Test collections are crucial for evaluating Information Retrieval (IR) systems. Creating a diverse set of user queries for these collections can be challenging, and obtaining relevance judgments, which indicate how well retrieved documents match a query, is often costly and resource-intensive. Recently, generating synthetic datasets using Large Language Models (LLMs) has gained attention in various applications. While previous work has used LLMs to generate synthetic queries or documents to improve ranking models, using LLMs to create synthetic test collections is still relatively unexplored. Previous work~\cite{rahmani2024synthetic} showed that synthetic test collections have the potential to be used for system evaluation, however, more analysis is needed to validate this claim. In this paper, we thoroughly investigate the reliability of synthetic test collections constructed using LLMs, where LLMs are used to generate synthetic queries, labels, or both. In particular, we examine the potential biases that might occur when such test collections are used for evaluation. We first empirically show the presence of such bias in evaluation results and analyse the effects it might have on system evaluation. We further validate the presence of such bias using a linear mixed-effects model. Our analysis shows that while the effect of bias present in evaluation results obtained using synthetic test collections could be significant, for e.g.~computing absolute system performance, its effect may not be as significant in comparing relative system performance. Codes and data are available at: https://github.com/rahmanidashti/BiasSyntheticData.

Promptagator demonstrated that Large Language Models (LLMs) with few-shot prompts can be used as task-specific query generators for fine-tuning domain-specialized dense retrieval models. However, the original Promptagator approach relied on proprietary and large-scale LLMs which users may not have access to or may be prohibited from using with sensitive data. In this work, we study the impact of open-source LLMs at accessible scales ($\leq$14B parameters) as an alternative. Our results demonstrate that open-source LLMs as small as 3B parameters can serve as effective Promptagator-style query generators. We hope our work will inform practitioners with reliable alternatives for synthetic data generation and give insights to maximize fine-tuning results for domain-specific applications.

Large language models (LLMs) inference relies heavily on KV-caches to accelerate autoregressive decoding, but the resulting memory footprint grows rapidly with sequence length, posing significant efficiency challenges. Current KV-cache compression methods suffer from a Procrustes' bed problem: they force diverse workloads into fixed compression ratios, leading to suboptimal resource allocation and inference performance. To this end, we present GVote, an adaptive KV-cache compression scheme that eliminates manual budget specification while achieving superior accuracy-efficiency trade-offs. GVote operates on the principle that the important keys are the aggregation of keys required by future queries. The method predicts future query attention demands by Monte-Carlo style sampling potential queries and aggregating selected keys to determine the optimal cache budget without manual specification. Experimental evaluation demonstrates GVote's effectiveness across multiple benchmarks, including GSM8K, RULER and Longbench. Compared to baselines, GVote exhibits 2$\times$ memory reduction while the accuracy maintains higher or comparable.

This paper presents a deployed, production-grade system designed to enhance and scale search query datasets for intent-based recommendation systems in digital banking. In real-world environments, the growing volume and complexity of user intents create substantial challenges for data management, resulting in suboptimal recommendations and delayed product onboarding. To overcome these challenges, our approach shifts the focus from model-centric enhancements to automated, data-centric strategies. The proposed system integrates three core modules: Synthetic Query Generation, Intent Disambiguation, and Intent Gap Analysis. Synthetic Query Generation produces diverse and realistic user queries. Our experiments reveal no statistically significant difference when using synthetic data for Clinc150, while Banking77 and a proprietary dataset show significant differences. We dig into the underlying factors driving these variations, demonstrating that our approach effectively alleviates the cold start problem (i.e. the challenge of recommending new products with limited historical data). Intent Disambiguation refines broad and overlapping intent categories into precise subintents, achieving an F1 score of 0.863 $\pm$ 0.127 against expert reannotations and leading to clearer differentiation and more precise recommendation mapping. Meanwhile, Intent Gap Analysis identifies latent customer needs by extracting novel intents from unlabeled queries; recovery rates reach up to 71\% in controlled evaluations. Deployed in a live banking environment, our system demonstrates significant improvements in recommendation precision and operation agility, ultimately delivering enhanced user experiences and strategic business benefits. This work underscores the role of high-quality, scalable data in modern AI-driven applications and advocates a proactive approach to data enhancement as a key driver of value.

This is the fifth year of the TREC Deep Learning track. As in previous years, we leverage the MS MARCO datasets that made hundreds of thousands of human-annotated training labels available for both passage and document ranking tasks. We mostly repeated last year's design, to get another matching test set, based on the larger, cleaner, less-biased v2 passage and document set, with passage ranking as primary and document ranking as a secondary task (using labels inferred from passage). As we did last year, we sample from MS MARCO queries that were completely held out, unused in corpus construction, unlike the test queries in the first three years. This approach yields a more difficult test with more headroom for improvement. Alongside the usual MS MARCO (human) queries from MS MARCO, this year we generated synthetic queries using a fine-tuned T5 model and using a GPT-4 prompt. The new headline result this year is that runs using Large Language Model (LLM) prompting in some way outperformed runs that use the "nnlm" approach, which was the best approach in the previous four years. Since this is the last year of the track, future iterations of prompt-based ranking can happen in other tracks. Human relevance assessments were applied to all query types, not just human MS MARCO queries. Evaluation using synthetic queries gave similar results to human queries, with system ordering agreement of $τ=0.8487$. However, human effort was needed to select a subset of the synthetic queries that were usable. We did not see clear evidence of bias, where runs using GPT-4 were favored when evaluated using synthetic GPT-4 queries, or where runs using T5 were favored when evaluated on synthetic T5 queries.

