Yixing Fan

Wenxiang Sun, Yixing Fan, Jiafeng Guo et al.

Visual Entity Linking (VEL) is a task to link regions of images with their corresponding entities in Knowledge Bases (KBs), which is beneficial for many computer vision tasks such as image retrieval, image caption, and visual question answering. While existing tasks in VEL either rely on textual data to complement a multi-modal linking or only link objects with general entities, which fails to perform named entity linking on large amounts of image data. In this paper, we consider a purely Visual-based Named Entity Linking (VNEL) task, where the input only consists of an image. The task is to identify objects of interest (i.e., visual entity mentions) in images and link them to corresponding named entities in KBs. Since each entity often contains rich visual and textual information in KBs, we thus propose three different sub-tasks, i.e., visual to visual entity linking (V2VEL), visual to textual entity linking (V2TEL), and visual to visual-textual entity linking (V2VTEL). In addition, we present a high-quality human-annotated visual person linking dataset, named WIKIPerson. Based on WIKIPerson, we establish a series of baseline algorithms for the solution of each sub-task, and conduct experiments to verify the quality of proposed datasets and the effectiveness of baseline methods. We envision this work to be helpful for soliciting more works regarding VNEL in the future. The codes and datasets are publicly available at https://github.com/ict-bigdatalab/VNEL.

Weichao Zhang, Ruqing Zhang, Jiafeng Guo et al.

As the scale of training corpora for large language models (LLMs) grows, model developers become increasingly reluctant to disclose details on their data. This lack of transparency poses challenges to scientific evaluation and ethical deployment. Recently, pretraining data detection approaches, which infer whether a given text was part of an LLM's training data through black-box access, have been explored. The Min-K\% Prob method, which has achieved state-of-the-art results, assumes that a non-training example tends to contain a few outlier words with low token probabilities. However, the effectiveness may be limited as it tends to misclassify non-training texts that contain many common words with high probabilities predicted by LLMs. To address this issue, we introduce a divergence-based calibration method, inspired by the divergence-from-randomness concept, to calibrate token probabilities for pretraining data detection. We compute the cross-entropy (i.e., the divergence) between the token probability distribution and the token frequency distribution to derive a detection score. We have developed a Chinese-language benchmark, PatentMIA, to assess the performance of detection approaches for LLMs on Chinese text. Experimental results on English-language benchmarks and PatentMIA demonstrate that our proposed method significantly outperforms existing methods. Our code and PatentMIA benchmark are available at https://github.com/zhang-wei-chao/DC-PDD.

Lu Chen, Ruqing Zhang, Jiafeng Guo et al.

Retrieval-augmented generation (RAG) has emerged as a popular solution to mitigate the hallucination issues of large language models. However, existing studies on RAG seldom address the issue of predictive uncertainty, i.e., how likely it is that a RAG model's prediction is incorrect, resulting in uncontrollable risks in real-world applications. In this work, we emphasize the importance of risk control, ensuring that RAG models proactively refuse to answer questions with low confidence. Our research identifies two critical latent factors affecting RAG's confidence in its predictions: the quality of the retrieved results and the manner in which these results are utilized. To guide RAG models in assessing their own confidence based on these two latent factors, we develop a counterfactual prompting framework that induces the models to alter these factors and analyzes the effect on their answers. We also introduce a benchmarking procedure to collect answers with the option to abstain, facilitating a series of experiments. For evaluation, we introduce several risk-related metrics and the experimental results demonstrate the effectiveness of our approach. Our code and benchmark dataset are available at https://github.com/ict-bigdatalab/RC-RAG.

Wenshan Wang, Yihang Wang, Yixing Fan et al.

In-context learning (ICL) capabilities are foundational to the success of large language models (LLMs). Recently, context compression has attracted growing interest since it can largely reduce reasoning complexities and computation costs of LLMs. In this paper, we introduce a novel Query-gUIded aTtention cOmpression (QUITO) method, which leverages attention of the question over the contexts to filter useless information. Specifically, we take a trigger token to calculate the attention distribution of the context in response to the question. Based on the distribution, we propose three different filtering methods to satisfy the budget constraints of the context length. We evaluate the QUITO using two widely-used datasets, namely, NaturalQuestions and ASQA. Experimental results demonstrate that QUITO significantly outperforms established baselines across various datasets and downstream LLMs, underscoring its effectiveness. Our code is available at https://github.com/Wenshansilvia/attention_compressor.

Yu-An Liu, Ruqing Zhang, Jiafeng Guo et al.

Recent advances in neural information retrieval (IR) models have significantly enhanced their effectiveness over various IR tasks. The robustness of these models, essential for ensuring their reliability in practice, has also garnered significant attention. With a wide array of research on robust IR being proposed, we believe it is the opportune moment to consolidate the current status, glean insights from existing methodologies, and lay the groundwork for future development. We view the robustness of IR to be a multifaceted concept, emphasizing its necessity against adversarial attacks, out-of-distribution (OOD) scenarios and performance variance. With a focus on adversarial and OOD robustness, we dissect robustness solutions for dense retrieval models (DRMs) and neural ranking models (NRMs), respectively, recognizing them as pivotal components of the neural IR pipeline. We provide an in-depth discussion of existing methods, datasets, and evaluation metrics, shedding light on challenges and future directions in the era of large language models. To the best of our knowledge, this is the first comprehensive survey on the robustness of neural IR models, and we will also be giving our first tutorial presentation at SIGIR 2024 \url{https://sigir2024-robust-information-retrieval.github.io}. Along with the organization of existing work, we introduce a Benchmark for robust IR (BestIR), a heterogeneous evaluation benchmark for robust neural information retrieval, which is publicly available at \url{https://github.com/Davion-Liu/BestIR}. We hope that this study provides useful clues for future research on the robustness of IR models and helps to develop trustworthy search engines \url{https://github.com/Davion-Liu/Awesome-Robustness-in-Information-Retrieval}.

