arXiv subjects

Jiaxin Mao

Publications and source records attributed to Jiaxin Mao.

At least 37 records · Page 2Linked to original sources

arXiv2026

Self-Compression of Chain-of-Thought via Multi-Agent Reinforcement Learning

The inference overhead induced by redundant reasoning undermines the interactive experience and severely bottlenecks the deployment of Large Reasoning Models. Existing reinforcement learning (RL)-based solutions tackle this problem by coupling a length penalty with outcome-based rewards. This simplistic reward weighting struggles to reconcile brevity with accuracy, as enforcing brevity may compromise critical reasoning logic. In this work, we address this limitation by proposing a multi-agent RL framework that selectively penalizes redundant chunks, while preserving essential reasoning logic. Our framework, Self-Compression via MARL (SCMA), instantiates redundancy detection and evaluation through two specialized agents: \textbf{a Segmentation Agent} for decomposing the reasoning process into logical chunks, and \textbf{a Scoring Agent} for quantifying the significance of each chunk. The Segmentation and Scoring agents collaboratively define an importance-weighted length penalty during training, incentivizing \textbf{a Reasoning Agent} to prioritize essential logic without introducing inference overhead during deployment. Empirical evaluations across model scales demonstrate that SCMA reduces response length by 11.1\% to 39.0\% while boosting accuracy by 4.33\% to 10.02\%. Furthermore, ablation studies and qualitative analysis validate that the synergistic optimization within the MARL framework fosters emergent behaviors, yielding more powerful LRMs compared to vanilla RL paradigms.

cs.AI↗

arXiv2026

MergeMix: Optimizing Mid-Training Data Mixtures via Learnable Model Merging

Optimizing data mixtures is essential for unlocking the full potential of large language models (LLMs), yet identifying the optimal composition remains computationally prohibitive due to reliance on heuristic trials or expensive proxy training. To address this, we introduce \textbf{MergeMix}, a novel approach that efficiently determines optimal data mixing ratios by repurposing model merging weights as a high-fidelity, low-cost performance proxy. By training domain-specific experts on minimal tokens and optimizing their merging weights against downstream benchmarks, MergeMix effectively optimizes the performance of data mixtures without incurring the cost of full-scale training. Extensive experiments on models with 8B and 16B parameters validate that MergeMix achieves performance comparable to or surpassing exhaustive manual tuning while drastically reducing search costs. Furthermore, MergeMix exhibits high rank consistency (Spearman $ρ> 0.9$) and strong cross-scale transferability, offering a scalable, automated solution for data mixture optimization.

cs.LG↗

arXiv2026

Beyond Monolithic Architectures: A Multi-Agent Search and Knowledge Optimization Framework for Agentic Search

Agentic search has emerged as a promising paradigm for complex information seeking by enabling Large Language Models (LLMs) to interleave reasoning with tool use. However, prevailing systems rely on monolithic agents that suffer from structural bottlenecks, including unconstrained reasoning outputs that inflate trajectories, sparse outcome-level rewards that complicate credit assignment, and stochastic search noise that destabilizes learning. To address these challenges, we propose \textbf{M-ASK} (Multi-Agent Search and Knowledge), a framework that explicitly decouples agentic search into two complementary roles: Search Behavior Agents, which plan and execute search actions, and Knowledge Management Agents, which aggregate, filter, and maintain a compact internal context. This decomposition allows each agent to focus on a well-defined subtask and reduces interference between search and context construction. Furthermore, to enable stable coordination, M-ASK employs turn-level rewards to provide granular supervision for both search decisions and knowledge updates. Experiments on multi-hop QA benchmarks demonstrate that M-ASK outperforms strong baselines, achieving not only superior answer accuracy but also significantly more stable training dynamics.\footnote{The source code for M-ASK is available at https://github.com/chenyiqun/M-ASK.}

cs.AI↗

arXiv2025

Entropy-Guided Token Dropout: Training Autoregressive Language Models with Limited Domain Data

