arXiv subjects

Zhichao Xu

Publications and source records attributed to Zhichao Xu.

At least 55 records · Page 3Linked to original sources

arXiv2025

Reinforcement Learning to Rank Using Coarse-grained Rewards

Learning to rank (LTR) plays a crucial role in various Information Retrieval (IR) tasks. Although supervised LTR methods based on fine-grained relevance labels (e.g., document-level annotations) have achieved significant success, their reliance on costly and potentially biased annotations limits scalability and alignment with realistic goals. In contrast, coarse-grained feedback signals, such as duration time and session-level engagement, are more accessible and affordable. Reinforcement Learning (RL) offers a promising framework to directly optimize these objectives using reward signals, but most existing Reinforcement Learning to Rank (RLTR) approaches suffer from high variance and low sample efficiency. Motivated by recent advances in large language models (LLMs), we re-examine the problem of RLTR with coarse-grained rewards and propose new RLTR methods based on widely used RL algorithms for LLMs. We systematically compare supervised learning and RL-based methods across various model architectures and coarse-grained reward functions on large-scale LTR benchmarks. Experimental results demonstrate that advanced RL methods can directly learn from coarse-grained rewards and outperform strong supervised learning baselines even with fine-grained labels. This shows the great potential of RLTR for metric-agnostic ranking optimization.

cs.IR↗

arXiv2025

Found in Translation: Measuring Multilingual LLM Consistency as Simple as Translate then Evaluate

Large language models (LLMs) provide detailed and impressive responses to queries in English. However, are they really consistent at responding to the same query in other languages? The popular way of evaluating for multilingual performance of LLMs requires expensive-to-collect annotated datasets. Further, evaluating for tasks like open-ended generation, where multiple correct answers may exist, is nontrivial. Instead, we propose to evaluate the predictability of model response across different languages. In this work, we propose a framework to evaluate LLM's cross-lingual consistency based on a simple Translate then Evaluate strategy. We instantiate this evaluation framework along two dimensions of consistency: information and empathy. Our results reveal pronounced inconsistencies in popular LLM responses across thirty languages, with severe performance deficits in certain language families and scripts, underscoring critical weaknesses in their multilingual capabilities. These findings necessitate cross-lingual evaluations that are consistent along multiple dimensions. We invite practitioners to use our framework for future multilingual LLM benchmarking.

cs.CL↗

arXiv2025

SLOT: Structuring the Output of Large Language Models

Structured outputs are essential for large language models (LLMs) in critical applications like agents and information extraction. Despite their capabilities, LLMs often generate outputs that deviate from predefined schemas, significantly hampering reliable application development. We present SLOT (Structured LLM Output Transformer), a model-agnostic approach that transforms unstructured LLM outputs into precise structured formats. While existing solutions predominantly rely on constrained decoding techniques or are tightly coupled with specific models, SLOT employs a fine-tuned lightweight language model as a post-processing layer, achieving flexibility across various LLMs and schema specifications. We introduce a systematic pipeline for data curation and synthesis alongside a formal evaluation methodology that quantifies both schema accuracy and content fidelity. Our results demonstrate that fine-tuned Mistral-7B model with constrained decoding achieves near perfect schema accuracy (99.5%) and content similarity (94.0%), outperforming Claude-3.5-Sonnet by substantial margins (+25 and +20 percentage points, respectively). Notably, even compact models like Llama-3.2-1B can match or exceed the structured output capabilities of much larger proprietary models when equipped with SLOT, enabling reliable structured generation in resource-constrained environments.

cs.CL↗

arXiv2025

State Space Models are Strong Text Rerankers

Transformers dominate NLP and IR; but their inference inefficiencies and challenges in extrapolating to longer contexts have sparked interest in alternative model architectures. Among these, state space models (SSMs) like Mamba offer promising advantages, particularly $O(1)$ time complexity in inference. Despite their potential, SSMs' effectiveness at text reranking -- a task requiring fine-grained query-document interaction and long-context understanding -- remains underexplored. This study benchmarks SSM-based architectures (specifically, Mamba-1 and Mamba-2) against transformer-based models across various scales, architectures, and pre-training objectives, focusing on performance and efficiency in text reranking tasks. We find that (1) Mamba architectures achieve competitive text ranking performance, comparable to transformer-based models of similar size; (2) they are less efficient in training and inference compared to transformers with flash attention; and (3) Mamba-2 outperforms Mamba-1 in both performance and efficiency. These results underscore the potential of state space models as a transformer alternative and highlight areas for improvement in future IR applications.

