IP Library Granted Patent US 12705679
Granted Patent B2
US 12705679 · App. 18/091,208 · Granted Aug 11, 2026

Machine learning models for patent valuation

Inventors: David Hwei-Yu Hsu (Philadelphia, PA); Prasanna Tambe (Wynnewood, PA); Dokyun Lee (Englewood Cliffs, NJ)
Assignees: The Trustees of the University of Pennsylvania; Carnegie Mellon University
G06Q50/184G06N3/0442G06N3/0464G06N3/08
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12705679
App. No.
18/091,208
Filed
Dec 29, 2022
Granted
Aug 11, 2026
Kind
B2
Art Unit
3626
USPC
705/310
Abstract

Methods, systems, and computer readable media for using machine learning models to determine predicted values of patent documents. In some examples, a method includes training, by at least one processor, a machine learning model to predict patent value based on unstructured text from training patents and, for each training patent, a measure of patent value. The method includes supplying, by the at least one processor, unstructured text from a patent document to the machine learning model. The method includes outputting, by the at least one processor, a predicted measure of value of the patent document.

Claims (23)

1 . A method comprising:

training, by at least one processor, a machine learning model to predict patent value based on unstructured text from a plurality of training patents and, for each training patent, a measure of patent value, wherein training the machine learning model includes creating a vocabulary of tokens corresponding to words that appear at least a predetermined number of times in the training patents, creating, using a pretrained embedding model and for a predetermined number of the tokens created from the summary section of each of the training patents, word embeddings for the summary section of each of the training patents and providing the word embeddings for the summary section of each of the training patents as input to the machine learning model, wherein the machine learning model comprises a plurality of convolutional neural network (CNN) layers, the machine learning model comprises a bidirectional long short-term memory (LSTM) network configured for tracking local-level features extracted by the CNN layers, and the machine learning model comprises a multi-layer perceptron layer with an activation function to produce a final patent representation from the bidirectional LSTM network;

supplying, by the at least one processor and to the machine learning model, unstructured text from a summary section of a patent document; and

outputting, by the at least one processor, a predicted measure of value of the patent document.

2 . The method of claim 1 , wherein training the machine learning model comprises determining the measure of patent value for each training patent based on one or more market reactions to firm patent grants.

3 . The method of claim 1 , wherein the predicted measure of value of the patent document is a predicted measure of economic value to be assigned to the patent document by a market for patents or firms.

4 . The method of claim 1 , wherein the predicted measure of value of the patent document is a predicted number of forward citations for the patent document.

5 . The method of claim 1 , wherein the machine learning model is a deep learning model configured to use natural language processing on the unstructured text.

6 . The method of claim 1 , wherein the patent document comprises a patent, patent application, or patent application publication.

7 . The method of claim 1 , wherein training the machine learning model comprises further training the machine learning model using structured data from the training patents.

8 . A system comprising:

at least one processor; and

a machine learning model implemented and trained by the at least one processor to predict patent value based on unstructured text from a plurality of training patents and, for each training patent, a measure of patent value, wherein training the machine learning model includes creating a vocabulary of tokens corresponding to words that appear at least a predetermined number of times in a summary section of each of the training patents, creating, using a pretrained embedding model and for a predetermined number of the tokens created from the summary section of each of the training patents, word embeddings for the summary section of each of the training patents and providing the word embeddings for the summary section of each of the training patents as input to the machine learning model, wherein the machine learning model comprises a plurality of convolutional neural network (CNN) layers, the machine learning model comprises a bidirectional long short-term memory (LSTM) network configured for tracking local-level features extracted by the CNN layers, and the machine learning model comprises a multi-layer perceptron layer with an activation function to produce a final patent representation from the bidirectional LSTM network;

the machine learning model being supplied, by the at least one processor, with unstructured text from a summary section of a patent document to and outputting a predicted measure of value of the patent document.

9 . The system of claim 8 , wherein training the machine learning model comprises determining the measure of patent value for each training patent based on one or more market reactions to firm patent grants.

10 . The system of claim 8 , wherein the predicted measure of value of the patent document is a predicted measure of economic value to be assigned to the patent document by a market for patents or firms.

11 . The system of claim 8 , wherein the predicted measure of value of the patent document is a predicted number of forward citations for the patent document.

12 . The system of claim 8 , wherein the machine learning model is a deep learning model configured to use natural language processing on the unstructured text.

13 . The system of claim 8 , wherein training the machine learning model comprises further training the machine learning model using structured data from the training patents.

14 . A non-transitory computer readable medium having stored thereon executable instructions hat when executed by a processor of a computer controls the computer to perform steps comprising

training, by the at least one processor, a machine learning model to predict patent value based on unstructured text from a plurality of training patents and, for each training patent, a measure of patent value, wherein training the machine learning model includes creating a vocabulary of tokens corresponding to words that appear at least a predetermined number of times in a summary section of each of the training patents, creating, using a pretrained embedding model and for a predetermined number of the tokens created from the summary section of each of the training patents, word embeddings for the summary section of each of the training patents, and providing the v word embeddings for the summary section of each of the training patents as input to the machine learning model, wherein the machine learning model comprises a plurality of convolutional neural network (CNN) layers, the machine learning model comprises a bidirectional long short-term memory (LSTM) network configured for tracking local-level features extracted by the CNN layers, and the machine learning model comprises a multi-layer perceptron layer with an activation function to produce a final patent representation from the bidirectional LSTM network;

supplying, by at least one processor and to the machine learning model, unstructured text from a summary section of a patent document; and

outputting, by the at least one processor, a predicted measure of value of the patent document.