As access to high-quality, domain-specific data grows increasingly scarce, multi-epoch training has become a practical strategy for adapting large language models (LLMs). However, autoregressive models often suffer from performance degradation under repeated data exposure, where overfitting leads to a marked decline in model capability. Through empirical analysis, we trace this degradation to an imbalance in learning dynamics: predictable, low-entropy tokens are learned quickly and come to dominate optimization, while the model's ability to generalize on high-entropy tokens deteriorates with continued training. To address this, we introduce EntroDrop, an entropy-guided token dropout method that functions as structured data regularization. EntroDrop selectively masks low-entropy tokens during training and employs a curriculum schedule to adjust regularization strength in alignment with training progress. Experiments across model scales from 0.6B to 8B parameters show that EntroDrop consistently outperforms standard regularization baselines and maintains robust performance throughout extended multi-epoch training. These findings underscore the importance of aligning regularization with token-level learning dynamics when training on limited data. Our approach offers a promising pathway toward more effective adaptation of LLMs in data-constrained domains.

cs.CL↗

arXiv2025

Deep Research: A Systematic Survey

Large language models (LLMs) have rapidly evolved from text generators into powerful problem solvers. Yet, many open tasks demand critical thinking, multi-source, and verifiable outputs, which are beyond single-shot prompting or standard retrieval-augmented generation. Recently, numerous studies have explored Deep Research (DR), which aims to combine the reasoning capabilities of LLMs with external tools, such as search engines, thereby empowering LLMs to act as research agents capable of completing complex, open-ended tasks. This survey presents a comprehensive and systematic overview of deep research systems, including a clear roadmap, foundational components, practical implementation techniques, important challenges, and future directions. Specifically, our main contributions are as follows: (i) we formalize a three-stage roadmap and distinguish deep research from related paradigms; (ii) we introduce four key components: query planning, information acquisition, memory management, and answer generation, each paired with fine-grained sub-taxonomies; (iii) we summarize optimization techniques, including prompting, supervised fine-tuning, and agentic reinforcement learning; and (iv) we consolidate evaluation criteria and open challenges, aiming to guide and facilitate future development. As the field of deep research continues to evolve rapidly, we are committed to continuously updating this survey to reflect the latest progress in this area.

cs.CL↗

arXiv2025

CardRewriter: Leveraging Knowledge Cards for Long-Tail Query Rewriting on Short-Video Platforms

Short-video platforms have rapidly become a new generation of information retrieval systems, where users formulate queries to access desired videos. However, user queries, especially long-tail ones, often suffer from spelling errors, incomplete phrasing, and ambiguous intent, resulting in mismatches between user expectations and retrieved results. While large language models (LLMs) have shown success in long-tail query rewriting within e-commerce, they struggle on short-video platforms, where proprietary content such as short videos, live streams, micro dramas, and user social networks falls outside their training distribution. To address this challenge, we introduce \textbf{CardRewriter}, an LLM-based framework that incorporates domain-specific knowledge to enhance long-tail query rewriting. For each query, our method aggregates multi-source knowledge relevant to the query and summarizes it into an informative and query-relevant knowledge card. This card then guides the LLM to better capture user intent and produce more effective query rewrites. We optimize CardRewriter using a two-stage training pipeline: supervised fine-tuning followed by group relative policy optimization, with a tailored reward system balancing query relevance and retrieval effectiveness. Offline experiments show that CardRewriter substantially improves rewriting quality for queries targeting proprietary content. Online A/B testing further confirms significant gains in long-view rate (LVR) and click-through rate (CTR), along with a notable reduction in initiative query reformulation rate (IQRR). Since September 2025, CardRewriter has been deployed on Kuaishou, one of China's largest short-video platforms, serving hundreds of millions of users daily.

cs.IR↗

arXiv2025

Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning

Retrieval-augmented generation (RAG) is widely utilized to incorporate external knowledge into large language models, thereby enhancing factuality and reducing hallucinations in question-answering (QA) tasks. A standard RAG pipeline consists of several components, such as query rewriting, document retrieval, document filtering, and answer generation. However, these components are typically optimized separately through supervised fine-tuning, which can lead to misalignments between the objectives of individual components and the overarching aim of generating accurate answers. Although recent efforts have explored using reinforcement learning (RL) to optimize specific RAG components, these approaches often focus on simple pipelines with only two components or do not adequately address the complex interdependencies and collaborative interactions among the modules. To overcome these limitations, we propose treating the complex RAG pipeline with multiple components as a multi-agent cooperative task, in which each component can be regarded as an RL agent. Specifically, we present MMOA-RAG, Multi-Module joint Optimization Algorithm for RAG, which employs multi-agent reinforcement learning to harmonize all agents' goals toward a unified reward, such as the F1 score of the final answer. Experiments conducted on various QA benchmarks demonstrate that MMOA-RAG effectively boost the overall performance of the pipeline and outperforms existing baselines. Furthermore, comprehensive ablation studies validate the contributions of individual components and demonstrate MMOA-RAG can be adapted to different RAG pipelines and benchmarks.