cs.CL↗

arXiv2025

A Systematic Survey of Automatic Prompt Optimization Techniques

Since the advent of large language models (LLMs), prompt engineering has been a crucial step for eliciting desired responses for various Natural Language Processing (NLP) tasks. However, prompt engineering remains an impediment for end users due to rapid advances in models, tasks, and associated best practices. To mitigate this, Automatic Prompt Optimization (APO) techniques have recently emerged that use various automated techniques to help improve the performance of LLMs on various tasks. In this paper, we present a comprehensive survey summarizing the current progress and remaining challenges in this field. We provide a formal definition of APO, a 5-part unifying framework, and then proceed to rigorously categorize all relevant works based on their salient features therein. We hope to spur further research guided by our framework.

cs.CL↗

arXiv2024

MAIN-RAG: Multi-Agent Filtering Retrieval-Augmented Generation

Large Language Models (LLMs) are becoming essential tools for various natural language processing tasks but often suffer from generating outdated or incorrect information. Retrieval-Augmented Generation (RAG) addresses this issue by incorporating external, real-time information retrieval to ground LLM responses. However, the existing RAG systems frequently struggle with the quality of retrieval documents, as irrelevant or noisy documents degrade performance, increase computational overhead, and undermine response reliability. To tackle this problem, we propose Multi-Agent Filtering Retrieval-Augmented Generation (MAIN-RAG), a training-free RAG framework that leverages multiple LLM agents to collaboratively filter and score retrieved documents. Specifically, MAIN-RAG introduces an adaptive filtering mechanism that dynamically adjusts the relevance filtering threshold based on score distributions, effectively minimizing noise while maintaining high recall of relevant documents. The proposed approach leverages inter-agent consensus to ensure robust document selection without requiring additional training data or fine-tuning. Experimental results across four QA benchmarks demonstrate that MAIN-RAG consistently outperforms traditional RAG approaches, achieving a 2-11% improvement in answer accuracy while reducing the number of irrelevant retrieved documents. Quantitative analysis further reveals that our approach achieves superior response consistency and answer accuracy over baseline methods, offering a competitive and practical alternative to training-based solutions.

cs.CL↗

arXiv2024

Multi-dimensional Evaluation of Empathetic Dialog Responses

Empathy is critical for effective and satisfactory conversational communication. Prior efforts to measure conversational empathy mostly focus on expressed communicative intents -- that is, the way empathy is expressed. Yet, these works ignore the fact that conversation is also a collaboration involving both speakers and listeners. In contrast, we propose a multi-dimensional empathy evaluation framework to measure both \emph{expressed intents from the speaker's perspective} and \emph{perceived empathy from the listener's perspective}. We apply our analytical framework to examine internal customer-service dialogues. We find the two dimensions (expressed intent types and perceived empathy) are inter-connected, while perceived empathy has high correlations with dialogue satisfaction levels. To reduce the annotation cost, we explore different options to automatically measure conversational empathy: prompting LLMs and training language model-based classifiers. Our experiments show that prompting methods with even popular models like GPT-4 and Flan family models perform relatively poorly on both public and our internal datasets. In contrast, instruction-finetuned classifiers based on Flan-T5 family models outperform prior works and competitive baselines. We conduct a detailed ablation study to give more insights into instruction finetuning method's strong performance.

cs.CL↗

arXiv2024

Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression

Increasingly, model compression techniques enable large language models (LLMs) to be deployed in real-world applications. As a result of this momentum towards local deployment, compressed LLMs will interact with a large population. Prior work on compression typically prioritize preserving perplexity, which is directly analogous to training loss. The impact of compression method on other critical aspects of model behavior\, -- \,particularly safety\, -- \,requires systematic assessment. To this end, we investigate the impact of model compression along four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; and(4) language modeling and downstream task performance. We examine a wide spectrum of LLM compression techniques, including unstructured pruning, semi-structured pruning, and quantization. Our analysis reveals that compression can lead to unexpected consequences. Although compression may unintentionally alleviate LLMs' degeneration harm, it can still exacerbate representational harm. Furthermore, increasing compression produces a divergent impact on different protected groups. Finally, different compression methods have drastically different safety impacts: for example, quantization mostly preserves bias while pruning degrades quickly. Our findings underscore the importance of integrating safety assessments into the development of compressed LLMs to ensure their reliability across real-world applications.\footnote{Our implementation and results are available here: \url{https://github.com/zhichaoxu-shufe/Beyond-Perplexity-Compression-Safety-Eval}}

