Date on Master's Thesis/Doctoral Dissertation

8-2019

Document Type

Doctoral Dissertation

Degree Name

Ph. D.

Department

Computer Engineering and Computer Science

Degree Program

Computer Science and Engineering, PhD

Committee Chair

Nasraoui, Olfa

Committee Co-Chair (if applicable)

Frigui, Hichem

Committee Member

Frigui, Hichem

Committee Member

Altiparmak, Nihat

Committee Member

Badia, Antonio

Committee Member

Sanders, Scott

Author's Keywords

Artificial Intelligence; machine learning; recommender system; collaborative filtering; explanation; semantic web

Abstract

Collaborative Filtering techniques provide the ability to handle big and sparse data to predict the ratings for unseen items with high accuracy. Matrix factorization is an accurate collaborative filtering method used to predict user preferences. However, it is a black box system that recommends items to users without being able to explain why. This is due to the type of information these systems use to build models. Although rich in information, user ratings do not adequately satisfy the need for explanation in certain domains. White box systems, in contrast, can, by nature, easily generate explanations. However, their predictions are less accurate than sophisticated black box models. Recent research has demonstrated that explanations are an essential component in bringing the powerful predictions of big data and machine learning methods to a mass audience without a compromise in trust. Explanations can take a variety of formats, depending on the recommendation domain and the machine learning model used to make predictions. Semantic Web (SW) technologies have been exploited increasingly in recommender systems in recent years. The SW consists of knowledge graphs (KGs) providing valuable information that can help improve the performance of recommender systems. Yet KGs, have not been used to explain recommendations in black box systems. In this dissertation, we exploit the power of the SW to build new explainable recommender systems. We use the SW's rich expressive power of linked data, along with structured information search and understanding tools to explain predictions. More specifically, we take advantage of semantic data to learn a semantically aware latent space of users and items in the matrix factorization model-learning process to build richer, explainable recommendation models. Our off-line and on-line evaluation experiments show that our approach achieves accurate prediction with the additional ability to explain recommendations, in comparison to baseline approaches. By fostering explainability, we hope that our work contributes to more transparent, ethical machine learning without sacrificing accuracy.

Download

Share

COinS