ConceptNet has published an open-source, fine-tuned distilbert-base-multilingual-cased model designed for enterprise voice intent classification. The system utilizes a four-layer hierarchy to categorize commands into basic, context-aware, predictive, and autonomous intents.

  • The architecture offers two inference paths: a fast TF-IDF + Logistic Regression path achieving 83% accuracy in under 5ms on CPU, and a neural DistilBERT path reaching 98.6% accuracy in under 100ms.
  • The model is constrained to output exactly four valid labels, preventing hallucination.
  • It supports nine languages: English, French, Spanish, German, Italian, Portuguese, Chinese, Arabic, and Russian.
  • The training dataset consists of 730 labeled examples across these languages.

The authors are seeking feedback from the NLP community and inviting enterprise teams to pilot the tool for free via a sandbox environment.