Nugget-based LLM judges evaluate Retrieval-Augmented Generation (RAG) systems using a bank of questions that capture the key facts and criteria an answer should address. These nugget banks are typically constructed through a combination of human input and LLM generation. System outputs are graded by how well they cover the nuggets. For cost and scalability reasons, the nugget bank should be small. However, a major limitation of current nugget generation approaches is that many questions are overly generic and fail to discriminate between top-performing RAG systems. Grounding nuggets in system responses or source documents can increase specificity, but typically leads to an explosion in the number of questions. Since every response is graded for every nugget question, a higher number of questions directly increases the amount of LLM prompts and/or tokens required, contributing to costs. Inspired by preference-based evaluation, we derive differential nuggets from winner-loser passage pairs, focusing on information that captures differences in topicality, level of detail, and evidential support between responses under an automatic preference judge. We examine how these contrastive signals can be leveraged to construct nugget banks that are both compact and discriminative, enabling reliable separation among top-performing RAG systems.

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.

The unjudged document problem, where systems that did not contribute to the original judgement pool may retrieve documents without a relevance judgement, is a key obstacle to the reuseability of test collections in information retrieval. While the de facto standard to deal with the problem is to treat unjudged documents as non-relevant, many alternatives have been proposed, such as the use of large language models (LLMs) as a relevance judge (LLM-as-a-judge). However, this has been criticized, among other things, as circular, since the same LLM can be used as the ranker and the judge. We propose to train topic-specific relevance classifiers instead: By finetuning monoT5 with independent LoRA weight adaptation on the judgments of a single assessor for a single topic's pool, we align it to that assessor's notion of relevance for the topic. The system rankings obtained through our classifier's relevance judgments achieve a Spearmans' $ρ$ correlation of $>0.94$ with ground truth system rankings. As little as 128 initial human judgments per topic suffice to improve the comparability of models, compared to treating unjudged documents as non-relevant, while achieving more reliability than existing LLM-as-a-judge approaches. Topic-specific relevance classifiers are thus a lightweight and straightforward way to tackle the unjudged document problem, while maintaining human judgments as the gold standard for retrieval evaluation. Code, models, and data are made openly available.

Long-form Retrieval-Augmented Generation (RAG) brings the challenge of coverage-based ranking, because ranking methods must ensure the inclusion of comprehensive relevant nuggets (i.e., facts), which can thereby be synthesized into a comprehensive output. In this work, we propose CoveR, a dense retrieval method optimized for coverage-aware retrieval scenarios. CoveR is a bi-encoder trained with the coverage-based contrastive and distillation objectives, which enables CoveR to capture diverse aspects of information needs. To train CoveR, we create the SCOPE dataset, which comprises 90K training pairs from Researchy Questions with synthetic coverage signals augmented from sub-question answerability judgments generated by LLMs. Our empirical experiments show that CoveR enhances nugget coverage by 10% over strong dense retrieval baselines without sacrificing its relevance-based retrieval capability. Further ablation studies validate the importance of our proposed learning method, showing that CoveR achieves a superior trade-off between relevance- and coverage-based ranking, which is essential for long-form RAG.

Advancements in retrieval models necessitate re-indexing, a computationally expensive process for large-scale production environments. While updating only the query encoder and continuing to use the old index could be a promising middle ground, dense retrieval systems suffer severe performance drops in this setting. We investigate whether Learned Sparse Retrieval (LSR) can mitigate this backward compatibility issue, as its lexical matching may provide a stable term-based anchor to preserve compatibility across model versions. Experiments on BEIR and the streaming settings of LoTTE show that upgrading only the query encoder causes only a small effectiveness drop in LSR when no mitigation applied, whereas dense retrieval fails severely. We explore lightweight query adaptation methods including ranking fusion, representation fusion, and minimal-training adapters to further improve compatibility. These approaches significantly improve backward compatibility on BEIR and effectively reduce performance loss in streaming retrieval. Code: https://github.com/JingfenQiao/LSR-BC.git

