Project Overview

This project aims to search uspto patents data, to find patents similar to a given patent.

Data

Refer to the data approach

Project Status

ETL

This stage of the project can be marked as 3.5/5 complete. We have normalised data. We have made a wrapper python class which can be used for direct population to PGSQL.

Unsupervised learning

A sample is created with minimal tuning.Accuracy may not be great currently. Refer to get_similar

To improve accuracy, one needs to do the following.

1. Create a validation procedure.
2. Create a cluster model, then a topic model and possibly within the existing patent classifications. One needs to iterate to see whats the best approach based on the specifics of the patent data as such.

Development Environment

conda env create -f dev_env.yml

source activate patentsearch