Latent Dirichlet Allocation Modeling for CPS Patent Topic Discovery

Abstract

Industry 4.0 is an organized framework to infuse the latest technology in the manufacturing sector. The inclusion of next-generation technologies such as Cyber-Physical Systems (CPS), cloud computing, big data and artificial intelligence approaches increases productivity and manufacturing output in today's dynamic industrial environments. This research is a Latent Dirichlet Allocation (LDA) topic modeling extension from a prior research on technology standards and patent portfolios for industrial CPS [1]. Topic modeling is a statistical approach for discovering topics that occur in a document corpus [2]. Latent Dirichlet Allocation (LDA) is an unsupervised technical approach in topic modeling for efficient and insightful data analysis. A collection of 1868 CPS patents from the US patent database has been used as input to group patents in several relevant topics for industrial CPS using LDA model in this research. Topic modeled patent groups allowed for the identification of relationships between terms and topics, enabling better visualizations of underlying intellectual property dynamics. Top assignees for each group are computed based on LDA results, these insights were unknown in prior investigations. Further, a graphical representation of the topic trend across groups present a direction of promising patents towards industrial application. The correlations presented enhances patent utilization and promotes cross-licensing commercialization.