Large Language Models (LLMs) have made query reformulation ubiquitous in modern retrieval and Retrieval-Augmented Generation (RAG) pipelines, enabling the generation of multiple semantically equivalent query variants. However, executing the full pipeline for every reformulation is computationally expensive, motivating selective execution: can we identify the best query variant before incurring downstream retrieval and generation costs? We investigate Query Performance Prediction (QPP) as a mechanism for variant selection across ad-hoc retrieval, and end-to-end RAG. Unlike traditional QPP, which estimates query difficulty across topics, we study intra-topic discrimination—selecting the optimal reformulation among competing variants of the same information need. Through large-scale experiments on TREC-RAG using both sparse and dense retrievers, we evaluate pre- and post-retrieval predictors under correlation- and decision-based metrics. Our results reveal a systematic divergence between retrieval and generation objectives: variants that maximize ranking metrics such as nDCG often fail to produce the best generated answers, exposing a "utility gap" between retrieval relevance and generation fidelity. Nevertheless, QPP can reliably identify variants that improve end-to-end quality over the original query. Notably, lightweight pre-retrieval predictors frequently match or outperform more expensive post-retrieval methods, offering a latency-efficient approach to robust RAG.

Peer review is central to scientific decision-making, yet it is rarely evaluated or audited at scale. Growing submission volumes and the increasing use of large language models (LLMs) in drafting reviews have introduced new challenges for transparency, accountability, and quality control. Today, peer reviews are often produced through hybrid human--AI workflows, where a reviewer may develop the core evaluative ideas while using an LLM to refine wording, restructure arguments, or improve fluency. This shift raises new questions beyond authorship detection alone: Are reviews constructive? Are reviewer claims grounded in the submitted paper? How can we quantify collaboration between human reasoning and AI-assisted writing, and distinguish whether the intellectual contribution or the surface text originates from humans or models?

Large Language Models (LLMs) are now widely used for query reformulation and expansion in Information Retrieval, with many studies reporting substantial effectiveness gains. However, these results are typically obtained under heterogeneous experimental conditions, making it difficult to assess which findings are reproducible and which depend on specific implementation choices. In this work, we present a systematic reproducibility and comparative study of ten representative LLM-based query reformulation methods under a unified and strictly controlled experimental framework. We evaluate methods across two architectural LLM families at two parameter scales, three retrieval paradigms (lexical, learned sparse, and dense), and nine benchmark datasets spanning TREC Deep Learning and BEIR. Our results show that reformulation gains are strongly conditioned on the retrieval paradigm, that improvements observed under lexical retrieval do not consistently transfer to neural retrievers, and that larger LLMs do not uniformly yield better downstream performance. These findings clarify the stability and limits of reported gains in prior work. To enable transparent replication and ongoing comparison, we release all prompts, configurations, evaluation scripts, and run files through QueryGym, an open-source reformulation toolkit with a public leaderboard.\footnote{https://leaderboard.querygym.com}

Peer review is central to scientific publishing, yet reviewers frequently include claims that are subjective, rhetorical, or misaligned with the submitted work. Assessing whether review statements are factual and verifiable is crucial for fairness and accountability. At the scale of modern conferences and journals, manually inspecting the grounding of such claims is infeasible. We present Peerispect, an interactive system that operationalizes claim-level verification in peer reviews by extracting check-worthy claims from peer reviews, retrieving relevant evidence from the manuscript, and verifying the claims through natural language inference. Results are presented through a visual interface that highlights evidence directly in the paper, enabling rapid inspection and interpretation. Peerispect is designed as a modular Information Retrieval (IR) pipeline, supporting alternative retrievers, rerankers, and verifiers, and is intended for use by reviewers, authors, and program committees. We demonstrate Peerispect through a live, publicly available demo at https://app.reviewer.ly/app/peerispect and API services at https://github.com/Reviewerly-Inc/Peerispect, accompanied by a video tutorial https://www.youtube.com/watch?v=pc9RkvkUh14.

Modern information retrieval pipelines combine retrieval, LLM-based generation, and LLM-based judgment, and a poor outcome may originate in any of the three stages. Existing work studies these failures in isolation. This paper instead asks whether query difficulty itself transfers across the three stages: are the same queries hard to retrieve, hard to generate for, and hard to judge? Using four years of TREC Deep Learning benchmarks (2019–2022), we define hard-to-retrieve, hard-to-generate, and hard-to-judge query sets under a unified quartile-based operationalization and analyze their overlap, their stability across system configurations, and the linguistic and semantic causes of failure in each task. We find that the three sets overlap only weakly; three-way overlap is at or below the level expected under independence, indicating that difficulty is largely task-conditioned and does not transfer reliably across stages. The overlap structure is nonetheless stable across retrievers, generators, and judging setups, suggesting that task-specific difficulty is driven by query characteristics interacting with each task's inductive biases rather than by model choice. We further induce a data-driven typology of failure causes and show that conditioning generation on task-relevant difficulty cues yields consistent gains in answer quality.

