基于物品交互约束的自编码器推荐模型  

Item-interaction constraint-based autoencoder model for recommendation

作　　者：李昌兵[1] 陈思彤 罗陈红 邓江洲 叶建梅 LI Changbing;CHEN Sitong;LUO Chenhong;DENG Jiangzhou;YE Jianmei(School of Economics and Management,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;Digital Intelligence Technology Innovation and Industrial Development Research Center,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学经济管理学院,重庆400065 [2]重庆邮电大学计算机科学与技术学院,重庆400065 [3]重庆邮电大学数智技术创新与产业发展研究中心,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2024年第5期1052-1061,共10页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：教育部人文社会科学研究一般项目(23YJC630215);2023年重庆市教育委员会人文社会科学研究基地项目(23SKJD066);中国博士后科学基金面上项目(2021M702321);重庆市教委科学技术项目(KJQN202100603)。

摘　　要：近年来,自编码器凭借其良好的数据压缩能力在推荐领域得到广泛应用。研究发现,受推荐系统中数据稀疏性的影响,自编码器模型会因为用户与物品之间缺少交互而在训练过程中产生偏差,从而影响模型的推荐结果。为解决此问题,提出一种基于物品交互约束的自编码器推荐模型。该模型以物品交互情况作为约束条件,设计了新的参数更新规则,规避了由数据稀疏性引入的模型训练偏差。在此基础上,推荐模型还将物品标签信息引入到训练中,通过增加新的数据源来缓解数据稀疏性的影响,提升推荐性能。在3个不同规模和稀疏度的数据集上的实验表明,模型对稀疏数据集具有良好的适应性,能够有效提高推荐的准确性,展现出良好的应用潜力。In recent years,autoencoders have been widely used in the recommendation domain due to their strong data compression capabilities.However,research has shown that due to the sparsity of data in recommendation systems,autoencoder models can develop biases during training caused by the lack of interactions between users and items,negatively impacting recommendation results.To address this issue,we propose a recommendation model based on item interaction constraints.The model uses item interaction as a constraint and designs new parameter update rules to avoid biases introduced by data sparsity in model training.Additionally,the model incorporates item tag information into the training process,leveraging new data sources to mitigate the effects of data sparsity and improve recommendation performance.Experiments on three datasets with varying sizes and sparsity levels demonstrate that the model is well-suited for sparse datasets,effectively enhancing recommendation accuracy and showing great application potential.

关 键 词：自编码器 推荐系统 物品交互约束 协同过滤 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]