Looking to improve the navigation of its online database of patents, the United States Patent and Trademark Office (USPTO) and the U.S. Department of Commerce sponsored an international competition. The winner has been revealed and the prize goes to the computer scientists at the University of Massachusetts (UMass) Amherst.
The goal of the competition was to improve access to the millions of USPTO records by designing creative new approaches that deliver better information about innovators and the new inventions they develop by disambiguating inventors’ names.
According to the USPTO, the competition attracted entries from throughout the U.S., as well as Germany, China, Australia, and Belgium.
The entry from the UMass Amherst team focused on enhancing the navigation so businesses, innovators, and policy makers could more easily locate the records they are seeking.
The result is a more effective and efficient US patent search. Called the entity disambiguation algorithm, it will be incorporated into PatentsView, the USPTO’s new online interactive patent search platform.
USPTO Director and Undersecretary of Commerce for Intellectual Property Michelle K. Lee said the invention “will provide users more efficient and effective searches of the country’s millions of inventors and patents.”
The team, advised by Andrew McCallum, a professor in the Information Extraction and Synthesis Laboratory and director of the UMass Center for Data Science, designed the computer algorithm to remove inventor ambiguity from patent records quickly.
The algorithm differentiates among many entities or individuals with patent applications containing similar attributes and groups them together correctly. Inventor disambiguation is extremely important to the USPTO because inventors could be listed in patent records with different names, spellings, or nicknames. Furthermore, multiple inventors may have the same name.
These types of ambiguities make the US patent search tool unreliable, plus necessitate time-consuming manual intervention.
The UMass team took a hierarchical approach to disambiguation, rather than using a pairwise comparison. The hierarchical method evaluates groups of two or more mentions to determine the disambiguation and then applies the disambiguation procedure.
For winning the competition, the UMass Information Extraction and Synthesis Laboratory will receive a $25,000 stipend to compensate team members for technical guidance as its computer algorithm is integrated into PatentsView.
According to a news release from UMass, the winning team members are graduate students Ari Kobren and Nicholas Monath; Michael Wick, who now works at Oracle Labs; Jack Sullivan, who now works at Cambridge Semantics; and Sameer Singh, currently at the University of Washington.
Monath, who co-authored the algorithm with McCallum, credited the team’s top finish to its quick and accurate solution of disambiguating the inventors listed in more than 12 million patent records filed between 1976 and 2014.
“We had the fastest system in the competition and the system with the highest accuracy score,” Monath noted.
The whole purpose of PatentsView is to make it possible for all people to understand USPTO data. The existing database, as shown in the above image, is difficult to use because of complicated data formats and inconsistent formatting.
PatentsView makes the data more accessible and easy to use with search filters that provide multiple viewing options. With the UMass team’s computer algorithm invention, the data will be easier to navigate, too.
