SciBorg Architecture and Chemical Text Mining Reference
Heather Phillips
Core Architecture and Primary Capabilities of SciBorg
The SciBorg project, developed as a collaborative initiative led by the University of Cambridge, established an advanced natural language processing framework specifically designed to extract structured scientific data, discourse structure, and chemical entity relations directly from unstructured academic literature.
Performance Across Chemical Named Entity and Discourse Extraction
Component Module
Target Task
Reported Metric
OSCAR Integration
Chemical Named Entity Recognition
85.3% F-Score
Discourse Tagger
Zone Attribution (15 Rhetorical Categories)
76.4% Accuracy
Dependency Parser
Bi-directional Semantic Extraction
81.2% Precision
Standard Execution Protocol for Literature Corpus Ingestion
Source acquisition: Normalize raw XML and PDF publications into unified SciXML schemas.
Tokenization and POS tagging: Execute domain-adapted lexical segmentation tailored for IUPAC nomenclatures.
Entity resolution and parsing: Apply deep semantic analysis to detect reaction targets and quantitative properties.
Evaluation Criteria Rubric
Criterion
Verification Standard
Primary Evidence
Cross-checked against archival & institutional records
Terminology
Aligned with domain-specific technical nomenclature