Procode

Procode assigns classification codes to occupational and industry free-text descriptions using Complement Naïve Bayes to map entries to the French PCS and NAF taxonomies.

Key Features:

• Algorithm: Uses Complement Naïve Bayes (CNB) as the primary machine learning classifier for automatic coding.
• Training data: Developed and evaluated on nearly 30,000 free-text entries manually coded to PCS (Classification des Professions et Catégories Sociales) and NAF (Nomenclature d'Activités Française).
• Evaluation protocol: Performance assessed with 5-fold cross-validation.
• Performance: Reported accuracy ranges of 57%–81% for PCS and 63%–83% for NAF under cross-validation.
• Recoding between classifications: Integrates recoding implemented as a simple search mechanism that uses existing crosswalks linking classification codes.
• Extensibility plans: Intends to collect additional datasets and extend automatic coding and recoding support to other classifications.

Scientific Applications:

• Automatic occupational coding: Assigns PCS and NAF codes to free-text occupational and industry descriptions.
• Cross-classification mapping: Translates and aligns occupational data between classification systems using crosswalks.
• Method benchmarking: Provides empirical performance data for evaluating machine-learning approaches to occupational coding using cross-validation.

Methodology:

Procode applies Complement Naïve Bayes trained on ~30,000 manually coded PCS and NAF entries and was evaluated using 5-fold cross-validation; recoding between classifications uses a search over existing crosswalks.