Jiangui Chen, Ruqing Zhang, Jiafeng Guo et al.

Knowledge-intensive language tasks (KILT) usually require a large body of information to provide correct answers. A popular paradigm to solve this problem is to combine a search system with a machine reader, where the former retrieves supporting evidences and the latter examines them to produce answers. Recently, the reader component has witnessed significant advances with the help of large-scale pre-trained generative models. Meanwhile most existing solutions in the search component rely on the traditional ``index-retrieve-then-rank'' pipeline, which suffers from large memory footprint and difficulty in end-to-end optimization. Inspired by recent efforts in constructing model-based IR models, we propose to replace the traditional multi-step search pipeline with a novel single-step generative model, which can dramatically simplify the search process and be optimized in an end-to-end manner. We show that a strong generative retrieval model can be learned with a set of adequately designed pre-training tasks, and be adopted to improve a variety of downstream KILT tasks with further fine-tuning. We name the pre-trained generative retrieval model as CorpusBrain as all information about the corpus is encoded in its parameters without the need of constructing additional index. Empirical results show that CorpusBrain can significantly outperform strong baselines for the retrieval task on the KILT benchmark and establish new state-of-the-art downstream performances. We also show that CorpusBrain works well under zero- and low-resource settings.

Jiangui Chen, Ruqing Zhang, Jiafeng Guo et al.

Fact verification (FV) is a challenging task which aims to verify a claim using multiple evidential sentences from trustworthy corpora, e.g., Wikipedia. Most existing approaches follow a three-step pipeline framework, including document retrieval, sentence retrieval and claim verification. High-quality evidences provided by the first two steps are the foundation of the effective reasoning in the last step. Despite being important, high-quality evidences are rarely studied by existing works for FV, which often adopt the off-the-shelf models to retrieve relevant documents and sentences in an "index-retrieve-then-rank" fashion. This classical approach has clear drawbacks as follows: i) a large document index as well as a complicated search process is required, leading to considerable memory and computational overhead; ii) independent scoring paradigms fail to capture the interactions among documents and sentences in ranking; iii) a fixed number of sentences are selected to form the final evidence set. In this work, we propose GERE, the first system that retrieves evidences in a generative fashion, i.e., generating the document titles as well as evidence sentence identifiers. This enables us to mitigate the aforementioned technical issues since: i) the memory and computational cost is greatly reduced because the document index is eliminated and the heavy ranking process is replaced by a light generative process; ii) the dependency between documents and that between sentences could be captured via sequential generation process; iii) the generative formulation allows us to dynamically select a precise set of relevant evidences for each claim. The experimental results on the FEVER dataset show that GERE achieves significant improvements over the state-of-the-art baselines, with both time-efficiency and memory-efficiency.

Yu-An Liu, Ruqing Zhang, Jiafeng Guo et al.

Neural ranking models (NRMs) and dense retrieval (DR) models have given rise to substantial improvements in overall retrieval performance. In addition to their effectiveness, and motivated by the proven lack of robustness of deep learning-based approaches in other areas, there is growing interest in the robustness of deep learning-based approaches to the core retrieval problem. Adversarial attack methods that have so far been developed mainly focus on attacking NRMs, with very little attention being paid to the robustness of DR models. In this paper, we introduce the adversarial retrieval attack (AREA) task. The AREA task is meant to trick DR models into retrieving a target document that is outside the initial set of candidate documents retrieved by the DR model in response to a query. We consider the decision-based black-box adversarial setting, which is realistic in real-world search engines. To address the AREA task, we first employ existing adversarial attack methods designed for NRMs. We find that the promising results that have previously been reported on attacking NRMs, do not generalize to DR models: these methods underperform a simple term spamming method. We attribute the observed lack of generalizability to the interaction-focused architecture of NRMs, which emphasizes fine-grained relevance matching. DR models follow a different representation-focused architecture that prioritizes coarse-grained representations. We propose to formalize attacks on DR models as a contrastive learning problem in a multi-view representation space. The core idea is to encourage the consistency between each view representation of the target document and its corresponding viewer via view-wise supervision signals. Experimental results demonstrate that the proposed method can significantly outperform existing attack strategies in misleading the DR model with small indiscernible text perturbations.

Yongkang Li, Panagiotis Eustratiadis, Yixing Fan et al.

Decoder-only large language models (LLMs) are increasingly replacing BERT-style architectures as the backbone for dense retrieval, achieving substantial performance gains and broad adoption. However, the robustness of these LLM-based retrievers remains underexplored. In this paper, we present the first systematic study of the robustness of state-of-the-art open-source LLM-based dense retrievers from two complementary perspectives: generalizability and stability. For generalizability, we evaluate retrieval effectiveness across four benchmarks spanning 30 datasets, using linear mixed-effects models to estimate marginal mean performance and disentangle intrinsic model capability from dataset heterogeneity. Our analysis reveals that while instruction-tuned models generally excel, those optimized for complex reasoning often suffer a ``specialization tax,'' exhibiting limited generalizability in broader contexts. For stability, we assess model resilience against both unintentional query variations~(e.g., paraphrasing, typos) and malicious adversarial attacks~(e.g., corpus poisoning). We find that LLM-based retrievers show improved robustness against typos and corpus poisoning compared to encoder-only baselines, yet remain vulnerable to semantic perturbations like synonymizing. Further analysis shows that embedding geometry (e.g., angular uniformity) provides predictive signals for lexical stability and suggests that scaling model size generally improves robustness. These findings inform future robustness-aware retriever design and principled benchmarking. Our code is publicly available at https://github.com/liyongkang123/Robust_LLM_Retriever_Eval.

