A framework for automatic TRIZ level of invention estimation of patents using natural language processing, knowledge-transfer and patent citation metrics

  • Authors:

  • Zhen Li;Derrick Tate;Christopher Lane;Christopher Adams

  • Affiliations:

  • Department of Mechanical Engineering, Box 41021, Texas Tech University, 7th and Boston, Lubbock, TX 79409, USA;Department of Mechanical Engineering, Box 41021, Texas Tech University, 7th and Boston, Lubbock, TX 79409, USA;Department of Mechanical Engineering, Box 41021, Texas Tech University, 7th and Boston, Lubbock, TX 79409, USA;Raytheon Company, 2501 W. University Drive, MS 8078, McKinney, TX 75071, USA

  • Venue:
  • Computer-Aided Design
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Patents provide a wealth of information about design concepts, their physical realization, and their relationship to prior designs in the form of citations. Patents can provide useful input for several goals of next-generation computer-aided design (CAD) systems, yet more efficient tools are needed to facilitate patent search and ranking. In this paper, a novel framework is presented and implemented for classifying patents according to level of invention (LOI) as defined in the theory of inventive problem solving (TRIZ). Level of invention characterizes the creativity of a design concept based on the resolution of a design conflict and the disciplines used in resolving the conflict. The assessment of LOI for a series of patents provides a useful input for screening and ranking patents in databases to identify high-impact patents. However, the manual effort required for assigning LOI to each patent is laborious and time-consuming. In this paper, a novel method that combines text mining, natural language processing, creation of knowledge-transfer metrics, and application of machine learning approaches is presented and implemented for classifying patents according to LOI. Two case studies are presented in which LOI data is compiled for patents: dynamic magnetic information storage or retrieval using Giant Magnetoresistive (GMR) or Colossal Magnetoresistive (CMR) sensors formed of multiple thin films (USPC 360/324) and arbitration for access to a channel (USPC 370/462). The peak performance in 5-fold stratified cross-validation was found to be 73.38% in the first case study and 77.12% for the second.