Research

My research is dedicated to building more intelligent and helpful ways for people to access information. I design and develop assistive AI technologies that help users find, understand, and interact with information more effectively. My work involves creating novel models and algorithms, always with a user-centric focus, with the ultimate goal of creating systems that are intuitive and trustworthy, giving users greater agency and control over the tools they use.

Current Research Focus

My recent work is concentrated on three interconnected areas that aim to create more effective, trustworthy, and transparent interactive information access systems.

Conversational Information Access

My research in conversational information access focuses on developing multi-turn, multi-modal systems that assist people in complex information-seeking scenarios. A key challenge is effectively understanding and eliciting user needs, which involves personalizing interactions by adapting to individual preferences and knowledge levels. Beyond understanding the user, it is crucial to generate responses that are factual and grounded in verifiable sources. To that end, I investigate how to improve the transparency of the generation process by providing users with explanations about the response’s source, the system’s confidence, and its potential limitations. Underpinning this work is the creation of large-scale resources and the development of novel data collection and annotation protocols, which are essential for advancing research in this area.

đź“š Selected publications:

Evaluation Methodologies

Modern interactive AI systems pose a significant evaluation challenge, as their dynamic, multi-turn nature renders traditional static test collections unsuitable. To address this, a significant part of my research is dedicated to pioneering work on user simulation—creating computational agents that mimic human interaction with AI systems. This research extends beyond the development of novel algorithmic solutions to also include a deep understanding of the requirements for user simulators, acknowledging their limitations, and developing robust methods for their verification. I strongly believe that user simulation is not merely a useful tool, but an indispensable technology required to accelerate progress towards Artificial General Intelligence (AGI) by providing scalable environments for evaluation and interactive learning. Beyond my own research, I am committed to promoting this methodology through the organization of workshops and seminars, and by sharing open resources. Advancing this area requires a truly interdisciplinary effort, integrating insights from machine learning, cognitive science, and human-computer interaction. To this end, we have established usersim.ai, a central hub intended to foster collaboration and build a vibrant research community around this critical topic.

My work in this area also includes the development of novel data collection protocols and the creation of reusable resources, such as benchmark test collections, to foster reproducible research.

đź“š Selected publications:

Transparent and Explainable AI

My research in this area focuses on improving the transparency and explainability of AI systems, particularly in the context of search and recommendation. One major thread of this work proposes a new approach to recommendation based on natural language user profiles. Instead of relying on opaque user representations like embedding vectors, this method uses explicit, scrutable summaries of a person’s interests written in plain language. This allows people to directly understand, inspect, and modify the basis for their recommendations, giving them more meaningful control over the personalization they receive. Another line of research investigates how to provide transparency for system-generated responses in conversational search. This work explores how disclosing key aspects of the generation process—such as the source of the information, the system’s confidence in its answer, and any potential limitations—can empower users to better verify and assess the quality of the responses they receive.

đź“š Selected publications:

Foundational Contributions

Over the past 20 years, my research has contributed to several key areas in information retrieval and its neighboring fields, including natural language processing, recommender systems, and human-computer interaction.

  • Evaluation Methodologies & Resources: A recurring theme throughout my career has been the development of novel evaluation methodologies and reusable resources. I have designed test collections and evaluation frameworks for a wide range of information access scenarios. Key examples include establishing evaluation methodology for entity retrieval as the lead organizer of the TREC Entity track (2009-2011) and creating the DBpedia-Entity test collection for entity retrieval, which continues to be heavily utilized as part of widely used evaluation bundles such as BEIR and MTEB. I also pioneered the “living labs” paradigm for IR evaluation—using production systems as live experimental platforms for academic research—to bridge the gap between academic research and industrial practice, which I advanced by co-organizing the Living Labs for Information Retrieval Evaluation (LL4IR) lab at CLEF (2015-2016) and the OpenSearch track at TREC (2016-2017). My current research in this area now centers on user simulation, a topic detailed further above.
  • Semantic and Entity-Oriented Search: My work in this area has focused on moving search beyond keywords to a deeper understanding of “things” (entities) and their relationships. This involved research on entity linking, retrieval, and leveraging knowledge graphs. Recognizing the fundamental limitation that public knowledge graphs only cover prominent entities, I proposed a novel direction by introducing the concept of Personal Knowledge Graphs—resources of structured information about entities personally related to a user. I synthesized research in this area in a 2018 open-access book Entity-Oriented Search, which has since garnered over 250,000 chapter downloads.
  • User Modeling: A central, cross-cutting theme in my research has been understanding user goals and information needs, modeling user knowledge and preferences, and personalizing interactions. My work includes developing ways for users to express their information needs beyond simple keywords, for instance, by providing examples of desired items or through scrutable natural language profiles. A related research thread explores intuitive and natural ways for users to provide feedback on system responses.
  • Recommender Systems: My work on recommender systems advances a user-centric perspective, with a strong focus on conversational interaction, transparency, and evaluation. This includes developing novel methods for preference elicitation and for enabling users to provide intuitive feedback within a dialogue. To enhance transparency and user control, I have pioneered the use of scrutable natural language user profiles and have designed methods for generating effective explanations. Furthermore, I have made significant contributions to the evaluation of these interactive systems through the creation of reusable test collections and developing novel user simulation methodologies.
  • Expertise Retrieval: My foundational PhD research (People Search in the Enterprise) developed influential probabilistic models for expert finding and profiling in organizational settings, establishing core methods for identifying and representing human expertise from unstructured text. I also co-authored the book Expertise Retrieval, which synthesized research across the field and serves as a key reference on the topic.