比传统的协同过滤算法在MAE和RMSE指标上下降了3%～4%,在数据较稀疏的情况下具有良好的推荐效果.
Abstract
Aiming at the problems of traditional collaborative filtering algorithm on user evaluation and serious data sparseness in the field of drug recommendation,a collaborative filtering hybrid recommendation algorithm based on convolutional neural net-work(CNN)and association rules are proposed.First,CNN is used to obtain the deep efficacy feature vector from the drug text da-ta,and then the Apriori algorithm is used to explore the association rules between drugs.And on this basis,this paper combines the corresponding similarity calculation formula from the two perspectives of drug efficacy and relevance to calculate the drug similarity,and then predicts the missing value of the score,and finally superimposes and fills the sparse matrix to realize the optimization of drug recommendation.Compared with the experiment,the algorithm in this paper reduces the MAE and RMSE indicators by 3%to 4%compared with the traditional collaborative filtering algorithm,and it has a good recommendation effect when the data is sparse.
关键词
卷积神经网络/关联规则/协同过滤/功效特征/数据稀疏性Key words
convolutional neural network/association rules/collaborative filtering/efficacy characteristics/data sparsity引用本文复制引用
基金项目
国家自然科学基金(61702383)
湖北省教育厅重大项目(17ZD014)
出版年
2024
国家科技期刊平台
NSTL
万方数据