07/15/19

Predicting Institution Decisions in Inter Partes Review Proceedings


Predicting Institution Decisions in Inter Partes Review Proceedings

Yuh-Harn Yang, Pu-Jen Cheng, and Feng-Chi Chen

Inter partes review (IPR) was introduced in Year 2012 as an adversarial, post-grant patent review process. The principle of claim construction (broadest reasonable interpretation), standard of proving unpatentability (preponderance of evidence), and shortened time to final decision (18-24 months) have made IPR a popular venue for patent challengers. Institution of an IPR mounts substantial pressures on the patentee because the challenged claims are highly likely to be invalidated in the final decision. Therefore, a reliable model to predict institution decisions is critical for patent and business management. In this study, we construct three support vector machine (SVM) models separately based on the contexts of IPR proceedings and features of the disputed patents. The ensemble model that incorporated the three SVMs can predict institution decisions with 79% accuracy and 0.85 Area under the ROC Curve. Separately, the IPR context-based models perform better than the patent feature-based model. Interestingly, most of the features traditionally regarded important for patent values are not significantly associated with institution decisions. Furthermore, models trained on earlier IPR documents can accurately predict the institution decisions in later proceedings. The prediction accuracy increases with the accumulation of training data. In addition, our approach can identify IPR context features that may influence institution decisions. Our results can provide an empirical basis for IPR policy making and business strategic planning. 

100 J. Pat. & Trademark Off. Soc’y 697 (2019)

If you would like to read the full article and other published articles, subscribe to the Journal of the Patent and Trademark Office Society, for more information click here.

© 2000-2019, Journal of the Patent & Trademark Office Society
Disclaimer & Privacy Policy