We present the Auto-Judge resource for the meta-evaluation of automated LLM judges, especially judges that evaluate Retrieval-Augmented Generation (RAG) systems that ground their response with citations. The resource couples (i) a data release of topics, pooled RAG responses, and human judgments, with (ii) a standardized protocol and software infrastructure for implementing "LLM-as-a-judge" methods in a reproducible and extensible way, including support for parameter sweeps and variant tracking.

Traditional short keyword queries are increasingly being replaced by longer, more detailed queries that reflect complex and nuanced user information needs, especially in conversational assistants equipped with web search capabilities. In this work, we present a methodology for automatically generating such human-style long-form queries (narratives) by clustering raw short queries to form synthetic search sessions, designed to reflect a real user's search behavior. Based on our human interpretation study of a 50-narrative set (comprising both human-written and automated narratives), 44% of the automated narratives are misidentified as human-written, underscoring not only the realism and complexity of the generated content but also its indistinguishability from authentic human narratives. Furthermore, we share a collection of automated narratives as a testbed for evaluating LLMs on long-form question answering (QA), which was used in the TREC 2025 RAG track. Our code is available at https://github.com/castorini/narrative-generation.

Large language models (LLMs) have emerged as powerful listwise rerankers but remain prohibitively slow for many real-world applications. What's more, training on the language modeling (LM) objective is not intrinsically aligned with reranking tasks. To address these challenges, FIRST, a novel approach for listwise reranking, integrates a learning-to-rank objective and leverages only the logits of the first generated token for reranking, significantly reducing computational overhead while preserving effectiveness. We systematically evaluate the capabilities and limitations of FIRST. By extending its evaluation to TREC Deep Learning collections (DL19-23), we show that FIRST achieves robust out-of-domain effectiveness. Through training FIRST on a variety of backbone models, we demonstrate its generalizability across different model architectures, and achieve effectiveness surpassing the original implementation. Further analysis of the interaction between FIRST and various first-stage retrievers reveals diminishing returns akin to traditional LLM rerankers. A comprehensive latency study confirms that FIRST consistently delivers a 40% efficiency gain over traditional rerankers without sacrificing effectiveness. Notably, while LM training implicitly improves zero-shot single-token reranking, our experiments also highlight potential conflicts between LM pre-training and subsequent fine-tuning on the FIRST objective. These findings pave the way for more efficient and effective listwise reranking in future applications. Our code is available at: https://rankllm.ai.

The Anserini IR toolkit has come a long way since efforts began in 2015. Although the goals of the project - to bridge research and practice in information retrieval, and to provide reproducible, easy-to-use baselines - have remained constant, the world has changed quite a bit. We discuss how Anserini has evolved in response to this changing environment, the most significant of which is the advent of transformer-based retrieval models that did not exist when the project started. The bi-encoder architecture provides a framework for understanding retrieval models based on dense and sparse vector representations, and offers a reference for conveying the capabilities of our toolkit. Anserini provides end-to-end first-stage retrieval based on single-vector learned dense and sparse representations, directly building on the open-source Lucene search library and the ONNX runtime. This minimal design accelerates the pace of research and fosters reproducibility, enabling ''two-click reproductions''. By better aligning research and practice, we increase the potential real-world impact of research innovations.

Large Language Models (LLMs) have significantly enhanced the capabilities of information access systems, especially with retrieval-augmented generation (RAG). Nevertheless, the evaluation of RAG systems remains a barrier to continued progress, a challenge we tackle in this work by proposing an automatic evaluation framework that is validated against human annotations. We believe that the nugget evaluation methodology provides a solid foundation for evaluating RAG systems. This approach, originally developed for the TREC Question Answering (QA) Track in 2003, evaluates systems based on atomic facts that should be present in good answers. Our efforts focus on ''refactoring'' this methodology, where we describe the AutoNuggetizer framework that specifically applies LLMs to both automatically create nuggets and automatically assign nuggets to system answers. In the context of the TREC 2024 RAG Track, we calibrate a fully automatic approach against strategies where nuggets are created manually or semi-manually by human assessors and then assigned manually to system answers. Based on results from a community-wide evaluation, we observe strong agreement at the run level between scores derived from fully automatic nugget evaluation and human-based variants. The agreement is stronger when individual framework components such as nugget assignment are automated independently. This suggests that our evaluation framework provides tradeoffs between effort and quality that can be used to guide the development of future RAG systems. However, further research is necessary to refine our approach, particularly in establishing robust per-topic agreement to diagnose system failures effectively.