Large Language Models (LLMs) are increasingly used in scientific peer review, assisting with drafting, rewriting, and refinement. However, existing peer-review LLM detection methods largely treat authorship as a binary problem (human vs. AI) without accounting for the hybrid nature of modern review workflows. In practice, evaluative ideas and surface realization may originate from different sources, creating a spectrum of human-AI collaboration. To address this, we introduce PeerPrism, a large-scale benchmark of 20,690 peer reviews explicitly designed to disentangle idea provenance from text provenance. We construct controlled generation regimes spanning fully human, fully synthetic, and multiple hybrid transformations. We benchmark state-of-the-art LLM text detection methods on PeerPrism. While several methods achieve high accuracy on the standard binary task, their predictions diverge sharply under hybrid regimes. In particular, when ideas originate from humans but the surface text is AI-generated, detectors frequently disagree and produce contradictory classifications. Our results show that current detection methods conflate surface realization with intellectual contribution. Rather than relying on this binary, authorship must be modeled as a multidimensional construct spanning semantic reasoning and stylistic realization. PeerPrism is the first benchmark evaluating human-AI collaboration in these settings. We release all code, data, prompts, and evaluation scripts to facilitate reproducible research at https://github.com/Reviewerly-Inc/PeerPrism.

While large language models (LLMs) have been widely studied in scholarly workflows, e.g., for citation recommendation and literature summarization and more, their role in supporting research integrity remains underexplored. In this talk, we share our experience building and deploying two real-world systems that audit peer reviews and verify their factual grounding at scale. We evaluate the capabilities and limitations of LLMs in two key tasks: (1) assessing review quality along dimensions like specificity and tone, and (2) verifying whether reviewer claims are supported by the submitted paper. Using expert-annotated benchmarks, we compare static metrics, ML baselines, zero-shot LLMs, and fine-tuned models to assess alignment with human judgment. The talk will highlight methodological choices, deployment lessons, and empirical insights into where LLMs succeed and where hybrid approaches with interpretable ML and retrieval perform more reliably. We conclude with reflections on what infrastructure is needed to make use LLMs as a robust foundation for research integrity at scale.

The varying performance of information retrieval (IR) methods, including state-of-the-art transformer-based neural retrievers, across diverse queries poses a significant challenge for achieving robust and reliable retrieval effectiveness. Query Performance Prediction (QPP) seeks to estimate the effectiveness of a retrieval method for individual queries, enabling adaptive strategies to improve retrieval outcomes, particularly for challenging queries. However, existing QPP approaches face fundamental challenges: pre-retrieval methods often rely on surface-level query features that fail to capture the nuanced relationship between queries and retrieval effectiveness, while post-retrieval methods depend heavily on the quality of retrieved documents, which can be unreliable for difficult queries. To this end, we propose the Query Space Distance-Based QPP ( QSD-QPP ) framework, which leverages the deterministic and consistent behavior of retrieval methods to estimate query performance by referencing historical queries with known effectiveness. The approach is motivated by the observation that semantically or syntactically similar queries often exhibit consistent retrieval performance, a property that can be exploited to make reliable predictions for unseen queries. QSD-QPP operates in two modes: (1) a lightweight pre-retrieval instantiation that dynamically constructs a query subspace based on embedding distances to interpolate the performance of proximate historical queries, and (2) an enriched post-retrieval instantiation that incorporates contextualized embeddings, document interactions, and historical query associations to enhance prediction accuracy. By utilizing large-scale contextualized embeddings derived from pre-trained language models, QSD-QPP efficiently identifies semantically similar queries and leverages their performance for robust predictions. By addressing the inherent limitations of prior approaches, QSD-QPP achieves a balanced trade-off between computational efficiency, prediction accuracy, and scalability. We evaluate QSD-QPP on four benchmark datasets, including MS MARCO Dev and TREC Deep Learning tracks (2019, 2020, and DL-Hard), demonstrating its superior accuracy and robustness compared to state-of-the-art baselines in both pre-retrieval and post-retrieval QPP tasks. To ensure reproducibility and encourage further research, we publicly release the implementation of our work.

Ad hoc retrieval, a cornerstone task in Information Retrieval (IR) , aims to rank documents in response to a user’s query, often without prior knowledge of the user’s specific information need. While transformer-based neural rankers have achieved state-of-the-art performance in ad hoc retrieval, their effectiveness varies significantly across queries. Certain queries—commonly referred to as hard queries —remain particularly challenging, highlighting critical gaps in retrieval models. Identifying these hard queries is essential for improving retrieval systems, motivating the task of Query Performance Prediction (QPP) , which aims to estimate the effectiveness of a query without requiring access to relevance judgments. In this article, we propose Context-aware Query Performance Prediction ( CA-QPP ) , a novel post-retrieval QPP method, which builds on the foundations of perturbation-based QPP methods that hypothesize a relationship between query sensitivity to small perturbations and query retrieval effectiveness. Building on this foundation, our approach exposes the given query to perturbations by constructing two query variations: an effective variation emphasizing terms that enhance retrieval and an ineffective variation accentuating terms that hinder it. By contrasting the retrieval outcomes of these variations using a cross-encoder model, CA-QPP captures the interplay of term contributions and predicts the performance for the given query. We evaluate CA-QPP on the widely used MS MARCO datasets and their associated query sets, including TREC DL 2019 , TREC DL 2020 , DL-Hard , TREC DL 2021 , and TREC DL 2022 , which feature extensive human-labeled relevance judgments. Our experiments demonstrate that CA-QPP consistently outperforms traditional and neural-based QPP baselines across standard correlation metrics, including Pearson’s \(\rho\) , Kendall’s \(\tau\) , and Spearman’s \(\rho\) . Through a detailed case study, we further illustrate the mechanics of CA-QPP and provide empirical evidence for its ability to model the contextual impact of individual query terms, making it a robust framework for query performance prediction.

