Human-AI Interaction for Exploratory Search & Recommender Systems with Application to Cultural Heritage

Abstract

Exploratory search and recommender systems are ubiquitous and central to information navigation. Yet, many pressing challenges remain surrounding the development of robust systems, from producing high-quality data and metadata to answering fundamental questions in human-AI interaction concerning the interactive affordances for search and recommendation. These challenges are exacerbated by 1) the ever-expanding wealth of information to be searched, and 2) the widespread incorporation of increasingly opaque and complex machine learning models into deployed systems. This thesis explores these challenges and investigates how we can improve interaction mechanisms in exploratory search and recommendation. Much of this dissertation adopts the setting of digital cultural heritage collections, where impoverished metadata redoubles challenges of searchability, with implications across disciplines.