Just noticed that the accepted papers of #SIGIR2024 has been released. Based the number of papers with faculty from each institute, it seems that Asia has contributed the most this year!🎉 (no data cleaning, so the numbers are by no means accurate😉)
🚀 Eager to improve speed and performance in LLM-based re-ranking?
Explore our #SIGIR2024 SIGIR 2024 long paper “Ranked List Truncation for Large Language Model-based Re-Ranking” w/ Negar Arabzadeh Arian Askari Mohammad Aliannejadi Maarten de Rijke
Preprint: arxiv.org/abs/2404.18185
Thread 👇 1/4
Our #SIGIR2024 paper about evaluating retrieval models in RAG pipelines is now on Arxiv! We also release a Python package too!
'Evaluating Retrieval Quality in Retrieval-Augmented Generation'
Hamed Zamani
github.com/alirezasalemi7…
arxiv.org/abs/2404.13781
Thread 🧵👇 1/7
How deep should we go for reranking?
Check out our recent paper SIGIR 2024 to find out setting a proper reranking depth on a per-query basis can improve reranking effectiveness and efficiency simultaneously.
arxiv.org/abs/2404.18185
#SIGIR2024
I'm incredibly proud of our #SIGIR2024 paper on approx. sparse vector retrieval. It marks a seismic shift in retrieval over learnt sparse embeddings of text collections such as Splade. Preprint at arxiv.org/abs/2404.18812 (w/ the fabulous Franco Maria Nardini, cosimorulli, Rossano Venturini)
I'm happy to share that our paper (arxiv.org/abs/2309.09013) studying clustering-based ANN search for sparse and hybrid vectors has been accepted to ACM TOIS (joint work w/ Franco Maria Nardini , Amir Ingber, and Edo Liberty). That work formed the basis for our #SIGIR2024 paper below.
Extremely happy to share the preprint and resources for our SIGIR 2024 2024 paper on developing error taxonomy for GPT-4 Rationales to Medical Board exam questions (USMLE)
arXiv: arxiv.org/abs/2404.13307
Codebase: github.com/roysoumya/usml…
CNeRG IIT KGP L3S Research Center @L3S_Research_Center@wisskomm Leibniz AI Lab
#SIGIR2024
Extremely happy to share the preprint and resources for our SIGIR 2024 2024 paper on developing error taxonomy for GPT-4 Rationales to Medical board exam questions (USMLE)
arXiv: arxiv.org/abs/2404.13307
Codebase: github.com/roysoumya/usml…
CNeRG IIT KGP L3S Research Center @L3S_Research_Center@wisskomm Leibniz AI Lab
#SIGIR2024
Deadline extended to May 2nd, 2024.
Published work, original work, and ongoing work on topics related to evaluation and LLMs in IR can be submitted to the workshop.
The submission length is flexible; check out the website for more details: llm4eval.github.io/cfp/
#SIGIR2024
Pre-print of our #SIGIR2024 paper titled 'Explainability for Transparent Conversational Information-Seeking' led by PhD student Weronika Łajewska (@UniStavanger @iai_group) w/ Johanne Krisztian Balog is now available:
damianospina.com/publication/la…
ADM+S Centre @RTMIComputing SIGIR 2024
Exciting news! We have two fantastic keynotes for our LLM4Eval workshop by Ian Soboroff | [email protected] and Don Metzler .
Stay tuned for more info!
More info about the workshop at llm4eval.github.io
#SIGIR2024
TL;DR: Unlike in language generation where multiple equally-acceptable outputs exist, each doc in GR is associated with a unique ID. If you miss one DocID token in beam search, you cannot recover. Solution: plan ahead in GR decoding! #SIGIR2024
Pre-print of our #SIGIR2024 paper 'Characterizing Information Seeking Processes with Multiple Physiological Signals' led by PhD student KAIXIN JI w/ F. Scholer Danula Hettiachchi Flora Salim is now available: damianospina.com/publication/ji…
ADM+S Centre RMIT School of Computing Technologies UNSW Computing SIGIR 2024
Interested in speeding up approximate retrieval over learnt sparse embeddings of up to 12x w.r.t. the winning methods of the Big ANN 2023 Sparse Track challenge? Check this out: arxiv.org/abs/2404.18812. Long paper @ #SIGIR2024 with Sebastian Bruch, cosimorulli, Rossano Venturini
Wondering how to evaluate the retrieval quality in RAG pipelines? check this out!
TL;DR: evaluating based on the passages that contain groundtruth answers or KILT's provenance labels has low correlations with end-to-end RAG performance. Try 'pip install erag' instead! #SIGIR2024
This work got accepted to #SIGIR2024 . Many thanks to Chen Luo and my advisor Hamed Zamani.