Recent research has shown that neural information retrieval techniques may be susceptible to adversarial attacks. Adversarial attacks seek to manipulate the ranking of documents, with the intention of exposing users to targeted content. In this article, we introduce the Embedding Perturbation Rank Attack ( EMPRA ) method, a novel approach designed to perform adversarial attacks on black-box Neural Ranking Models (NRMs) . EMPRA manipulates sentence-level embeddings, guiding them towards pertinent context related to the query while preserving semantic integrity. This process generates adversarial texts that seamlessly integrate with the original content and remain imperceptible to humans. Our extensive evaluation conducted on the widely used MS MARCO V1 passage collection as well as the TREC DL 2019 and TREC DL 2020 benchmarks, demonstrate the effectiveness of EMPRA against a wide range of state-of-the-art baselines in promoting a specific set of target documents within a given ranked results. Specifically, on MS MARCO Dev set queries, EMPRA successfully achieves a re-ranking of almost 96% of target documents originally ranked between 51 and 100 to rank within the top 10. Furthermore, EMPRA does not rely on surrogate models for generating adversarial documents, enhancing its robustness against various victim NRMs in realistic settings.

Peer review is foundational to academic publishing, yet the quality of reviews remains difficult to assess at scale due to subjectivity, inconsistency, and the lack of standardized evaluation mechanisms. This talk presents our experience developing and deploying a scalable framework for assessing review quality in operational settings. We combine two complementary approaches: interpretable machine learning models built on quantifiable review- and reviewer-level features, and the application of large language models (LLMs), including Qwen, Phi, and GPT-4o, in zero- and few-shot configurations for textual quality evaluation. We also explore the fine-tuning of LLMs on expert-annotated datasets to examine their upper-bound capabilities. To benchmark these methods, we constructed a dataset of over 700 paper-review pairs labeled by domain experts across multiple quality dimensions. Our findings demonstrate that transparent, feature-based models consistently outperform LLMs in reliability and generalization, particularly when evaluating conceptual depth and argumentative structure. The talk will highlight key engineering choices, deployment challenges, and broader implications for integrating automated review evaluation into scholarly workflows.

The quality of peer review plays a critical role in scientific publishing, yet remains poorly understood and challenging to evaluate at scale. In this work, we introduce RottenReviews, a benchmark designed to facilitate systematic assessment of review quality. RottenReviews comprises over 15,000 submissions from four distinct academic venues enriched with over 9,000 reviewer scholarly profiles and paper metadata. We define and compute a diverse set of quantifiable review-dependent and reviewer-dependent metrics, and compare them against structured assessments from large language models (LLMs) and expert human annotations. Our human-annotated subset includes over 700 paper-review pairs labeled across 13 explainable and conceptual dimensions of review quality. Our empirical findings reveal that LLMs, both zero-shot and fine-tuned, exhibit limited alignment with human expert evaluations of peer review quality. Surprisingly, simple interpretable models trained on quantifiable features outperform fine-tuned LLMs in predicting overall review quality. We publicly release all data, code, and models at https://github.com/Reviewerly-Inc/RottenReviews to support further research in this area.

Entity retrieval plays a critical role in information access systems, yet the development and evaluation of retrieval models remain constrained by the limited availability of high-quality supervision. While recent work has demonstrated the utility of large language models (LLMs) as relevance assessors in passage and document retrieval, their reliability in the context of entity retrieval-where targets are abstract, underspecified, and often semantically sparse-remains unexplored. In this work, we evaluate LLM-based judgments against two complementary supervision signals: human-annotated relevance labels from the DBpedia-Entity benchmark and implicit feedback from user clicks in the LaQuE dataset. We show that LLMs exhibit strong agreement with expert annotations and replicate user click patterns with over 91% agreement, suggesting alignment with behavioral judgments despite noisy input queries. We further identify and analyze systematic mismatches for user clicks on irrelevant entities. Our findings establish LLMs not only as effective annotators for entity relevance judgment-even when given only the entity title-but also as powerful tools for predicting click-through behavior and simulating explainable user intent. Our code, prompts, and data are publicly available at: https://github.com/17shiraz/ClickLLM

Large Language Models (LLMs) exhibit strong capabilities across various Information Retrieval (IR) and natural language processing tasks. However, they are highly sensitive to prompt variations, where slight rephrasings can significantly alter responses, leading to inconsistent or incorrect outputs. This variability poses challenges for response reliability in real-world applications. Inspired by Query Performance Prediction (QPP) in IR, we focus on Prompt Performance Prediction (PPP), which estimates whether an LLM will generate a correct response for a given prompt before execution. We propose VAP3 (Variation-Aware Prompt Performance Prediction), a novel pre-generation PPP approach that integrates prompt variations with adversarial training to enhance robustness against trivial modifications and better capture prompt sensitivity. Evaluating VAP3 against LLM-based self-evaluation, QPP-inspired baselines, and supervised classification models on the PromptSET-HotpotQA and PromptSET-TriviaQA datasets, we demonstrate that VAP3 consistently outperforms all baselines, achieving stable and reliable performance across datasets.

