Monday, August 31, 2020

How accurate is our Art Unit Predictor?

 We launched our Group and Art Unit Predictor at the beginning of the year, and we are excited to announce that we just deployed an update that greatly increases the accuracy of our predictions. 

To refresh your memory, our Predictor processes a draft of your patent application, predicts the group and art unit that will be assigned, and helps you steer your application to a preferred group and art unit. Here is an example art unit prediction result:

Of course, such predictions are only useful if they are reasonably accurate. We computed the accuracy of our Predictor over a set of 10,000 patent applications (separate from our training data), and here are the results:

  • Group Prediction (57 groups)
    • Top prediction has 67% accuracy
    • Top 3 prediction has 89% accuracy
  • Art Unit Prediction (504 art units)
    • Top prediction has 48% accuracy
    • Top 5 prediction has 80% accuracy

We worked extremely hard to achieve this level of accuracy, and are very pleased with the results. This prediction task is very challenging because of (i) the large number of groups and art units and (ii) inconsistencies in the assignments done by the USPTO (humans are involved).

As far as we know, no other providers of similar services have disclosed the accuracies of their predictions. We believe transparency is important, and we also want to show off the hard work that we've done. At Patent Bots, we are experts in big data and machine learning, and our expertise allows us to provide advanced tools, such as our Group and Art Unit Predictor