Organizations: Dept. of Computer Science, Stony Brook University. · Dept. of Materials Science and Chemical Engineering, Stony Brook University. · Dept. of Materials Engineering, Indian Institute of Technology Jodhpur. · Dept. of Biomedical Informatics, Stony Brook University.
Electrochemistry is central to solving many modern-day challenges, including energy storage systems for a sustainable future, and its advancement relies heavily on the data from past findings. Existing automated pipelines are limited by text-only architectures, context fragmentation, and numerical hallucinations - collectively preventing accurate, large-scale curation of property databases. We introduce ElectrochemLLM, a multimodal information extraction pipeline that jointly identifies structured electrochemical records encompassing electrolyte compositions, electrode materials, cell configurations, and their measured physicochemical properties from unstructured scientific PDFs at scale. By combining hybrid document parsing with vision-capable LLM orchestration, the pipeline extracts verifiable, structured data across text, tables, and figures. Evaluated on the battery-electrolyte subset of our manually annotated benchmark of 232 papers, ElectrochemLLM achieves an F-1 score of 0.97, and has been applied to curate ElectroChemDB, a structured database spanning 15,000 electrochemical research articles. We demonstrate the utility of this resource by training machine learning models for electrolyte property prediction, illustrating a complete pathway from unstructured literature to predictive materials discovery.
Figures & tables
Figure 1: ElectroChemLLM workflow for constructing ElectroChemDB. Scientific PDFs are parsed into multimodal document representations comprising text, figures, and tables. Prompt-guided large language models aggregate information across modalities to extract complete electrolyte compositions (salts, solvents, additives, and concentrations), electrode and cell-chemistry information, physicochemical properties, and electrochemical performance metrics. Extracted records are standardised using external chemistry tools and databases, verified by domain experts, and consolidated into ElectroChemDB, enabling downstream machine-learning tasks including property prediction and materials discovery.
Capability
Direct LLM
CDE
BDE
MERMaid
LLMB
ElectroChem LLM
Heterogeneous PDF ingestion
✓
∼
✗
✓
∼
✓
Full-document contextual extraction
✓
∼
∼
✓
✓
✓
Table extraction
∼
✗
✗
✓
✗
✓
Figure understanding
∼
✗
✗
✓
✓
✓
Structured property schema
✓
✓
✓
✓
✓
✓
Chemical and unit normalisation
∼
∼
∼
∼
✓
✓
Table 1: Capability comparison of direct multimodal LLM extraction, ChemDataExtractor 2.0 Swain and Cole (2016) ; Mavracic et al. (2021) , BatteryDataExtractor Huang and Cole (2022b) , MERMaid Leong et al. (2025) , LLMB Lee et al. (2026) , and ElectroChemLLM .
Configuration
Precision
Recall
F1
Prior Extraction Pipelines
BatteryDataExtractor †
0.18
0.14
0.16
ChemDataExtractor †
0.37
0.28
0.32
LLM Baselines
ChemLLM-20B-SFT
0.57
0.45
0.50
GPT-4o (Direct PDF)
0.75
0.44
0.55
Table 2: End-to-end property-extraction performance on the 121-paper benchmark. Direct-PDF models receive the original PDFs without the document-processing and visual-summarization components of ElectroChemLLM . † Because the prior extraction pipelines support only specific property types, their results are calculated on the corresponding restricted property subsets; the LLM-based systems are evaluated on the complete output schema.
Statistic
Value
Total papers processed
232
Battery electrolytes
121
CO 2 electroreduction
111
Total electrolyte entries
1,686
Battery electrolytes
526
CO 2 electroreduction
1,160
Table 3: Summary statistics for the manually validated benchmark subset of ElectroChemDB used for evaluation, comprising battery electrolyte and CO 2 electroreduction records with unit-standardised compositions and properties.
Figure 2: Distribution of records and distinct electrolyte compositions by electrolyte class in the full battery-electrolyte corpus (41,144 records). (a)Records per class: The solid segment counts records with at least one of five properties (two electrolyte properties, ionic conductivity and viscosity, and three cell-level performance metrics that depend on both electrolyte and electrodes: cycle life, Coulombic efficiency and specific capacity) within its plausible physical range; the pale segment is the remainder. Labels give usable/total records and the class’s share of all records. (b)Number of distinct electrolyte compositions per class. unknown class shows unmapped compositions.
Figure 3: Number of values per property and electrolyte class in the full battery-electrolyte corpus, after removing values outside each property’s plausible physical range. Excludes unknown class.
Figure 4: Distributions of cycle life, ionic conductivity, Coulombic efficiency, specific capacity and viscosity by electrolyte class in the full battery-electrolyte corpus. Boxes span the interquartile range, the thick line marks the median, and whiskers extend to the 5th and 95th percentiles. All axes are logarithmic except Coulombic efficiency.
Figure 5: Source provenance of ElectroChemDB records from the multimodal extraction run: 56,925 records from 15k papers, split into battery electrolytes ( n=40,261 ) and CO 2 electroreduction ( n=16,664 ). (a) Distribution of records by source modality, as a percentage of each domain’s records. Mixed records draw their fields from more than one of text, tables and figures; table only and figure only records come entirely from a single table or figure. (b) The ten most frequently cited sources. Labels were normalized; numbered tables and figures were collapsed into Table and Figure / Graph.
Figure 6: ML readiness of the Li-ion electrolyte slice of ElectroChemDB, evaluated on ionic conductivity. (a–c) In-sample parity plots for (a) Random Forest, (b) Ridge regression, and (c) XGBoost on concentration-weighted ECFP composition features; the dashed line indicates y=x , and the shaded band corresponds to ±0.5log10 units. (d) log10σ by solvent class ( n≥3 , ordered by descending median). (e) Random Forest feature importance aggregated by descriptor type across the top 20 features.