Large Language Models (LLMs) are increasingly used to automate relevance judgments for information retrieval (IR) tasks, often demonstrating agreement with human labels that approaches inter-human agreement. To assess the robustness and reliability of LLM-based relevance judgments, we systematically investigate impact of prompt sensitivity on the task. We collected prompts for relevance assessment from 15 human experts and 15 LLMs across three tasks-binary, graded, and pairwise-yielding 90 prompts in total. We compare LLM-generated labels with TREC official human labels using Cohen's κ and pairwise agreement measures. In addition, we compare human- and LLM-generated prompts and analyze differences among different LLMs as judges. We release all data and prompts at https://github.com/Narabzad/prompt-sensitivity-relevance-judgements/.

Large Language Models (LLMs) are increasingly deployed in both academic and industry settings to automate the evaluation of information seeking systems, particularly by generating graded relevance judgments. Several studies report Kendall τ correlations exceeding 0.85 when comparing system rankings derived from human versus LLM-generated relevance labels. Previous work on LLM-based relevance assessment has primarily focused on replicating graded human relevance judgments through various prompting strategies. However, there has been limited exploration of alternative assessment methods or comprehensive comparative studies. In this paper, we systematically compare multiple LLM-based relevance assessment methods, including binary relevance judgments, graded relevance assessments, pairwise preference-based methods, and two nugget-based evaluation methods~-~document-agnostic and document-dependent. Wherever possible, we employ state-of-the-art tools and optimized prompts tailored for these methods. In addition to a traditional comparison based on system rankings using Kendall correlations, we also examine how well LLM judgments align with human preferences, as inferred from relevance grades. We conduct extensive experiments on datasets from three TREC Deep Learning tracks 2019, 2020 and 2021 as well as the ANTIQUE dataset, which focuses on non-factoid open-domain question answering. Beyond dataset-specific results, our work offers a practical methodology for evaluating diverse LLM-based relevance assessment methods. As part of our data release, we include relevance judgments generated by both an open-source (Llama3.2b) and a commercial (gpt-4o) model. Our goal is to reproduce various LLM-based relevance judgment methods to provide a comprehensive comparison. We release all the relevance judgments as a resource that establishes a baseline for future work, ensuring a level playing field for evaluation of LLM-based relevance judgments. All code, data, and resources are publicly available in our GitHub Repository at https://github.com/Narabzad/llm-relevance-judgement-comparison

In recent years, Retrieval-Augmented Generation (RAG) systems have become a cornerstone of artificial intelligence, attracting considerable attention in a variety of fields. By integrating the strengths of information retrieval and generative models, these systems have shown immense potential to push the boundaries of machine learning applications. Nevertheless, RAG systems still face significant challenges and offer ample room for advancement and innovation.

Seamless interaction between AI agents and humans using natural language remains a key goal in AI research. This paper addresses the challenges of developing interactive agents capable of understanding and executing grounded natural language instructions through the IGLU competition. Despite advancements, challenges such as a scarcity of appropriate datasets and the need for effective evaluation platforms persist. We introduce a scalable data collection tool for gathering interactive grounded language instructions within a Minecraft-like environment, resulting in a Multi-Modal dataset with around 9,000 utterances and over 1,000 clarification questions. Additionally, we present a Human-in-the-Loop interactive evaluation platform for qualitative analysis and comparison of agent performance through multi-turn communication with human annotators. We offer to the community these assets referred to as IDAT (IGLU Dataset And Toolkit) which aim to advance the development of intelligent, interactive AI agents and provide essential resources for further research.

Query performance prediction (QPP) is a key task in information retrieval (IR), focusing on estimating the retrieval quality of a given query without relying on human-labeled relevance judgments. Over the decades, QPP has gained increasing significance, with a surge in research activity in recent years. It has proven to benefit various aspects of retrieval, such as optimizing retrieval effectiveness by selecting the most appropriate ranking function for each query.

Query performance prediction (QPP) aims to estimate the retrieval quality of a search system for a query without human relevance judgments. Previous QPP methods typically return a single scalar value and do not require the predicted values to approximate a specific information retrieval (IR) evaluation measure, leading to certain drawbacks: (i) a single scalar is insufficient to accurately represent different IR evaluation measures, especially when metrics do not highly correlate, and (ii) a single scalar limits the interpretability of QPP methods because solely using a scalar is insufficient to explain QPP results. To address these issues, we propose a QPP framework using automatically gen erated re levance judgments (QPP-GenRE), which decomposes QPP into independent subtasks of predicting the relevance of each item in a ranked list to a given query. This allows us to predict any IR evaluation measure using the generated relevance judgments as pseudo-labels. This also allows us to interpret predicted IR evaluation measures, and identify, track, and rectify errors in generated relevance judgments to improve QPP quality. We predict an item’s relevance by using open source large language models (LLMs) to ensure scientific reproducibility. We face two main challenges: (i) excessive computational costs of judging an entire corpus for predicting a metric considering recall, and (ii) limited performance in prompting open source LLMs in a zero-/few-shot manner. To solve the challenges, we devise an approximation strategy to predict an IR measure considering recall and propose to fine-tune open source LLMs using human-labeled relevance judgments. Experiments on the TREC 2019–2022 deep learning tracks and CAsT-19–20 datasets show that QPP-GenRE achieves state-of-the-art QPP quality for both lexical and neural rankers.

