%0 Journal Article
%T 一种超图和自监督的协同训练算法<br>A Collaborative Training Algorithm Using Hypergraph and Self-Supervision
%A 徐一波
%A 李昊昱
%J Computer Science and Application
%P 396-407
%@ 2161-881X
%D 2024
%I Hans Publishing
%R 10.12677/CSA.2024.142040
%X <span style=\"font-size:14px;\">为了针对会话推荐中节点数量与图规模不断扩增而产生的节点稀疏问题，提出一种超图与自监督协同训练算法——DHCN-COTREC (Dual-Channel Convolutional for Hypergraphs and Collaborative Training Recommendation)。首先，构建两个不同的图编码器，将数据进行编码用于自监督学习，产生正负样本信息；其次，引入协同训练算法，并加入发散约束进行对比学习，从而最大化两个编码器之间的互信息；最后，根据序列中最后一个项目与正样本之间的余弦距离，向用户推荐TOP-K个项目。本文在Tmall与Diginetica真实数据集上进行实验，结果表明，DHCN-COTREC算法的准确度和平均倒数排名比目前最优的DHCN算法分别提高了26%和1.92%。证实了DHCN-COTREC算法的有效性。</span><br />
<span style=\"font-size:14px;\">In order to address the problem of sparse nodes in recommendation systems caused by the in-creasing number of nodes and graph size in session-based recommendation, a cotraining algorithm called DHCN-COTREC using hypergraphs and self-supervised learning is proposed. Firstly, two different graph encoders are constructed to encode the data for self-supervised learning, generating positive and negative sample information. Secondly, a cotraining algorithm is introduced, and divergent constraints are added for contrastive learning to maximize the mutual information between the two encoders. Finally, TOP-K items are recommended to users based on the cosine distance between the last item in the sequence and positive samples. The experiment was conducted on Tmall and Diginetica real datasets, and the results showed that the accuracy and average recip-rocal rank of DHCN-COTREC algorithm were improved by 26% and 1.92% respectively compared with the current optimal DHCN algorithm, which confirms the effectiveness of the DHCN-COTREC algorithm.</span>
%K 超图，会话推荐，自监督，协同训练，发散约束<br>Hypergraph
%K Session-Based Recommendation
%K Self-Supervision
%K Collaborative Training
%K Divergent Constraints
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=81948