This paper investigates synthetic data generation strategies in developing generative retrieval models for domain-specific corpora, thereby addressing the scalability challenges inherent in manually annotating in-domain queries. We study the data strategies for a two-stage training framework: in the first stage, which focuses on learning to decode document identifiers from queries, we investigate LLM-generated queries across multiple granularity (e.g. chunks, sentences) and domain-relevant search constraints that can better capture nuanced relevancy signals. In the second stage, which aims to refine document ranking through preference learning, we explore the strategies for mining hard negatives based on the initial model's predictions. Experiments on public datasets over diverse domains demonstrate the effectiveness of our synthetic data generation and hard negative sampling approach.

In the field of legal information retrieval, effective embedding-based models are essential for accurate question-answering systems. However, the scarcity of large annotated datasets poses a significant challenge, particularly for Vietnamese legal texts. To address this issue, we propose a novel approach that leverages large language models to generate high-quality, diverse synthetic queries for Vietnamese legal passages. This synthetic data is then used to pre-train retrieval models, specifically bi-encoder and ColBERT, which are further fine-tuned using contrastive loss with mined hard negatives. Our experiments demonstrate that these enhancements lead to strong improvement in retrieval accuracy, validating the effectiveness of synthetic data and pre-training techniques in overcoming the limitations posed by the lack of large labeled datasets in the Vietnamese legal domain.

Despite the recent advancements in information retrieval (IR), zero-shot IR remains a significant challenge, especially when dealing with new domains, languages, and newly-released use cases that lack historical query traffic from existing users. For such cases, it is common to use query augmentations followed by fine-tuning pre-trained models on the document data paired with synthetic queries. In this work, we propose a novel Universal Document Linking (UDL) algorithm, which links similar documents to enhance synthetic query generation across multiple datasets with different characteristics. UDL leverages entropy for the choice of similarity models and named entity recognition (NER) for the link decision of documents using similarity scores. Our empirical studies demonstrate the effectiveness and universality of the UDL across diverse datasets and IR models, surpassing state-of-the-art methods in zero-shot cases. The developed code for reproducibility is included in https://github.com/eoduself/UDL

Language models (LMs) are known to suffer from hallucinations and misinformation. Retrieval augmented generation (RAG) that retrieves verifiable information from an external knowledge corpus to complement the parametric knowledge in LMs provides a tangible solution to these problems. However, the generation quality of RAG is highly dependent on the relevance between a user's query and the retrieved documents. Inaccurate responses may be generated when the query is outside of the scope of knowledge represented in the external knowledge corpus or if the information in the corpus is out-of-date. In this work, we establish a statistical framework that assesses how well a query can be answered by an RAG system by capturing the relevance of knowledge. We introduce an online testing procedure that employs goodness-of-fit (GoF) tests to inspect the relevance of each user query to detect out-of-knowledge queries with low knowledge relevance. Additionally, we develop an offline testing framework that examines a collection of user queries, aiming to detect significant shifts in the query distribution which indicates the knowledge corpus is no longer sufficiently capable of supporting the interests of the users. We demonstrate the capabilities of these strategies through a systematic evaluation on eight question-answering (QA) datasets, the results of which indicate that the new testing framework is an efficient solution to enhance the reliability of existing RAG systems.

Dense retrieval systems are commonly used for information retrieval (IR). They rely on learning text representations through an encoder and usually require supervised modeling via labelled data which can be costly to obtain or simply unavailable. In this study, we introduce a novel unsupervised text representation learning technique via instruction-tuning the pre-trained encoder-decoder large language models (LLM) under the dual-encoder retrieval framework. We demonstrate the corpus representation can be augmented by the representations of relevant synthetic queries generated by the instruct-tuned LLM founded on the Rao-Blackwell theorem. Furthermore, we effectively align the query and corpus text representation with self-instructed-tuning. Specifically, we first prompt an open-box pre-trained LLM to follow defined instructions (i.e. question generation and keyword summarization) to generate synthetic queries. Next, we fine-tune the pre-trained LLM with defined instructions and the generated queries that passed quality check. Finally, we generate synthetic queries with the instruction-tuned LLM for each corpora and represent each corpora by weighted averaging the synthetic queries and original corpora embeddings. We evaluate our proposed method under low-resource settings on three English and one German retrieval datasets measuring NDCG@10, MRR@100, Recall@100. We significantly improve the average zero-shot retrieval performance on all metrics, increasing open-box FLAN-T5 model variations by [3.34%, 3.50%] in absolute and exceeding three competitive dense retrievers (i.e. mDPR, T-Systems, mBART-Large), with model of size at least 38% smaller, by 1.96%, 4.62%, 9.52% absolute on NDCG@10.