The peer review process is a fundamental aspect of academic publishing, ensuring the quality and credibility of scholarly work. In this talk, we will explore the critical challenges associated specifically with the assignment of reviewers to submitted papers. We will introduce Reviewerly, our innovative solution designed to enhance the efficiency and effectiveness of reviewer assignments by leveraging data from diverse sources, including OpenAlex, PubMed, and DBLP. By modeling the reviewer assignment problem as an information retrieval task, we focus on retrieving a pool of relevant and diverse reviewers for each paper.

Prior research has demonstrated that reformulation of queries can significantly enhance retrieval effectiveness. Despite notable successes in neural-based query reformulation methods, identifying optimal reformulations that cover the same information need while enhancing retrieval effectiveness is still challenging. This paper introduces a two-step query reformulation framework for generating and selecting optimal target query variants which not only achieve higher retrieval performance but also preserve the original query's information need. Our comprehensive evaluations on the MS MARCO dataset and TREC Deep Learning tracks demonstrate substantial improvements over original query's performance.

The rapid advancement of natural language processing, information retrieval (IR), computer vision, and other technologies has presented significant challenges in evaluating the performance of these systems.One of the main challenges is the scarcity of human-labeled data, which hinders the fair and accurate assessment of these systems.In this work, we specifically focus on evaluating IR systems with sparse labels, borrowing from recent research on evaluating computer vision tasks.taking inspiration from the success of using Fréchet Inception Distance (FID) in assessing text-to-image generation systems.We propose leveraging the Fréchet Distance to measure the distance between the distributions of relevant judged items and retrieved results.Our experimental results on MS MARCO V1 dataset and TREC Deep Learning Tracks query sets demonstrate the effectiveness of the Fréchet Distance as a metric for evaluating IR systems, particularly in settings where a few labels are available.This approach contributes to the advancement of evaluation methodologies in real-world scenarios such as the assessment of generative IR systems.

Negar Arabzadeh, Siqing Huo, Nikhil Mehta, Qingyun Wu, Chi Wang, Ahmed Hassan Awadallah, Charles L. A. Clarke, Julia Kiseleva. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

We study ranked list truncation (RLT) from a novel retrieve-then-re-rank perspective, where we optimize re-ranking by truncating the retrieved list (i.e., trim re-ranking candidates). RLT is crucial for re-ranking as it can improve re-ranking efficiency by sending variable-length candidate lists to a re-ranker on a per-query basis. It also has the potential to improve re-ranking effectiveness. Despite its importance, there is limited research into applying RLT methods to this new perspective. To address this research gap, we reproduce existing RLT methods in the context of re-ranking, especially newly emerged large language model (LLM)-based re-ranking. In particular, we examine to what extent established findings on RLT for retrieval are generalizable to the "retrieve-then-re-rank" setup from three perspectives: (i) assessing RLT methods in the context of LLM-based re-ranking with lexical first-stage retrieval, (ii) investigating the impact of different types of first-stage retrievers on RLT methods, and (iii) investigating the impact of different types of re-rankers on RLT methods. We perform experiments on the TREC 2019 and 2020 deep learning tracks, investigating 8 RLT methods for pipelines involving 3 retrievers and 2 re-rankers. We reach new insights into RLT methods in the context of re-ranking.

Estimating query difficulty, also known as Query Performance Prediction (QPP), is concerned with assessing the retrieval quality of a ranking method for an input query. Most traditional unsupervised frequency-based models and many recent supervised neural methods have been designed specifically for predicting the performance of sparse retrievers such as BM25. In this paper we propose an unsupervised QPP method for dense neural retrievers which operates by redefining the well-known concept of query robustness i.e., a more robust query to perturbations is an easier query to handle. We propose to generate query perturbations for measuring query robustness by systematically injecting noise into the contextualized neural representation of each query. We then compare the retrieved list for the original query with that of the perturbed query as a way to measure query robustness. Our experiments on four different query sets including MS MARCO, TREC Deep Learning track 2019 and 2020 and TREC DL-Hard show consistently improved performance on linear and ranking correlation metrics over the state of the art.

Researchers have shown that the retrieval effectiveness of queries may depend on other factors in addition to the semantics of the query. In other words, several queries expressed with the same intent, and even using overlapping keywords, may exhibit completely different degrees of retrieval effectiveness. As such, the objective of our work in this paper is to propose a neural disentanglement method that is able to disentangle query semantics from query difficulty. The disentangled query semantics representation provides the means to determine semantic association between queries whereas the disentangled query difficulty representation would allow for the estimation of query effectiveness. We show through our experiments on the query performance prediction; and, query similarity calculation tasks that our proposed disentanglement method is able to show better performance compared to the state of the art.