Hongru Song, Yu-An Liu, Ruqing Zhang et al.

Retrieval-augmented generation (RAG) enhances large language model (LLM) reasoning by retrieving external documents, but also opens up new attack surfaces. We study knowledge-base poisoning attacks in RAG, where an attacker injects malicious content into the retrieval corpus, which is then naturally surfaced by the retriever and consumed by the LLM during reasoning. Unlike prior work that floods the corpus with poisoned documents, we propose AdversarialCoT, a query-specific attack that poisons only a single document in the corpus. AdversarialCoT first extracts the target LLM's reasoning framework to guide the construction of an initial adversarial chain-of-thought (CoT). The adversarial document is iteratively refined through interactions with the LLM, progressively exposing and exploiting critical reasoning vulnerabilities. Experiments on benchmark LLMs show that a single adversarial document can significantly degrade reasoning accuracy, revealing subtle yet impactful weaknesses. This study exposes security risks in RAG systems and provides actionable insights for designing more robust LLM reasoning pipelines.

Yihang Wang, Xu Huang, Bowen Tian et al.

Generative LLM have achieved remarkable success in various industrial applications, owing to their promising In-Context Learning capabilities. However, the issue of long context in complex tasks poses a significant barrier to their wider adoption, manifested in two main aspects: (i) The excessively long context leads to high costs and inference delays. (ii) A substantial amount of task-irrelevant information introduced by long contexts exacerbates the "lost in the middle" problem. Existing methods compress context by removing redundant tokens using metrics such as self-information or PPL, which is inconsistent with the objective of retaining the most important tokens when conditioning on a given query. In this study, we introduce information bottleneck theory (IB) to model the problem, offering a novel perspective that thoroughly addresses the essential properties required for context compression. Additionally, we propose a cross-attention-based approach to approximate mutual information in IB, which can be flexibly replaced with suitable alternatives in different scenarios. Extensive experiments on four datasets demonstrate that our method achieves a 25% increase in compression rate compared to the state-of-the-art, while maintaining question answering performance. In particular, the context compressed by our method even outperform the full context in some cases.

Yixing Fan, Qiang Yan, Wenshan Wang et al.

\Ac{RAG} has emerged as a crucial technique for enhancing large models with real-time and domain-specific knowledge. While numerous improvements and open-source tools have been proposed to refine the \ac{RAG} framework for accuracy, relatively little attention has been given to improving the trustworthiness of generated results. To address this gap, we introduce TrustRAG, a novel framework that enhances \ac{RAG} from three perspectives: indexing, retrieval, and generation. Specifically, in the indexing stage, we propose a semantic-enhanced chunking strategy that incorporates hierarchical indexing to supplement each chunk with contextual information, ensuring semantic completeness. In the retrieval stage, we introduce a utility-based filtering mechanism to identify high-quality information, supporting answer generation while reducing input length. In the generation stage, we propose fine-grained citation enhancement, which detects opinion-bearing sentences in responses and infers citation relationships at the sentence-level, thereby improving citation accuracy. We open-source the TrustRAG framework and provide a demonstration studio designed for excerpt-based question answering tasks \footnote{https://huggingface.co/spaces/golaxy/TrustRAG}. Based on these, we aim to help researchers: 1) systematically enhancing the trustworthiness of \ac{RAG} systems and (2) developing their own \ac{RAG} systems with more reliable outputs.

Zhihao Zhang, Yixing Fan, Ruqing Zhang et al.

The advancement of LLMs has significantly boosted the performance of complex long-form question answering tasks. However, one prominent issue of LLMs is the generated "hallucination" responses that are not factual. Consequently, attribution for each claim in responses becomes a common solution to improve the factuality and verifiability. Existing researches mainly focus on how to provide accurate citations for the response, which largely overlook the importance of identifying the claims or statements for each response. To bridge this gap, we introduce a new claim decomposition benchmark, which requires building system that can identify atomic and checkworthy claims for LLM responses. Specifically, we present the Chinese Atomic Claim Decomposition Dataset (CACDD), which builds on the WebCPM dataset with additional expert annotations to ensure high data quality. The CACDD encompasses a collection of 500 human-annotated question-answer pairs, including a total of 4956 atomic claims. We further propose a new pipeline for human annotation and describe the challenges of this task. In addition, we provide experiment results on zero-shot, few-shot and fine-tuned LLMs as baselines. The results show that the claim decomposition is highly challenging and requires further explorations. All code and data are publicly available at \url{https://github.com/FBzzh/CACDD}.

Xinyu Ma, Jiafeng Guo, Ruqing Zhang et al.

Recently pre-trained language representation models such as BERT have shown great success when fine-tuned on downstream tasks including information retrieval (IR). However, pre-training objectives tailored for ad-hoc retrieval have not been well explored. In this paper, we propose Pre-training with Representative wOrds Prediction (PROP) for ad-hoc retrieval. PROP is inspired by the classical statistical language model for IR, specifically the query likelihood model, which assumes that the query is generated as the piece of text representative of the "ideal" document. Based on this idea, we construct the representative words prediction (ROP) task for pre-training. Given an input document, we sample a pair of word sets according to the document language model, where the set with higher likelihood is deemed as more representative of the document. We then pre-train the Transformer model to predict the pairwise preference between the two word sets, jointly with the Masked Language Model (MLM) objective. By further fine-tuning on a variety of representative downstream ad-hoc retrieval tasks, PROP achieves significant improvements over baselines without pre-training or with other pre-training methods. We also show that PROP can achieve exciting performance under both the zero- and low-resource IR settings. The code and pre-trained models are available at https://github.com/Albert-Ma/PROP.