cs.CL↗

arXiv2025

FinS-Pilot: A Benchmark for Online Financial RAG System

Large language models (LLMs) have demonstrated remarkable capabilities across various professional domains, with their performance typically evaluated through standardized benchmarks. In the financial field, the stringent demands for professional accuracy and real-time data processing often necessitate the use of retrieval-augmented generation (RAG) techniques. However, the development of financial RAG benchmarks has been constrained by data confidentiality issues and the lack of dynamic data integration. To address this issue, we introduce FinS-Pilot, a novel benchmark for evaluating RAG systems in online financial applications. Constructed from real-world financial assistant interactions, our benchmark incorporates both real-time API data and text data, organized through an intent classification framework covering critical financial domains. The benchmark enables comprehensive evaluation of financial assistants' capabilities in handling both static knowledge and time-sensitive market information.Through systematic experiments with multiple Chinese leading LLMs, we demonstrate FinS-Pilot's effectiveness in identifying models suitable for financial applications while addressing the current gap in specialized evaluation tools for the financial domain. Our work contributes both a practical evaluation framework and a curated dataset to advance research in financial NLP systems. The code and dataset are accessible on GitHub.

cs.CL↗

arXiv2025

Investigating the Robustness of Counterfactual Learning to Rank Models: A Reproducibility Study

Counterfactual learning to rank (CLTR) has attracted extensive attention in the IR community for its ability to leverage massive logged user interaction data to train ranking models. While the CLTR models can be theoretically unbiased when the user behavior assumption is correct and the propensity estimation is accurate, their effectiveness is usually empirically evaluated via simulation-based experiments due to a lack of widely available, large-scale, real click logs. However, many previous simulation-based experiments are somewhat limited because they may have one or more of the following deficiencies: 1) using a weak production ranker to generate initial ranked lists, 2) relying on a simplified user simulation model to simulate user clicks, and 3) generating a fixed number of synthetic click logs. As a result, the robustness of CLTR models in complex and diverse situations is largely unknown and needs further investigation. To address this problem, in this paper, we aim to investigate the robustness of existing CLTR models in a reproducibility study with extensive simulation-based experiments that (1) use production rankers with different ranking performance, (2) leverage multiple user simulation models with different user behavior assumptions, and (3) generate different numbers of synthetic sessions for the training queries. We find that the IPS-DCM, DLA-PBM, and UPE models show better robustness under various simulation settings than other CLTR models. Moreover, existing CLTR models often fail to outperform naive click baselines when the production ranker is strong and the number of training sessions is limited, indicating a pressing need for new CLTR algorithms tailored to these conditions.

cs.LG↗

arXiv2025

Addressing Personalized Bias for Unbiased Learning to Rank

Unbiased learning to rank (ULTR), which aims to learn unbiased ranking models from biased user behavior logs, plays an important role in Web search. Previous research on ULTR has studied a variety of biases in users' clicks, such as position bias, presentation bias, and outlier bias. However, existing work often assumes that the behavior logs are collected from an ``average'' user, neglecting the differences between different users in their search and browsing behaviors. In this paper, we introduce personalized factors into the ULTR framework, which we term the user-aware ULTR problem. Through a formal causal analysis of this problem, we demonstrate that existing user-oblivious methods are biased when different users have different preferences over queries and personalized propensities of examining documents. To address such a personalized bias, we propose a novel user-aware inverse-propensity-score estimator for learning-to-rank objectives. Specifically, our approach models the distribution of user browsing behaviors for each query and aggregates user-weighted examination probabilities to determine propensities. We theoretically prove that the user-aware estimator is unbiased under some mild assumptions and shows lower variance compared to the straightforward way of calculating a user-dependent propensity for each impression. Finally, we empirically verify the effectiveness of our user-aware estimator by conducting extensive experiments on two semi-synthetic datasets and a real-world dataset.