We wish to measure the information coverage of an ad hoc retrieval algorithm, that is, how much of the range of available relevant information is covered by the search results. Information coverage is a central aspect for retrieval, especially when the retrieval system is integrated with generative models in a retrieval-augmented generation (RAG) system. The classic metrics for ad hoc retrieval, precision and recall, reward a system as more relevant documents are retrieved. However, since relevance in ad hoc test collections is defined for a document without any relation to other documents that might contain the same information, high recall is sufficient but not necessary to ensure coverage. The same is true for other metrics such as rank-biased precision (RBP), normalized discounted cumulative gain (nDCG), and mean average precision (MAP). Test collections developed around the notion of diversity ranking in web search incorporate multiple aspects that support a concept of coverage in the web domain. In this work, we construct a benchmark, CoverageBench, for evaluating information coverage made from existing collections. This suite offers researchers a unified testbed spanning multiple genres and tasks. All topics, nuggets, relevance labels, and baseline rankings are released on Hugging Face Datasets, along with instructions for accessing the publicly available document collections.

Prior work has shown that vector-based pseudo relevance feedback (PRF) is an effective technique for query expansion for improving retrieval results in dense information retrieval. In dense retrieval, ColBERT-PRF has emerged as a novel mechanism, using cluster centroids built from feedback documents as PRF expansion tokens and leveraging statistical information from the closest neighboring token ids to dictate how useful these expansion tokens are. While this approach has been shown to work well in the monolingual retrieval setting for English using the original ColBERT infrastructure, such systems have since evolved to improve inference speed, reduce storage and memory usage, and support cross-language (CLIR) and multilingual (MLIR) retrieval. As a result, many of these advancements have reduced the ability to utilize token-level statistics. In this work, we aim to explore how well this type of approach can adapt to dense retrieval models when it is not feasible to use surface-form information to pick discriminating expansion tokens. Furthermore, we explore alternative clustering mechanisms, such as HDBScan, to compare how different clustering methods perform at building clusters that can be useful for PRF. Experiments on MLIR, CLIR, and Report Generation tasks, such as those in the TREC 2024 NeuCLIR Report Generation Pilot Task, show that even without access to these token statistics, the use of cluster centroids for PRF can still improve nDCG and α-nDCG by up to 12%.

Generation of citation-backed reports is a primary use case for retrieval-augmented generation (RAG) systems. While open-source evaluation tools exist for various RAG tasks, tools designed for report generation are lacking. Accordingly, we introduce Auto-ARGUE, a robust LLM-based implementation of the recently proposed ARGUE framework for report generation evaluation. We present analysis of Auto-ARGUE on the report generation pilot task from the TREC 2024 NeuCLIR track and on two tasks from the TREC 2024 RAG track, showing good system-level correlations with human judgments. Additionally, we release ARGUE-viz, a web app for visualization and fine-grained analysis of Auto-ARGUE judgments and scores1.

While ColBERT is an effective neural retrieval architecture, it requires a heavy index structure to support candidate set retrieval based on approximated token embeddings, gathering and decompressing document token embeddings, and applying the MaxSim operation. Indexes in PLAID and similar ColBERT implementations require five to ten times the disk storage of the original raw text, which limits their scalability. Furthermore, prior work has identified that the gathering and decompression stages are the primary inefficiencies at query time. Limiting the number of document tokens that must be gathered by thresholding and score approximation does not eliminate the need for the entire index to support ad hoc queries. In this work, we propose an embedding quantization approach that turns a ColBERT index into a true inverted index. We show that, theoretically, ColBERT with embedding quantization is equivalent to learned-sparse retrieval except for the scoring mechanism. Empirically, we demonstrate that our index is 50-70% smaller than a one-bit PLAID index while retaining retrieval effectiveness.

Multilingual retrieval has become a core functionality in information systems, such as retrieval-augmented generation, since the retrieved document can be post-processed by a large language model into a form that the users can easily consume. Whether the user can directly read the source documents becomes less vital to the search process. However, ad hoc retrieval has primarily been developed in English, limiting the capability of these complex systems to retrieve information across different languages. To address this challenge, we introduce a multilingual retrieval task at WSDM Cup 2026 that requires systems to search a multilingual collection using English queries.

