Recent advancements have successfully harnessed the power of Large Language Models (LLMs) for zero-shot document ranking, exploring a variety of prompting strategies. Comparative approaches like pairwise and listwise achieve high effectiveness but are computationally intensive and thus less practical for larger-scale applications. Scoring-based pointwise approaches exhibit superior efficiency by independently and simultaneously generating the relevance scores for each candidate document. However, this independence ignores critical comparative insights between documents, resulting in inconsistent scoring and suboptimal performance. In this paper, we aim to improve the effectiveness of pointwise methods while preserving their efficiency through two key innovations: (1) We propose a novel Global-Consistent Comparative Pointwise Ranking (GCCP) strategy that incorporates global reference comparisons between each candidate and an anchor document to generate contrastive relevance scores. We strategically design the anchor document as a query-focused summary of pseudo-relevant candidates, which serves as an effective reference point by capturing the global context for document comparison. (2) These contrastive relevance scores can be efficiently Post-Aggregated with existing pointwise methods, seamlessly integrating essential Global Context information in a training-free manner (PAGC). Extensive experiments on the TREC DL and BEIR benchmark demonstrate that our approach significantly outperforms previous pointwise methods while maintaining comparable efficiency. Our method also achieves competitive performance against comparative methods that require substantially more computational resources. More analyses further validate the efficacy of our anchor construction strategy.
Recent studies have shown that large language models (LLMs) can assess relevance and support information retrieval (IR) tasks such as document ranking and relevance judgment generation. However, the internal mechanisms by which off-the-shelf LLMs understand and operationalize relevance remain largely unexplored. In this paper, we systematically investigate how different LLM modules contribute to relevance judgment through the lens of mechanistic interpretability. Using activation patching techniques, we analyze the roles of various model components and identify a multi-stage, progressive process in generating either pointwise or pairwise relevance judgment. Specifically, LLMs first extract query and document information in the early layers, then process relevance information according to instructions in the middle layers, and finally utilize specific attention heads in the later layers to generate relevance judgments in the required format. Our findings provide insights into the mechanisms underlying relevance assessment in LLMs, offering valuable implications for future research on leveraging LLMs for IR tasks.
We present a novel approach for training small language models for reasoning-intensive document ranking that combines knowledge distillation with reinforcement learning optimization. While existing methods often rely on expensive human annotations or large black-box language models, our methodology leverages web data and a teacher LLM to automatically generate high-quality training examples with relevance explanations. By framing document ranking as a reinforcement learning problem and incentivizing explicit reasoning capabilities, we train a compact 3B parameter language model that achieves state-of-the-art performance on the BRIGHT benchmark. Our model ranks third on the leaderboard while using substantially fewer parameters than other approaches, outperforming models that are over 20 times larger. Through extensive experiments, we demonstrate that generating explanations during inference, rather than directly predicting relevance scores, enables more effective reasoning with smaller language models. The self-supervised nature of our method offers a scalable and interpretable solution for modern information retrieval systems.
While question-like queries are gaining popularity and search engines' users increasingly adopt them, keyphrase search has traditionally been the cornerstone of web search. This query type is also prevalent in specialised search tasks such as academic or professional search, where experts rely on keyphrases to articulate their information needs. However, current dense retrieval models often fail with keyphrase-like queries, primarily because they are mostly trained on question-like ones. This paper introduces a novel model that employs the ColBERT architecture to enhance document ranking for keyphrase queries. For that, given the lack of large keyphrase-based retrieval datasets, we first explore how Large Language Models can convert question-like queries into keyphrase format. Then, using those keyphrases, we train a keyphrase-based ColBERT ranker (ColBERTKP_QD) to improve the performance when working with keyphrase queries. Furthermore, to reduce the training costs associated with training the full ColBERT model, we investigate the feasibility of training only a keyphrase query encoder while keeping the document encoder weights static (ColBERTKP_Q). We assess our proposals' ranking performance using both automatically generated and manually annotated keyphrases. Our results reveal the potential of the late interaction architecture when working under the keyphrase search scenario.
Learned Sparse Retrieval (LSR) models use vocabularies from pre-trained transformers, which often split entities into nonsensical fragments. Splitting entities can reduce retrieval accuracy and limits the model's ability to incorporate up-to-date world knowledge not included in the training data. In this work, we enhance the LSR vocabulary with Wikipedia concepts and entities, enabling the model to resolve ambiguities more effectively and stay current with evolving knowledge. Central to our approach is a Dynamic Vocabulary (DyVo) head, which leverages existing entity embeddings and an entity retrieval component that identifies entities relevant to a query or document. We use the DyVo head to generate entity weights, which are then merged with word piece weights to create joint representations for efficient indexing and retrieval using an inverted index. In experiments across three entity-rich document ranking datasets, the resulting DyVo model substantially outperforms state-of-the-art baselines.