Yu-An Liu, Ruqing Zhang, Jiafeng Guo et al.

Adversarial ranking attacks have gained increasing attention due to their success in probing vulnerabilities, and, hence, enhancing the robustness, of neural ranking models. Conventional attack methods employ perturbations at a single granularity, e.g., word or sentence level, to target documents. However, limiting perturbations to a single level of granularity may reduce the flexibility of adversarial examples, thereby diminishing the potential threat of the attack. Therefore, we focus on generating high-quality adversarial examples by incorporating multi-granular perturbations. Achieving this objective involves tackling a combinatorial explosion problem, which requires identifying an optimal combination of perturbations across all possible levels of granularity, positions, and textual pieces. To address this challenge, we transform the multi-granular adversarial attack into a sequential decision-making process, where perturbations in the next attack step build on the perturbed document in the current attack step. Since the attack process can only access the final state without direct intermediate signals, we use reinforcement learning to perform multi-granular attacks. During the reinforcement learning process, two agents work cooperatively to identify multi-granular vulnerabilities as attack targets and organize perturbation candidates into a final perturbation sequence. Experimental results show that our attack method surpasses prevailing baselines in both attack effectiveness and imperceptibility.

Jiafeng Guo, Changjiang Zhou, Ruqing Zhang et al.

Knowledge-intensive language tasks (KILTs) typically require retrieving relevant documents from trustworthy corpora, e.g., Wikipedia, to produce specific answers. Very recently, a pre-trained generative retrieval model for KILTs, named CorpusBrain, was proposed and reached new state-of-the-art retrieval performance. However, most existing research on KILTs, including CorpusBrain, has predominantly focused on a static document collection, overlooking the dynamic nature of real-world scenarios, where new documents are continuously being incorporated into the source corpus. To address this gap, it is crucial to explore the capability of retrieval models to effectively handle the dynamic retrieval scenario inherent in KILTs. In this work, we first introduce the continual document learning (CDL) task for KILTs and build a novel benchmark dataset named KILT++ based on the original KILT dataset for evaluation. Then, we conduct a comprehensive study over the use of pre-trained CorpusBrain on KILT++. Unlike the promising results in the stationary scenario, CorpusBrain is prone to catastrophic forgetting in the dynamic scenario, hence hampering the retrieval performance. To alleviate this issue, we propose CorpusBrain++, a continual generative pre-training framework. Empirical results demonstrate the significant effectiveness and remarkable efficiency of CorpusBrain++ in comparison to both traditional and generative IR methods.

Run-Ze Fan, Yixing Fan, Jiangui Chen et al.

Automatic mainstream hashtag recommendation aims to accurately provide users with concise and popular topical hashtags before publication. Generally, mainstream hashtag recommendation faces challenges in the comprehensive difficulty of newly posted tweets in response to new topics, and the accurate identification of mainstream hashtags beyond semantic correctness. However, previous retrieval-based methods based on a fixed predefined mainstream hashtag list excel in producing mainstream hashtags, but fail to understand the constant flow of up-to-date information. Conversely, generation-based methods demonstrate a superior ability to comprehend newly posted tweets, but their capacity is constrained to identifying mainstream hashtags without additional features. Inspired by the recent success of the retrieval-augmented technique, in this work, we attempt to adopt this framework to combine the advantages of both approaches. Meantime, with the help of the generator component, we could rethink how to further improve the quality of the retriever component at a low cost. Therefore, we propose RetrIeval-augmented Generative Mainstream HashTag Recommender (RIGHT), which consists of three components: 1) a retriever seeks relevant hashtags from the entire tweet-hashtags set; 2) a selector enhances mainstream identification by introducing global signals; and 3) a generator incorporates input tweets and selected hashtags to directly generate the desired hashtags. The experimental results show that our method achieves significant improvements over state-of-the-art baselines. Moreover, RIGHT can be easily integrated into large language models, improving the performance of ChatGPT by more than 10%.

Haosheng Qian, Yixing Fan, Ruqing Zhang et al.

Retrieval-augmented generation (RAG) appears as a promising method to alleviate the "hallucination" problem in large language models (LLMs), since it can incorporate external traceable resources for response generation. The essence of RAG in combating the hallucination issue lies in accurately attributing claims in responses to the corresponding retrieved documents. However, most of existing works focus on improving the quality of generated responses from the LLM, while largely overlooked its ability to attribute sources accurately. In this study, we conduct a systematic analysis about the capabilities of LLMs in generating citations within response generation, and further introduce a novel method to enhance their citation generation abilities. Specifically, we evaluate both the correctness and citation quality for seven widely-used LLMs on two benchmark datasets. Meanwhile, we introduce new citation evaluation metrics to eliminate the over-penalization of unnecessary and excessive citations in existing metrics. Furthermore, we propose a Generate-then-Refine method that completes relevant citations and removes irrelevant ones without altering the response text. The results on WebGLM-QA, ASQA and ELI5 datasets show that our method substantially improves the quality of citations in responses generated by LLMs.

Yixing Fan, Xiaohui Xie, Yinqiong Cai et al.

The core of information retrieval (IR) is to identify relevant information from large-scale resources and return it as a ranked list to respond to the user's information need. In recent years, the resurgence of deep learning has greatly advanced this field and leads to a hot topic named NeuIR (i.e., neural information retrieval), especially the paradigm of pre-training methods (PTMs). Owing to sophisticated pre-training objectives and huge model size, pre-trained models can learn universal language representations from massive textual data, which are beneficial to the ranking task of IR. Recently, a large number of works, which are dedicated to the application of PTMs in IR, have been introduced to promote the retrieval performance. Considering the rapid progress of this direction, this survey aims to provide a systematic review of pre-training methods in IR. To be specific, we present an overview of PTMs applied in different components of an IR system, including the retrieval component, the re-ranking component, and other components. In addition, we also introduce PTMs specifically designed for IR, and summarize available datasets as well as benchmark leaderboards. Moreover, we discuss some open challenges and highlight several promising directions, with the hope of inspiring and facilitating more works on these topics for future research.