The adoption of large language models (LLMs) as rerankers in multi-stage retrieval systems has gained significant traction in academia and industry. These models refine a candidate list of retrieved documents, often through carefully designed prompts, and are typically used in applications built on retrieval-augmented generation (RAG). This paper introduces RankLLM, an open-source Python package for reranking that is modular, highly configurable, and supports both proprietary and open-source LLMs in customized reranking workflows. To improve usability, RankLLM features optional integration with Pyserini for retrieval and provides integrated evaluation for multi-stage pipelines. Additionally, RankLLM includes a module for detailed analysis of input prompts and LLM responses, addressing reliability concerns with LLM APIs and non-deterministic behavior in Mixture-of-Experts (MoE) models. This paper presents the architecture of RankLLM, along with a detailed step-by-step guide and sample code. We reproduce results from RankGPT, LRL, RankVicuna, RankZephyr, and other recent models. RankLLM integrates with common inference frameworks and a wide range of LLMs. This compatibility allows for quick reproduction of reported results, helping to speed up both research and real-world applications. The complete repository is available at rankllm.ai, and the package can be installed via PyPI.

Retrieval-augmented generation (RAG) enables large language models (LLMs) to generate answers with citations from source documents containing ''ground truth''. A crucial factor in RAG evaluation is ''support'', or whether the information in the cited documents supports the answer. We conducted a comparative study of submissions to the TREC 2024 RAG Track, evaluating an automatic LLM judge (GPT-4o) against human judges for support assessment. We considered two conditions: (1) fully manual assessments from scratch and (2) manual assessments with post-editing of LLM predictions. Our results indicate good agreement between human and GPT-4o predictions. Further analysis of the disagreements shows that an independent human judge correlates better with GPT-4o than a human judge, suggesting that LLM judges can be a reliable alternative for support assessment. We provide a qualitative analysis of human and GPT-4o errors to help guide future evaluations.

Junxiong Wang, Ali Mousavi, Omar Attia, Ronak Pradeep, Saloni Potdar, Alexander Rush, Umar Farooq Minhas, Yunyao Li. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

We provide a reproducible, end-to-end demonstration of vector search with OpenAI embeddings using Lucene on the popular MS MARCO passage ranking test collection. The main goal of our work is to challenge the prevailing narrative that a dedicated vector store is necessary to take advantage of recent advances in deep neural networks as applied to search. Quite the contrary, we show that hierarchical navigable small-world network (HNSW) indexes in Lucene are adequate to provide vector search capabilities in a standard bi-encoder architecture. This suggests that, from a simple cost-benefit analysis, there does not appear to be a compelling reason to introduce a dedicated vector store into a modern "AI stack" for search, since such applications have already received substantial investments in existing, widely deployed infrastructure.

The emerging paradigm of generative retrieval re-frames the classic information retrieval problem into a sequence-to-sequence modeling task, forgoing external indices and encoding an entire document corpus within a single Transformer.Although many different approaches have been proposed to improve the effectiveness of generative retrieval, they have only been evaluated on document corpora on the order of 100K in size.We conduct the first empirical study of generative retrieval techniques across various corpus scales, ultimately scaling up to the entire MS MARCO passage ranking task with a corpus of 8.8M passages and evaluating model sizes up to 11B parameters.We uncover several findings about scaling generative retrieval to millions of passages; notably, the central importance of using synthetic queries as document representations during indexing, the ineffectiveness of existing proposed architecture modifications when accounting for compute cost, and the limits of naively scaling model parameters with respect to retrieval performance.While we find that generative retrieval is competitive with stateof-the-art dual encoders on small corpora, scaling to millions of passages remains an important and unsolved challenge.We believe these findings will be valuable for the community to clarify the current state of generative retrieval, highlight the unique challenges, and inspire new research directions.

