This repository contains the project 2 of EPFL CS-433 Machine Learning.
Team name: Novae
Team members: Giacomo Orsi, Vittorio Rossi, Chun-Tso Tsai
Lab: Entrepreneurship and Technology Commercialization Chair, EPFL
Despite providing a publicly accessible definition of novelty, the United State's Patents and Trademarks Office (USPTO) receives tens of thousands of non-compliant utility applications every year. This paper provides information regarding the non-linear correlations between a patent application's data and its novelty, and describes a model which predicts whether a patent will be approved.
The paper which describes this project can be found here
This repository contains the source code of the analysis carried out for the project.
This notebook contains the query executed on Google Cloud BigQuery in order to build the datasets as described in the paper.
This notebook contains the process of converting patent claims into vector representations using word2vec. The trained word2vec model can be found here
This notebook contains the neural network design, training, and evaluation. The testing can be executed using the trained neural network that can be found here
The datasets that we generated are accessible in this Google Drive folder (EPFL authentication required).
The computed vectorization of the patent applications included in the dataset we generated is accessible here.