Intellectual property is central to corporations' growth and prosperity. The number of patent applications has risen drastically in the past decade reaching close to 1 million new applications per year in the United States and Europe alone. Patent application filings have risen at an annual rate of 4.7% since 1995. To be granted, a patent application has to demonstrate novelty and non-obviousness. To determine whether a new patent application is indeed novel and non-obvious, the patent examiner searches for related prior-art in other patent documents and public sources.
What is the problem?
The search for prior-art spans million of patent documents. The main challenge for automated prior-art search is how to identify scientific relations, based on textual features for large scale datasets of patent documents. Related scientific ideas usually include different terminology. Determining the conceptual relatedness of words and technical phrases requires specialized professional knowledge and evaluations of hundred of thousands related technologies.
What is the current market solution?
Several for-fee and not for-fee patent search engines have emerged in the past few years. Leading prior-art search engines are: USPTO, EPO, Google Patent, Dialog, Delphion, and FreePatentsOnline.com. These engines are based on textual similarity indices (mostly key-words) which assume that patents are more related as their textual overlap rises.
Why does the market solution fail?
There are three main reasons for why existing textual search engines perform poorly. First, inventors have an incentive to phrase their inventions in a manner that would be as distant as possible from the text of the most related prior-art, hoping this would mitigate the risk the application would be rejected by the examiner. Second, the textual domain used to describe scientific concepts is typically large. Therefore, similar inventions are likely to be described in different ways. Third, in numerous cases the prior-art stated by the patent examiner is from different technology areas than the application itself, where the textual overlap between the prior-art and the application is usually low.
What is our solution?
We develop a novel proprietary database that includes concepts that have been proven to be scientifically related. This database is a "dictionary" of terms that indicates the relevancy of one term to the technical problem that is described by other terms. Using artificial intelligence algorithms we develop a "trained" search engine that investigates the conceptual relationship between patent documents as implied by the ideas they contain, as opposed to their textual overlap. Alpha-version tests show that our search engine substantially improves the predictive ability of competing engines in identifying patent applications that are rejected by examiners.
--
Jon Fjeld, Ph.D.
Professor of the Practice, Strategy
Executive Director
Center for Entrepreneurship & Innovation
The Fuqua School of Business
Duke University
(919) 660-8052
http://www.fuqua.duke.edu/centers/cei/
Entrepreneurship at Duke: www.EntrepreneurshipAtDuke.com
A Duke Resource for Entrepreneurs: http://www.dukeven.com/
--
Howie Rhee, MBA
Managing Director, Center for Entrepreneurship and Innovation
Fuqua School of Business, Duke University
Office A236
919-617-1123 Google Voice / Mobile
919-660-1929 Office
Learn more about Entrepreneurship at Duke - www.EntrepreneurshipAtDuke.com
and read our Duke Entrepreneurship Manual - www.dukeven.com.
Twitter: @EshipAtDuke
4 comments:
Hi Howie,
Usually I do not read article on blogs, but I would like to say that this write-up very pressured me to check out and do it! Your writing taste has been surprised me. Thank you, quite nice article.
Wilson Peter
matchmaker Chicago
Hello Howie,
Your place is valuable for me.Thanks!
discount make up
this is nice blog. you are doing great work on this blog....
Ballast water management plan
I like this blog and information
about events....
cattle panels
Post a Comment