 Prediction of patent grant and interpreting the key determinants: an application of interpretable machine learning approachLi Yao and
He Ni ()
Scientometrics, 2023, vol. 128, issue 9, No 2, 4933-4969 Abstract: Abstract Patents are valuable intellectual property only when granted by the governments, and failing to receive an official grant means disclosing valuable technologies and information, which otherwise would be kept as commercial secrets. Yet, a typical patent application process takes years to complete and the outcome is uncertain. This study implements machine learning models to predict patent examination outcomes based on early information disclosed at patent publication and interpret the mechanism of how these models make predictions, highlighting the key determinants to patent grant and delineating the relationships between the patent features and the examination outcome. The predictive models that integrate patent-level variables with textual information accomplish the best prediction performances with a 0.854 ROC-AUC score and 77% accuracy rate. A number of interpretable machine learning methods are applied. The permutation-based feature importance metric identifies key determinants such as applicants’ prior experience, page length, backward citation, claim counts, number of patent family, etc. SHAP (SHapley Additive exPlanations), a local interpretability method, describes the marginal contributions to the model prediction of key predictors using two actual patent examples. Our study provides several valuable findings with important theoretical insights and practical applications. Specifically, we show that patent-level information can serve as a predictor of examination outcomes and the relationships between the predictors and outcome variables are complex. Knowledge accumulation and technology complexity positively affect the likelihood of patent grants, albeit with a curvilinear relationship. At lower levels, both factors significantly increase the chance of a grant, but beyond a certain threshold, the marginal effect becomes less pronounced. Additionally, prior experience, patent family size, and engagement with the patent agency have a monotonic and positive relationship with the grant likelihood, whereas the impact of patent scope on patent grants remains uncertain. While a narrower and more specific patent claim is associated with a higher grant rate, the number of claims increases it. Moreover, technology range, inventor team size, and examination duration have little effect on the patent grant results. From a practical standpoint, the accurate prediction of patent grants has significant potential applications. For instance, it could help firms better prioritize resources on the patent applications of high grant potentials to secure the final grant, as failure means a waste of R &D effort and disclosure of technology without IP protection. Additionally, patent examiners could utilize our predictive results as prior knowledge to enhance their judgment and accelerate the examination process. Keywords: Patent grant; Interpretable machine learning; Predictive models (search for similar items in EconPapers) Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:spr:scient:v:128:y:2023:i:9:d:10.1007_s11192-023-04736-z Ordering information: This journal article can be ordered from DOI: 10.1007/s11192-023-04736-z Access Statistics for this article Scientometrics is currently edited by Wolfgang Glänzel More articles in Scientometrics from Springer, Akadémiai Kiadó |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.