Information access systems mediate how we find and discover information in nearly every walk of life. The ranking models powering these systems base their predictions on users' historical interactions to cater to the wide variety of users and workflows that leverage them. However, during a task session, personalized predictions often fall short of user's expectations, with users desiring greater control over a system. Greater control, in turn, leads to greater user trust and satisfaction in using a system. In this thesis, I explore methods to dynamically update personalized rankings through user interaction with ranking models. I explore control in various retrieval tasks through 1) expressive natural language queries, 2) control over latent user representations, and 3) control over both queries and latent user representations.
First, I explore long-form narrative queries as a way for users to express rich context-dependent preferences in a narrative-driven recommendation (NDR) task. Here, I propose MINT - a data augmentation strategy leveraging LLMs to generate long-form narrative queries from historical user interactions to allow the training of effective NDR models. Next, I propose LACE, a text recommendation model that represents users with a transparent concept-based user profile inferred from historical user documents. The concepts function as an interpretable bottleneck within a neural recommender, allowing users to control the underlying model. Effective performance is ensured through personalized concept values paired with the induced concept profiles. Users may specify positive or negative preferences for concepts and directly edit them, with these actions reflected in the generated recommendations.
In proposed work, I will explore two directions. First, I will leverage concept-value profiles introduced in LACE to personalize crossencoder models for personalized search tasks. Specifically, we treat concept-value profiles as editable memory representations of a user's historical documents and augment a transformer crossencoder with these memories. These may present an effective pathway to personalizing crossencoders, allowing them to condition on large amounts of user data while allowing users effective control over personalization. Second, we will study the effects of LACE in a realistic application scenario: paper-reviewer matching for peer review. Besides serving as a deeper evaluation of LACE, this study will point in the direction of future design and methodological work.
Advisor: Andrew McCallum