Abstract

Filtering, classifying and organizing patents (into a taxonomy, say) is a costly, time-consuming and error-prone process when undertaken by human beings.

Automating the process is not easy either. Following are some of the challenges faced when automating patent filtering and classification:
1. Document length diversity: The lengths of the documents vary considerably
2. Multiple languages: Patents are written in multiple languages, and their translations are often inconsistent
3. Inconsistent terminology: Different inventors use widely differing terminology even within the same domain/topic
4. Classification validation: Validating the quality of the classification for different domains is time-consuming
5. Tuning parameters: When applying an automated process across multiple domains, the tuning parameters to be used could vary considerably

Completely automating the process, thus, would give patent classifications which are overfitted, underfitted or has low confidence levels.

A different approach - supervised learning - also is not very effective: patents are not homogeneous, and the training set for a particular patent collection may be significantly different from the rest of the data set.

In our experience, transductive* and semi-supervised* learning algorithms produce significantly superior results.

We will demonstrate several approaches we have used to address the challenges described above with the help of such algorithms, and their applicability to different patent sets.

* will be explained in presentation.

Speaker Biography

Ankit Biyani is an Engagement Manager with Dolcera. He has worked with several Fortune 500 companies in industries like Pharmaceuticals, Consumer Packaged Goods, Chemicals, Food & Beverages to design innovative solutions and frameworks that aid their R&D teams focus on white spaces and IP teams to formulate their IP strategies.

He has a Master's degree in Management from the Management Development Institute, a top management school in India and also has a Degree in Electrical Engineering. He has leveraged his technical background in the past to lead quality assurance teams for testing of software at Fidelity Investments.