Large language models (LLMs) can be effective at retrieval but are generally too expensive to use as first-stage rankers.As a consequence, several approaches to their use as rerankers of less expensive first-stage retrieval results have been suggested.Zhuang et al. recommend setwise approaches.In particular, they use heapsort to efficiently return top-ranked documents by obtaining a partial order from each LLM call.Utilizing rerankers requires setting parameters such as number of input documents, number of documents to be reranked, evaluation depth, number of tokens per document, and selection of LLM, all of which contribute to latency and effectiveness.In this work, we reproduce the batching and reranking of Zhuang et al. with a larger comparison window size.Furthermore, we determine that the document truncation used in the original implementation is suboptimal.By providing more context to the LLM, we show that the reranker is more effective than originally reported.

Most pretrained language models that support neural information retrieval are fine-tuned on the MS MARCO dataset.MS MARCO is expressed in English, so it naturally supports monolingual English retrieval.However, for Cross-Language Information Retrieval (CLIR), no similar training data naturally exists that matches the languages of the query and the documents.The main ways to address this problem have been to continue to fine-tune with English data, or to translate MS MARCO queries and/or documents to match the CLIR setting.Machine translation often introduces errors that reduce retrieval effectiveness.It is usually easy to find target language documents suitable for training, but difficult to find naturally-occurring queries in the query language.An alternative is to train on naturally-occurring documents and syntheticallygenerated queries.Generate-Distill uses this approach with stateof-the-art distillation methods to match the effectiveness of training with translated MS MARCO across different domains.

The fundamental property of Cranfield-style evaluations, that system rankings are stable even when assessors disagree on individual relevance decisions, was validated on traditional test collections. However, the paradigm shift towards neural retrieval models affected the characteristics of modern test collections, e.g., documents are short, judged with four grades of relevance, and information needs have no descriptions or narratives. Under these changes, it is unclear whether assessor disagreement remains negligible for system comparisons. We investigate this aspect under the additional condition that the few modern test collections are heavily re-used. Given more possible query interpretations due to less formalized information needs, an ''expiration date'' for test collections might be needed if top-effectiveness requires overfitting to a single interpretation of relevance. We run a reproducibility study and re-annotate the relevance judgments of the 2019~TREC Deep Learning track. We can reproduce prior work in the neural retrieval setting, showing that assessor disagreement does not affect system rankings. However, we observe that some models substantially degrade with our new relevance judgments, and some have already reached the effectiveness of humans as rankers, providing evidence that test collections can expire.

Videos inherently contain multiple modalities, including visual events, text overlays, sounds, and speech, all of which are important for retrieval. However, state-of-the-art multimodal language models like VAST and LanguageBind are built on vision-language models (VLMs), and thus overly prioritize visual signals. Retrieval benchmarks further reinforce this bias by focusing on visual queries and neglecting other modalities. We create a search system MMMORRF that extracts text and features from both visual and audio modalities and integrates them with a novel modality-aware weighted reciprocal rank fusion. MMMORRF is both effective and efficient, demonstrating practicality in searching videos based on users' information needs instead of visual descriptive queries. We evaluate MMMORRF on MultiVENT 2.0 and TVR, two multimodal benchmarks designed for more targeted information needs, and find that it improves nDCG@20 by 81% over leading multimodal encoders and 37% over single-modality retrieval.

Recent work has shown that Large Language Models (LLMs) can produce relevance judgements for English retrieval that are useful as a basis for system comparison, and they do so at vastly reduced cost compared to human assessors. Using relevance judgements and ranked retrieval runs from the TREC NeuCLIR track, this paper shows that LLMs can also produce reliable assessments in other languages, even when the topic description or the prompt are in a language different from the documents. Results with Chinese, Persian and Russian documents show that although document language affects both agreement with human assessors on graded relevance and on preference ordering among systems, prompt-language and topic-language effects are negligible. This has implications for the design of multilingual test collections, suggesting that prompts and topic descriptions can be developed in any convenient language.

Retrieval-augmented generation (RAG) summarizes retrieved documents into a text passage that fulfills the information need expressed by the user. Such generated responses should faithfully distill the relevant information and provide sufficient attribution back to the source documents. Nugget-based evaluation was proposed for text summarization and has been adapted to evaluate RAG output in recent shared tasks such as 2024 TREC RAG, BioGen, and NeuCLIR tracks. However, annotating such detailed and nuanced information is complex and errorful. Multiple pieces of information need to be labeled, extracted, linked, and cross-referenced. In this work, we present an annotation protocol and tool tailored to collecting information for evaluating RAG systems. Our tool has four steps: nugget creation, nugget revision, document support assessment, and finally, nugget alignment. Each step aims to minimize the annotator's cognitive load, improving the efficiency and reliability.

