Ontology based hierarchical recommendation model on Deep Neural Networks

Xinghao Li, Armin Haller, Yingnan Shi

Keywords: Machine Learning, Recommender Systems, Knowledge Graph

Abstract: Traditional algorithms in recommender systems aim to solve the recommendation problem by finding the latent connections between users and items. They mainly consider the information provided by the input datasets without exploring the inherent connections between items such as taxonomic and hierarchical relationships in a specific field. In this paper, we propose an extension to an existing deep neural network algorithm, by using an associated domain ontology as auxiliary information to support the training of the model, called Ontology-based Neural Collaborative Filtering (ONCF). Specifically, our method exploits the hierarchical properties of the item set as defined by an ontology and integrates the structural information into the training process of the deep neural network. Consequently, our algorithm is not only able to predict the exact item but also offers a recommendation to a specific class of candidate items based on the ontological relations in the knowledge graph. We demonstrate our OCNF model's comparable accuracy in terms of the quality of predictions, while at the same time having a lower computational complexity to recommend potential matches.

Subtrack: Machine Learning

First Author Is Student: Yes