Jiangui Chen, Ruqing Zhang, Jiafeng Guo et al.

Question Answering (QA), a popular and promising technique for intelligent information access, faces a dilemma about data as most other AI techniques. On one hand, modern QA methods rely on deep learning models which are typically data-hungry. Therefore, it is expected to collect and fuse all the available QA datasets together in a common site for developing a powerful QA model. On the other hand, real-world QA datasets are typically distributed in the form of isolated islands belonging to different parties. Due to the increasing awareness of privacy security, it is almost impossible to integrate the data scattered around, or the cost is prohibited. A possible solution to this dilemma is a new approach known as federated learning, which is a privacy-preserving machine learning technique over distributed datasets. In this work, we propose to adopt federated learning for QA with the special concern on the statistical heterogeneity of the QA data. Here the heterogeneity refers to the fact that annotated QA data are typically with non-identical and independent distribution (non-IID) and unbalanced sizes in practice. Traditional federated learning methods may sacrifice the accuracy of individual models under the heterogeneous situation. To tackle this problem, we propose a novel Federated Matching framework for QA, named FedMatch, with a backbone-patch architecture. The shared backbone is to distill the common knowledge of all the participants while the private patch is a compact and efficient module to retain the domain information for each participant. To facilitate the evaluation, we build a benchmark collection based on several QA datasets from different domains to simulate the heterogeneous situation in practice. Empirical studies demonstrate that our model can achieve significant improvements against the baselines over all the datasets.

Chen Wu, Ruqing Zhang, Jiafeng Guo et al.

Recently, we have witnessed the bloom of neural ranking models in the information retrieval (IR) field. So far, much effort has been devoted to developing effective neural ranking models that can generalize well on new data. There has been less attention paid to the robustness perspective. Unlike the effectiveness which is about the average performance of a system under normal purpose, robustness cares more about the system performance in the worst case or under malicious operations instead. When a new technique enters into the real-world application, it is critical to know not only how it works in average, but also how would it behave in abnormal situations. So we raise the question in this work: Are neural ranking models robust? To answer this question, firstly, we need to clarify what we refer to when we talk about the robustness of ranking models in IR. We show that robustness is actually a multi-dimensional concept and there are three ways to define it in IR: 1) The performance variance under the independent and identically distributed (I.I.D.) setting; 2) The out-of-distribution (OOD) generalizability; and 3) The defensive ability against adversarial operations. The latter two definitions can be further specified into two different perspectives respectively, leading to 5 robustness tasks in total. Based on this taxonomy, we build corresponding benchmark datasets, design empirical experiments, and systematically analyze the robustness of several representative neural ranking models against traditional probabilistic ranking models and learning-to-rank (LTR) models. The empirical results show that there is no simple answer to our question. While neural ranking models are less robust against other IR models in most cases, some of them can still win 1 out of 5 tasks. This is the first comprehensive study on the robustness of neural ranking models.

Yinqiong Cai, Yixing Fan, Jiafeng Guo et al.

Similar question retrieval is a core task in community-based question answering (CQA) services. To balance the effectiveness and efficiency, the question retrieval system is typically implemented as multi-stage rankers: The first-stage ranker aims to recall potentially relevant questions from a large repository, and the latter stages attempt to re-rank the retrieved results. Most existing works on question retrieval mainly focused on the re-ranking stages, leaving the first-stage ranker to some traditional term-based methods. However, term-based methods often suffer from the vocabulary mismatch problem, especially on short texts, which may block the re-rankers from relevant questions at the very beginning. An alternative is to employ embedding-based methods for the first-stage ranker, which compress texts into dense vectors to enhance the semantic matching. However, these methods often lose the discriminative power as term-based methods, thus introduce noise during retrieval and hurt the recall performance. In this work, we aim to tackle the dilemma of the first-stage ranker, and propose a discriminative semantic ranker, namely DenseTrans, for high-recall retrieval. Specifically, DenseTrans is a densely connected Transformer, which learns semantic embeddings for texts based on Transformer layers. Meanwhile, DenseTrans promotes low-level features through dense connections to keep the discriminative power of the learned representations. DenseTrans is inspired by DenseNet in computer vision (CV), but poses a new way to use the dense connectivity which is totally different from its original design purpose. Experimental results over two question retrieval benchmark datasets show that our model can obtain significant gain on recall against strong term-based methods as well as state-of-the-art embedding-based methods.

Xinyu Ma, Jiafeng Guo, Ruqing Zhang et al.

Pre-training and fine-tuning have achieved remarkable success in many downstream natural language processing (NLP) tasks. Recently, pre-training methods tailored for information retrieval (IR) have also been explored, and the latest success is the PROP method which has reached new SOTA on a variety of ad-hoc retrieval benchmarks. The basic idea of PROP is to construct the \textit{representative words prediction} (ROP) task for pre-training inspired by the query likelihood model. Despite its exciting performance, the effectiveness of PROP might be bounded by the classical unigram language model adopted in the ROP task construction process. To tackle this problem, we propose a bootstrapped pre-training method (namely B-PROP) based on BERT for ad-hoc retrieval. The key idea is to use the powerful contextual language model BERT to replace the classical unigram language model for the ROP task construction, and re-train BERT itself towards the tailored objective for IR. Specifically, we introduce a novel contrastive method, inspired by the divergence-from-randomness idea, to leverage BERT's self-attention mechanism to sample representative words from the document. By further fine-tuning on downstream ad-hoc retrieval tasks, our method achieves significant improvements over baselines without pre-training or with other pre-training methods, and further pushes forward the SOTA on a variety of ad-hoc retrieval tasks.