With doc2query, we train a neural sequence-to-sequence model that, given an input span of text, predicts a natural language query that the text might answer. These predictions can be viewed as document expansions that feed standard bag-of-words term weighting models such as BM25 or neural retrieval models based on learned sparse lexical representations such as uniCOIL. Previous experiments on the MS MARCO datasets have demonstrated the effectiveness of these methods, and they serve as baselines that are widely used by the community today. Following the recent release of the MS MARCO V2 passage and document ranking test collections, we have refreshed our doc2query and uniCOIL models. This work describes a number of resources that support competitive, reproducible baselines for both the MS MARCO V1 and V2 test collections using our Anserini and Pyserini IR toolkits. Together, they provide a solid foundation for future research on neural retrieval models using the MS MARCO datasets and beyond.

In this work, we propose an effective multi-stage neural ranking system for the clinical trial matching problem. First, we introduce NQS, a neural query synthesis method that leverages a zero-shot document expansion model to generate multiple sentence-long queries from lengthy patient descriptions. These queries are independently issued to a search engine and the results are fused. We find that on the TREC 2021 Clinical Trials Track, this method outperforms strong traditional baselines like BM25 and BM25 + RM3 by about 12 points in [email protected], a relative improvement of 34%. This simple method is so effective that even a state-of-the-art neural relevance ranking method trained on the medical subset of MS MARCO passage, when reranking the results of NQS, fails to improve on the ranked list. Second, we introduce a two-stage neural reranking pipeline trained on clinical trial matching data using tailored ranking templates. In this setting, we can train a pointwise reranker using just 1.1k positive examples and obtain effectiveness improvements over NQS by 24 points. This end-to-end multi-stage system demonstrates a 20% relative effectiveness gain compared to the second-best submission at TREC 2021, making it an important step towards better automated clinical trial matching.

Pyserini is a Python toolkit for reproducible information retrieval research with sparse and dense representations. It aims to provide effective, reproducible, and easy-to-use first-stage retrieval in a multi-stage ranking architecture. Our toolkit is self-contained as a standard Python package and comes with queries, relevance judgments, pre-built indexes, and evaluation scripts for many commonly used IR test collections. We aim to support, out of the box, the entire research lifecycle of efforts aimed at improving ranking with modern neural approaches. In particular, Pyserini supports sparse retrieval (e.g., BM25 scoring using bag-of-words representations), dense retrieval (e.g., nearest-neighbor search on transformer-encoded representations), as well as hybrid retrieval that integrates both approaches. This paper provides an overview of toolkit features and presents empirical results that illustrate its effectiveness on two popular ranking tasks. Around this toolkit, our group has built a culture of reproducibility through shared norms and tools that enable rigorous automated testing.

The COVID-19 pandemic has brought about a proliferation of harmful news articles online, with sources lacking credibility and misrepresenting scientific facts. Misinformation has real consequences for consumer health search, i.e., users searching for health information. In the context of multi-stage ranking architectures, there has been little work exploring whether they prioritize correct and credible information over misinformation. We find that, indeed, training models on standard relevance ranking datasets like MS MARCO passage---which have been curated to contain mostly credible information---yields models that might also promote harmful misinformation. To rectify this, we propose a label prediction technique that can separate helpful from harmful content. Our design leverages pretrained sequence-to-sequence transformer models for both relevance ranking and label prediction. Evaluated at the TREC 2020 Health Misinformation Track, our techniques represent the top-ranked system: Our best submitted run was 19.2 points higher than the second-best run based on the primary metric, a 68% relative improvement. Additional post-hoc experiments show that we can boost effectiveness by another 3.5 points.

Chatty Goose is an open-source Python conversational search framework that provides strong, reproducible reranking pipelines built on recent advances in neural models. The framework comprises extensible modular components that integrate with popular libraries such as Transformers by HuggingFace and ParlAI by Facebook. Our aim is to lower the barrier of entry for research in conversational search by providing reproducible baselines that researchers can build on top of. We provide an overview of the framework and demonstrate how to instantiate a new system from scratch. Chatty Goose incorporates improvements to components that we introduced in the TREC 2019 Conversational Assistance Track (CAsT), where our submission represented the top-performing system. Using our framework, a comparable run can be reproduced with just a few lines of code.