SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning
Open accessAbstract
Retrieval-Augmented Generation (RAG) is widely employed to mitigate risks such as hallucinations and knowledge obsolescence in medical question answering, yet its predominantly single-round, static retrieval paradigm misaligns with the multi-stage process of clinical reasoning.This compressed workflow induces two structural deficiencies: question-to-query translation often lacks clinically grounded semantic interpretation, and retrieval lacks iterative sufficiency feedback, making it difficult to form reliable evidence chains.We argue that both issues stem from a deeper cause-overloading a single reasoning chain with heterogeneous tasks of interpretation, exploration, and adjudication-and that the remedy is to reconstruct the workflow via task decoupling and dynamic multi-round exploration.To this end, we propose SEMA-RAG, a Self-Evolving Multi-Agent RAG framework for medical question answering, which assigns these roles to three specialist agents: the Interpreter Agent for clinical schema interpretation, the Explorer Agent for sufficiency-driven self-evolving retrieval, and the Arbiter Agent for evidence adjudication and answer selection.Across five benchmarks and five LLM backbones, SEMA-RAG improves the strongest baseline by +6.46 accuracy points on average, measured per backbone.
AI snapshot
Written from the paper’s full text.
TL;DR
SEMA-RAG is a self-evolving multi-agent RAG framework for medical question answering that decouples clinical reasoning into Interpreter, Explorer, and Arbiter agents. Across five medical QA benchmarks and five LLM backbones, it improves the strongest baseline by an average of +6.46 accuracy points per backbone.
Key findings
- SEMA-RAG achieves the best average accuracy within every model block across five benchmarks and five LLMs, with an average improvement of +6.46 accuracy points over the strongest baseline.
- Role-wise removal on MedQA-US / PubMedQA* with deepseek-v3.1: full SEMA-RAG scores 89.95 / 59.20; removing I-Agent gives 85.47 / 54.20, removing E-Agent gives 83.58 / 50.80, and removing A-Agent gives 86.49 / 53.60. Removing E-Agent causes the largest drop.
- Query breadth with Tmax = 2 on MedQA-US: m = 1 gives 86.72, m = 2 gives 89.00, and m = 3 gives 89.95. Most benefit is captured within two rounds, with performance peaking around Tmax ∈ {2, 3}; the two-round setting already outperforms the reproduced i-MedRAG baseline using three fixed retrieval rounds.
- Efficiency on MedQA-US with deepseek-v3.1: SEMA-RAG uses 4.8 LLM calls, 3.4 retrievals, 9.5 s, 89.95% accuracy, and 19,488.4 tokens/question, compared with i-MedRAG at 3.0 calls, 9.0 retrievals, 8.8 s, 74.78%, and 21,516.6 tokens/question.
Methods
- SEMA-RAG assigns three roles to agents sharing the same underlying LLM but differentiated by prompts: I-Agent maps the question into a clinical schema tuple (clinical intent, medical entities, clinical constraints, initial query) and linearizes it; E-Agent performs sufficiency-driven self-evolving multi-round dense retrieval, predicts a sufficiency flag, gap description, and next queries, and stops when evidence is sufficient, Tmax is reached, or stagnation occurs; A-Agent adjudicates the converged evidence into a traceable report and selects the final answer.
- Evaluation uses five MIRAGE benchmarks: MMLU-Med, MedQA-US, MedMCQA, PubMedQA*, and BioASQ-Y/N. Five LLM backbones are tested: deepseek-v3.1, kimi-k2, qwen3-coder-plus, gemini-2.0-flash, and glm-4.0-flash. Baselines are CoT, MedCPT, MedRAG, and i-MedRAG.
- Retrieval uses MedCPT as the dense retriever with FAISS over Textbooks and StatPearls. All methods are zero-shot; default SEMA-RAG settings are Tmax = 2, k = 16, m = 3, temperature 1.0 for I/E-Agent and 0.0 for A-Agent. Analyses include role-wise removal, depth and breadth variation, cost/efficiency, retriever robustness, smaller-model robustness, and open-ended/interactive benchmarks.
Limitations
- The paper states that evaluation remains limited to benchmark-based settings rather than realistic clinical workflows such as longitudinal EHR reasoning or record-grounded decision support.
- The paper states that the framework depends on retrieval corpus quality and coverage; if critical evidence is missing, outdated, or only partially retrieved, the self-evolving loop may still converge to incomplete grounding.
- The paper states that the sufficiency criterion is not explicitly designed for option-level separability or generative completeness, and that SEMA-RAG adds inference cost over single-round methods and lacks explicit relevance filtering during evidence accumulation, making performance sensitive to stopping criteria and exploration hyperparameters. The error analysis further shows a case where retrieval supports class-level elimination but not option-level discrimination, and A-Agent commits to an unsupported candidate.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
2 of 2 cited sources checked · 2 supported · 0 flagged
- Findingreasoning
These gains come with a moderate overhead relative to single-pass methods.
“Second, these gains come with a moderate overhead relative to singlepass methods, as expected from multi-agent, multiround inference.”
Quote in the paperSubstantial claim - Findingown data
The full SEMA-RAG consistently performs best across both MedQA-US and PubMedQA*.
“Table 2 shows that the full SEMA-RAG consistently performs best across both MedQA-US and PubMedQA*.”
Quote in the paperSubstantial claim - Findingown data
Ablating any single agent leads to a clear drop.
“Ablating any single agent leads to a clear drop”
Quote in the paperSubstantial claim - Findingreasoning
The gains are larger on benchmarks where premature commitment under incomplete evidence is costly.
“A consistent pattern is that the gains are larger on benchmarks where premature commitment under incomplete evidence is costly.”
Quote in the paperSubstantial claim - Findingown data
SEMA-RAG achieves a more favorable accuracy-efficiency trade-off compared with i-MedRAG.