Learned Sparse Retrieval (LSR) techniques use neural machinery to represent queries and documents as learned bags of words. In contrast with other neural retrieval techniques, such as generative retrieval and dense retrieval, LSR has been shown to be a remarkably robust, transferable, and efficient family of methods for retrieving high-quality search results. This half-day tutorial aims to provide an extensive overview of LSR, ranging from its fundamentals to the latest emerging techniques. By the end of the tutorial, attendees will be familiar with the important design decisions of an LSR system, know how to apply them to text and other modalities, and understand the latest techniques for retrieving with them efficiently. Website: https://lsr-tutorial.github.io

Review for responsiveness is a recall-oriented document classification task central to civil litigation. In large legal matters, it may involve the coding of millions of documents by teams of dozens to hundreds of contract attorneys. We describe a prototype document review system based on a large language model (LLM) for replacing the first level of attorney review. Our system accepts the same guidance—a written review protocol—that would be provided to a human review team. We tested our prototype in the context of a live legal matter, evaluating both human review and our LLM-based system against a gold standard coded by expert senior attorneys. Our prototype achieved an estimated 96% recall and 60% precision without matter-specific tuning, and has numerous avenues for further improvement.

High Recall Retrieval (HRR) tasks, including eDiscovery in the law, systematic literature reviews, and sunshine law requests focus on efficiently prioritizing relevant documents for human review.Technology-assisted review (TAR) refers to iterative human-in-the-loop workflows that combine human review with IR and AI techniques to minimize both time and manual effort while maximizing recall. This full-day tutorial provides a comprehensive introduction to TAR. The morning session presents an overview of the key technologies and workflow designs used, the basics of practical evaluation methods, and the social and ethical implications of TAR deployment. The afternoon session provides more technical depth on the implications of TAR workflows for supervised learning algorithm design, how generative AI is can be applied in TAR, more sophisticated statistical evaluation techniques, and a wide range of open research questions.

PLAID, an efficient implementation of the ColBERT late interaction bi-encoder using pretrained language models for ranking, consistently achieves state-of-the-art performance in monolingual, cross-language, and multilingual retrieval. PLAID differs from ColBERT by assigning terms to clusters and representing those terms as cluster centroids plus compressed residual vectors. While PLAID is effective in batch experiments, its performance degrades in streaming settings where documents arrive over time because representations of new tokens may be poorly modeled by the earlier tokens used to select cluster centroids. PLAID Streaming Hierarchical Indexing that Runs on Terabytes of Temporal Text (PLAID HIRTTT) addresses this concern using multi-phase incremental indexing based on hierarchical sharding. Experiments on ClueWeb09 and the multilingual NeuCLIR collection demonstrate the effectiveness of this approach both for the largest collection indexed to date by the ColBERT architecture and in the multilingual setting, respectively.

Large Language Models (LLMs) have enabled new ways to satisfy information needs. Although great strides have been made in applying them to settings like document ranking and short-form text generation, they still struggle to compose complete, accurate, and verifiable long-form reports. Reports with these qualities are necessary to satisfy the complex, nuanced, or multi-faceted information needs of users. In this perspective paper, we draw together opinions from industry and academia, and from a variety of related research areas, to present our vision for automatic report generation, and---critically---a flexible framework by which such reports can be evaluated. In contrast with other summarization tasks, automatic report generation starts with a detailed description of an information need, stating the necessary background, requirements, and scope of the report. Further, the generated reports should be complete, accurate, and verifiable. These qualities, which are desirable---if not required---in many analytic report-writing settings, require rethinking how to build and evaluate systems that exhibit these qualities. To foster new efforts in building these systems, we present an evaluation framework that draws on ideas found in various evaluations. To test completeness and accuracy, the framework uses nuggets of information, expressed as questions and answers, that need to be part of any high-quality generated report. Additionally, evaluation of citations that map claims made in the report to their source documents ensures verifiability.