The information retrieval community has observed significant performance improvements over various tasks due to the introduction of neural architectures. However, such improvements do not necessarily seem to have happened uniformly across a range of queries. As we will empirically show in this paper, the performance of neural rankers follow a long-tail distribution where there are many subsets of queries, which are not effectively satisfied by neural methods. Despite this observation, performance is often reported using standard retrieval metrics, such as MRR or nDCG, which capture average performance over all queries. As such, it is not clear whether reported improvements are due to incremental boost on a small subset of already well-performing queries or addressing queries that have been difficult to address by existing methods. In this paper, we propose the Task Subspace Coverage (TaSC /tAHsk/) metric, which systematically quantifies whether and to what extent improvements in retrieval effectiveness happen on similar or disparate query subspaces for different rankers. Our experiments show that the consideration of our proposed TaSC metric in conjunction with existing ranking metrics provides deeper insight into ranker performance and their contribution to overall advances on a given task.

Query performance prediction (QPP) is a core task in information retrieval. The QPP task is to predict the retrieval quality of a search system for a query without relevance judgments. Research has shown the effectiveness and usefulness of QPP for ad-hoc search. Recent years have witnessed considerable progress in conversational search (CS). Effective QPP could help a CS system to decide an appropriate action to be taken at the next turn. Despite its potential, QPP for CS has been little studied. We address this research gap by reproducing and studying the effectiveness of existing QPP methods in the context of CS. While the task of passage retrieval remains the same in the two settings, a user query in CS depends on the conversational history, introducing novel QPP challenges. In particular, we seek to explore to what extent findings from QPP methods for ad-hoc search generalize to three CS settings: (i) estimating the retrieval quality of different query rewriting-based retrieval methods, (ii) estimating the retrieval quality of a conversational dense retrieval method, and (iii) estimating the retrieval quality for top ranks vs. deeper-ranked lists. Our findings can be summarized as follows: (i) supervised QPP methods distinctly outperform unsupervised counterparts only when a large-scale training set is available; (ii) point-wise supervised QPP methods outperform their list-wise counterparts in most cases; and (iii) retrieval score-based unsupervised QPP methods show high effectiveness in assessing the conversational dense retrieval method, ConvDR.

Preference judgments have been established as an effective method for offline evaluation of information retrieval systems with advantages to graded or binary relevance judgments. Graded judgments assign each document a pre-defined grade level, while preference judgments involve assessing a pair of items presented side by side and indicating which is better. However, leveraging preference judgments may require a more extensive number of judgments, and there are limitations in terms of evaluation measures. In this study, we present a new preference judgment tool called JUDGO, designed for expert assessors and researchers. The tool is supported by a new heap-like preference judgment algorithm that assumes transitivity and allows for ties. An earlier version of the tool was employed by NIST to determine up to the top-10 best items for each of the 38 topics for the TREC 2022 Health Misinformation track, with over 2,200 judgments collected. The current version has been applied in a separate research study to collect almost 10,000 judgments, with multiple assessors completing each topic. The code and resources are available at https://judgo-system.github.io.

Despite recent progress on conversational systems, they still do not perform smoothly when faced with ambiguous requests. When questions are unclear, conversational systems should have the ability to ask clarifying questions, rather than assuming a particular interpretation or simply responding that they do not understand. While the research community has paid substantial attention to the problem of predicting query ambiguity in traditional search contexts, researchers have paid relatively little attention to predicting when this ambiguity is sufficient to warrant clarification in the context of conversational systems. In this paper, we propose an unsupervised method for predicting the need for clarification. This method is based on the measured coherency of results from an initial answer retrieval step, under the assumption that a less ambiguous query is more likely to retrieve more coherent results when compared to an ambiguous query. We build a graph from retrieved items based on their context similarity, treating measures of graph connectivity as indicators of ambiguity. We evaluate our approach on two open-domain conversational question answering datasets, ClariQ and AmbigNQ, comparing it with neural and non-neural baselines. Our unsupervised approach performs as well as supervised approaches while providing better generalization.

Despite the advantages of their low-resource settings, traditional sparse retrievers depend on exact matching approaches between high-dimensional bag-of-words (BoW) representations of both the queries and the collection. As a result, retrieval performance is restricted by semantic discrepancies and vocabulary gaps. On the other hand, transformer-based dense retrievers introduce significant improvements in information retrieval tasks by exploiting low-dimensional contextualized representations of the corpus. While dense retrievers are known for their relative effectiveness, they suffer from lower efficiency and lack of generalization issues, when compared to sparse retrievers. For a lightweight retrieval task, high computational resources and time consumption are major barriers encouraging the renunciation of dense models despite potential gains. In this work, we propose boosting the performance of sparse retrievers by expanding both the queries and the documents with linked entities in two formats for the entity names: 1) explicit and 2) hashed. We employ a zero-shot end-to-end dense entity linking system for entity recognition and disambiguation to augment the corpus. By leveraging the advanced entity linking methods, we believe that the effectiveness gap between sparse and dense retrievers can be narrowed. We conduct our experiments on the MS MARCO passage dataset. Since we are concerned with the early stage retrieval in cascaded ranking architectures of large information retrieval systems, we evaluate our results using [email protected] Our approach is also capable of retrieving documents for query subsets judged to be particularly difficult in prior work. We further demonstrate that the non-expanded and the expanded runs with both explicit and hashed entities retrieve complementary results. Consequently, we adopt a run fusion approach to maximize the benefits of entity linking.

