The classification criteria for patent classification vary depending on the purpose of the analysis, and it is very important to obtain a result that meets the purpose.
WIPS' PRISM is a service that generates and provides a classification engine optimized for the user's criteria by learning the user's correct answer data using 5 specialized AI algorithms.
In patent research and analysis, denoising (which filter patents that is not fit for the analyzing purpose) and classifying works are very essential but costs a lot and
time-consuming because most of it is done by hand. It is really hard work. Although we have IPC (International Patent Classification), IPC is insufficient to respond to the rapidly changing technology system, so it’s not suitable for use as a technology classification system for patent. WIPS' PRISM is a customized classification engine that has been trained with a classification system (correct answer data) created through users' long-valued experience and practice, and is a unique classifier that only users can own.
It uses 5 specialized AI algorithms to learn user’s criteria and individually create a classification engine. Companies with internal classification criteria can be more tailored to the service. It enables consistent short-time processing of review, denoising, and classification tasks that must be repeated over and over again.
Combination of 5 algorithms creates - 70 untrained models (INDUCER) - Input 3 data types creates - 210 individual classification models - Select the best 15 classification models based on measured values – apply 3 ensembles – final best model is generated.
Through this process, the custom classification engine is created.
Here, ‘train’ (also learning) refers to the process of creating a classification model by applying user-defined correct answer data to an untrained model (INDUCER).
The classification engine created by learning the correct answer infers the unclassified data and performs classification based on the user's answer data.
The inference result provides three categories per patent and the probability of each inference result. Over 90% accuracy have been confirmed by numerous tests.
Two files are required for the free trial.
Requirements for PRISM Free Trial
① Correct answer data for AI to learn ② unclassified data to be inferred by the classifier that learned the correct answer.