Personalized recommendation with knowledge graph via dual-autoencoder

Yang Yang,Yi Zhu,Yun Li

DOI: https://doi.org/10.1007/s10489-021-02647-1

IF: 5.3

2021-09-03

Applied Intelligence

Abstract:In the past decades, personalized recommendation systems have attracted a vast amount of attention and researches from multiple disciplines. Recently, for the powerful ability of feature representation learning, deep neural networks have achieved sound performance in the recommendation. However, most of the existing deep recommendation approaches require a large number of labeled data, which is often expensive and labor-some in applications. Meanwhile, the side information of users and items that can extend the feature space effectively is usually scarce. To address these problems, we propose a Personalized Recommendation method, which extends items’ feature representations with Knowledge Graph via dual-autoencoder (short for PRKG). More specifically, we first extract items’ side information from open knowledge graph like DBpedia as items’ feature extension. Secondly, we learn the low-dimensional representations of additional features collected from DBpedia via the autoencoder module and then integrate the processed features into the original feature space. Finally, the reconstructed features is incorporated into the semi-autoencoder for personalized recommendations. Extensive experiments conducted on several real-world datasets validate the effectiveness of our proposed methods compared to several state-of-the-art models.

computer science, artificial intelligence