“Third, compared with the iterative baseline i-MedRAG, SEMA-RAG achieves a more favorable accuracy-efficiency trade-off”
Quote in the paperSubstantial claim
- Findingown data
SEMA-RAG improves the strongest baseline by +6.46 accuracy points on average across five benchmarks and five LLM backbones.
“Across five benchmarks and five LLM backbones, SEMA-RAG improves the strongest baseline by +6.46 accuracy points on average, measured per backbone.”
- Also stated in: § 1 Introduction · plain
“Across five benchmarks and five LLM backbones, it improves the strongest baseline by an average of +6.46 accuracy points, validating convergent evidence-chain construction via task decoupling, role specialization, and evidence sufficiency-driven self-evolving retrieval.”
- Also stated in: § 6 Conclusion · plain
“Across five medical QA benchmarks and five LLM backbones, SEMA-RAG consistently outperforms strong baselines, improving the strongest baseline by an average of +6.46 accuracy points, while ablations verify the necessity of the interpret-explore-adjudicate loop for reliable evidence-chain construction.”
Quote in the paperSubstantial claim - Also stated in: § 1 Introduction · plain
- Findingown data
Removing E-Agent causes the largest performance drop.
“Table 2 shows that removing E-Agent causes the largest performance drop”
Quote in the paperSubstantial claim - Findingreasoning
SEMA-RAG introduces additional inference cost due to role specialization and multi-round exploration.
“Finally, SEMA-RAG introduces additional inference cost due to role specialization and multi-round exploration.”
Quote in the paperNot a substantial claim - Findingown data
SEMA-RAG consistently improves decision quality over single-round baselines.
“First, SEMA-RAG consistently improves decision quality over single-round baselines.”
Quote in the paperSubstantial claim - Findingown data
The two-round setting outperforms the reproduced i-MedRAG baseline (which uses three fixed retrieval rounds).
“Notably, the two-round setting already outperforms our reproduced i-MedRAG baseline (which uses three fixed retrieval rounds)”
Quote in the paperSubstantial claim - Findingown data
Additional experiments further support SEMA-RAG's robustness across retrievers, smaller models, and more open-ended interactive settings.
“Additional experiments further support its robustness across retrievers, smaller models, and more open-ended interactive settings.”
Quote in the paperNot a substantial claim - Findingown data
SEMA-RAG consistently outperforms strong baselines across five medical QA benchmarks and five LLM backbones.
“Across five medical QA benchmarks and five LLM backbones, SEMA-RAG consistently outperforms strong baselines”
Quote in the paperSubstantial claim - Findingown data
SEMA-RAG maintains clear gains over MedRAG on all four metrics (Accuracy, Robustness, Relevance, Empathy) on MAQuE using deepseek-v3.1.
“Table 11 shows that SEMA-RAG maintains clear gains over MedRAG on all four metrics.”
Quote in the paperSubstantial claim - Findingreasoning
MedCPT remains the stronger default choice in this clinical setting.
“Meanwhile, MedCPT remains the stronger default choice in this clinical setting, suggesting that domain-specific retrieval is still advantageous for medical QA.”
Quote in the paperNot a substantial claim - Interpretationown data
The core gain comes from a self-evolving, sufficiency-driven closed-loop retrieval rather than static retrieval.
“highlighting that the core gain comes from a self-evolving, sufficiency-driven closed-loop retrieval rather than static retrieval.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Interpretationown data
The improvement is backbone-agnostic rather than tied to a particular LLM.
“indicating that the improvement is backbone-agnostic rather than tied to a particular LLM.”
Quote in the paperSubstantial claim- Rests on an assumption: The five LLM backbones are representative of LLMs for medical QA, so the backbone-agnostic conclusion generalizes.
- Interpretationown data
The gain comes from self-evolving, sufficiency-driven closed-loop exploration rather than simply increasing the number of iterations.
“suggesting that the gain comes from self-evolving, sufficiency-driven closed-loop exploration rather than simply increasing the number of iterations.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Interpretationown data
Sufficiency-driven early stopping allocates additional computation more effectively than fixed-step iteration.
“indicating that sufficiency-driven early stopping allocates additional computation more effectively than fixedstep iteration.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Interpretationown data
Question-to-query translation benefits from clinically grounded semantic interpretation.
“indicating that question-to-query translation benefits from clinically grounded semantic interpretation.”
Quote in the paperSubstantial claim- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
- Interpretationown data
The gains do not come from one isolated module but from role-specialized collaboration that mirrors the staged clinical workflow.
“suggesting that the gains do not come from one isolated module but from role-specialized collaboration that mirrors the staged clinical workflow”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
- Rests on an assumption: The A-Agent's evidence adjudication reliably identifies relevant evidence and resolves conflicts.
- Interpretationreasoning
SEMA-RAG improves decision quality in a practically affordable regime.
“Taken together, these results suggest that SEMA-RAG improves decision quality in a practically affordable regime”
Quote in the paperSubstantial claim - Interpretationown data
Sufficiency-driven exploration remains beneficial in interactive clinical scenarios, where the system must sustain grounded response generation over multiple turns.
“This improvement suggests that sufficiency-driven exploration remains beneficial in interactive clinical scenarios, where the system must sustain grounded response generation over multiple turns.”
Quote in the paperSubstantial claim - Interpretationreasoning
The failure in Q0060 arises when moving from class identification to option selection.
“The failure arises when moving from class identification to option selection.”
Quote in the paperSubstantial claim - Interpretationown data
The SEMA-RAG framework generalizes beyond discrete medical QA to more open-ended and interactive forms of clinical response generation.
“Overall, these results suggest that the SEMA-RAG framework generalizes beyond discrete medical QA to more open-ended and interactive forms of clinical response generation.”
Quote in the paperSubstantial claim - Interpretationreasoning
The loop may stop once it reaches a correct coarse conclusion, even though an additional round is still needed to uniquely determine the answer option.
“The loop may stop once it reaches a correct coarse conclusion, even though an additional round is still needed to uniquely determine the answer option.”