Recent studies have shown that it is possible for stereotypical gender biases to find their way into representational and algorithmic aspects of retrieval methods; hence, exhibit themselves in retrieval outcomes. In this tutorial, we inform the audience of various studies that have systematically reported the presence of stereotypical gender biases in Information Retrieval (IR) systems. We further classify existing work on gender biases in IR systems as being related to (1) relevance judgement datasets, (2) structure of retrieval methods, and (3) representations learnt for queries and documents. We present how each of these components can be impacted by or cause intensified biases during retrieval. Based on these identified issues, we then present a collection of approaches from the literature that have discussed how such biases can be measured, controlled, or mitigated. Additionally, we introduce publicly available datasets that are often used for investigating gender biases in IR systems as well as evaluation methodology adopted for determining the utility of gender bias mitigation strategies.

While neural rankers continue to show notable performance improvements over a wide variety of information retrieval tasks, there have been recent studies that show such rankers may intensify certain stereotypical biases. In this paper, we investigate whether neural rankers introduce retrieval effectiveness (performance) disparities over queries related to different genders. We specifically study whether there are significant performance differences between male and female queries when retrieved by neural rankers. Through our empirical study over the MS MARCO collection, we find that such performance disparities are notable and that the performance disparities may be due to the difference between how queries and their relevant judgements are collected and distributed for different gendered queries. More specifically, we observe that male queries are more closely associated with their relevant documents compared to female queries and hence neural rankers are able to more easily learn associations between male queries and their relevant documents. We show that it is possible to systematically balance relevance judgment collections in order to reduce performance disparity between different gendered queries without negatively compromising overall model performance.

Researchers have already shown that it is possible to improve retrieval effectiveness through the systematic reformulation of users' queries. Traditionally, most query reformulation techniques relied on unsupervised approaches such as query expansion through pseudo-relevance feedback. More recently and with the increasing effectiveness of neural sequence-to-sequence architectures, the problem of query reformulation has been studied as a supervised query translation problem, which learns to rewrite a query into a more effective alternative. While quite effective in practice, such supervised query reformulation methods require a large number of training instances. In this paper, we present three large-scale query reformulation datasets, namely Diamond, Platinum and Gold datasets, based on the queries in the MS MARCO dataset. The Diamond dataset consists of over 188,000 query pairs where the original source query is matched with an alternative query that has a perfect retrieval effectiveness (an average precision of 1). To the best of our knowledge, this is the first set of datasets for supervised query reformulation that offers perfect query reformulations for a large number of queries. The implementation of our fully automated tool, which is based on a transformer architecture, and our three datasets are made publicly available. We also establish a neural query reformulation baseline performance on our datasets by reporting the performance of strong neural query reformulation baselines. It is our belief that our datasets will significantly impact the development of supervised query reformulation methods in the future.

Query Performance Prediction (QPP) is focused on estimating the difficulty of satisfying a user query for a certain retrieval method. While most state of the art QPP methods are based on term frequency and corpus statistics, more recent work in this area have started to explore the utility of pretrained neural embeddings, neural architectures and contextual embeddings. Such approaches extract features from pretrained or contextual embeddings for the sake of training a supervised performance predictor. In this paper, we adopt contextual embeddings to perform performance prediction, but distinguish ourselves from the state of the art by proposing to directly fine-tune a contextual embedding, i.e., BERT, specifically for the task of query performance prediction. As such, our work allows the fine-tuned contextual representations to estimate the performance of a query based on the association between the representation of the query and the retrieved documents. We compare the performance of our approach with the state-of-the-art based on the MS MARCO passage retrieval corpus and its three associated query sets: (1) MS MARCO development set, (2) TREC DL 2019, and (3) TREC DL 2020. We show that our approach not only shows significant improved prediction performance compared to all the state-of-the-art methods, but also, unlike past neural predictors, it shows significantly lower latency, making it possible to use in practice.

During the recent years and with the growing influence of neural architectures, tasks such as ad hoc retrieval have witnessed an impressive improvement in performance. For instance, the performance of rankers on the passage retrieval task on the MS MARCO dataset has improved by an order of magnitude in less than two years. In this paper, we go beyond the overall performance of the state of the art rankers and empirically study their performance from a finer-grained perspective. We find that while neural rankers have been able to consistently improve performance, this has been in part thanks to a specific set of queries from within the larger query set. We systematically show that there are subsets of queries that are difficult for each and every one of the neural rankers, which we refer to as obstinate queries. We show the obstinate queries are similar to easier queries in terms of their number of available relevant judgement documents and the length of the query itself but they are extremely more difficult to satisfy by existing rankers. Furthermore, we observe that query reformulation methods cannot help these queries. On this basis, we present three datasets derived from the MS MARCO Dev set, called the MS MARCO Chameleon datasets. We believe that the next breakthrough in performance would need to necessarily consider the queries in the MS MARCO Chameleons, as such, propose that a well-rounded evaluation strategy for any new ranker would need to include performance measures on both the overall MS MARCO dataset as well as the proposed MS MARCO Chameleon datasets.

