Towards Scrutable Decision Tree-based User Model utilising Interactive and Interpretable Machine Learning (SUM-IML)

Abstract

Personalising user models has gained considerable attention in recent literature. In an information-rich environment, it is crucial not only to provide the information at any time, at any place, and in any form, but to also minimise information overload for the user and ease their ability to access relevant information in their user model. Supplying tailored information and providing offerings that suit each user's interests may be enhanced by involving the user. Recently, much research has been concerned with employing Machine Learning for user modelling. Machine Learning (ML) tries to mimic and predict a user s activities, however, it cannot model the user themselves. Users can benefit from involvement in the modelling process by incorporating their input as considerations in modelling through employing interactive Machine Learning or human-in-theloop controls. Such user involvement needs the user to understand the model behaviour to ensure their inputs are effective. This can be achieved by utilising Machine Learning interpretability techniques. This work proposes an interpretation of the model to the user in order to provide the user with better understanding for the model behaviour. In this study, the proposed approach, termed SUM-IML (Scrutable User Modelling using Interactive and Interpretable Machine Learning)., implements model scrutability by combining the benefits of interactive ML as well as Interpretable ML in user modelling. This thesis presents the research question driving this work, a state-ofthe-art review of user modelling, scrutability, interactive, interpretable Machine Learning literature and the evaluation methodologies required. It then presents two related experiments that demonstrate the exploration of research question through their results and the conclusion.