The Retrieval Augmented Generation (RAG) framework utilizes a combination of parametric knowledge and external knowledge to demonstrate state-of-the-art performance on open-domain question answering tasks. However, the RAG framework suffers from performance degradation when the query is accompanied by irrelevant contexts. In this work, we propose the RE-RAG framework, which introduces a relevance estimator (RE) that not only provides relative relevance between contexts as previous rerankers did, but also provides confidence, which can be used to classify whether given context is useful for answering the given question. We propose a weakly supervised method for training the RE simply utilizing question-answer data without any labels for correct contexts. We show that RE trained with a small generator (sLM) can not only improve the sLM fine-tuned together with RE but also improve previously unreferenced large language models (LLMs). Furthermore, we investigate new decoding strategies that utilize the proposed confidence measured by RE such as choosing to let the user know that it is "unanswerable" to answer the question given the retrieved contexts or choosing to rely on LLM's parametric knowledge rather than unrelated contexts.
Utilizing large language models (LLMs) for zero-shot document ranking is done in one of two ways: (1) prompt-based re-ranking methods, which require no further training but are only feasible for re-ranking a handful of candidate documents due to computational costs; and (2) unsupervised contrastive trained dense retrieval methods, which can retrieve relevant documents from the entire corpus but require a large amount of paired text data for contrastive training. In this paper, we propose PromptReps, which combines the advantages of both categories: no need for training and the ability to retrieve from the whole corpus. Our method only requires prompts to guide an LLM to generate query and document representations for effective document retrieval. Specifically, we prompt the LLMs to represent a given text using a single word, and then use the last token's hidden states and the corresponding logits associated with the prediction of the next token to construct a hybrid document retrieval system. The retrieval system harnesses both dense text embedding and sparse bag-of-words representations given by the LLM. Our experimental evaluation on the MSMARCO, TREC deep learning and BEIR zero-shot document retrieval datasets illustrates that this simple prompt-based LLM retrieval method can achieve a similar or higher retrieval effectiveness than state-of-the-art LLM embedding methods that are trained with large amounts of unsupervised data, especially when using a larger LLM.
Transformer structure has achieved great success in multiple applied machine learning communities, such as natural language processing (NLP), computer vision (CV) and information retrieval (IR). Transformer architecture's core mechanism -- attention requires $O(n^2)$ time complexity in training and $O(n)$ time complexity in inference. Many works have been proposed to improve the attention mechanism's scalability, such as Flash Attention and Multi-query Attention. A different line of work aims to design new mechanisms to replace attention. Recently, a notable model structure -- Mamba, which is based on state space models, has achieved transformer-equivalent performance in multiple sequence modeling tasks. In this work, we examine \mamba's efficacy through the lens of a classical IR task -- document ranking. A reranker model takes a query and a document as input, and predicts a scalar relevance score. This task demands the language model's ability to comprehend lengthy contextual inputs and to capture the interaction between query and document tokens. We find that (1) Mamba models achieve competitive performance compared to transformer-based models with the same training recipe; (2) but also have a lower training throughput in comparison to efficient transformer implementations such as flash attention. We hope this study can serve as a starting point to explore Mamba models in other classical IR tasks. Our code implementation and trained checkpoints are made public to facilitate reproducibility (https://github.com/zhichaoxu-shufe/RankMamba).
Query augmentation is a crucial technique for refining semantically imprecise queries. Traditionally, query augmentation relies on extracting information from initially retrieved, potentially relevant documents. If the quality of the initially retrieved documents is low, then the effectiveness of query augmentation would be limited as well. We propose Brain-Aug, which enhances a query by incorporating semantic information decoded from brain signals. BrainAug generates the continuation of the original query with a prompt constructed with brain signal information and a ranking-oriented inference approach. Experimental results on fMRI (functional magnetic resonance imaging) datasets show that Brain-Aug produces semantically more accurate queries, leading to improved document ranking performance. Such improvement brought by brain signals is particularly notable for ambiguous queries.
In search settings, calibrating the scores during the ranking process to quantities such as click-through rates or relevance levels enhances a system's usefulness and trustworthiness for downstream users. While previous research has improved this notion of calibration for low complexity learning-to-rank models, the larger data demands and parameter count specific to modern neural text rankers produce unique obstacles that hamper the efficacy of methods intended for the learning-to-rank setting. This paper proposes exploiting large language models (LLMs) to provide relevance and uncertainty signals for these neural text rankers to produce scale-calibrated scores through Monte Carlo sampling of natural language explanations (NLEs). Our approach transforms the neural ranking task from ranking textual query-document pairs to ranking corresponding synthesized NLEs. Comprehensive experiments on two popular document ranking datasets show that the NLE-based calibration approach consistently outperforms past calibration methods and LLM-based methods for ranking, calibration, and query performance prediction tasks.