cs.CL↗

arXiv2024

Counterfactual Editing for Search Result Explanation

Search Result Explanation (SeRE) aims to improve search sessions' effectiveness and efficiency by helping users interpret documents' relevance. Existing works mostly focus on factual explanation, i.e. to find/generate supporting evidence about documents' relevance to search queries. However, research in cognitive sciences has shown that human explanations are contrastive i.e. people explain an observed event using some counterfactual events; such explanations reduce cognitive load and provide actionable insights. Though already proven effective in machine learning and NLP communities, there lacks a strict formulation on how counterfactual explanations should be defined and structured, in the context of web search. In this paper, we first discuss the possible formulation of counterfactual explanations in the IR context. Next, we formulate a suite of desiderata for counterfactual explanation in SeRE task and corresponding automatic metrics. With this desiderata, we propose a method named \textbf{C}ounter\textbf{F}actual \textbf{E}diting for Search Research \textbf{E}xplanation (\textbf{CFE2}). CFE2 provides pairwise counterfactual explanations for document pairs within a search engine result page. Our experiments on five public search datasets demonstrate that CFE2 can significantly outperform baselines in both automatic metrics and human evaluations.

cs.IR↗

arXiv2024

In-Context Example Ordering Guided by Label Distributions

By allowing models to predict without task-specific training, in-context learning (ICL) with pretrained LLMs has enormous potential in NLP. However, a number of problems persist in ICL. In particular, its performance is sensitive to the choice and order of in-context examples. Given the same set of in-context examples with different orderings, model performance may vary between near random to near state-of-the-art. In this work, we formulate in-context example ordering as an optimization problem. We examine three problem settings that differ in the assumptions they make about what is known about the task. Inspired by the idea of learning from label proportions, we propose two principles for in-context example ordering guided by model's probability predictions. We apply our proposed principles to thirteen text classification datasets and nine different autoregressive LLMs with 700M to 13B parameters. We demonstrate that our approach outperforms the baselines by improving the classification accuracy, reducing model miscalibration, and also by selecting better in-context examples.

cs.CL↗

arXiv2024

An In-depth Investigation of User Response Simulation for Conversational Search

Conversational search has seen increased recent attention in both the IR and NLP communities. It seeks to clarify and solve users' search needs through multi-turn natural language interactions. However, most existing systems are trained and demonstrated with recorded or artificial conversation logs. Eventually, conversational search systems should be trained, evaluated, and deployed in an open-ended setting with unseen conversation trajectories. A key challenge is that training and evaluating such systems both require a human-in-the-loop, which is expensive and does not scale. One strategy is to simulate users, thereby reducing the scaling costs. However, current user simulators are either limited to only responding to yes-no questions from the conversational search system or unable to produce high-quality responses in general. In this paper, we show that existing user simulation systems could be significantly improved by a smaller finetuned natural language generation model. However, rather than merely reporting it as the new state-of-the-art, we consider it a strong baseline and present an in-depth investigation of simulating user response for conversational search. Our goal is to supplement existing work with an insightful hand-analysis of unsolved challenges by the baseline and propose our solutions. The challenges we identified include (1) a blind spot that is difficult to learn, and (2) a specific type of misevaluation in the standard setup. We propose a new generation system to effectively cover the training blind spot and suggest a new evaluation setup to avoid misevaluation. Our proposed system leads to significant improvements over existing systems and large language models such as GPT-4. Additionally, our analysis provides insights into the nature of user simulation to facilitate future work.