Jiafeng Guo, Yinqiong Cai, Yixing Fan et al.

Multi-stage ranking pipelines have been a practical solution in modern search systems, where the first-stage retrieval is to return a subset of candidate documents, and latter stages attempt to re-rank those candidates. Unlike re-ranking stages going through quick technique shifts during past decades, the first-stage retrieval has long been dominated by classical term-based models. Unfortunately, these models suffer from the vocabulary mismatch problem, which may block re-ranking stages from relevant documents at the very beginning. Therefore, it has been a long-term desire to build semantic models for the first-stage retrieval that can achieve high recall efficiently. Recently, we have witnessed an explosive growth of research interests on the first-stage semantic retrieval models. We believe it is the right time to survey current status, learn from existing methods, and gain some insights for future development. In this paper, we describe the current landscape of the first-stage retrieval models under a unified framework to clarify the connection between classical term-based retrieval methods, early semantic retrieval methods and neural semantic retrieval methods. Moreover, we identify some open challenges and envision some future directions, with the hope of inspiring more researches on these important yet less investigated topics.

Yixing Fan, Jiafeng Guo, Xinyu Ma et al.

Relevance plays a central role in information retrieval (IR), which has received extensive studies starting from the 20th century. The definition and the modeling of relevance has always been critical challenges in both information science and computer science research areas. Along with the debate and exploration on relevance, IR has already become a core task in many real-world applications, such as Web search engines, question answering systems, conversational bots, and so on. While relevance acts as a unified concept in all these retrieval tasks, the inherent definitions are quite different due to the heterogeneity of these tasks. This raises a question to us: Do these different forms of relevance really lead to different modeling focuses? To answer this question, in this work, we conduct an empirical study on relevance modeling in three representative IR tasks, i.e., document retrieval, answer retrieval, and response retrieval. Specifically, we attempt to study the following two questions: 1) Does relevance modeling in these tasks really show differences in terms of natural language understanding (NLU)? We employ 16 linguistic tasks to probe a unified retrieval model over these three retrieval tasks to answer this question. 2) If there do exist differences, how can we leverage the findings to enhance the relevance modeling? We proposed three intervention methods to investigate how to leverage different modeling focuses of relevance to improve these IR tasks. We believe the way we study the problem as well as our findings would be beneficial to the IR community.

Chen Wu, Ruqing Zhang, Jiafeng Guo et al.

Ranked list truncation is of critical importance in a variety of professional information retrieval applications such as patent search or legal search. The goal is to dynamically determine the number of returned documents according to some user-defined objectives, in order to reach a balance between the overall utility of the results and user efforts. Existing methods formulate this task as a sequential decision problem and take some pre-defined loss as a proxy objective, which suffers from the limitation of local decision and non-direct optimization. In this work, we propose a global decision based truncation model named AttnCut, which directly optimizes user-defined objectives for the ranked list truncation. Specifically, we take the successful transformer architecture to capture the global dependency within the ranked list for truncation decision, and employ the reward augmented maximum likelihood (RAML) for direct optimization. We consider two types of user-defined objectives which are of practical usage. One is the widely adopted metric such as F1 which acts as a balanced objective, and the other is the best F1 under some minimal recall constraint which represents a typical objective in professional search. Empirical results over the Robust04 and MQ2007 datasets demonstrate the effectiveness of our approach as compared with the state-of-the-art baselines.

Ruqing Zhang, Jiafeng Guo, Yixing Fan et al.

Query understanding is a fundamental problem in information retrieval (IR), which has attracted continuous attention through the past decades. Many different tasks have been proposed for understanding users' search queries, e.g., query classification or query clustering. However, it is not that precise to understand a search query at the intent class/cluster level due to the loss of many detailed information. As we may find in many benchmark datasets, e.g., TREC and SemEval, queries are often associated with a detailed description provided by human annotators which clearly describes its intent to help evaluate the relevance of the documents. If a system could automatically generate a detailed and precise intent description for a search query, like human annotators, that would indicate much better query understanding has been achieved. In this paper, therefore, we propose a novel Query-to-Intent-Description (Q2ID) task for query understanding. Unlike those existing ranking tasks which leverage the query and its description to compute the relevance of documents, Q2ID is a reverse task which aims to generate a natural language intent description based on both relevant and irrelevant documents of a given query. To address this new task, we propose a novel Contrastive Generation model, namely CtrsGen for short, to generate the intent description by contrasting the relevant documents with the irrelevant documents given a query. We demonstrate the effectiveness of our model by comparing with several state-of-the-art generation models on the Q2ID task. We discuss the potential usage of such Q2ID technique through an example application.

Lixin Su, Jiafeng Guo, Ruqing Zhang et al.

Machine reading comprehension (MRC) has become a core component in a variety of natural language processing (NLP) applications such as question answering and dialogue systems. It becomes a practical challenge that an MRC model needs to learn in non-stationary environments, in which the underlying data distribution changes over time. A typical scenario is the domain drift, i.e. different domains of data come one after another, where the MRC model is required to adapt to the new domain while maintaining previously learned ability. To tackle such a challenge, in this work, we introduce the \textit{Continual Domain Adaptation} (CDA) task for MRC. So far as we know, this is the first study on the continual learning perspective of MRC. We build two benchmark datasets for the CDA task, by re-organizing existing MRC collections into different domains with respect to context type and question type, respectively. We then analyze and observe the catastrophic forgetting (CF) phenomenon of MRC under the CDA setting. To tackle the CDA task, we propose several BERT-based continual learning MRC models using either regularization-based methodology or dynamic-architecture paradigm. We analyze the performance of different continual learning MRC models under the CDA task and show that the proposed dynamic-architecture based model achieves the best performance.

