Corrosion inhibitor dataset from 5,597 literature
Description
Corrosion inhibitor dataset from 5,597 literature, IE_datasets.xlsx file is the dataset file after manual verification and processing, where fields include Reference.Title Reference.Authors Reference.Author_Affiliations Reference.Journal Reference.Publication_Date Reference.Keywords Corrosion_Inhibitor.Name Corrosion_Inhibitor.Composition Corrosion_Inhibitor.Type_Anode_Cathode_Mixed Corrosion_Inhibitor.Type_Film_Mechanism Corrosion_Inhibitor.SMILES Corrosion_Material.Name Corrosion_Material.Grade Corrosion_Material.Composition Corrosion_Material.Processing Corrosion_Material.Heat_Treatment Corrosion_Environment.Medium Corrosion_Environment.Medium_Type Corrosion_Environment.Concentration Corrosion_Environment.Temperature Corrosion_Evaluation.Test_Temperature Corrosion_Evaluation.Test_Time Corrosion_Evaluation.Inhibitor_Concentration Corrosion_Evaluation.Test_Method Corrosion_Evaluation.Inhibition_Efficiency_% json_extracted.zip file is the json files that extracted by LLM. IE_PH_4_materials.xlsx file is the dataset used for construting models.
Files
Steps to reproduce
All data extraction code is available at https://github.com/MarrytheToilet/SPED. The repository contains the multi-agent extraction pipeline (schema handling, extraction, merging, review and evidence verification) and the analysis scripts that regenerate every statistic and figure in this Descriptor, covering corpus ingestion, gold-standard alignment and scoring, evidence localization, the schema ablation and the confidence model. Document parsing used MinerU. Language-model inference used deepseek-v4-flash at temperature 0.1 with a 65,536-token output budget through an OpenAI-compatible interface. Sequence alignment used the RapidFuzz library. Package versions are pinned in the repository.
Institutions
- University of Science and Technology BeijingBeijing, Beijing