Discovery of Technological Innovation Systems: Implications for Predicting Future Innovation

ABSTRACT

In contrast with the accelerating trend of boundary-spanning (horizontal) technological innovation, the current Cooperative Patent Classification (CPC) scheme applies a hierarchical (vertical) structure to innovation output in terms of patents. For this reason, we argue that the CPC can be complemented with dynamic technological innovation system (TIS) discovery through machine learning that accounts for horizontal relationships across seemingly disparate technologies. Using a design science approach, we propose a framework to discover boundary-spanning TISs by leveraging the textual information from millions of patents. We validate our framework in terms of the ability of discovered relationships to predict future innovation quantity and quality in different technology classes. Our novel TIS-based innovation metrics that leverage patenting activity in related technology classes are significantly associated with future innovation intensity in focal technologies. We conduct experiments with machine learning models to further tease out the predictive utility of our TIS discovery framework.

KEYWORDS:

• technological innovation system (TIS)
• machine learning
• patents
• predictive analytics

Disclosure statement

No potential conflicts of interest are reported by the authors(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/07421222.2023.2301172

Notes

^1. In this paper, we use the term “technology class” to denote the patent subclass (four-character CPC) as defined by the USPTO. An example of a “technology class” in our research is “H01L,” which corresponds to the category of “Semiconductor” within the CPC structure.

^2. The authors use the term “technology ecosystem” [4], which is conceptually consistent with the notion of TIS when all the components of the TIS are individual technologies [60].

^3. https://maartengr.github.io/BERTopic/index.html

^4. We find our main results to be consistent with another popular topic modeling technique called LDA.

^5. In addition to a five-year window, we conducted the main analysis in Equation 9(9) $\mathit{F}{P}_{\mathit{i},\mathit{t}}={\beta }_0+{\beta }_1\cdot \mathit{P}{P}_{\mathit{i},\mathit{t}-1}+{\beta }_2\cdot \mathit{CRT}{I}_{\mathit{i},\mathit{t}-1}+{\beta }_3\cdot \mathit{RT}{I}_{\mathit{i},\mathit{t}-1}+{\beta }_4\cdot \mathit{BS}{I}_{\mathit{i},\mathit{t}-1}\text{break}+{\beta }_5\cdot \mathit{RT}{I}_{\mathit{i},\mathit{t}-1}\cdot \mathit{BS}{I}_{\mathit{i},\mathit{t}-1}+{\varphi }_i+{ϵ}_{\mathit{i},\mathit{t}}\mathit{\hspace{0.17em}}$(9) with a three-year window. The results remain consistent with those from a five-year window.

^6. According to Hall et al. (2001), about 98.4% of the filed patents in 1990-92 were granted within four years. Given that our dataset covers until 2021, it is reasonable to stop in 2017 to account for the truncation problem.

^7. Verspagen and De Loo (1999) find that the average citation lag between technology sectors is about 4.67 years. Hence, it is conservative to consider a five-year citation lag window after the peak within the first two years.

^8. We can expect older patents to be cited more.

^9. The predictor variables for each year still accumulate information from the past 5 years.

^10. For details, see Section 2 of [67].

Additional information

Notes on contributors

Junho Yoon

Junho Yoon ([email protected]) is a Ph.D. candidate at Tippie College of Business, University of Iowa. His research is focused on technology innovation and interfirm collaborations by leveraging machine learning, econometrics, and network analysis.

Gautam Pant

Gautam Pant ([email protected]) is a Professor at the Gies College of Business, University of Illinois Urbana-Champaign. His research focuses on using machine learning and data science to gain insights into inter-firm competition, human capital, innovation, and sustainability. He serves as a senior editor for Information Systems Research.

Shagun Pant

Shagun Pant ([email protected]) is a Teaching Associate Professor at the Gies College of Business, University of Illinois Urbana-Champaign. Her research interests include corporate finance, corporate governance, innovation, and business analytics. Dr. Pant’s work has appeared in such journals as Journal of Finance and Research Policy.