Cross-Domain Recommendation based on Heterogeneous information network with Adversarial learning

Abstract

In this paper, based on the heterogeneous information network, we propose a cross-domain recommendation model by integrating adversarial learning (Cross-Domain Recommendation based on Heterogeneous information network with Adversarial learning, CDR-HA). Using information from other domains to alleviate the target data sparseness of the domain can improve the accuracy and performance of recommendations. In this paper, we focus on the cross-domain recommendation. Firstly, due to the differences in the feature distributions of the same users in different domains, we use the HIN2Vec algorithm to extract the user's feature distribution in the network based on the heterogeneous information network. Secondly, we propose a multi-domain feature filtering method, which maximizes the difference in the distribution of different domains based on Wasserstein Distance to preserve the differences in the feature distributions of users in different domains. Then, separately establish a classifier for each domain, we consider the results of the two classifiers comprehensively, and take the best as the final result. We apply the proposed model to two datasets and experimental results demonstrate that our approach outperforms state-of-the-art recommender baselines.