We have previously disclosed details of the accuracy of our Art Unit Predictor, and we expect that changes to USPTO procedures will make our Art Unit Predictor even more accurate in the long term.
Here is a summary of in progress changes to the USPTO's procedures for assigning patent applications to examiners:
- Changing to the classification system (USPC, CPC, etc.)
- Assigning applications directly to examiners instead of to art units and then to examiners.
- Using computers instead of humans.
Data Don't Lie
At a high level, we don't care about changes behind the scenes at the USPTO. For the most part, we don't even need to know how they are doing it. We have the data, and that is all we need.
We have lots of public data that tells us which applications were assigned to which groups and art units. The beauty of machine learning is that we don't need much else. We crunch the data and generate a machine learning model that learns directly from the data. After we have trained this machine learning model, we give it a new patent application and it predicts the group and art unit that will be assigned based on what it learned from the data.
The classification codes don't really matter. For each patent application, we have the full text and we have the group and art unit that were assigned. This is the most import information that is needed to learn how the USPTO assigns patent applications to groups and art units. The classification codes are mostly irrelevant.
It also doesn't matter that applications may be assigned directly to examiners instead of first to an art unit and then to an examiner. Either way, we know the group and art unit of the application and that is all we need to train our models and do our predictions.
USPTO Changes Will Improve Our Accuracy
We actually can't wait for the USPTO to fully automate the assignment of patent applications to groups and art units.
Our current prediction task is quite hard. We need to predict how a human will assign a patent application to classification codes, a group, an art unit, or an examiner. Humans make mistakes and are hard to predict. Essentially, there is a lot of randomness when humans are involved, and you cannot predict randomness.
Once the USPTO starts using computers to do the assignment, the accuracy of our Art Unit Predictor will improve and will be even more accurate than it is now. When computers are doing the assignments, there is no longer any randomness. The computer assignments are thus much more predictable so that is much easier for our computers to learn what the USPTO computers do. We might even be able to achieve near perfect accuracy in the future.
With computers at the USPTO doing the assignments, we may even be able to predict the specific examiner that will be assigned to an application. Since there are nearly 10,000 patent examiners, this will be a very difficult task!
In summary, the USPTO is changing how it assigns patent applications, but these changes should not prevent our Art Unit Predictor from making accurate predictions and should even allow us to do even better in the future.