Yuelyu Ji

Yuelyu Ji, Zhuochun Li, Hui Ji et al.

We present \textbf{StepGap}, a hybrid NLI-LLM decision tree that detects step-level evidence gaps in multi-hop QA and emits one of three typed labels: \textsc{Contradicted Claim} (CC), \textsc{Irrelevant Evidence} (IE), or \textsc{Missing Bridge} (MB), each tied to a concrete repair action. On 82 multi-hop questions (181 annotated steps, $κ{=}0.704$), StepGap reaches sF1$=$72.0, within the bootstrap confidence interval of an LLM-only baseline (70.1) but with a more decomposable structure: every StepGap stage \emph{hurts} F1 when removed, while three of four LLM-only removals \emph{improve} F1 -- a sign of \emph{competing-error cancellation}, where internal stages mask each other's errors. We further expose a \emph{Q-F1 trap}: question-level F1 is mechanically inflated by checkers that flag every step, making step-level F1 the necessary diagnostic. Used as a typed GRPO process reward, StepGap improves Qwen2.5-7B-Instruct Exact Match from $32.1{\pm}0.3$ to $35.4{\pm}0.9$ across three seeds, with the single-run comparison showing a $+5.6$ Avg EM gain over the matched Search-R1 GRPO reproduction.

Chenyu Li, Zohaib Akhtar, Mingu Kwak et al.

As large language models (LLMs) increasingly generate and process clinical text, scalable evaluation has become critical. LLM-as-a-Judge (LaaJ), which uses LLMs to evaluate model outputs, offers a scalable alternative to costly expert review, but its healthcare adoption raises safety and bias concerns. We conducted a PRISMA-ScR scoping review of six databases (January 2020-January 2026), screening 11,727 studies and including 49. The landscape was dominated by evaluation and benchmarking applications (n=37, 75.5%), pointwise scoring (n=42, 85.7%), and GPT-family judges (n=36, 73.5%). Despite growing adoption, validation rigor was limited: among 36 studies with human involvement, the median number of expert validators was 3, while 13 (26.5%) used none. Risk of bias testing was absent in 36 studies (73.5%), only 1 (2.0%) examined demographic fairness, and none assessed temporal stability or patient context. Deployment remained limited, with 1 study (2.0%) reaching production and four (8.2%) prototype stage. Importantly, these gaps may interact: when judges and evaluated systems share training data or architectures, they may inherit similar blind spots, and agreement metrics may fail to distinguish true validity from shared errors. Minimal human oversight, limited bias assessment, and model monoculture together represent a governance gap where current validation may miss clinically significant errors. To address this, we propose MedJUDGE (Medical Judge Utility, De-biasing, Governance and Evaluation), a risk-stratified three-pillar framework organized around validity, safety, and accountability across clinical risk tiers, providing deployment-oriented evaluation guidance for healthcare LaaJ systems.

Xizhi Wu, Nelly Estefanie Garduno-Rapp, Justin F Rousseau et al.

Unlike most primary headaches, secondary headaches need specialized care and can have devastating consequences if not treated promptly. Clinical guidelines highlight several 'red flag' features, such as thunderclap onset, meningismus, papilledema, focal neurologic deficits, signs of temporal arteritis, systemic illness, and the 'worst headache of their life' presentation. Despite these guidelines, determining which patients require urgent evaluation remains challenging in primary care settings. Clinicians often work with limited time, incomplete information, and diverse symptom presentations, which can lead to under-recognition and inappropriate care. We present a large language model (LLM)-based multi-agent clinical decision support system built on an orchestrator-specialist architecture, designed to perform explicit and interpretable secondary headache diagnosis from free-text clinical vignettes. The multi-agent system decomposes diagnosis into seven domain-specialized agents, each producing a structured and evidence-grounded rationale, while a central orchestrator performs task decomposition and coordinates agent routing. We evaluated the multi-agent system using 90 expert-validated secondary headache cases and compared its performance with a single-LLM baseline across two prompting strategies: question-based prompting (QPrompt) and clinical practice guideline-based prompting (GPrompt). We tested five open-source LLMs (Qwen-30B, GPT-OSS-20B, Qwen-14B, Qwen-8B, and Llama-3.1-8B), and found that the orchestrated multi-agent system with GPrompt consistently achieved the highest F1 scores, with larger gains in smaller models. These findings demonstrate that structured multi-agent reasoning improves accuracy beyond prompt engineering alone and offers a transparent, clinically aligned approach for explainable decision support in secondary headache diagnosis.