In the field of information retrieval, Query Likelihood Models (QLMs) rank documents based on the probability of generating the query given the content of a document. Recently, advanced large language models (LLMs) have emerged as effective QLMs, showcasing promising ranking capabilities. This paper focuses on investigating the genuine zero-shot ranking effectiveness of recent LLMs, which are solely pre-trained on unstructured text data without supervised instruction fine-tuning. Our findings reveal the robust zero-shot ranking ability of such LLMs, highlighting that additional instruction fine-tuning may hinder effectiveness unless a question generation task is present in the fine-tuning dataset. Furthermore, we introduce a novel state-of-the-art ranking system that integrates LLM-based QLMs with a hybrid zero-shot retriever, demonstrating exceptional effectiveness in both zero-shot and few-shot scenarios. We make our codebase publicly available at https://github.com/ielab/llm-qlm.
We propose a novel zero-shot document ranking approach based on Large Language Models (LLMs): the Setwise prompting approach. Our approach complements existing prompting approaches for LLM-based zero-shot ranking: Pointwise, Pairwise, and Listwise. Through the first-of-its-kind comparative evaluation within a consistent experimental framework and considering factors like model size, token consumption, latency, among others, we show that existing approaches are inherently characterised by trade-offs between effectiveness and efficiency. We find that while Pointwise approaches score high on efficiency, they suffer from poor effectiveness. Conversely, Pairwise approaches demonstrate superior effectiveness but incur high computational overhead. Our Setwise approach, instead, reduces the number of LLM inferences and the amount of prompt token consumption during the ranking procedure, compared to previous methods. This significantly improves the efficiency of LLM-based zero-shot ranking, while also retaining high zero-shot ranking effectiveness. We make our code and results publicly available at \url{https://github.com/ielab/llm-rankers}.
Common document ranking pipelines in search systems are cascade systems that involve multiple ranking layers to integrate different information step-by-step. In this paper, we propose a novel re-ranker Fusion-in-T5 (FiT5), which integrates text matching information, ranking features, and global document information into one single unified model via templated-based input and global attention. Experiments on passage ranking benchmarks MS MARCO and TREC DL show that FiT5, as one single model, significantly improves ranking performance over complex cascade pipelines. Analysis finds that through attention fusion, FiT5 jointly utilizes various forms of ranking information via gradually attending to related documents and ranking features, and improves the detection of subtle nuances. Our code is open-sourced at https://github.com/OpenMatch/FiT5.
Pre-trained language models have achieved great success in various large-scale information retrieval tasks. However, most of pretraining tasks are based on counterfeit retrieval data where the query produced by the tailored rule is assumed as the user's issued query on the given document or passage. Therefore, we explore to use large-scale click logs to pretrain a language model instead of replying on the simulated queries. Specifically, we propose to use user behavior features to pretrain a debiased language model for document ranking. Extensive experiments on Baidu desensitization click logs validate the effectiveness of our method. Our team on WSDM Cup 2023 Pre-training for Web Search won the 1st place with a Discounted Cumulative Gain @ 10 (DCG@10) score of 12.16525 on the final leaderboard.
Knowledge distillation is often used to transfer knowledge from a strong teacher model to a relatively weak student model. Traditional methods include response-based methods and feature-based methods. Response-based methods are widely used but suffer from lower upper limits of performance due to their ignorance of intermediate signals, while feature-based methods have constraints on vocabularies, tokenizers and model architectures. In this paper, we propose a liberal feature-based distillation method (LEAD). LEAD aligns the distribution between the intermediate layers of teacher model and student model, which is effective, extendable, portable and has no requirements on vocabularies, tokenizers, or model architectures. Extensive experiments show the effectiveness of LEAD on widely-used benchmarks, including MS MARCO Passage Ranking, TREC 2019 DL Track, MS MARCO Document Ranking and TREC 2020 DL Track. Our code is available in https://github.com/microsoft/SimXNS/tree/main/LEAD.
Information Retrieval evaluation has traditionally focused on defining principled ways of assessing the relevance of a ranked list of documents with respect to a query. Several methods extend this type of evaluation beyond relevance, making it possible to evaluate different aspects of a document ranking (e.g., relevance, usefulness, or credibility) using a single measure (multi-aspect evaluation). However, these methods either are (i) tailor-made for specific aspects and do not extend to other types or numbers of aspects, or (ii) have theoretical anomalies, e.g. assign maximum score to a ranking where all documents are labelled with the lowest grade with respect to all aspects (e.g., not relevant, not credible, etc.). We present a theoretically principled multi-aspect evaluation method that can be used for any number, and any type, of aspects. A thorough empirical evaluation using up to 5 aspects and a total of 425 runs officially submitted to 10 TREC tracks shows that our method is more discriminative than the state-of-the-art and overcomes theoretical limitations of the state-of-the-art.
