From Papers to Mechanisms: An Evidence-Grounded Knowledge Substrate for Scientific Language Models

Qiuhui Chen; Yibo Liu; Tao Dai; Jiafan Lu; Zhenglei Zhou; Weimin Zhong

cs.AIOct 5, 2026

From Papers to Mechanisms: An Evidence-Grounded Knowledge Substrate for Scientific Language Models

Authors: Qiuhui Chen, Yibo Liu, Tao Dai, Jiafan Lu, Zhenglei Zhou, Weimin Zhong

Organizations: School of Information Science and Engineering, East China University of Science and Technology · School of Computer Science, Shanghai Jiao Tong University · Tencent

Abstract

Scientific language models often access literature through untyped text chunks, which fragment the functional and evidential structure required for mechanism-rich questions. We introduce an evidence-grounded mechanism knowledge substrate that organizes scientific literature into provenance-linked evidence units, role-typed entities, and directed mechanism paths. We instantiate it as MS $^3$ , a Material-Sensor-Signal-System schema for conductive-fiber flexible sensors, over 13,689 papers, 131,083 evidence items, and 26,648 mechanism objects. On in-domain and coverage-shift question-answering benchmarks, we compare closed-book generation, Web search, Raw-PDF RAG, and MS $^3$ retrieval across ten language models. MS $^3$ improves macro-averaged scientific correctness. It also improves citation entailment and answer completeness. These results support mechanism substrates as a reliable representation layer for scientific language models and motivate a source-repair workflow in which insufficient MS $^3$ evidence triggers targeted retrieval from its linked papers rather than assuming that a user has already supplied the correct PDFs.

Figures & tables

Figure 1: Evidence-grounded construction of an MS 3 mechanism object.

Figure 2: Overview of the MS 3 knowledge substrate, including corpus scale, domain coverage, evidence composition, structural coverage, and multimodal provenance.

Figure 3: Expert-guided supervision, paper-disjoint benchmark construction, and decomposed extraction of evidence-grounded MS 3 mechanism objects.

Figure 4: Mechanism-centered MS 3 retrieval and grounding. A scientific query is answered either from parametric knowledge alone or through dual-index retrieval, provenance expansion, and evidence-aware context packing over the MS 3 substrate, enabling mechanism-grounded responses with claim-level citations and explicit support boundaries.

			In-domain					Coverage-shift
Generator	Params.	Inference	Direct	Web	Raw-PDF	Verified-MS 3	$\Delta_{V-R}$	Web	Verified-MS 3	Hybrid	$\Delta_{H-B}$
Qwen2.5-14B	14B	Local-4bit	37.5	58.3	80.8	80.8	0.0	51.9	52.1	54.2	+2.1
Qwen2.5-72B	72B	Local-4bit	34.2	60.0	81.7	86.7	+5.0	52.2	54.6	57.8	+3.2
Qwen3-30B-A3B	30B/3B active	Local-4bit	23.3	53.3	46.7	48.3	+1.7	48.9	51.1	53.8	+2.7
Mistral-Small-24B	24B	Local-4bit	25.0	54.2	65.0	70.0	+5.0	53.3	52.1	55.6	+2.3
DeepSeek-V4-Flash	Undisc.	API	25.8	51.7	80.0	89.2	+9.2	56.2	58.3	68.8	+10.4

Table 1: Scientific correctness (%).

V

R

, and

H

denote Verified-MS 3 , Raw-PDF RAG, and Hybrid retrieval, respectively. The Raw-PDF condition is a source-resolved oracle supplied with the benchmark-matched paper, rather than a user-facing retrieval baseline.

\Delta_{V-R}=V-R

measures the improvement of Verified-MS 3 over Raw-PDF RAG. For the coverage-shift evaluation,

B=\max(\mathrm{Web},V)

denotes the strongest single-source retrieval condition, and

\Delta_{H-B}=H-B

measures the additional benefit of combining Web and Verified-MS 3 . Bold indicates the best result within each evaluation setting.

