Accurate and justifiable : new algorithms for explainable recommendations.

Abstract

Studies for providing me the opportunity to pursue my degree and for their unfailing support and assistance throughout my graduate studies. I would also like to thank the Kentucky Science and Engineering Foundation for partially providing the funding for the work. My sincere thanks goes to my fellow lab-mates, with a special mention to Wenlong Sun and Gopi Chand Nutakki. Our conversations on research and their good-hearted support and friendship will not be forgotten. Last but by no means least, I am so grateful to my family and friends who have supported me in every possible way unconditionally and selflessly. Many thanks to my sister Behnaz Abdollahi for encouraging me and expressing confidence in my abilities when I could only do the opposite. Moreover, thanks to Milad Nikoukar for motivating me to keep reaching for excellence. iii

Studying and handling iterated algorithmic biases in human and machine learning interaction.

“…Recommender systems (RS), on the other hand, generally do not await an explicit query to provide results [8][9][10][11][12][13][14][15][16][17][18][19][20], consisting of recommend items which are believed to be of interest to the users. Recommender systems can be categorizeed based on which data they use and how they predict user interests.…”

“…17: Results of the Mann-Whitney U test and the t-test comparing boundary shift for the three forms of iterated algorithmic bias with class-dependent human action probability ratio 1:1 and 10:1. The effect size is calculated as (Boundary| ratio=1:1 − Boundary| ratio=10:1 )/standard.dev at time t=200.…”

An explainable sequence-based deep learning predictor with applications to song recommendation and text classification.

“…Thus, recommendation is becoming a common part of our daily lives. The ML systems that generate these personalized recommendations are called recommender systems [3]. These systems have known a tremendous and increasing interest by the ML research community during the last few decades.…”

“…Aside from accuracy and explainability, the cold start problem characterizes a significant issue that recommender systems [3], and especially collaborative filtering recommender systems, usually suffer from. The latter problem consists of generating recommendations for new users (user cold start) or recommending new items (item cold start) that are newly added to the system.…”

“…In addition, we demonstrate how our explanation mechanism can be used with generic sequential data beyond music, namely unstructured free text in two application domains: sentiment classification of online user reviews and delineating potential child abuse instances from medical examination reports. [3]. Their applications range from self driving cars to finance and health care.…”