High Recall Retrieval (HRR), such as eDiscovery and medical systematic review, is a search problem that optimizes the cost of retrieving most relevant documents in a given collection. Iterative approaches, such as iterative relevance feedback and uncertainty sampling, are shown to be effective under various operational scenarios. Despite neural models demonstrating success in other text-related tasks, linear models such as logistic regression, in general, are still more effective and efficient in HRR since the model is trained and retrieves documents from the same fixed collection. In this work, we leverage SPLADE, an efficient retrieval model that transforms documents into contextualized sparse vectors, for HRR. Our approach combines the best of both worlds, leveraging both the contextualization from pretrained language models and the efficiency of linear models. It reduces 10% and 18% of the review cost in two HRR evaluation collections under a one-phase review workflow with a target recall of 80%. The experiment is implemented with TARexp and is available at https://github.com/eugene-yang/LSR-for-TAR.

Multilingual information retrieval (MLIR) considers the problem of ranking documents in several languages for a query expressed in a language that may differ from any of those languages. Recent work has observed that approaches such as combining ranked lists representing a single document language each or using multilingual pretrained language models demonstrate a preference for one language over others. This results in systematic unfair treatment of documents in different languages. This work proposes a language fairness metric to evaluate whether documents across different languages are fairly ranked through statistical equivalence testing using the Kruskal-Wallis test. In contrast to most prior work in group fairness, we do not consider any language to be an unprotected group. Thus our proposed measure, PEER (Probability of Equal Expected Rank), is the first fairness metric specifically designed to capture the language fairness of MLIR systems. We demonstrate the behavior of PEER on artificial ranked lists. We also evaluate real MLIR systems on two publicly available benchmarks and show that the PEER scores align with prior analytical findings on MLIR fairness. Our implementation is compatible with ir-measures and is available at http://github.com/hltcoe/peer_measure.

Recent work in cross-language information retrieval (CLIR), where queries and documents are in different languages, has shown the benefit of the Translate-Distill framework that trains a cross-language neural dual-encoder model using translation and distillation. However, Translate-Distill only supports a single document language. Multilingual information retrieval (MLIR), which ranks a multilingual document collection, is harder to train than CLIR because the model must assign comparable relevance scores to documents in different languages. This work extends Translate-Distill and propose Multilingual Translate-Distill (MTD) for MLIR. We show that ColBERT-X models trained with MTD outperform their counterparts trained with Multilingual Translate-Train, which is the previous state-of-the-art training approach, by 5% to 25% in nDCG@20 and 15% to 45% in MAP. We also show that the model is robust to the way languages are mixed in training batches. Our implementation is available on GitHub.

While there are many test collections for Cross-Language Information Retrieval (CLIR), none of the large public test collections focus on short informal text documents. This paper introduces a new pair of CLIR test collections with millions of Chinese or Persian Tweets or Tweet threads as documents, sixty event-motivated topics written both in English and in each of the two document languages, and three-point graded relevance judgments constructed using interactive search and active learning. The design and construction of these new test collections are described, and baseline results are presented that demonstrate the utility of the collections for system evaluation. Shallow pooling is used to assess the efficacy of active learning to select documents for judgment.

Learning sparse representations using pretrained language models enhances the monolingual ranking effectiveness. Such representations are sparse vectors in the vocabulary of a language model projected from document terms. Extending such approaches to Cross-Language Information Retrieval (CLIR) using multilingual pretrained language models poses two challenges. First, the larger vocabularies of multilingual models affect both training and inference efficiency. Second, the representations of terms from different languages with similar meanings might not be sufficiently similar. To address these issues, we propose a learned sparse representation model, BLADE, combining vocabulary pruning with intermediate pre-training based on cross-language supervision. Our experiments reveal BLADE significantly reduces indexing time compared to its monolingual counterpart, SPLADE, on machine-translated documents, and it generates rankings with strengths complementary to those of other efficient CLIR methods.

This half day tutorial introduces the participant to the basic concepts underlying neural Cross-Language Information Retrieval (CLIR). It discusses the most common algorithmic approaches to CLIR, focusing on modern neural methods; the history of CLIR; where to find and how to use CLIR training collections, test collections and baseline systems; how CLIR training and test collections are constructed; and open research questions in CLIR.