	Scientific correctness				Citation entailment			Completeness
Generator	Direct	Web	Raw-PDF	Verified-MS 3	Web	Raw-PDF	Verified-MS 3	Direct	Web	Raw-PDF	Verified-MS 3
Qwen2.5-14B	37.5	58.3	80.8	80.8	58.3	87.5	85.0	20.8	42.5	69.2	73.3
Qwen2.5-72B	34.2	60.0	81.7	86.7	61.7	86.7	90.0	20.0	45.0	75.8	79.2
Qwen3-30B-A3B	23.3	53.3	46.7	48.3	49.2	50.0	49.2	24.2	48.3	45.8	49.2
Mistral-Small-24B	25.0	54.2	65.0	70.0	50.8	70.0	75.0	20.0	40.0	59.2	64.2
DeepSeek-V4-Flash	25.8	51.7	80.0	89.2	68.3	90.8	97.5	18.3	36.7	77.5	87.5

Table 2: Per-generator results on the in-domain evaluation (%). Citation entailment is not applicable to Closed-book Direct. Bold indicates the best method within each metric group for a given generator.

Figure 5: Ablation of the MS 3 substrate construction pipeline. Error bars denote 95% confidence intervals.

Explore similar work

Large Knowledge Model: A Knowledge Foundation for Agentic Science at Scale
Sep 23, 2026Yuan Huang, Sihan Hu, Hongyu Gu +18Scientific Discovery Knowledge Graphs
DP Technology, Beijing, China · Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing, China · International Center for Quantum Materials, School of Physics, Peking University, Beijing, China +4
TRACES: A Benchmark for Epistemic Reliability in Scientific Reasoning by LLMs
Aug 11, 2026Valentin Rodionov, Shamil AssylbekovLarge Language Model Reliability Scientific Discovery
Intellicat Cleveland, OH 44106
LitTraceQA: A Benchmark for Multi-Stage Grounding and Verification in Scientific Question Answering
Aug 7, 2026Xuye Liu, Yimu Wang, Peng Shi +7Literature Review
University of Waterloo · 2Amazon · 3City University of Hong Kong +1

Sep 23, 2026, cs.AIJ/K move · Enter open · S save

Large Knowledge Model: A Knowledge Foundation for Agentic Science at Scale

Yuan Huang, Sihan Hu, Hongyu Gu, Chao Ma+17

DP Technology, Beijing, China · Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing, China · International Center for Quantum Materials, School of Physics, Peking University, Beijing, China+4

Scientific Discovery Knowledge Graphs

Open paper AskPDF arXiv

Agentic science envisions many autonomous agents investigating concurrently while building on a shared, evolving body of scientific knowledge. This requires a knowledge foundation that supports high-concurrency access, preserves traceable and reusable reasoning, and grows incrementally. We propose the Large Knowledge Model (LKM), a growing, agent-native knowledge foundation that provides a general representation of scientific knowledge across disciplines. LKM organizes the scientific literature into reasoning graphs, with claims as the core nodes and associated reasoning chains that make explicit how premises and evidence support conclusions. These source-grounded objects are persistent and addressable; cross-paper links organize them into aligned question, workflow, and evidence views. Newly extracted papers extend the foundation incrementally while preserving existing object identities. Building on this foundation, we develop an agent-native, reasoning-aware scientific retrieval system that retrieves claims together with their reasoning chains and sources, enabling agents to inspect and reuse the evidence underlying scientific conclusions. Across benchmarks, agents using LKM retrieve more evidence, cite more faithfully, and answer scientific questions more accurately: LKM nearly doubles the known supporting and contradicting evidence retrieved on SciFact-Open (818 versus 443 claim-paper pairs), reasoning graphs raise citation F1 on ScholarQABench by more than 5 points over the same retrieved papers, and LKM retrieval improves a fixed answering model by 9.3, 4.2, and 14.7 points over no retrieval on ChemBench, PubMedQA, and SciBench. LKM lays the foundation for a scientific ecosystem in which AI scientists not only recall accumulated knowledge but also extend it, returning new questions, workflows, and evidence to a memory that every subsequent investigation can build on.

Abstract

Explore similar work

Large Knowledge Model: A Knowledge Foundation for Agentic Science at Scale

TRACES: A Benchmark for Epistemic Reliability in Scientific Reasoning by LLMs

LitTraceQA: A Benchmark for Multi-Stage Grounding and Verification in Scientific Question Answering