Users' search tasks have become increasingly complicated, requiring multiple queries and interactions with the results. Recent studies have demonstrated that modeling the historical user behaviors in a session can help understand the current search intent. Existing context-aware ranking models primarily encode the current session sequence (from the first behavior to the current query) and compute the ranking score using the high-level representations. However, there is usually some noise in the current session sequence (useless behaviors for inferring the search intent) that may affect the quality of the encoded representations. To help the encoding of the current user behavior sequence, we propose to use a decoder and the information of future sequences and a supplemental query. Specifically, we design three generative tasks that can help the encoder to infer the actual search intent: (1) predicting future queries, (2) predicting future clicked documents, and (3) predicting a supplemental query. We jointly learn the ranking task with these generative tasks using an encoder-decoder structured approach. Extensive experiments on two public search logs demonstrate that our model outperforms all existing baselines, and the designed generative tasks can actually help the ranking task. Besides, additional experiments also show that our approach can be easily applied to various Transformer-based encoder-decoder models and improve their performance.
Contextual information in search sessions is important for capturing users' search intents. Various approaches have been proposed to model user behavior sequences to improve document ranking in a session. Typically, training samples of (search context, document) pairs are sampled randomly in each training epoch. In reality, the difficulty to understand user's search intent and to judge document's relevance varies greatly from one search context to another. Mixing up training samples of different difficulties may confuse the model's optimization process. In this work, we propose a curriculum learning framework for context-aware document ranking, in which the ranking model learns matching signals between the search context and the candidate document in an easy-to-hard manner. In so doing, we aim to guide the model gradually toward a global optimum. To leverage both positive and negative examples, two curricula are designed. Experiments on two real query log datasets show that our proposed framework can improve the performance of several existing methods significantly, demonstrating the effectiveness of curriculum learning for context-aware document ranking.
We evaluated 20+ Transformer models for ranking of long documents (including recent LongP models trained with FlashAttention) and compared them with a simple FirstP baseline, which applies the same model to the truncated input (at most 512 tokens). We used MS MARCO Documents v1 as a primary training set and evaluated both the zero-shot transferred and fine-tuned models. On MS MARCO, TREC DLs, and Robust04 no long-document model outperformed FirstP by more than 5% in NDCG and MRR (when averaged over all test sets). We conjectured this was not due to models' inability to process long context, but due to a positional bias of relevant passages, whose distribution was skewed towards the beginning of documents. We found direct evidence of this bias in some test sets, which motivated us to create MS MARCO FarRelevant (based on MS MARCO Passages) where the relevant passages were not present among the first 512 tokens. Unlike standard collections where we saw both little benefit from incorporating longer contexts and limited variability in model performance (within a few %), experiments on MS MARCO FarRelevant uncovered dramatic differences among models. The FirstP models performed roughly at the random-baseline level in both zero-shot and fine-tuning scenarios. Simple aggregation models including MaxP and PARADE Attention had good zero-shot accuracy, but benefited little from fine-tuning. Most other models had poor zero-shot performance (sometimes at a random baseline level), but outstripped MaxP by as much as 13-28% after fine-tuning. Thus, the positional bias not only diminishes benefits of processing longer document contexts, but also leads to model overfitting to positional bias and performing poorly in a zero-shot setting when the distribution of relevant passages changes substantially. We make our software and data available.
Contextual ranking models based on BERT are now well established for a wide range of passage and document ranking tasks. However, the robustness of BERT-based ranking models under adversarial inputs is under-explored. In this paper, we argue that BERT-rankers are not immune to adversarial attacks targeting retrieved documents given a query. Firstly, we propose algorithms for adversarial perturbation of both highly relevant and non-relevant documents using gradient-based optimization methods. The aim of our algorithms is to add/replace a small number of tokens to a highly relevant or non-relevant document to cause a large rank demotion or promotion. Our experiments show that a small number of tokens can already result in a large change in the rank of a document. Moreover, we find that BERT-rankers heavily rely on the document start/head for relevance prediction, making the initial part of the document more susceptible to adversarial attacks. More interestingly, we find a small set of recurring adversarial words that when added to documents result in successful rank demotion/promotion of any relevant/non-relevant document respectively. Finally, our adversarial tokens also show particular topic preferences within and across datasets, exposing potential biases from BERT pre-training or downstream datasets.