Quote in the paperSubstantial claim- Rests on an assumption: The MedQA-US error Q0060 is representative of SEMA-RAG's failure modes.
- Interpretationreasoning
If critical evidence is missing, outdated, or only partially retrieved, the self-evolving loop may still converge to incomplete grounding.
“If critical evidence is missing, outdated, or only partially retrieved, the self-evolving loop may still converge to incomplete grounding.”
Quote in the paperNot a substantial claim- Rests on an assumption: The retrieval corpus contains the necessary evidence to answer the benchmark questions.
- Interpretationown data
Domain-specific retrieval is still advantageous for medical QA.
“Meanwhile, MedCPT remains the stronger default choice in this clinical setting, suggesting that domain-specific retrieval is still advantageous for medical QA.”
Quote in the paperNot a substantial claim - Interpretationreasoning
The current sufficiency and adjudication design has a limitation exposed by the Q0060 error.
“This exposes a limitation of the current sufficiency and adjudication design.”
Quote in the paperSubstantial claim- Rests on an assumption: The MedQA-US error Q0060 is representative of SEMA-RAG's failure modes.
- Interpretationreasoning
The consistent and substantial advantage directly stems from SEMA-RAG's successful simulation of expert clinical reasoning by decoupling interpretation, exploration, and adjudication, thereby alleviating the cognitive overload bottleneck diagnosed in the Introduction.
“Overall, this consistent and substantial advantage is not accidental.It directly stems from SEMA-RAG's successful simulation of expert clinical reasoning by decoupling interpretation, exploration, and adjudication, thereby alleviating the cognitive overload bottleneck diagnosed in the Introduction.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
- Rests on an assumption: The A-Agent's evidence adjudication reliably identifies relevant evidence and resolves conflicts.
- Interpretationown data
Task decoupling remains beneficial even when the base model has weaker instruction-following ability.
“This result suggests that task decoupling remains beneficial even when the base model has weaker instruction-following ability.”
Quote in the paperSubstantial claim- Rests on an assumption: gemma3:4b has weaker instruction-following ability than the larger backbones.
- Study design and conductnone given
The current study is limited to benchmark-based settings rather than realistic clinical workflows such as longitudinal EHR reasoning or record-grounded decision support.
“Although we extend the evaluation beyond discretechoice medical QA with additional open-ended and multi-turn benchmarks, the current study is still limited to benchmark-based settings rather than realistic clinical workflows such as longitudinal EHR reasoning or record-grounded decision support.”
Quote in the paperNot a substantial claim - Interpretationown data
Ablations verify the necessity of the interpret-explore-adjudicate loop for reliable evidence-chain construction.
“while ablations verify the necessity of the interpret-explore-adjudicate loop for reliable evidence-chain construction.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
- Rests on an assumption: The A-Agent's evidence adjudication reliably identifies relevant evidence and resolves conflicts.
- Interpretationreasoning
When the evidence report lacks an explicit bridge from the clinical setting to the discriminative phenotype expected in blood culture, A-Agent can be pulled toward salient surface descriptors and commit to an unsupported candidate.
“When the evidence report lacks an explicit bridge from the clinical setting to the discriminative phenotype expected in blood culture, A-Agent can be pulled toward salient surface descriptors and commit to an unsupported candidate.”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 5
- May combine two claims. 14
- The quote only partly states this claim. 3
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationCompared with single-turn QA, MAQuE places greater emphasis on maintaining robust, relevant, and contextually appropriate responses across iterative interactions.
“Compared with single-turn QA, it places greater emphasis on maintaining robust, rele-vant, and contextually appropriate responses across iterative interactions.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Gong et al., 2025 source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationMAQuE is a multi-turn medical dialogue benchmark that evaluates response quality in simulated clinical communication settings.
“MAQuE is a multi-turn medical dialogue benchmark that evaluates response quality in simulated clinical communication settings”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Gong et al., 2025 source
May combine two claims.
FindingreasoningSEMA-RAG makes the temporal constraint retrieval-actionable and carries it through to option selection, yielding the correct choice Staphylococcus aureus (Option D).
“In contrast, SEMA-RAG makes the temporal constraint retrieval-actionable and carries it through to option selection.The I-Agent surfaces hospital day 7 as a key constraint, the E-Agent detects the missing distinction between community-and hospital-acquired spectra and performs targeted follow-up retrieval, and the A-Agent consolidates the resulting evidence and maps it to the candidate set, yield”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataI-Agent makes hospital day 7 retrieval-actionable, anchoring evidence in the late-onset inpatient context and enabling option-level discrimination for S. aureus.
“As shown in Table 4 (Step 1), I-Agent makes hospital day 7 retrieval-actionable, anchoring evidence in the late-onset inpatient context and enabling option-level discrimination for S. aureus.”
Quote in the paperSubstantial claimThe quote only partly states this claim.
Findingown dataSEMA-RAG achieves the best average accuracy within every model block across five medical QA benchmarks and five LLMs.
“Overall, SEMA-RAG achieves the best average accuracy within every model block”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataMost of the benefit is captured within two rounds, with performance peaking around T_max ∈ {2,3}; beyond that, deeper exploration saturates and can introduce noise.
“Figure 3 suggests that most of the benefit is captured within two rounds, with performance peaking around T max ∈ {2, 3}; beyond that, deeper exploration saturates and can introduce noise.”
Quote in the paperSubstantial claimMay combine two claims.
Findingnone givenThis overhead is more efficient than fixed-step iterative baselines, but remains higher than single-round methods.
“Although this overhead is more efficient than fixed-step iterative baselines, it remains higher than single-round methods.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataIncreasing m improves accuracy, but the gains quickly taper.
“Table 3 shows a clear monotonic trend: increasing m improves accuracy, but the gains quickly taper.”
Quote in the paperSubstantial claimThe quote only partly states this claim.
Findingown dataRemoving the I-Agent consistently degrades performance on MedQA-US and PubMedQA*.