Cross-lingual information retrieval (CLIR) aims to provide access to information across languages. Recent pre-trained multilingual language models brought large improvements to the natural language tasks, including cross-lingual adhoc retrieval. However, pseudo-relevance feedback (PRF), a family of techniques for improving ranking using the contents of top initially retrieved items, has not been explored with neural CLIR retrieval models. Two of the challenges are incorporating feedback from long documents, and cross-language knowledge transfer. To address these challenges, we propose a novel neural CLIR architecture, NCLPRF, capable of incorporating PRF feedback from multiple potentially long documents, which enables improvements to query representation in the shared semantic space between query and document languages. The additional information that the feedback documents provide in a target language, can enrich the query representation, bringing it closer to relevant documents in the embedding space. The proposed model performance across three CLIR test collections in Chinese, Russian, and Persian languages, exhibits significant improvements over traditional and SOTA neural CLIR baselines across all three collections.

Technology-assisted review (TAR) is an important industrial application of information retrieval (IR) and machine learning (ML). While a small TAR research community exists, the complexity of TAR software and workflows is a major barrier to entry. Drawing on past open source TAR efforts, as well as design patterns from the IR and ML open source software, we present an open source Python framework for conducting experiments on TAR algorithms. Key characteristics of this framework are declarative representations of workflows and experiment plans, the ability for components to play variable numbers of workflow roles, and state maintenance and restart capabilities. Users can draw on reference implementations of standard TAR algorithms while incorporating novel components to explore their research interests. The framework is available at https://github.com/eugene-yang/tarexp.

Pretrained language models have improved effectiveness on numerous tasks, including ad-hoc retrieval. Recent work has shown that continuing to pretrain a language model with auxiliary objectives before fine-tuning on the retrieval task can further improve retrieval effectiveness. Unlike monolingual retrieval, designing an appropriate auxiliary task for cross-language mappings is challenging. To address this challenge, we use comparable Wikipedia articles in different languages to further pretrain off-the-shelf multilingual pretrained models before fine-tuning on the retrieval task. We show that our approach yields improvements in retrieval effectiveness.

Technology-assisted review (TAR) workflows based on iterative active learning are widely used in document review applications. Most stopping rules for one-phase TAR workflows lack valid statistical guarantees, which has discouraged their use in some legal contexts. Drawing on the theory of quantile estimation, we provide the first broadly applicable and statistically valid sample-based stopping rules for one-phase TAR. We further show theoretically and empirically that overshooting a recall target, which has been treated as innocuous or desirable in past evaluations of stopping rules, is a major source of excess cost in one-phase TAR workflows. Counterintuitively, incurring a larger sampling cost to reduce excess recall leads to lower total cost in almost all scenarios.

Technology-assisted review (TAR) refers to human-in-the-loop machine learning workflows for document review in legal discovery and other high recall review tasks. Attorneys and legal technologists have debated whether review should be a single iterative process (one-phase TAR workflows) or whether model training and review should be separate (two-phase TAR workflows), with implications for the choice of active learning algorithm. The relative cost of manual labeling for different purposes (training vs. review) and of different documents (positive vs. negative examples) is a key and neglected factor in this debate. Using a novel cost dynamics analysis, we show analytically and empirically that these relative costs strongly impact whether a one-phase or two-phase workflow minimizes cost. We also show how category prevalence, classification task difficulty, and collection size impact the optimal choice not only of workflow type, but of active learning method and stopping point.

Technology-assisted review (TAR) refers to human-in-the-loop active learning workflows for finding relevant documents in large collections. These workflows often must meet a target for the proportion of relevant documents found (i.e. recall) while also holding down costs. A variety of heuristic stopping rules have been suggested for striking this tradeoff in particular settings, but none have been tested against a range of recall targets and tasks. We propose two new heuristic stopping rules, Quant and QuantCI based on model-based estimation techniques from survey research. We compare them against a range of proposed heuristics and find they are accurate at hitting a range of recall targets while substantially reducing review costs.

Discriminative learning algorithms such as logistic regression excel when training data are plentiful, but falter when it is meager. An extreme case is text retrieval (zero training data), where discriminative learning is impossible and heuristics such as BM25, which combine domain knowledge (a topical keyword query) with generative learning (Naive Bayes), are dominant. Building on past work, we show that BM25-inspired Gaussian priors for Bayesian logistic regression based on topical keywords provide better effectiveness than the usual L2 (zero mode, uniform variance) Gaussian prior. On two high recall retrieval datasets, the resulting models transition smoothly from BM25 level effectiveness to discriminative effectiveness as training data volume increases, dominating L2 regularization even when substantial training data is available.