cs.IR↗

arXiv2025

Leveraging Large Language Models for Relevance Judgments in Legal Case Retrieval

Determining which legal cases are relevant to a given query involves navigating lengthy texts and applying nuanced legal reasoning. Traditionally, this task has demanded significant time and domain expertise to identify key Legal Facts and reach sound juridical conclusions. In addition, existing data with legal case similarities often lack interpretability, making it difficult to understand the rationale behind relevance judgments. With the growing capabilities of large language models (LLMs), researchers have begun investigating their potential in this domain. Nonetheless, the method of employing a general large language model for reliable relevance judgments in legal case retrieval remains largely unexplored. To address this gap in research, we propose a novel few-shot approach where LLMs assist in generating expert-aligned interpretable relevance judgments. The proposed approach decomposes the judgment process into several stages, mimicking the workflow of human annotators and allowing for the flexible incorporation of expert reasoning to improve the accuracy of relevance judgments. Importantly, it also ensures interpretable data labeling, providing transparency and clarity in the relevance assessment process. Through a comparison of relevance judgments made by LLMs and human experts, we empirically demonstrate that the proposed approach can yield reliable and valid relevance assessments. Furthermore, we demonstrate that with minimal expert supervision, our approach enables a large language model to acquire case analysis expertise and subsequently transfers this ability to a smaller model via annotation-based knowledge distillation.

cs.AI↗

arXiv2025

WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training

Recent advances in learning rate (LR) scheduling have demonstrated the effectiveness of decay-free approaches that eliminate the traditional decay phase while maintaining competitive performance. Model merging techniques have emerged as particularly promising solutions in this domain. We present Warmup-Stable and Merge (WSM), a general framework that establishes a formal connection between learning rate decay and model merging. WSM provides a unified theoretical foundation for emulating various decay strategies-including cosine decay, linear decay and inverse square root decay-as principled model averaging schemes, while remaining fully compatible with diverse optimization methods. Through extensive experiments, we identify merge duration-the training window for checkpoint aggregation-as the most critical factor influencing model performance, surpassing the importance of both checkpoint interval and merge quantity. Our framework consistently outperforms the widely-adopted Warmup-Stable-Decay (WSD) approach across multiple benchmarks, achieving significant improvements of +3.5% on MATH, +2.9% on HumanEval, and +5.5% on MMLU-Pro. The performance advantages extend to supervised fine-tuning scenarios, highlighting WSM's potential for long-term model refinement.

cs.CL↗

arXiv2025

MAO-ARAG: Multi-Agent Orchestration for Adaptive Retrieval-Augmented Generation

In question-answering (QA) systems, Retrieval-Augmented Generation (RAG) has become pivotal in enhancing response accuracy and reducing hallucination issues. The architecture of RAG systems varies significantly, encompassing single-round RAG, iterative RAG, and reasoning RAG, each tailored to address different types of queries. Due to the varying complexity of real-world queries, a fixed RAG pipeline often struggles to balance performance and cost efficiency across different queries. To address this challenge, we propose an adaptive RAG framework called MAO-ARAG, which leverages multi-agent orchestration. Our adaptive RAG is conceived as a multi-turn framework. Specifically, we define multiple executor agents, representing typical RAG modules such as query reformulation agents, document selection agent, and generation agents. A planner agent intelligently selects and integrates the appropriate agents from these executors into a suitable workflow tailored for each query, striving for high-quality answers while maintaining reasonable costs. During each turn, the planner agent is trained using reinforcement learning, guided by an outcome-based reward (F1 score) and a cost-based penalty, continuously improving answer quality while keeping costs within a reasonable range. Experiments conducted on multiple QA datasets demonstrate that our approach, which dynamically plans workflows for each query, not only achieves high answer quality but also maintains both cost and latency within acceptable limits.The code of MAO-ARAG is on https://github.com/chenyiqun/Agentic-RAG.