“Table 2 shows that removing the I-Agent consistently degrades performance”
Quote in the paperSubstantial claimMay combine two claims.
FindingreasoningThe retrieved evidence supports ruling out the Gram-positive options, but it does not provide option-separating signals within the remaining Gram-negative candidates.
“The retrieved evidence supports ruling out the Gram-positive options, but it does not provide option-separating signals within the remaining Gram-negative candidates.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataOn MedQA-US with gemma3:4b, although absolute performance of all methods drops, SEMA-RAG still maintains a clear advantage over the strongest single-round baseline.
“Table 9 shows that although the absolute performance of all methods drops on the smaller model, SEMA-RAG still maintains a clear advantage over the strongest single-round baseline.”
Quote in the paperSubstantial claimMay combine two claims.
FindingreasoningE-Agent retrieves evidence linking Lipid A to LPS and therefore to Gram-negative organisms, which is sufficient to eliminate the Gram-positive distractors in Q0060.
“Guided by these anchors, E-Agent retrieves evidence linking Lipid A to LPS and therefore to Gram-negative organisms, which is sufficient to eliminate the Gram-positive distractors.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataIn MedQA-US Q0060, SEMA-RAG retrieves the correct biochemical cue at the class level but still makes an incorrect discrete choice among the remaining candidates.
“This case is informative because the system retrieves the correct biochemical cue at the class level, yet still makes an incorrect discrete choice among the remaining candidates.”
Quote in the paperNot a substantial claimThe quote only partly states this claim.
Findingown dataSEMA-RAG consistently improves over the corresponding single-round retrieval baseline under both qwen3embedding-4b and MedCPT retrievers on MedQA-US.
“8 shows that SEMA-RAG consistently improves over the corresponding single-round retrieval baseline under both retrievers.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataExpanding the query set helps cover complementary evidence gaps in early exploration, while additional branches beyond a moderate breadth tend to introduce overlapping or low-yield retrievals.
“This suggests that expanding the query set helps cover complementary evidence gaps in early exploration, while additional branches beyond a moderate breadth tend to introduce overlapping or low-yield retrievals.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataTask decoupling with sufficiency-driven self-evolving retrieval helps form a more reliable evidence chain and prevents premature decisions under insufficient evidence.
“Overall, the case shows how task decoupling with sufficiency-driven self-evolving retrieval helps form a more reliable evidence chain and prevents premature decisions under insufficient evidence.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
- Rests on an assumption: The A-Agent's evidence adjudication reliably identifies relevant evidence and resolves conflicts.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone givenThe current design lacks explicit relevance filtering during evidence accumulation, making performance sensitive to stopping criteria and exploration hyperparameters.
“The current design also lacks explicit relevance filtering during evidence accumulation, making performance sensitive to stopping criteria and exploration hyperparameters.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Study design and conductnone givenThe current sufficiency criterion is not explicitly designed for option-level separability or generative completeness.
“In addition, the current sufficiency criterion is not explicitly designed for option-level separability or generative completeness.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningI-Agent keeps retrieval aligned with the intended clinical scenario and provides a cleaner substrate for subsequent exploration and adjudication.
“By extracting a structured clinical schema and making such key conditions retrievable, I-Agent keeps retrieval aligned with the intended clinical scenario and provides a cleaner substrate for subsequent exploration and adjudication.”
Quote in the paperSubstantial claim- Rests on an assumption: The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
May combine two claims.
InterpretationreasoningWithout E-Agent, the system loses explicit sufficiency feedback and is more prone to stop with partially covered evidence, leaving key constraints unresolved.
“Without E-Agent, the system loses explicit sufficiency feedback and is more prone to stop with partially covered evidence, leaving key constraints unresolved.”
Quote in the paperSubstantial claim- Rests on an assumption: The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
Classified as probably not a substantial claim.
Study design and conductnone givenThe framework depends on the quality and coverage of the retrieval corpus.
“Our framework also depends on the quality and coverage of the retrieval corpus.”
Quote in the paperSubstantial claim- Rests on an assumption: The retrieval corpus contains the necessary evidence to answer the benchmark questions.
The paper relies on this without support.
The E-Agent's sufficiency flag reliably indicates whether evidence is sufficient.
“where s t ∈ {0, 1} indicates evidence sufficiency: if s t = 1, the evidence is sufficient and we set Q t+1 = ∅; otherwise (s t = 0), evidence gaps remain”
The paper relies on this without support.
The I-Agent's clinical schema extraction accurately captures clinical intent and constraints.
“I-Agent does not merely rephrase the input medical question Q; instead, it semantically structures Q and projects it onto an explicit Clinical Schema.This process externalizes latent clinical intent and key constraints, providing stable anchors for subsequent retrieval and reasoning.”
The paper relies on this without support.
The A-Agent's evidence adjudication reliably identifies relevant evidence and resolves conflicts.
“A-Agent first adjudicates C * by removing irrelevant or duplicated content, identifying consistencies and conflicts, and organizing supporting and refuting clues into a structured evidence report R.”
The paper relies on this without support.
The retrieval corpus contains the necessary evidence to answer the benchmark questions.
“Our framework also depends on the quality and coverage of the retrieval corpus.”
The paper relies on this without support.
The five LLM backbones are representative of LLMs for medical QA, so the backbone-agnostic conclusion generalizes.
“These models originate from different providers, encompass diverse pretraining configurations and capability focuses.”
The paper relies on this without support.
The five benchmarks are representative of medical QA tasks, so the improvements generalize.
“Together, they cover general medical knowledge, clinical examinations, and biomedical literature inference.”
The paper relies on this without support.
The MedQA-US error Q0060 is representative of SEMA-RAG's failure modes.
“we analyze a representative MedQA-US error (Question 0060) in Table 7.”
The paper relies on this without support.
gemma3:4b has weaker instruction-following ability than the larger backbones.
