Machine Learning Algorithm to Predict Favorable Art Unit to be Assigned to Patent Application

Project Description

About Us

We are a software and services company. We have serviced IPlaw firms for the past 15 years and have a deep understanding of the business and technology required to be successful in a competitive marketplace. We have approximately 50 employees, with an even split between software developers and implementation engineers. Our PLink platform enables firms to optimize efficiencies, reduce risk, and increase client satisfaction. PLink has an analytics module, and the algorithm developed here will ensure that we remain ahead of the market. 

Project overview

Intellectual Property law firms are applying for patents every day.  This process is called ‘patent prosecution.’ A major part of this process is surviving the review of the USPTO patent agents. It’s their job to find problems with the patent application – it’s too broad (you can’t patent a table) or too specific or already exists. Certain patent categories (or ‘art units’) are more difficult to achieve success than others.  We want to develop a machine learning algorithm that analyzes existing approved patent abstracts and their assigned art units. We would then expose new patent abstracts to this algorithm to predict the most likely art unit to be assigned to this patent application.  We need the Experfy expert to develop this algorithm and train one of our resources to maintain it (i.e. continue to ‘teach’ it).

Objectives:

Based on a prior catalog of application claims, the classes and art units in which they were examined, and the class and subclasses that make up each art unit, extrapolate the most likely class and art unit for the proposed new application.

Present global statistics for the art unit as well as analytics on the examiners in the art unit.  Allow for adjustment of proposed language to re-evaluate the projected classification and art unit.

Assumptions & Results

The outcome of this project will be a web service hosted in Azure that will allow PLink to pass proposed application abstract and claims language and return a proposed class, subclass, and art unit.

None of the proposed language entered by the practitioner will be stored. 

User Experience

Upon submission of the proposed application language, the PLink interface will present the user with the likeliest Class, Subclass and Art Unit, a summary of global and internal statistics for the Art Unit to include: overall allowance rate, allowance rate before final, allowance rate after final, allowance rate with appeal, allowance rate without appeal, and a list of examiners that have worked in the art unit.    

Data Sources: See attachment

In your proposal please tell us:

1. Tell us more about the technique and methodologies you will use
2. What technologies would you suggest? We are a .NET shop.
3. How will you work with our in-house tech staff to ensure that we can maintain the algorithm moving forward?
4. How would you design the solution so that we could build on it?  Our next project would be to recommend language changes to move a patent application into a ‘friendlier’ art unit.