Back to Projects
demoML/AImldata

Word Sense Disambiguation Router

An NLP router that decides which sense of an ambiguous academic noun a sentence means — 90.35% accuracy above 0.70 confidence at 2.1 ms median latency.

2025-09 — 2025-12
View Source

Key Highlights

  • Disambiguates 10 ambiguous academic nouns over a 2,000-sentence arXiv/Brown corpus
  • TF-IDF with MultinomialNB and LinearSVC: 90.35% accuracy above 0.70 confidence
  • 2.1 ms median latency

Overview

Words like "model", "kernel" or "network" mean different things in different fields. This router reads a sentence and picks the right sense for 10 ambiguous academic nouns, trained on a 2,000-sentence corpus drawn from arXiv and the Brown corpus.

Approach

  • Features: TF-IDF over the sentence context, with WordNet senses as labels.
  • Models: MultinomialNB and LinearSVC, with a confidence threshold so the router abstains instead of guessing.
  • Result: 90.35% accuracy on predictions above 0.70 confidence, at 2.1 ms median latency.