基于时间编码和GNN的商品个性化推荐算法研究  

Research on the Personalized Product Recommendation Algorithm Based on Time Coding and GNN

作　　者：马俊 王贤来[2] MA Jun;WANG Xianlai(Anhui Vocational and Technical College of Water Resources and Hydropower,Hefei 230000,China;Anhui University,Hefei 230000,China)

机构地区：[1]安徽水利水电职业技术学院,安徽合肥230000 [2]安徽大学,安徽合肥230000

出　　处：《佳木斯大学学报（自然科学版）》2024年第11期26-29,共4页Journal of Jiamusi University：Natural Science Edition

基　　金：2022年度安徽省省级质量工程项目基于(2022jyxm577);2023年度安徽省高等学校科学研究项目重点项目(2023AH052897)。

摘　　要：虽然推荐系统能帮助用户筛选出所需产品,进而缓解信息过载问题,但由于当前的推荐系统大多都忽视了数据的稀疏性问题,导致精度欠佳。因此,为了提高推荐精度,研究提出了基于图神经网络和时间编码的会话推荐模型。实验结果显示,在1/4 Yoochoose数据集中,基于图神经网络和时间编码的会话推荐模型Top 20个性化推荐结果的精度和平均倒数排名分别为74.3%和33.9,均高于其他算法。在去除时间编码和语义知识库后,前20个项目的精度分别由74.5%下降至了71.6%和71.1%,平均倒数排名由34.4分别下降至32.3和30.8。上述结果表明,研究提出的会话推荐模型性能优越,且时间编码和语义知识库能有效提高模型的推荐精度。Although the recommendation system can help users to screen out the desired products,and then alleviate the information overload problem.However,most of the current recommendation systems ignore the sparsity of the data,resulting in poor accuracy.Therefore,to improve recommendation accuracy,we propose session recommendation models based on graph neural network and temporal coding.The experimental results show that in the 1/4 Yoochoose data set,the accuracy and average reciprocal ranking of the session recommendation results based on graph neural network and temporal coding were 74.3%and 33.9,respectively,which are higher than the other algorithms.Moreover,after removing the temporal coding and semantic knowledge base,the accuracy of the top 20 items decreased from 74.5%to 71.6%and 71.1%,respectively,and the average reciprocal ranking decreased from 34.4 to 32.3 and 30.8,respectively.The above results show that the proposed session recommendation model performs well,and the time coding and semantic knowledge base can effectively improve the model recommendation accuracy.

关 键 词：电子商务 会话推荐 时间编码 图神经网络 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]