Figure 7: Out-of-fold predictions of log10 specific capacity for Li-ion (top) and Li-metal (bottom) cells. Columns show Random Forest, Ridge, XGBoost, and a pretrained GNN using electrolyte, electrode, and test-condition inputs. Each panel reports R2 and MAE (in log10 units) for the first of five cross-validation repeats, with n the number of entries; dashed lines indicate perfect agreement. Both axes show log10 of the reported specific capacity; each point is one extracted entry, predicted by a model that did not see it during training. GNN, pretrained graph neural network.
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
Field
Content
Paper ID
c5ee03874j.grobid.tei
metadata
Title
Cesium-containing triple cation perovskite solar cells: improved stability […]
Authors
Michael Saliba, Taisuke Matsui, Ji-Youn Seo, […]
DOI
10.1039/c5ee03874j
Publisher
Royal Society of Chemistry (RSC)
Appendix
Table 4: Example S2ORC-Grobid output structure for a representative paper. The parser recovers structured bibliographic metadata, abstract, and section-keyed body text with inline figure/table references preserved. Visual content and table structure are not recovered at this stage.
Figure 8: Sub Figure extraction via YOLOv8 Processing.
Figure 9: YOLOv8 SubFigure Extraction to isolate individual sub-panels from a composite image.
Input Figure Crop
Summariser Output (JSON)
{ "summary": "A cyclic voltammetry (CV) plot showing the electrochemical performance of different electrodes [...]", "key_findings": [ "The bare GC electrode exhibits negligible current response.", "The Cu.MOF electrode shows a pair of redox peaks [...]", "The (GO 8 wt%)Cu.MOF electrode demonstrates significantly enhanced current density [...]", "Increasing the weight percentage of GO generally leads to an increase in current response [...]", "[...]" ] }
Appendix
Table 5: Example input–output pair from the image summariser. The left column shows a figure crop extracted by the fine-tuned YOLO detector 5.2.3 ; the right column shows the corresponding JSON summary produced by the vision-capable LLM, capturing figure-level context, key quantitative findings, and reference metadata.
Property
Captured Fields
Electrolyte Composition
Total composition
Full electrolyte description
Salt(s)
Name, concentration, unit
Solvent(s)
Name, ratio
Additive(s)
Name, concentration, unit
Cell Information
Appendix
Table 6: Target properties and captured fields in the extraction schema. Properties are grouped into composition, cell configuration, electrochemical, physicochemical, and provenance categories. Every field is nullable to avoid forcing hallucinated values when a property is not reported in the source document.
Figure 10: Chemical entity standardization workflow. Extracted literature-reported chemical and material names are normalized, expanded through dictionary and synonym-based candidate generation, resolved to molecular structures using external chemistry resources, and validated to produce canonical machine-readable identifiers for downstream database construction and machine-learning workflows.
Category
Property
F1
Composition
Total electrolyte composition
94.37
Salt(s)
98.61
Solvent(s)
99.55
Additive(s)
93.28
Cell Information
Cathode
98.47
Anode
98.47
Appendix
Table 7: Property-wise extraction performance (F1 score) of ElectroChemLLM on the battery-electrolyte benchmark. Properties are grouped into composition, cell information, electrochemical, and physicochemical categories.
Figure 11: Extraction cost vs. accuracy across configurations on the 121-paper benchmark (Table 2 ). Points are colored by pipeline type: prior extraction pipelines (gray, local tools with no API cost), direct-PDF LLM baselines (blue), and ElectroChemLLM (orange). The left panel shows configurations with no API cost on a separate axis; the right panel shows per-paper API cost on a log scale. Gemini-2.5-Flash under ElectroChemLLM achieves the highest F1 (0.97) at the lowest cost among API-based methods ( ∼ $0.004/paper), dominating all other configurations on both axes.
Detector
Accuracy
Image Quality
Image Complexity
Fine-tuned YOLO
91%
1.92
1.93
Appendix
Table 8: Cropped sub-panel quality and complexity scores (Component B.3) on 60 manually annotated figures. Each dimension is rated on a 1-3 scale by domain experts; see Appendix for the full rubric.
Dimension
Mean Score
Accuracy
2.4
Faithfulness
2.7
Coverage
1.8
Overall
2.28
Appendix
Table 9: Image summarisation quality 5.3 on 60 manually annotated graph-type figure-summary pairs. Each dimension is rated on a 1-3 scale (higher is better); see Appendix for the full rubric.
Property
LLMs
CDE2
BDE
Electrolyte composition
4,035
3,518
–
Coulombic efficiency
1,227
793
793
Cell information
929
–
–
Cell operating voltage range
688
–
508
Ionic conductivity
602
602
602
Electrolyte stability window
472
327
–
Appendix
Table 10: Ground-truth claims included in each system’s evaluation scope. Dashes indicate properties outside a system’s extraction scope. The schema contains a pH field, but it is omitted because the benchmark contains no non-null pH claims.
Laboratory of Organic and Macromolecular Chemistry (IOMC), Friedrich Schiller University Jena, Humboldtstrasse 10, 07743 Jena, Germany · Helmholtz Institute for Polymers in Energy Applications Jena (HIPOLE Jena), Lessingstrasse 12-14, 07743 Jena, Germany · Center for Energy and Environmental Chemistry Jena (CEEC Jena), Friedrich Schiller University Jena, Philosophenweg 7a, 07743 Jena, Germany +1
Department of Metallurgical and Materials Engineering, Indian Institute of Technology Madras, Chennai 600036, India · Center for Atomistic Modelling and Materials Design, IIT Madras, Chennai 600036, India · Wadhwani School of Data Science & Artificial Intelligence, IIT Madras, Chennai 600036, India