cs.CL↗

arXiv2025

Leveraging LLMs to Evaluate Usefulness of Document

The conventional Cranfield paradigm struggles to effectively capture user satisfaction due to its weak correlation between relevance and satisfaction, alongside the high costs of relevance annotation in building test collections. To tackle these issues, our research explores the potential of leveraging large language models (LLMs) to generate multilevel usefulness labels for evaluation. We introduce a new user-centric evaluation framework that integrates users' search context and behavioral data into LLMs. This framework uses a cascading judgment structure designed for multilevel usefulness assessments, drawing inspiration from ordinal regression techniques. Our study demonstrates that when well-guided with context and behavioral information, LLMs can accurately evaluate usefulness, allowing our approach to surpass third-party labeling methods. Furthermore, we conduct ablation studies to investigate the influence of key components within the framework. We also apply the labels produced by our method to predict user satisfaction, with real-world experiments indicating that these labels substantially improve the performance of satisfaction prediction models.

cs.IR↗

arXiv2025

LLM4Ranking: An Easy-to-use Framework of Utilizing Large Language Models for Document Reranking

Utilizing large language models (LLMs) for document reranking has been a popular and promising research direction in recent years, many studies are dedicated to improving the performance and efficiency of using LLMs for reranking. Besides, it can also be applied in many real-world applications, such as search engines or retrieval-augmented generation. In response to the growing demand for research and application in practice, we introduce a unified framework, \textbf{LLM4Ranking}, which enables users to adopt different ranking methods using open-source or closed-source API-based LLMs. Our framework provides a simple and extensible interface for document reranking with LLMs, as well as easy-to-use evaluation and fine-tuning scripts for this task. We conducted experiments based on this framework and evaluated various models and methods on several widely used datasets, providing reproducibility results on utilizing LLMs for document reranking. Our code is publicly available at https://github.com/liuqi6777/llm4ranking.

cs.IR↗

arXiv2025

Exploring Human-Like Thinking in Search Simulations with Large Language Models

Simulating user search behavior is a critical task in information retrieval, which can be employed for user behavior modeling, data augmentation, and system evaluation. Recent advancements in large language models (LLMs) have opened up new possibilities for generating human-like actions including querying, browsing, and clicking. In this work, we explore the integration of human-like thinking into search simulations by leveraging LLMs to simulate users' hidden cognitive processes. Specifically, given a search task and context, we prompt LLMs to first think like a human before executing the corresponding action. As existing search datasets do not include users' thought processes, we conducted a user study to collect a new dataset enriched with users' explicit thinking. We investigate the impact of incorporating such human-like thinking on simulation performance and apply supervised fine-tuning (SFT) to teach LLMs to emulate both human thinking and actions. Our experiments span two dimensions in leveraging LLMs for user simulation: (1) with or without explicit thinking, and (2) with or without fine-tuning on the thinking-augmented dataset. The results demonstrate the feasibility and potential of incorporating human-like thinking in user simulations, though performance improvements on some metrics remain modest. We believe this exploration provides new avenues and inspirations for advancing user behavior modeling in search simulations.

cs.IR↗

arXiv2025

How do Large Language Models Understand Relevance? A Mechanistic Interpretability Perspective

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.

cs.IR↗

arXiv2025

PanguIR Technical Report for NTCIR-18 AEOLLM Task

As large language models (LLMs) gain widespread attention in both academia and industry, it becomes increasingly critical and challenging to effectively evaluate their capabilities. Existing evaluation methods can be broadly categorized into two types: manual evaluation and automatic evaluation. Manual evaluation, while comprehensive, is often costly and resource-intensive. Conversely, automatic evaluation offers greater scalability but is constrained by the limitations of its evaluation criteria (dominated by reference-based answers). To address these challenges, NTCIR-18 introduced the AEOLLM (Automatic Evaluation of LLMs) task, aiming to encourage reference-free evaluation methods that can overcome the limitations of existing approaches. In this paper, to enhance the evaluation performance of the AEOLLM task, we propose three key methods to improve the reference-free evaluation: 1) Multi-model Collaboration: Leveraging multiple LLMs to approximate human ratings across various subtasks; 2) Prompt Auto-optimization: Utilizing LLMs to iteratively refine the initial task prompts based on evaluation feedback from training samples; and 3) In-context Learning (ICL) Optimization: Based on the multi-task evaluation feedback, we train a specialized in-context example retrieval model, combined with a semantic relevance retrieval model, to jointly identify the most effective in-context learning examples. Experiments conducted on the final dataset demonstrate that our approach achieves superior performance on the AEOLLM task.

cs.CL↗