Over the last few years, contextualized pre-trained transformer models such as BERT have provided substantial improvements on information retrieval tasks. Traditional sparse retrieval methods such as BM25 rely on high-dimensional, sparse, bag-of-words query representations to retrieve documents. On the other hand, recent approaches based on pre-trained transformer models such as BERT, fine-tune dense low-dimensional contextualized representations of queries and documents in embedding space. While these dense retrievers enjoy substantial retrieval effectiveness improvements compared to sparse retrievers, they are computationally intensive, requiring substantial GPU resources, and dense retrievers are known to be more expensive from both time and resource perspectives. In addition, sparse retrievers have been shown to retrieve complementary information with respect to dense retrievers, leading to proposals for hybrid retrievers. These hybrid retrievers leverage low-cost, exact-matching based sparse retrievers along with dense retrievers to bridge the semantic gaps between query and documents. In this work, we address this trade-off between the cost and utility of sparse vs dense retrievers by proposing a classifier to select a suitable retrieval strategy (i.e., sparse vs. dense vs. hybrid) for individual queries. Leveraging sparse retrievers for queries which can be answered with sparse retrievers decreases the number of calls to GPUs. Consequently, while utility is maintained, query latency decreases. Although we use less computational resources and spend less time, we still achieve improved performance. Our classifier can select between sparse and dense retrieval strategies based on the query alone. We conduct experiments on the MS MARCO passage dataset demonstrating an improved range of efficiency/effectiveness trade-offs between purely sparse, purely dense or hybrid retrieval strategies, allowing an appropriate strategy to be selected based on a target latency and resource budget.

Various researchers have recently explored the impact of different types of biases on information retrieval tasks such as ad hoc retrieval and question answering. While the impact of bias needs to be controlled in order to avoid increased prejudices, the literature has often viewed the relationship between increased retrieval utility (effectiveness) and reduced bias as a tradeoff where one can suffer from the other. In this paper, we empirically study this tradeoff and explore whether it would be possible to reduce bias while maintaining similar retrieval utility. We show this would be possible by revising the input query through a bias-aware pseudo-relevance feedback framework. We report our findings based on four widely used TREC corpora namely Robust04, Gov2, ClueWeb09 and ClueWeb12 and using two classes of bias metrics. The findings of this paper are significant as they are among the first to show that decrease in bias does not necessarily need to come at the cost of reduced utility.

We introduce Vis-Trec, an open-source cross-platform system, which provides the capability to perform in-depth analysis of the results obtained from trec-style evaluation campaigns. Vis-Trec allows researchers to dig deeper in their evaluations by providing various visualizations of the results based on performance percentiles, query difficulty, and comparative analysis of different methods using help-hurt diagrams at the query level. It also automatically organizes the obtained results in tabular LaTeX format that can be used for reporting evaluation findings. The added benefit for Vis-Trec is that it has been developed in Python and is extensible by other developers. The source code along with a functional version of the program are released to the public.

Specificity is the level of detail at which a given term is represented. Existing approaches to estimating term specificity are primarily dependent on corpus-level frequency statistics. In this work, we explore how neural embeddings can be used to define corpus-independent specificity metrics. Particularly, we propose to measure term specificity based on the distribution of terms in the neighborhood of the given term in the embedding space. The intuition is that a term that is surrounded by other terms in the embedding space is more likely to be specific while a term surrounded by less closely related terms is more likely to be generic. On this basis, we leverage geometric properties between embedded terms to define three groups of metrics: (1) neighborhood-based, (2) graph-based and (3) cluster-based metrics. Moreover, we employ learning-to-rank techniques to estimate term specificity in a supervised approach by employing the three proposed groups of metrics. We curate and publicly share a test collection of term specificity measurements defined based on Wikipedia's category hierarchy. We report on our experiments through metric performance comparison, ablation study and comparison against the state-of-the-art baselines.

The accurate prediction of users' future topics of interests on social networks can facilitate content recommendation and platform engagement. However, researchers have found that future interest prediction, especially on social networks such as Twitter, is quite challenging due to the rapid changes in community topics and evolution of user interactions. In this context, temporal collaborative filtering methods have already been used to perform user interest prediction, which benefit from similar user behavioral patterns over time to predict how a user's interests might evolve in the future. In this paper, we propose that instead of considering the whole user base within a collaborative filtering framework to predict user interests, it is possible to much more accurately predict such interests by only considering the behavioral patterns of the most influential users related to the user of interest. We model influence as a form of causal dependency between users. To this end, we employ the concept of Granger causality to identify causal dependencies. We show through extensive experimentation that the consideration of only one causally dependent user leads to much more accurate prediction of users' future interests in a host of measures including ranking and rating accuracy metrics.