Yuelyu Ji, Yuhe Gao, Runxue Bao et al.

The coronavirus disease 2019 (COVID-19) has led to a global pandemic of significant severity. In addition to its high level of contagiousness, COVID-19 can have a heterogeneous clinical course, ranging from asymptomatic carriers to severe and potentially life-threatening health complications. Many patients have to revisit the emergency room (ER) within a short time after discharge, which significantly increases the workload for medical staff. Early identification of such patients is crucial for helping physicians focus on treating life-threatening cases. In this study, we obtained Electronic Health Records (EHRs) of 3,210 encounters from 13 affiliated ERs within the University of Pittsburgh Medical Center between March 2020 and January 2021. We leveraged a Natural Language Processing technique, ScispaCy, to extract clinical concepts and used the 1001 most frequent concepts to develop 7-day revisit models for COVID-19 patients in ERs. The research data we collected from 13 ERs may have distributional differences that could affect the model development. To address this issue, we employed a classic deep transfer learning method called the Domain Adversarial Neural Network (DANN) and evaluated different modeling strategies, including the Multi-DANN algorithm, the Single-DANN algorithm, and three baseline methods. Results showed that the Multi-DANN models outperformed the Single-DANN models and baseline models in predicting revisits of COVID-19 patients to the ER within 7 days after discharge. Notably, the Multi-DANN strategy effectively addressed the heterogeneity among multiple source domains and improved the adaptation of source data to the target domain. Moreover, the high performance of Multi-DANN models indicates that EHRs are informative for developing a prediction model to identify COVID-19 patients who are very likely to revisit an ER within 7 days after discharge.

Yuhe Gao, Runxue Bao, Yuelyu Ji et al.

Knowledge sharing is crucial in healthcare, especially when leveraging data from multiple clinical sites to address data scarcity, reduce costs, and enable timely interventions. Transfer learning can facilitate cross-site knowledge transfer, but a major challenge is heterogeneity in clinical concepts across different sites. Large Language Models (LLMs) show significant potential of capturing the semantic meaning of clinical concepts and reducing heterogeneity. This study analyzed electronic health records from two large healthcare systems to assess the impact of semantic embeddings from LLMs on local, shared, and transfer learning models. Results indicate that domain-specific LLMs, such as Med-BERT, consistently outperform in local and direct transfer scenarios, while generic models like OpenAI embeddings require fine-tuning for optimal performance. However, excessive tuning of models with biomedical embeddings may reduce effectiveness, emphasizing the need for balance. This study highlights the importance of domain-specific embeddings and careful model tuning for effective knowledge transfer in healthcare.

Yuelyu Ji, Min Gu Kwak, Hang Zhang et al.

Medical RAG needs evidence-grounded claims, so plugging a claim-level NLI checker into retrieval-augmented RL is intuitive. \textbf{We find that the checker's \emph{output distribution} during training, not its held-out accuracy, decides whether it provides trainable gradient.} We compare four NLI checker back-ends as process rewards inside a GRPO-trained medical RAG agent (Qwen2.5-7B, replicated on Qwen3-4B and Llama-3.1-8B) across four held-out medical QA benchmarks. Three diagnostic findings emerge. \textbf{(i)} Signal collapse is log-prob-specific: LLM log-probability scoring labels over 97\% of claims neutral -- collapsing the RL gradient to zero -- while a calibrated MedNLI classifier scores the same pairs non-degenerately. \textbf{(ii)} Moderate signal beats strong signal on answer quality: a strong proprietary checker triggers a three-step reward-hacking cascade -- ultra-short answers, search avoidance, language collapse -- so a moderate-signal local classifier trains a higher-quality model (\textbf{+12\% BERTScore over zero-shot, no GPT dependency}). \textbf{(iii)} Signal strength is policy-dependent: the same checker registers as moderate on one policy but strong on another without triggering the cascade end-state. We frame these as boundary conditions for verifier-as-reward systems.

Zhuochun Li, Yong Zhang, Ming Li et al.

