First Workshop on Proactive and Agent-Supported Information Retrieval

Organized as a part of CIKM 2022

Get Started

Workshop Motivation and Description


Users typically interact with a digital system with the goal of accomplishing a task. While working on their task the user may encounter an information need which they must resolve in order to advance their activity. In this situation, the user will generally make use of an information retrieval (IR) system by manually specifying and entering their best attempt at a search query to express their information need. Using this query, the IR system makes its best attempt to returns a ranked list of documents potentially relevant to the user. While a user is often able to very effectively address their information needs by engaging with IR systems in this classical reactive mode, it is not without challenges and cost to the user, and may not expose the user to all of the potentially useful information available to them. In order to make use of the IR system, the user must first realize that they have an information need. They then need to make the effort to break off from their task to engage with the IR system. While using it, they may encounter difficulty in constructing an effective query, and must then navigate their way to relevant information within the retrieved content. Beyond the traditional active search paradigm, there may be other information available within the IR system which the user is unaware of or has forgotten about, which, if they were aware of, would enable them to complete their task more efficiently or better.


In this workshop, we will explore the alternative search framework of proactive and agent supported information retrieval (PASIR). A proactive agent should seek to automatically anticipate a user's information needs from their current activities, context or previous engagements with a system. The idea for the PASIR system is to support the user in completing their task without the user always needing to actively engage with the IR system. In this workshop we aim to address the limitations of current PASIR research (e.g. Proactive IR, Conversational IR, Interactive IR) and explore the opportunities for improved user experience using PASIR as outlined above. The workshop aims to accomplish two goals: a) gather researchers from the IR and NLP communities to present their recent work relevant to proactive information retrieval (PIR) and explore future research directions, and b) provide a platform for new PhD students or early researchers to develop potential research activities in this area and build datasets that can be used for supporting their research.

Call For Papers

We solicit submissions on the following research topics. We also comment on the scope and research challenges for each topic.
  1. Investigating types of tasks for which a system can be proactive or an agent can initiate the interaction.

    • Tasks which typically involve carrying out a certain set of activities in sequence may be more amenable to PIR support, which could perhaps be automated more effectively than than other unstructured tasks which may not be amenable to proactive retrieval.

  2. Investigating the scope of query performance prediction (QPP) for PASIR systems.

    • Estimating the quality of search results without the presence of relevance assessments (and also, as in contrast to standard QPP, without the presence of a user formulated query) is a major research challenge.
    • To investigate if QPP estimators can help decide on selective proactivity is also an interesting research direction.

  3. Investigating opportunities for providing proactive support to the user, i.e., when could users be notified of proactively retrieved information without distracting their current activities.
    • A system that is able to predict user actions from their patterns of interactions (e.g., is the user reading, writing an important email etc.) could potentially be used to detect such opportunities.
  4. Developing novel user-friendly interfaces for a PASIR system.
    • An important question in the development of PIR is to investigate if users should best be provided with a standard ranked list of documents, or should the information be presented in a more organized manner, e.g., as clusters of topics etc.
    • User studies could also be conducted to find out if users would like to be presented with links to whole documents, or snippets of relevant information, information cards, etc.
  5. Use of simulation methods in PASIR research.
    • Simulation can be used to generate large amounts of estimated user-system interactions on interaction models, which could facilitate development of supervised approaches.
    • Effectiveness of PIR models can also be investigated under a multi-agent simulated environment.
    • Reinforcement learning driven approaches that seek to maximize future rewards of a sequence of actions can also be applied for PIR models, since proactive retrieval usually involves a sequence of retrieval steps executed against a sequence of query formulation actions.
  6. Finding out ways of providing explanations for each proactively retrieved document or topic.
    • Investigate the required explanation units for information seeking tasks (may be different explanations for different granularity of information need).
    • Counterfactual explanations from a user's activities.
    • Attention based explanations from a user's activities.
    • Organize the information and link it back (somewhat similar to the background linking task of TREC) to one of the several information seeking tasks that a user had been involved with (users interactions typically correspond to multi-task activities).
  7. Investigating methods for PIR model evaluation.
    • How to evaluate the sub-components of a PIR system, e.g., evaluating the effectiveness of automatic query formulation, and of the downstream retrieval performance for one query or a sequence of queries, etc?
    • Investigating if there is a necessary correlation between the evaluation of each sub-component and the satisfaction of users.
    • Investigating the role of other user-centric aspects, such as the convenience with which users can relate to specific pieces of proactively retrieved information
  8. Investigating methods for effective query formulation in proactive IR models.
    • Exploring the query formulation approach in different PIR setup (e.g. asking clarifying questions in conversational search setup, query formulation in zero query scenario).
    • Investigating the effect of query enrichment in PIR models.

Submission Guidelines

Papers submitted should be 4-10 pages of content (plus any number of additional pages for references). Papers must be submitted in PDF according to the CEUR format. The review process is double-blind. Paper should be uploaded via Easychair via this submission link. Accepted papers will be included in the PASIR 2022 proceedings and at least one author of each accepted contribution must attend the workshop. After acceptance, no additional authors can be added.


Tentative Schedule

  • Keynote Talk by Hamed Zamani, University of Massachusetts, Amherst. 21st October, 13.00 EDT - 14.30 EDT [slides]
  • Sean MacAvaney, Nicola Tonellotto and Craig Macdonald, Adaptive Re-Ranking as an Information-Seeking Agent 21st October, 14.30 EDT - 15.00 EDT [slides]
  • Coffee break, 21st October, 15.00 EDT - 15.15 EDT.
  • Pruthvi Raj Venkatesh, Chaitanya K, Rishu Kumar and P Radha Krishna, Conversational Information Retrieval using Knowledge Graphs. 21st October, 15.15 EDT - 15.45 EDT [slides]
  • Tabish Ahmed and Sahan Bulathwela, Towards Proactive Information Retrieval with Wikipedia Concepts. 21st October, 15.45 EDT - 16.15 EDT [slides]

Important Dates

  • Paper Submission Deadline - 18th Autgust, 30th August, 2022 2nd September, 2022.
  • Review Notifications - 25th September, 2022
  • Camera Ready - 1st October, 2022
  • Workshop Day - 21st October, 2022

Contact US

Please reach out to the organizers for any questions. You can also mail to