An essential concept in technology management is the early detection of valuable patents. Traditional classification approaches have been utilized to identify effective patents based on the extracted patent topics and indices. However, they cannot consider detailed contextual information or relatively long word sequences in patent documents. In this study, we propose a patent grade evaluation framework based on a deep learning model that can capture the detailed semantic features of patent text. Therefore, this study adopts both a convolution neural network and bidirectional long short-term memory with structured patent text data consisting of abstracts and claims for the classification of three levels of patent grades measured in terms of the average number of forward citations per annum. We further exploit the patent indices identified in the early stage as additional inputs to the model to increase the accuracy. Our model has realized over 75% precision and recall in identifying top-grade semiconductor patents granted by the USPTO from 2000 to 2015. We anticipate that our deep learning-based framework with patent text and indices will play a significant supporting role in mergers and acquisitions, investment decisions, and corporate planning through the early-stage evaluation of a large number of patents.
Bibliographical noteFunding Information:
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) ( 2016R1A2A1A05005270 ) and (MSIT) ( 2020R1A2C2005026 )
All Science Journal Classification (ASJC) codes
- Business and International Management
- Applied Psychology
- Management of Technology and Innovation