1. Classify/
  2. Directory/
  3. Classification/
  4. Jev walks a CPC patent tree to a leaf with beam search

Jev walks a CPC patent tree to a leaf with beam search

Author: TypeSafe

Classification

TypeSafe

TypeSafe cookbook

docs.typesafe.ai

Classify this use case

  • Classification1.00
  • Shipped0.93
  • Measured0.15
  • Usefulness0.45
  • Cookbook1.00
  • Ready0.94
  • Pattern generic0.95
  • Copy quality0.54
  • Creativity0.62
  • Distinct intent0.83
Result
Leaf labels from parallel beam search over Choice probabilities

Ask Jev parallel Choice questions at each taxonomy level and keep the best beam paths to a leaf.

A CPC patent tree, a retail catalogue, a biomedical heading, or a code tree is a stack of Jev Choice questions.

Greedy search takes the best child at each node. Beam search asks several paths in one Jev call and keeps the best K by geometric-mean probability.

The leaf of the surviving path is the label. The cookbook is the cited source.

How it works

  1. Start at the root of the hierarchy
  2. Ask Jev to score the next-level options
  3. Keep the best K paths by geometric-mean probability
  4. Stop at a leaf and return that label