Zizhen Wang, Yixing Fan, Jiafeng Guo et al.

Community Question Answering (CQA) has become a primary means for people to acquire knowledge, where people are free to ask questions or submit answers. To enhance the efficiency of the service, similar question identification becomes a core task in CQA which aims to find a similar question from the archived repository whenever a new question is asked. However, it has long been a challenge to properly measure the similarity between two questions due to the inherent variation of natural language, i.e., there could be different ways to ask a same question or different questions sharing similar expressions. To alleviate this problem, it is natural to involve the existing answers for the enrichment of the archived questions. Traditional methods typically take a one-side usage, which leverages the answer as some expanded representation of the corresponding question. Unfortunately, this may introduce unexpected noises into the similarity computation since answers are often long and diverse, leading to inferior performance. In this work, we propose a two-side usage, which leverages the answer as a bridge of the two questions. The key idea is based on our observation that similar questions could be addressed by similar parts of the answer while different questions may not. In other words, we can compare the matching patterns of the two questions over the same answer to measure their similarity. In this way, we propose a novel matching over matching model, namely Match$^2$, which compares the matching patterns between two question-answer pairs for similar question identification. Empirical experiments on two benchmark datasets demonstrate that our model can significantly outperform previous state-of-the-art methods on the similar question identification task.

Jiafeng Guo, Yixing Fan, Xiang Ji et al.

Text matching is the core problem in many natural language processing (NLP) tasks, such as information retrieval, question answering, and conversation. Recently, deep leaning technology has been widely adopted for text matching, making neural text matching a new and active research domain. With a large number of neural matching models emerging rapidly, it becomes more and more difficult for researchers, especially those newcomers, to learn and understand these new models. Moreover, it is usually difficult to try these models due to the tedious data pre-processing, complicated parameter configuration, and massive optimization tricks, not to mention the unavailability of public codes sometimes. Finally, for researchers who want to develop new models, it is also not an easy task to implement a neural text matching model from scratch, and to compare with a bunch of existing models. In this paper, therefore, we present a novel system, namely MatchZoo, to facilitate the learning, practicing and designing of neural text matching models. The system consists of a powerful matching library and a user-friendly and interactive studio, which can help researchers: 1) to learn state-of-the-art neural text matching models systematically, 2) to train, test and apply these models with simple configurable steps; and 3) to develop their own models with rich APIs and assistance.

Lixin Su, Jiafeng Guo, Yixing Fan et al.

Web question answering (QA) has become an indispensable component in modern search systems, which can significantly improve users' search experience by providing a direct answer to users' information need. This could be achieved by applying machine reading comprehension (MRC) models over the retrieved passages to extract answers with respect to the search query. With the development of deep learning techniques, state-of-the-art MRC performances have been achieved by recent deep methods. However, existing studies on MRC seldom address the predictive uncertainty issue, i.e., how likely the prediction of an MRC model is wrong, leading to uncontrollable risks in real-world Web QA applications. In this work, we first conduct an in-depth investigation over the risk of Web QA. We then introduce a novel risk control framework, which consists of a qualify model for uncertainty estimation using the probe idea, and a decision model for selectively output. For evaluation, we introduce risk-related metrics, rather than the traditional EM and F1 in MRC, for the evaluation of risk-aware Web QA. The empirical results over both the real-world Web QA dataset and the academic MRC benchmark collection demonstrate the effectiveness of our approach.

Ruqing Zhang, Jiafeng Guo, Yixing Fan et al.

In this paper, we introduce and tackle the Outline Generation (OG) task, which aims to unveil the inherent content structure of a multi-paragraph document by identifying its potential sections and generating the corresponding section headings. Without loss of generality, the OG task can be viewed as a novel structured summarization task. To generate a sound outline, an ideal OG model should be able to capture three levels of coherence, namely the coherence between context paragraphs, that between a section and its heading, and that between context headings. The first one is the foundation for section identification, while the latter two are critical for consistent heading generation. In this work, we formulate the OG task as a hierarchical structured prediction problem, i.e., to first predict a sequence of section boundaries and then a sequence of section headings accordingly. We propose a novel hierarchical structured neural generation model, named HiStGen, for the task. Our model attempts to capture the three-level coherence via the following ways. First, we introduce a Markov paragraph dependency mechanism between context paragraphs for section identification. Second, we employ a section-aware attention mechanism to ensure the semantic coherence between a section and its heading. Finally, we leverage a Markov heading dependency mechanism and a review mechanism between context headings to improve the consistency and eliminate duplication between section headings. Besides, we build a novel WIKIOG dataset, a public collection which consists of over 1.75 million document-outline pairs for research on the OG task. Experimental results on our benchmark dataset demonstrate that our model can significantly outperform several state-of-the-art sequential generation models for the OG task.

Jiafeng Guo, Yixing Fan, Liang Pang et al.

Ranking models lie at the heart of research on information retrieval (IR). During the past decades, different techniques have been proposed for constructing ranking models, from traditional heuristic methods, probabilistic methods, to modern machine learning methods. Recently, with the advance of deep learning technology, we have witnessed a growing body of work in applying shallow or deep neural networks to the ranking problem in IR, referred to as neural ranking models in this paper. The power of neural ranking models lies in the ability to learn from the raw text inputs for the ranking problem to avoid many limitations of hand-crafted features. Neural networks have sufficient capacity to model complicated tasks, which is needed to handle the complexity of relevance estimation in ranking. Since there have been a large variety of neural ranking models proposed, we believe it is the right time to summarize the current status, learn from existing methodologies, and gain some insights for future development. In contrast to existing reviews, in this survey, we will take a deep look into the neural ranking models from different dimensions to analyze their underlying assumptions, major design principles, and learning strategies. We compare these models through benchmark tasks to obtain a comprehensive empirical understanding of the existing techniques. We will also discuss what is missing in the current literature and what are the promising and desired future directions.