“This result suggests that task decoupling remains beneficial even when the base model has weaker instruction-following ability.”
Cites
Chain-Of-Thought Prompting Elicits Reasoning in Large Language Models
Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q. V., & Zhou, D. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-1800 · Cited by 4,505 · see citing works
Kung, T. H., Cheatham, M., Medenilla, A., Sillos, C., De Leon, L., Elepaño, C., Madriaga, M., Aggabao, R., Diaz-Candido, G., Maningo, J., et al. (2023) · PLOS Digital Health · doi:10.1371/journal.pdig.0000198 · Cited by 4,115 · see citing works
Large language models encode clinical knowledge
Singhal, K., Azizi, S., Tu, T., Mahdavi, S. S., Wei, J., Chung, H. W., Scales, N., Tanwani, A., Cole-Lewis, H., Pfohl, S., et al. (2023) · Nature · doi:10.1038/s41586-023-06291-2 · Cited by 4,083 · see citing works
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-T., Rocktäschel, T., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.11401 · Cited by 3,101 · cited 1× · see citing works
“Retrieval-Augmented Generation (RAG), which incorporates external authoritative evidence to support the generation process, has been widely adopted to mitigate these risks (Lewis et al., 2020).”
Jin, D., Pan, E., Oufattole, N., Weng, W.-H., Fang, H., & Szolovits, P. (2020) · Applied Sciences · doi:10.48550/arxiv.2009.13081 · Cited by 2,320 · cited 10× · see citing works
“1 find the correct answer on MedQA-US (Question 0024) by making the key clinical constraint explicit and retrieving decision-critical evidence, while MedRAG’s single-round retrieval leads to a misleading rationale. such as MedRAG (Xiong et al., 2024) and Med-CPT (Jin et al., 2023) improve domain retrieval with medical…”
InfluentialChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All ToolsPreprint
Zeng, A.-H., Xu, B., Wang, B.-W., Zhang, C., Yin, D., Rojas, D., Feng, G., Zhao, H., Lai, H., Yu, H., et al. (2024) · arXiv.org · doi:10.48550/arxiv.2406.12793 · Cited by 1,884 · see citing works
Self-citationLost in the Middle: How Language Models Use Long Contexts
Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P. (2024) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00638 · Cited by 1,612 · cited 1× · see citing works
“…evidence exploration, and answer adjudication are tightly coupled, the cognitive load increases and the steps become interdependent, making it hard for the model to promptly adjust retrieval and reasoning when evidence is insufficient or conflicting (Wang et al., 2023; Liu et al., 2024).”
Gemma Team, Kamath, A., Ferret, J., Pathak, S., Vieillard, N., Merhej, R., Perrin, S., Matejovicova, T., Ramé, A., Rivière, M., et al. (2025) · arXiv · doi:10.48550/arxiv.2503.19786 · Cited by 1,156
BioGPT: generative pre-trained transformer for biomedical text generation and mining
Luo, R., Sun, L., Xia, Y., Qin, T., Zhang, S., Poon, H., & Liu, T.-Y. (2022) · Briefings in Bioinformatics · doi:10.1093/bib/bbac409 · Cited by 1,150 · cited 1× · see citing works
“Early medical LLMs systems largely depended on parametric knowledge or single-round retrieval (Luo et al., 2022; Singhal et al., 2023).”
Hupe, M. (2024) · Journal of Electronic Resources in Medical Libraries · doi:10.1080/15424065.2024.2389325 · Cited by 879
Retrieval-Augmented Generation for Large Language Models: A Survey
Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., Dai, Y., Sun, J., Wang, M., & Wang, H. (2023) · arXiv · doi:10.48550/arxiv.2312.10997 · Cited by 876 · see citing works
Large Language Model Based Multi-agents: A Survey of Progress and Challenges
Guo, T., Chen, X., Wang, Y., Chang, R., Pei, S., Chawla, N. V., Wiest, O., & Zhang, X. (2024) · Proceedings of the Thirty-ThirdInternational Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2024/890 · Cited by 851 · see citing works
Jin, D., Pan, E., Oufattole, N., Weng, W.-H., Fang, H., & Szolovits, P. (2021) · Applied Sciences · doi:10.3390/app11146421 · Cited by 718 · see citing works
Jin, D., Pan, E., Oufattole, N., Weng, W.-H., Fang, H., & Szolovits, P. (2021) · doi:10.20944/preprints202105.0498.v1 · Cited by 718 · cited 10× · see citing works
“1 find the correct answer on MedQA-US (Question 0024) by making the key clinical constraint explicit and retrieving decision-critical evidence, while MedRAG’s single-round retrieval leads to a misleading rationale. such as MedRAG (Xiong et al., 2024) and Med-CPT (Jin et al., 2023) improve domain retrieval with medical…”
InfluentialPubMedQA: A Dataset for Biomedical Research Question Answering
Jin, Q., Dhingra, B., Liu, Z., Cohen, W., & Lu, X. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1259 · Cited by 685 · cited 2× · see citing works
“…benchmark (Xiong et al., 2024): three medical examination datasets (MMLU-Med (Hendrycks et al., 2021), MedQA-US (Jin et al., 2020), MedMCQA (Pal et al., 2022)) and two biomedical research QA datasets (PubMedQA* (Jin et al., 2019) and BioASQ-Y/N (Tsatsaronis et al., 2015; Krithara et al., 2023)).”
ReAct: Synergizing Reasoning and Acting in Language Models
Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., & Cao, Y. (2022) · arXiv · doi:10.48550/arxiv.2210.03629 · Cited by 608 · cited 1× · see citing works
“ReAct (Yao et al., 2023) further couples reasoning with tool use, allowing agents to act and retrieve information during inference.”
Tsatsaronis, G., Balikas, G., Malakasiotis, P., Partalas, I., Zschunke, M., Alvers, M. R., Weissenborn, D., Krithara, A., Petridis, S., Polychronopoulos, D., et al. (2015) · BMC Bioinformatics · doi:10.1186/s12859-015-0564-6 · Cited by 598 · cited 3× · see citing works
“BioASQ-Y/N is a subset of the BioASQ Task B benchmark (Tsatsaronis et al., 2015; Krithara et al., 2023).”
