Heterogeneous Information Network-Based Recommendation with Metapath Search and Memory Network Architecture Search

Recommendation systems are now widely used on the Internet. In recommendation systems, user preferences predicted by interaction of users with products, such as clicks or purchases. Usually, heterogeneous information network is to capture semantic in data, which can be solve sparsity problem and cold-start problem. a more complex network, types nodes edges very large, so there lots metagraphs network. At same time, machine learning tasks networks have large number parameters neural architectures that need set artificially. The main goal find optimal hyperparameter settings for performance task space. To address this problem, we propose metapath search method based architecture search, metapaths suitable different tasks. We conducted experiments Amazon Yelp datasets compared obtained from an automatic manually structures verify effectiveness algorithm.