Large language models (LLMs) are widely used as reference-free evaluators via prompting, but this "LLM-as-a-Judge" paradigm is costly, opaque, and sensitive to prompt design. In this work, we investigate whether smaller models can serve as efficient evaluators by leveraging internal representations instead of surface generation. We uncover a consistent empirical pattern: small LMs, despite with weak generative ability, encode rich evaluative signals in their hidden states. This motivates us to propose the Semantic Capacity Asymmetry Hypothesis: evaluation requires significantly less semantic capacity than generation and can be grounded in intermediate representations, suggesting that evaluation does not necessarily need to rely on large-scale generative models but can instead leverage latent features from smaller ones. Our findings motivate a paradigm shift from LLM-as-a-Judge to Representation-as-a-Judge, a decoding-free evaluation strategy that probes internal model structure rather than relying on prompted output. We instantiate this paradigm through INSPECTOR, a probing-based framework that predicts aspect-level evaluation scores from small model representations. Experiments on reasoning benchmarks (GSM8K, MATH, GPQA) show that INSPECTOR substantially outperforms prompting-based small LMs and closely approximates full LLM judges, while offering a more efficient, reliable, and interpretable alternative for scalable evaluation.

Yuelyu Ji, Zhuochun Li, Rui Meng et al.

Multi-hop question answering (QA) requires systems to iteratively retrieve evidence and reason across multiple hops. While recent RAG and agentic methods report strong results, the underlying retrieval--reasoning \emph{process} is often left implicit, making procedural choices hard to compare across model families. This survey takes the execution procedure as the unit of analysis and introduces a four-axis framework covering (A) overall execution plan, (B) index structure, (C) next-step control (strategies and triggers), and (D) stop/continue criteria. Using this schema, we map representative multi-hop QA systems and synthesize reported ablations and tendencies on standard benchmarks (e.g., HotpotQA, 2WikiMultiHopQA, MuSiQue), highlighting recurring trade-offs among effectiveness, efficiency, and evidence faithfulness. We conclude with open challenges for retrieval--reasoning agents, including structure-aware planning, transferable control policies, and robust stopping under distribution shift.

Yuelyu Ji, Yuheng Song, Wei Wang et al.

Text generation in image-based platforms, particularly for music-related content, requires precise control over text styles and the incorporation of emotional expression. However, existing approaches often need help to control the proportion of external factors in generated text and rely on discrete inputs, lacking continuous control conditions for desired text generation. This study proposes Continuous Parameterization for Controlled Text Generation (CPCTG) to overcome these limitations. Our approach leverages a Language Model (LM) as a style learner, integrating Semantic Cohesion (SC) and Emotional Expression Proportion (EEP) considerations. By enhancing the reward method and manipulating the CPCTG level, our experiments on playlist description and music topic generation tasks demonstrate significant improvements in ROUGE scores, indicating enhanced relevance and coherence in the generated text.

Hang Zhang, Ruheng Wang, Yuelyu Ji et al.

Large language models have achieved strong performance on medical reasoning benchmarks, yet their deployment in clinical settings demands rigorous verification to ensure factual accuracy. While reward models offer a scalable approach for reasoning trace verification, existing methods face two limitations: they produce only scalar reward values without explicit justification, and they rely on single-pass retrieval that precludes adaptive knowledge access as verification unfolds. We introduce $\method$, an agentic framework that addresses these limitations by training medical reasoning verifiers to iteratively query external medical corpora during evaluation. Our approach combines tool-augmented verification with an iterative reinforcement learning paradigm that requires only trace-level supervision, alongside an adaptive curriculum mechanism that dynamically adjusts training data distribution. Across four medical reasoning benchmarks, $\method$ achieves substantial gains over existing methods, improving MedQA accuracy by 23.5% and MedXpertQA by 32.0% relative to the base generator in particular. Crucially, $\method$ demonstrates an $\mathbf{8\times}$ reduction in sampling budget requirement compared to prior reward model baselines. These findings establish that grounding verification in dynamically retrieved evidence offers a principled path toward more reliable medical reasoning systems.

Yuelyu Ji, Zhuochun Li, Rui Meng et al.