MedAgents: Large Language Models as Collaborators for Zero-shot Medical ReasoningPreprint
Tang, X., Zou, A.-N., Zhang, Z., Li, Z.-M., Zhao, Y., Zhang, X., Cohan, A., & Gerstein, M. (2023) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2311.10537 · Cited by 571 · cited 1× · see citing works
“In healthcare, MedAgents (Tang et al., 2024) and Agent-Hospital (Li et al., 2025) similarly show that multi-role clinical collaboration improves diagnosis and decision quality.”
Active Retrieval Augmented Generation
Jiang, Z., Xu, F., Gao, L., Sun, Z., Liu, Q., Dwivedi-Yu, J., Yang, Y., Callan, J., & Neubig, G. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.495 · Cited by 497 · see citing works
Large Language Models in Medicine: The Potentials and Pitfalls
Omiye, J. A., Gui, H., Rezaei, S. J., Zou, J., & Daneshjou, R. (2024) · Annals of Internal Medicine · doi:10.7326/m23-2772 · Cited by 394 · see citing works
Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions
Trivedi, H., Balasubramanian, N., Khot, T., & Sabharwal, A. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.557 · Cited by 349 · see citing works
Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions
Trivedi, H., Balasubramanian, N., Khot, T., & Sabharwal, A. (2022) · arXiv · doi:10.48550/arxiv.2212.10509 · Cited by 349 · cited 1× · see citing works
“…et al., 2023) improve domain retrieval with medical dual encoders and RAG pipelines, yet still follow a retrieve-once-then-answer pattern, which often falls short on questions requiring multi-hop evidence integration and clinically constrained reasoning (Jiang et al., 2023; Trivedi et al., 2023).”
Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models
Wang, L., Xu, W., Lan, Y., Hu, Z., Lan, Y., Lee, R. K.-W., & Lim, E.-P. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.147 · Cited by 344 · see citing works
Mallen, A., Asai, A., Zhong, V., Das, R., Khashabi, D., & Hajishirzi, H. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.546 · Cited by 340 · cited 1× · see citing works
“…explicitly (Soldaini et al., 2017); and (ii) the retrieval process lacks mechanisms for sufficiency assessment and feed-back , hindering self-evolving iterative convergence under insufficient evidence and thus weakening reliable evidence-chain formation (Mallen et al., 2023; Shi et al., 2023).”
Camel: Communicative agents for" mind" exploration of large language model society
Li, G., Hammoud, H., Itani, H., Khizbullin, D., & Ghanem, B. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-2264 · Cited by 314 · cited 1× · see citing works
“Multi-agent systems enhance complex-task solving through role specialization and coordination (Guo et al., 2024; Zong et al., 2024; Tran et al., 2025) (e.g., CAMEL (Li et al., 2023), MetaGPT (Hong et al., 2024)).”
Benchmarking Retrieval-Augmented Generation for Medicine
Xiong, G., Jin, Q., Lu, Z., & Zhang, A. (2024) · arXiv · doi:10.48550/arxiv.2402.13178 · Cited by 312 · cited 5× · see citing works
“…and generalisation capabilities across diverse medical question-answering scenarios, we select five widely used datasets from the MIRAGE benchmark (Xiong et al., 2024): three medical examination datasets (MMLU-Med (Hendrycks et al., 2021), MedQA-US (Jin et al., 2020), MedMCQA (Pal et al., 2022))…”
InfluentialBenchmarking Retrieval-Augmented Generation for Medicine
Xiong, G., Jin, Q., Lu, Z., & Zhang, A. (2024) · Findings of the Association for Computational Linguistics ACL 2024 · doi:10.18653/v1/2024.findings-acl.372 · Cited by 310 · see citing works
DeepSeek-AI, Liu, A., Feng, B., Xue, B., Wang, B., Wu, B., Lu, C., Zhao, C., Deng, C., Zhang, C., et al. (2024) · arXiv · doi:10.48550/arxiv.2412.19437 · Cited by 294 · see citing works
Self-citationMeasuring Massive Multitask Language Understanding
Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., & Steinhardt, J. (2020) · arXiv · doi:10.48550/arxiv.2009.03300 · Cited by 291 · cited 2× · see citing works
“…scenarios, we select five widely used datasets from the MIRAGE benchmark (Xiong et al., 2024): three medical examination datasets (MMLU-Med (Hendrycks et al., 2021), MedQA-US (Jin et al., 2020), MedMCQA (Pal et al., 2022)) and two biomedical research QA datasets (PubMedQA* (Jin et al.,…”
Retrieval augmented generation for large language models in healthcare: A systematic review
Amugongo, L. M., Mascheroni, P., Brooks, S., Doering, S., & Seidel, J. (2025) · PLOS Digital Health · doi:10.1371/journal.pdig.0000877 · Cited by 267 · see citing works
Large Language Models for Information Retrieval: A Survey
Zhu, Y., Yuan, H., Wang, S., Liu, J., Liu, W., Deng, C., Chen, H., Liu, Z., Dou, Z., & Wen, J.-R. (2026) · ACM Transactions on Information Systems · doi:10.1145/3748304 · Cited by 225 · cited 1× · see citing works
“However, without explicit clinical intent and constraints modeling, iterations may devolve into shallow rewrites, yielding inefficient retrieval, drift, and uncontrolled expansion (Zhao et al., 2024; Zhu et al., 2026).”
MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework
Hong, S., Zhuge, M., Chen, J., Zheng, X., Cheng, Y., Zhang, C., Wang, J., Wang, Z., Yau, S. K. S., Lin, Z., et al. (2023) · arXiv · doi:10.48550/arxiv.2308.00352 · Cited by 214 · see citing works
The clinicians’ guide to large language models: a general perspective with a focus on hallucinations
Roustan, D., & Bastardot, F. (2025) · Interactive Journal of Medical Research · doi:10.2196/59823 · Cited by 191 · see citing works
PubMedQA: A Dataset for Biomedical Research Question AnsweringPreprint
Jin, Q., Dhingra, B., Liu, Z., Cohen, W. W., & Lu, X. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.06146 · Cited by 182 · cited 2× · see citing works
“…benchmark (Xiong et al., 2024): three medical examination datasets (MMLU-Med (Hendrycks et al., 2021), MedQA-US (Jin et al., 2020), MedMCQA (Pal et al., 2022)) and two biomedical research QA datasets (PubMedQA* (Jin et al., 2019) and BioASQ-Y/N (Tsatsaronis et al., 2015; Krithara et al., 2023)).”
Corrective Retrieval Augmented Generation
Yan, S.-Q., Gu, J.-C., Zhu, Y., & Ling, Z.-H. (2024) · arXiv · doi:10.48550/arxiv.2401.15884 · Cited by 182 · see citing works
Jin, Q., Kim, W., Chen, Q., Comeau, D. C., Yeganova, L., Wilbur, W. J., & Lu, Z. (2023) · Bioinformatics · doi:10.1093/bioinformatics/btad651 · Cited by 175 · cited 7× · see citing works
“1 find the correct answer on MedQA-US (Question 0024) by making the key clinical constraint explicit and retrieving decision-critical evidence, while MedRAG’s single-round retrieval leads to a misleading rationale. such as MedRAG (Xiong et al., 2024) and Med-CPT (Jin et al., 2023) improve domain retrieval with medical…”
InfluentialMedMCQA : A Large-scale Multi-Subject Multi-Choice Dataset for Medical domain Question Answering
Pal, A., Umapathi, L. K., & Sankarasubbu, M. (2022) · arXiv · doi:10.48550/arxiv.2203.14371 · Cited by 157 · see citing works
A Survey of LLM-based Agents in Medicine: How far are we from Baymax?Preprint
Wang, W.-X., Ma, Z.-Z., Wang, Z., et al. (2025) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2502.11211 · Cited by 154 · cited 1× · see citing works
“However, prior healthcare multi-agent work largely centers on deliberation under the assumption that key evidence is already in-context, leaving evidence acquisition unsystematic (Chen et al., 2025; Gorenshtein et al., 2025; Wang et al., 2025).”
Zhao, S.-Y., Yang, Y.-Q., Wang, Z.-L., et al. (2024) · arXiv.org · doi:10.48550/arxiv.2409.14924 · Cited by 151 · cited 2× · see citing works
“Recent work has shifted from one-shot retrieval to iterative RAG (Gao et al., 2024; Zhao et al., 2024).”
Large Language Models Can Be Easily Distracted by Irrelevant Context
Shi, F., Chen, X., Misra, K., Scales, N., Dohan, D., Chi, E., Schärli, N., & Zhou, D. (2023) · arXiv · doi:10.48550/arxiv.2302.00093 · Cited by 134 · cited 1× · see citing works
“…explicitly (Soldaini et al., 2017); and (ii) the retrieval process lacks mechanisms for sufficiency assessment and feed-back , hindering self-evolving iterative convergence under insufficient evidence and thus weakening reliable evidence-chain formation (Mallen et al., 2023; Shi et al., 2023).”
MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning
Tang, X., Zou, A., Zhang, Z., Li, Z., Zhao, Y., Zhang, X., Cohan, A., & Gerstein, M. (2024) · Findings of the Association for Computational Linguistics ACL 2024 · doi:10.18653/v1/2024.findings-acl.33 · Cited by 133 · see citing works
Yang, A., Li, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Gao, C., Huang, C., Lv, C., et al. (2025) · arXiv · doi:10.48550/arxiv.2505.09388 · Cited by 111 · see citing works
Asai, A., Wu, Z., Wang, Y., Sil, A., & Hajishirzi, H. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2310.11511 · Cited by 101 · cited 1× · see citing works
“In general domains, Self-RAG (Asai et al., 2023) and CRAG (Yan et al., 2024) use self-reflection to trigger re-retrieval, while in healthcare i -MedRAG (Xiong et al., 2025) iteratively refines queries via follow-up questions.”
Evaluating large language models and agents in healthcare: key challenges in clinical applications
Chen, X., Xiang, J., Lu, S., Liu, Y., He, M., & Shi, D. (2025) · Intelligent Medicine · doi:10.1016/j.imed.2025.03.002 · Cited by 87 · cited 1× · see citing works
“However, prior healthcare multi-agent work largely centers on deliberation under the assumption that key evidence is already in-context, leaving evidence acquisition unsystematic (Chen et al., 2025; Gorenshtein et al., 2025; Wang et al., 2025).”
A Survey of LLM-based Agents in Medicine: How far are we from Baymax?
Wang, W., Ma, Z., Wang, Z., Chenghan, W., Ji, J., Chen, W., Li, X., & Yuan, Y. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.539 · Cited by 78 · see citing works
BioASQ-QA: A manually curated corpus for Biomedical Question Answering
Krithara, A., Nentidis, A., Bougiatiotis, K., & Παλιούρας, Γ. (2023) · Scientific Data · doi:10.1038/s41597-023-02068-4 · Cited by 75 · cited 2× · see citing works
“BioASQ-Y/N is a subset of the BioASQ Task B benchmark (Tsatsaronis et al., 2015; Krithara et al., 2023).”
Improving Retrieval-Augmented Generation in Medicine with Iterative Follow-up Questions
Xiong, G., Jin, Q., Wang, X., Zhang, M., Lu, Z., & Zhang, A. (2024) · Pacific Symposium on Biocomputing (PSB) · doi:10.1142/9789819807024_0015 · Cited by 64 · cited 4× · see citing works
“The iterative retrieval setting utilizes i -MedRAG (Xiong et al., 2025), which generates subsequent queries to drive multi-round retrieval and accumulate evidence.”