Yan Xiao, Jiafeng Guo, Yixing Fan et al.

Vocabulary mismatch is a central problem in information retrieval (IR), i.e., the relevant documents may not contain the same (symbolic) terms of the query. Recently, neural representations have shown great success in capturing semantic relatedness, leading to new possibilities to alleviate the vocabulary mismatch problem in IR. However, most existing efforts in this direction have been devoted to the re-ranking stage. That is to leverage neural representations to help re-rank a set of candidate documents, which are typically obtained from an initial retrieval stage based on some symbolic index and search scheme (e.g., BM25 over the inverted index). This naturally raises a question: if the relevant documents have not been found in the initial retrieval stage due to vocabulary mismatch, there would be no chance to re-rank them to the top positions later. Therefore, in this paper, we study the problem how to employ neural representations to improve the recall of relevant documents in the initial retrieval stage. Specifically, to meet the efficiency requirement of the initial stage, we introduce a neural index for the neural representations of documents, and propose two hybrid search schemes based on both neural and symbolic indices, namely the parallel search scheme and the sequential search scheme. Our experiments show that both hybrid index and search schemes can improve the recall of the initial retrieval stage with small overhead.

Yixing Fan, Jiafeng Guo, Yanyan Lan et al.

Assessing relevance between a query and a document is challenging in ad-hoc retrieval due to its diverse patterns, i.e., a document could be relevant to a query as a whole or partially as long as it provides sufficient information for users' need. Such diverse relevance patterns require an ideal retrieval model to be able to assess relevance in the right granularity adaptively. Unfortunately, most existing retrieval models compute relevance at a single granularity, either document-wide or passage-level, or use fixed combination strategy, restricting their ability in capturing diverse relevance patterns. In this work, we propose a data-driven method to allow relevance signals at different granularities to compete with each other for final relevance assessment. Specifically, we propose a HIerarchical Neural maTching model (HiNT) which consists of two stacked components, namely local matching layer and global decision layer. The local matching layer focuses on producing a set of local relevance signals by modeling the semantic matching between a query and each passage of a document. The global decision layer accumulates local signals into different granularities and allows them to compete with each other to decide the final relevance score. Experimental results demonstrate that our HiNT model outperforms existing state-of-the-art retrieval models significantly on benchmark ad-hoc retrieval datasets.

Jiafeng Guo, Yixing Fan, Qingyao Ai et al.

In recent years, deep neural networks have led to exciting breakthroughs in speech recognition, computer vision, and natural language processing (NLP) tasks. However, there have been few positive results of deep models on ad-hoc retrieval tasks. This is partially due to the fact that many important characteristics of the ad-hoc retrieval task have not been well addressed in deep models yet. Typically, the ad-hoc retrieval task is formalized as a matching problem between two pieces of text in existing work using deep models, and treated equivalent to many NLP tasks such as paraphrase identification, question answering and automatic conversation. However, we argue that the ad-hoc retrieval task is mainly about relevance matching while most NLP matching tasks concern semantic matching, and there are some fundamental differences between these two matching tasks. Successful relevance matching requires proper handling of the exact matching signals, query term importance, and diverse matching requirements. In this paper, we propose a novel deep relevance matching model (DRMM) for ad-hoc retrieval. Specifically, our model employs a joint deep architecture at the query term level for relevance matching. By using matching histogram mapping, a feed forward matching network, and a term gating network, we can effectively deal with the three relevance matching factors mentioned above. Experimental results on two representative benchmark collections show that our model can significantly outperform some well-known retrieval models as well as state-of-the-art deep matching models.

Yixing Fan, Jiafeng Guo, Yanyan Lan et al.

When applying learning to rank algorithms to Web search, a large number of features are usually designed to capture the relevance signals. Most of these features are computed based on the extracted textual elements, link analysis, and user logs. However, Web pages are not solely linked texts, but have structured layout organizing a large variety of elements in different styles. Such layout itself can convey useful visual information, indicating the relevance of a Web page. For example, the query-independent layout (i.e., raw page layout) can help identify the page quality, while the query-dependent layout (i.e., page rendered with matched query words) can further tell rich structural information (e.g., size, position and proximity) of the matching signals. However, such visual information of layout has been seldom utilized in Web search in the past. In this work, we propose to learn rich visual features automatically from the layout of Web pages (i.e., Web page snapshots) for relevance ranking. Both query-independent and query-dependent snapshots are considered as the new inputs. We then propose a novel visual perception model inspired by human's visual search behaviors on page viewing to extract the visual features. This model can be learned end-to-end together with traditional human-crafted features. We also show that such visual features can be efficiently acquired in the online setting with an extended inverted indexing scheme. Experiments on benchmark collections demonstrate that learning visual features from Web page snapshots can significantly improve the performance of relevance ranking in ad-hoc Web retrieval tasks.

Yixing Fan, Liang Pang, JianPeng Hou et al.

In recent years, deep neural models have been widely adopted for text matching tasks, such as question answering and information retrieval, showing improved performance as compared with previous methods. In this paper, we introduce the MatchZoo toolkit that aims to facilitate the designing, comparing and sharing of deep text matching models. Specifically, the toolkit provides a unified data preparation module for different text matching problems, a flexible layer-based model construction process, and a variety of training objectives and evaluation metrics. In addition, the toolkit has implemented two schools of representative deep text matching models, namely representation-focused models and interaction-focused models. Finally, users can easily modify existing models, create and share their own models for text matching in MatchZoo.