From Hypothesis to Victory: RMIT‑ADM+S Wins SIGIR 2025 LiveRAG Challenge
Speaker: Kun Ran
Date: 01 Sept, 2025
Time: 11:00 AM - 12:00 PM (AEDT)
Location: B080.07.009 at RMIT University & MS Teams
Abstract: This talk presents the RMIT-ADM+S participation in the SIGIR 2025 LiveRAG Challenge. Our Generation-Retrieval-Augmented Generation (GRAG) approach relies on generating a hypothetical answer that is used in the retrieval phase, alongside the original question. GRAG also incorporates a pointwise large language model (LLM)-based re-ranking step prior to final answer generation. We describe the system architecture and the rationale behind our design choices. In particular, a systematic evaluation using the Grid of Points (GoP) framework and N-way ANOVA enabled comparison across multiple configurations, including query variant generation, question decomposition, rank fusion strategies, and prompting techniques for answer generation. Our system achieved a Relevance score of 1.199 and a Faithfulness score of 0.477 on the private leaderboard, placing the first place in the LiveRAG 2025 Challenge 🏆.
Chuffing~ 👋 We are a group of researchers and students who are passionate about Information Retrieval, Recommender Systems, Natural Language Processing, Human-Computer Interaction, Large Language Models, and beyond.


Kun Ran will share insights from RMIT-ADM+S’s winning approach in the SIGIR 2025 LiveRAG Challenge, focusing on their Generation-Retrieval-Augmented Generation (GRAG) method and its innovative features.

Amin Sadri, Principal Data Scientist at ANZ, will discuss the latest in AI, including LLMs and their potential for “infinite intelligence”. He’ll also share his journey from a PhD in machine learning to industry, offering insights and advice for those looking to transition from academia to practical applications in data science.

Sebastiano Vigna presents a groundbreaking weighted generalization of Kendall’s $\tau$ that brilliantly handles tied ranks, powered by an $O(nlogn)$ algorithm and validated on massive social and web graphs.

Marwah Alaofi presents her paper on how Large Language Models can be fooled into labelling a document as relevant. The paper won the Best Paper Honorable Mention award at SIGIR-AP'24.