InfluentialClinical reasoning - a guide to improving teaching and practice.
Linn, A. M. J., Khaw, C., Kildea, H., et al. (2012) · PubMed · Cited by 61 · see citing works
AI Agents in Clinical Medicine: A Systematic ReviewPreprint
Gorenshtein, A., Omar, M., Glicksberg, B. S., Nadkarni, G. N., & Klang, E. (2025) · medRxiv · doi:10.1101/2025.08.22.25334232 · Cited by 59 · cited 1× · see citing works
“However, prior healthcare multi-agent work largely centers on deliberation under the assumption that key evidence is already in-context, leaving evidence acquisition unsystematic (Chen et al., 2025; Gorenshtein et al., 2025; Wang et al., 2025).”
Omar, M., Nadkarni, G. N., Klang, E., & Glicksberg, B. S. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000662 · Cited by 49 · cited 1× · see citing works
“In recent years, large language models (LLMs) have shown specific capabilities in understanding and reasoning about medical knowledge when applied in healthcare (Kung et al., 2023; Omar et al., 2024).”
Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents
Li, J., Lai, Y., Li, W., Ren, J., Zhang, M., Kang, X., Wang, S., Li, P., Zhang, Y.-Q., Ma, W., et al. (2024) · arXiv · doi:10.48550/arxiv.2405.02957 · Cited by 46 · see citing works
Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation ModelsPreprint
Zhang, Y., Li, M., Long, D., Zhang, X., Lin, H., Yang, B., Xie, P., Yang, A., Liu, D., Lin, J., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.05176 · Cited by 31 · see citing works
HealthBench: Evaluating Large Language Models Towards Improved Human Health
Arora, R. K., Wei, J., Hicks, R. S., Bowman, P., Quiñonero-Candela, J., Tsimpourlas, F., Sharman, M., Shah, M., Vallone, A., Beutel, A., et al. (2025) · arXiv · doi:10.48550/arxiv.2505.08775 · Cited by 28 · see citing works
Multi-Agent Collaboration Mechanisms: A Survey of LLMs
Tran, K.-T., Dao, D., Nguyen, M.-D., Pham, Q.-V., O'Sullivan, B., & Nguyen, H. D. (2025) · arXiv · doi:10.48550/arxiv.2501.06322 · Cited by 26 · cited 1× · see citing works
“Multi-agent systems enhance complex-task solving through role specialization and coordination (Guo et al., 2024; Zong et al., 2024; Tran et al., 2025) (e.g., CAMEL (Li et al., 2023), MetaGPT (Hong et al., 2024)).”
Learning to reformulate long queries for clinical decision support
Soldaini, L., Yates, A., & Goharian, N. (2017) · Journal of the Association for Information Science and Technology · doi:10.1002/asi.23924 · Cited by 18 · cited 1× · see citing works
“…(i) the translation from question to query lacks clinical semantic interpretation , making implicit constraints difficult to articulate explicitly (Soldaini et al., 2017); and (ii) the retrieval process lacks mechanisms for sufficiency assessment and feed-back , hindering self-evolving iterative…”
Kimi K2: Open Agentic Intelligence
Kimi Team, Bai, Y., Bao, Y., Charles, Y., Chen, C., Chen, G., Chen, H., Chen, H., Chen, J., Chen, N., et al. (2025) · arXiv · doi:10.48550/arxiv.2507.20534 · Cited by 15 · see citing works
Self-citationGao, H.-A., Geng, J., Hua, W., Hu, M., Juan, X., Liu, H., Liu, S., Qiu, J., Qi, X., Wu, Y., et al. (2025) · arXiv · doi:10.48550/arxiv.2507.21046 · Cited by 3 · see citing works
The Dialogue That Heals: A Comprehensive Evaluation of Doctor Agents' Inquiry Capability
Gong, L., Wang, A., Lai, Y., Ma, W., & Liu, Y. (2025) · arXiv · doi:10.48550/arxiv.2509.24958 · Cited by 1 · see citing works
Zong, C. F., Yan, Y., Lü, W., et al. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2402.14320 · Cited by 1 · see citing works
Li, X. (2024) · arXiv · doi:10.48550/arxiv.2406.05804 · cited 2×
“In particular, gap identification, sufficiency-driven termination, and evidence adjudication/integration are often missing, weakening reliable external evidence grounding and closed evidence-chain formation in real clinical tasks (Li, 2025; Amugongo et al., 2025).”
Large Language Models for Information Retrieval: A Survey
Zhu, Y., Yuan, H., Wang, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2308.07107
BioGPT: Generative Pre-trained Transformer for Biomedical Text Generation and Mining
Luo, R., Sun, L., Xia, Y., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.10341
Flagged first, then by citations
Research works printed but not linked: 1
- Shahram Yazdani, Mohammad Hosseinzadeh, and Fakhrolsadat Hosseini. 2017. Models of clinical rea- soning with a focus on general practice: A critical review. Journal of Advances in Medical Education & Professionalism, 5(4):177-184.
- Authors
- Huang, Y., Chen, R., & Cheng, J.
- Affiliations
- Chinese University of Hong Kong (HK)
- Wuhan University of Technology (CN)
- ORCID
- Chen, Ruiying
- Venue
- Findings of the Association for Computational Linguistics: ACL 2026
- Published
- 2026
- Pages
- 18423-18442
- OpenAlex
- W7166882537
- Semantic Scholar
- 9009f25f9c214f5b584f2b2f62ba86aa4cc682ce
- Cited by
- 0
- References
- 52
- FWCI
- 0.00
- Type
- conference paper
- Open access
- gold
- doi.org
- Licence
- cc-by
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- retrieval-augmented generation, multi-agent systems, medical question answering, clinical reasoning, hallucination mitigation, task decoupling, large language models
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-07.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-07.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