This paper introduces the RAG-RLRC-LaySum framework, designed to make complex biomedical research understandable to laymen through advanced Natural Language Processing (NLP) techniques. Our Retrieval Augmented Generation (RAG) solution, enhanced by a reranking method, utilizes multiple knowledge sources to ensure the precision and pertinence of lay summaries. Additionally, our Reinforcement Learning for Readability Control (RLRC) strategy improves readability, making scientific content comprehensible to non-specialists. Evaluations using the publicly accessible PLOS and eLife datasets show that our methods surpass Plain Gemini model, demonstrating a 20% increase in readability scores, a 15% improvement in ROUGE-2 relevance scores, and a 10% enhancement in factual accuracy. The RAG-RLRC-LaySum framework effectively democratizes scientific knowledge, enhancing public engagement with biomedical discoveries.

Yuelyu Ji, Hang Zhang, Shiven Verma et al.

We propose DeepRAG, a novel framework that integrates DeepSeek hierarchical question decomposition capabilities with RAG Gym unified retrieval-augmented generation optimization using process level supervision. Targeting the challenging MedHopQA biomedical question answering task, DeepRAG systematically decomposes complex queries into precise sub-queries and employs concept level reward signals informed by the UMLS ontology to enhance biomedical accuracy. Preliminary evaluations on the MedHopQA dataset indicate that DeepRAG significantly outperforms baseline models, including standalone DeepSeek and RAG Gym, achieving notable improvements in both Exact Match and concept level accuracy.

Yuelyu Ji, Rui Meng, Zhuochun Li et al.

Multi-hop question answering (QA) requires models to retrieve and reason over multiple pieces of evidence. While Retrieval-Augmented Generation (RAG) has made progress in this area, existing methods often suffer from two key limitations: (1) fixed or overly frequent retrieval steps, and (2) ineffective use of previously retrieved knowledge. We propose MIND (Memory-Informed and INteractive Dynamic RAG), a framework that addresses these challenges through: (i) prompt-based entity extraction to identify reasoning-relevant elements, (ii) dynamic retrieval triggering based on token-level entropy and attention signals, and (iii) memory-aware filtering, which stores high-confidence facts across reasoning steps to enable consistent multi-hop generation.

Yuelyu Ji, Zeshui Yu, Yanshan Wang

In this study, we aim to address the task of assertion detection when extracting medical concepts from clinical notes, a key process in clinical natural language processing (NLP). Assertion detection in clinical NLP usually involves identifying assertion types for medical concepts in the clinical text, namely certainty (whether the medical concept is positive, negated, possible, or hypothetical), temporality (whether the medical concept is for present or the past history), and experiencer (whether the medical concept is described for the patient or a family member). These assertion types are essential for healthcare professionals to quickly and clearly understand the context of medical conditions from unstructured clinical texts, directly influencing the quality and outcomes of patient care. Although widely used, traditional methods, particularly rule-based NLP systems and machine learning or deep learning models, demand intensive manual efforts to create patterns and tend to overlook less common assertion types, leading to an incomplete understanding of the context. To address this challenge, our research introduces a novel methodology that utilizes Large Language Models (LLMs) pre-trained on a vast array of medical data for assertion detection. We enhanced the current method with advanced reasoning techniques, including Tree of Thought (ToT), Chain of Thought (CoT), and Self-Consistency (SC), and refine it further with Low-Rank Adaptation (LoRA) fine-tuning. We first evaluated the model on the i2b2 2010 assertion dataset. Our method achieved a micro-averaged F-1 of 0.89, with 0.11 improvements over the previous works. To further assess the generalizability of our approach, we extended our evaluation to a local dataset that focused on sleep concept extraction. Our approach achieved an F-1 of 0.74, which is 0.31 higher than the previous method.

Yuelyu Ji

Multimodal Retrieval-Augmented Generation (Visual RAG) significantly advances question answering by integrating visual and textual evidence. Yet, current evaluations fail to systematically account for query difficulty and ambiguity. We propose MRAG-Suite, a diagnostic evaluation platform integrating diverse multimodal benchmarks (WebQA, Chart-RAG, Visual-RAG, MRAG-Bench). We introduce difficulty-based and ambiguity-aware filtering strategies, alongside MM-RAGChecker, a claim-level diagnostic tool. Our results demonstrate substantial accuracy reductions under difficult and ambiguous queries, highlighting prevalent hallucinations. MM-RAGChecker effectively diagnoses these issues, guiding future improvements in Visual RAG systems.