GENERATED AUTOMATICALLY WITH GITHUB COPILOT

Publication Model Classifier

A machine learning system that classifies scientific publications by matching them to appropriate NASA models: ECCO, RAPID, ISSM, CMS-Flux, CARDAMOM, and MOMO-CHEM.

Overview

This classifier uses a combination of NLP techniques to identify which scientific model a publication is associated with:

• Deep learning classification models for research areas, science keywords, and divisions
• Semantic matching using sentence transformers
• Fuzzy keyword matching with context validation
• Data-driven model affinities for different scientific domains

The system provides confidence scores with model-specific thresholds and detailed visualizations for performance analysis.

Features

• Multi-model classification with confidence scores
• Research area, science keyword, and division classification
• Context validation against known model profiles
• Data-driven affinity analysis
• Comprehensive evaluation metrics including F1 scores, MCC, and calibration

Requirements

• Python 3.8+
• PyTorch 1.8+
• Transformers
• SentenceTransformers
• spaCy (with en_core_web_lg model)
• scikit-learn
• RapidFuzz
• NumPy
• Matplotlib

Getting Started

1. Install dependencies:

   pip install torch transformers sentence-transformers spacy scikit-learn rapidfuzz numpy matplotlib
   python -m spacy download en_core_web_lg
   

2. Download required data files:

   • curated_publications.json: Ground truth publication-model mappings
   • model_keywords.json: Model-specific keywords
   • model_descriptions.json: Model descriptions

3. Run the classifier:

   python pubclassifier.ipynb
   

Evaluation

The system includes comprehensive evaluation tools:

• Performance metrics (precision, recall, F1 score, MCC)
• Threshold optimization
• Generalization analysis (train/test split, cross-validation)
• Robustness assessment
• Visualization of all metrics

Models

The classifier works with the following NASA models:

• ECCO: Estimating the Circulation and Climate of the Ocean
• RAPID: Routing Application for Parallel Computation of Discharge
• ISSM: Ice-sheet and Sea-level System Model
• CMS-Flux: Carbon Monitoring System Flux
• CARDAMOM: CARbon Data MOdel fraMework
• MOMO-CHEM: Multi-Model Multi-Constituent Chemical...

License

Open source for scientific research purposes.