cs.IR↗

arXiv2023

Vertical Allocation-based Fair Exposure Amortizing in Ranking

Result ranking often affects consumer satisfaction as well as the amount of exposure each item receives in the ranking services. Myopically maximizing customer satisfaction by ranking items only according to relevance will lead to unfair distribution of exposure for items, followed by unfair opportunities and economic gains for item producers/providers. Such unfairness will force providers to leave the system and discourage new providers from coming in. Eventually, fewer purchase options would be left for consumers, and the utilities of both consumers and providers would be harmed. Thus, to maintain a balance between ranking relevance and fairness is crucial for both parties. In this paper, we focus on the exposure fairness in ranking services. We demonstrate that existing methods for amortized fairness optimization could be suboptimal in terms of fairness-relevance tradeoff because they fail to utilize the prior knowledge of consumers. We further propose a novel algorithm named Vertical Allocation-based Fair Exposure Amortizing in Ranking, or VerFair, to reach a better balance between exposure fairness and ranking performance. Extensive experiments on three real-world datasets show that VerFair significantly outperforms state-of-the-art fair ranking algorithms in fairness-performance trade-offs from both the individual level and the group level.

cs.IR↗

arXiv2023

FARA: Future-aware Ranking Algorithm for Fairness Optimization

Ranking systems are the key components of modern Information Retrieval (IR) applications, such as search engines and recommender systems. Besides the ranking relevance to users, the exposure fairness to item providers has also been considered an important factor in ranking optimization. Many fair ranking algorithms have been proposed to jointly optimize both ranking relevance and fairness. However, we find that most existing fair ranking methods adopt greedy algorithms that only optimize rankings for the next immediate session or request. As shown in this paper, such a myopic paradigm could limit the upper bound of ranking optimization and lead to suboptimal performance in the long term. To this end, we propose \textbf{FARA}, a novel \textbf{F}uture-\textbf{A}ware \textbf{R}anking \textbf{A}lgorithm for ranking relevance and fairness optimization. Instead of greedily optimizing rankings for the next immediate session, FARA plans ahead by jointly optimizing multiple ranklists together and saving them for future sessions. Specifically, FARA first uses the Taylor expansion to investigate how future ranklists will influence the overall fairness of the system. Then, based on the analysis of the Taylor expansion, FARA adopts a two-phase optimization algorithm where we first solve an optimal future exposure planning problem and then construct the optimal ranklists according to the optimal future exposure planning. Theoretically, we show that FARA is optimal for ranking relevance and fairness joint optimization. Empirically, our extensive experiments on three semi-synthesized datasets show that FARA is efficient, effective, and can deliver significantly better ranking performance compared to state-of-the-art fair ranking methods. We make our implementation public at \href{https://github.com/Taosheng-ty/QP_fairness/}{https://github.com/Taosheng-ty/QP\_fairness/}.

cs.IR↗

arXiv2023

A Lightweight Constrained Generation Alternative for Query-focused Summarization

Query-focused summarization (QFS) aims to provide a summary of a document that satisfies information need of a given query and is useful in various IR applications, such as abstractive snippet generation. Current QFS approaches typically involve injecting additional information, e.g. query-answer relevance or fine-grained token-level interaction between a query and document, into a finetuned large language model. However, these approaches often require extra parameters \& training, and generalize poorly to new dataset distributions. To mitigate this, we propose leveraging a recently developed constrained generation model Neurological Decoding (NLD) as an alternative to current QFS regimes which rely on additional sub-architectures and training. We first construct lexical constraints by identifying important tokens from the document using a lightweight gradient attribution model, then subsequently force the generated summary to satisfy these constraints by directly manipulating the final vocabulary likelihood. This lightweight approach requires no additional parameters or finetuning as it utilizes both an off-the-shelf neural retrieval model to construct the constraints and a standard generative language model to produce the QFS. We demonstrate the efficacy of this approach on two public QFS collections achieving near parity with the state-of-the-art model with substantially reduced complexity.

cs.IR↗

arXiv2023

Reward-free Policy Imitation Learning for Conversational Search

Existing conversational search studies mainly focused on asking better clarifying questions and/or improving search result quality. These works aim at retrieving better responses according to the search context, and their performances are evaluated on either single-turn tasks or multi-turn tasks under naive conversation policy settings. This leaves some questions about their applicability in real-world multi-turn conversations where realistically, each and every action needs to be made by the system itself, and search session efficiency is often an important concern of conversational search systems. While some recent works have identified the need for improving search efficiency in conversational search, they mostly require extensive data annotations and use hand-crafted rewards or heuristics to train systems that can achieve reasonable performance in a restricted number of turns, which has limited generalizability in practice. In this paper, we propose a reward-free conversation policy imitation learning framework, which can train a conversation policy without annotated conversation data or manually designed rewards. The trained conversation policy can be used to guide the conversational retrieval models to balance conversational search quality and efficiency. To evaluate the proposed conversational search system, we propose a new multi-turn-multi-response conversational evaluation metric named Expected Conversational Reciprocal Rank (ECRR). ECRR is designed to evaluate entire multi-turn conversational search sessions towards comprehensively evaluating both search result quality and search efficiency.

