Submission Guidelines

We invite authors to submit papers written in English. Submissions may range in length from a minimum of 6 pages to a maximum of 12 pages; however, references and supplementary materials may exceed this page count without limitation. In order to facilitate a double-blind review process, authors must ensure that submissions are fully anonymized. Please note that we do not impose a specific anonymity period prior to submission.

The papers (.pdf format) should be submitted using the EasyChair submission system at https://easychair.org/conferences/?conf=keirecir2025. Authors should consult Springer’s authors’ guidelines and use their proceedings templates to prepare the submission. The Microsoft Word and LaTeX versions of the template can be found at https://www.springer.com/gp/computer-science/lncs/conference-proceedings-guidelines. Submissions to KEIR @ ECIR 2025 will be peer-reviewed on the basis of technical quality, relevance to workshop topics, originality, significance, clarity, etc.

We accept submissions of the following types:

• Original work that is not published or submitted elsewhere.
• Work that is submitted elsewhere and is still under review. In this case, the authors should make sure that they are not violating the submission guidelines and anonymity requirements of the other venue(s).
• Work that has been rejected at ECIR 2025.

Good News: Our KEIR @ ECIR 2025 workshop has been accepted as a post-proceedings volume in Springer's Lecture Notes in Computer Science (LNCS). Accepted papers will have the opportunity to be published in the LNCS series!

Important dates

• Submission Deadline: January 12, 2025 January 31, 2025
• Acceptance Notification: February 23, 2025
• KEIR Workshop: April 10, 2025
• Deadlines refer to 23:59 (11:59 pm) in the AoE (Anywhere on Earth) time zone.

• Hui Feng, Yuntzu Yin, Emiliano Reynares, and Jay Nanavati.
  OntologyRAG: Better and Faster Biomedical Code Mapping with Retrieval-Augmented Generation (RAG) Leveraging Ontology Knowledge Graphs and Large Language Models.


• Jaspinder Singh and Carlo Merola.
  Reconstructing Context: Evaluating Advanced Chunking Strategies for Retrieval-Augmented Generation.


• Jingfen Qiao, Thong Nguyen, Evangelos Kanoulas, and Andrew Yates.
  Going Beyond Encoders: Leveraging Decoder Architectures for Learned Sparse Retrieval.


• Shen Dong, Iadh Ounis, and Zaiqiao Meng.
  To Personalise or Not? Why Personalisation Knowledge Can Fall Short in Enhancing Conversational Information Seeking.

• Alessandro Lenci, Full Professor, University of Pisa



Invited Talk 1: The Semantic Gap: Understanding What Large Language Models Still Fail to Understand

Abstract: The unprecedented success of Large Language Models (LLMs) in carrying out linguistic interactions disguises the fact that, on closer inspection, their knowledge of meaning and their inference abilities are still quite limited and different from human ones, especially if we consider the super-human amounts of training data. They generate human-like texts, but still fall short of fully understanding them. I will refer to this as the “semantic gap” of LLMs. They learn highly complex association spaces that correspond only partially to truly semantic and inferential ones. In this talk, I will present current research probing the limits of LLMs on undestanding various kinds of semantic relations, with the aim of investigating the missing links to bridge the gap between LLMs as sophisticated statistical engines and full-fledged semantic agents.



• Andrew Yates, Senior Research Scientist, Johns Hopkins University



Invited Talk 2: Lexical Representations and Test-time Compute for Knowledge-Enhanced IR

Abstract: In this talk, I will describe some of my recent work on knowledge-enhanced IR from two perspectives: building lexical representations for first-stage retrieval and reranking with reasoning LLMs distilled from DeepSeek-R1. To improve first-stage retrieval, the DyVo model incorporates knowledge of entities and concepts into lexical representations made of wordpieces, whereas the LENS model demonstrates that lexical representations can perform competitively with dense representations across a range of tasks. Given this strong retrieval foundation, I will describe Rank1, a recent model that uses test-time compute to perform reranking by reasoning about a document's relevance.

Location: Affreschi Room – San Micheletto Campus

• Zihan Wang, University of Amsterdam
• Jinyuan Fang, University of Glasgow
• Giacomo Frisoni, University of Bologna
• Zhuyun Dai, Google DeepMind
• Zaiqiao Meng, University of Glasgow
• Gianluca Moro, University of Bologna
• Emine Yilmaz, University College London

KEIR @ ECIR 2024: https://keirworkshop.github.io