cs.IR↗

arXiv2022

Marginal-Certainty-aware Fair Ranking Algorithm

Ranking systems are ubiquitous in modern Internet services, including online marketplaces, social media, and search engines. Traditionally, ranking systems only focus on how to get better relevance estimation. When relevance estimation is available, they usually adopt a user-centric optimization strategy where ranked lists are generated by sorting items according to their estimated relevance. However, such user-centric optimization ignores the fact that item providers also draw utility from ranking systems. It has been shown in existing research that such user-centric optimization will cause much unfairness to item providers, followed by unfair opportunities and unfair economic gains for item providers. To address ranking fairness, many fair ranking methods have been proposed. However, as we show in this paper, these methods could be suboptimal as they directly rely on the relevance estimation without being aware of the uncertainty (i.e., the variance of the estimated relevance). To address this uncertainty, we propose a novel Marginal-Certainty-aware Fair algorithm named MCFair. MCFair jointly optimizes fairness and user utility, while relevance estimation is constantly updated in an online manner. In MCFair, we first develop a ranking objective that includes uncertainty, fairness, and user utility. Then we directly use the gradient of the ranking objective as the ranking score. We theoretically prove that MCFair based on gradients is optimal for the aforementioned ranking objective. Empirically, we find that on semi-synthesized datasets, MCFair is effective and practical and can deliver superior performance compared to state-of-the-art fair ranking methods. To facilitate reproducibility, we release our code https://github.com/Taosheng-ty/WSDM22-MCFair.

cs.IR↗

arXiv2022

1st Place Solution of The Robust Vision Challenge 2022 Semantic Segmentation Track

This report describes the winning solution to the Robust Vision Challenge (RVC) semantic segmentation track at ECCV 2022. Our method adopts the FAN-B-Hybrid model as the encoder and uses SegFormer as the segmentation framework. The model is trained on a composite dataset consisting of images from 9 datasets (ADE20K, Cityscapes, Mapillary Vistas, ScanNet, VIPER, WildDash 2, IDD, BDD, and COCO) with a simple dataset balancing strategy. All the original labels are projected to a 256-class unified label space, and the model is trained using a cross-entropy loss. Without significant hyperparameter tuning or any specific loss weighting, our solution ranks the first place on all the testing semantic segmentation benchmarks from multiple domains (ADE20K, Cityscapes, Mapillary Vistas, ScanNet, VIPER, and WildDash 2). The proposed method can serve as a strong baseline for the multi-domain segmentation task and benefit future works. Code will be available at https://github.com/lambert-x/RVC_Segmentation.

cs.CV↗

arXiv2022

Learning to Rank Rationales for Explainable Recommendation

State-of-the-art recommender system (RS) mostly rely on complex deep neural network (DNN) model structure, which makes it difficult to provide explanations along with RS decisions. Previous researchers have proved that providing explanations along with recommended items can help users make informed decisions and improve their trust towards the uninterpretable blackbox system. In model-agnostic explainable recommendation, system designers deploy a separate explanation model to take as input from the decision model, and generate explanations to meet the goal of persuasiveness. In this work, we explore the task of ranking textual rationales (supporting evidences) for model-agnostic explainable recommendation. Most of existing rationales ranking algorithms only utilize the rationale IDs and interaction matrices to build latent factor representations; and the semantic information within the textual rationales are not learned effectively. We argue that such design is suboptimal as the important semantic information within the textual rationales may be used to better profile user preferences and item features. Seeing this gap, we propose a model named Semantic-Enhanced Bayesian Personalized Explanation Ranking (SE-BPER) to effectively combine the interaction information and semantic information. SE-BPER first initializes the latent factor representations with contextualized embeddings generated by transformer model, then optimizes them with the interaction data. Extensive experiments show that such methodology improves the rationales ranking performance while simplifying the model training process (fewer hyperparameters and faster convergence). We conclude that the optimal way to combine semantic and interaction information remains an open question in the task of rationales ranking.

cs.IR↗