Patent documents are hard to read, no doubt, but does it depend also on the applicant? A comprehensive study in computer linguistics of the University of Queensland, published at the beginning of October 2020, suggests significant differences between corporate and university patent applications.

The findings of Kong et al. (2020) in a nutshell:

• On average, corporate patents use more complicated sentence structure and elaborate vocabulary.
• Corporate patents require 1.1 to 1.6 more years of education to comprehend compared to university patents.
• Corporate patents tend to list more examples than university patents. The reason can be not to reveal the “best mode” of the invention. 
• Top applicants’ patents tend to be harder to read than patents from other applicants - the measured readability gap is 2.4 times.

Thus, obscuring - and even obfuscation - of the technical content of patents is a real phenomenon, making them not only harder to read, but also harder to find in the first place, and many patent search professionals confirm that.

It is time to think beyond keyword search and Boolean operators. IPRally’s Graph-AI uses a different approach to the prior art search - the AI understands the technology as a whole and offers you astonishing results. Let’s see a practical example of how IPRally helps in such cases.

We search for prior art for “a diaper comprising a QR-code”. This converts automatically or manually to the following graph:

When we launch the search, the AI compares this graph at a deep level to millions of full specification graphs and we get a list of most relevant results. 

Well, the first hit is obvious, as it explicitly mentions diapers and QR-codes. You could expect to find that with a boolean search too. The second hit, however,  illustrates the power of the graph AI search: “Paper diaper service time management method based on two-dimensional code”, CN111354173A:

The text is a machine translation from Chinese and it does not use the word “QR-code” at all. It mentions only “two-dimensional codes”. The AI is able to not only to find the document but also show you the passage it finds the most relevant:

A deeper look at the publication and its drawings reveals that it is all about QR-codes:

The example is a simple one, yet it illustrates well a key advantage of Graph AI: a deeper level understanding of technology. 

The search engine is trained with real patent cases and their prosecution history. That is, the work done by professional patent attorneys and examiners. This makes it possible for the AI to learn several various ways of describing technology - also obscured and obfuscated ones.

IPRally offers a new way of doing a prior art search - one with more freedom, structure, speed and relevant results which other engines due to their restrictive nature of Boolean search miss.

Our customers report a shorter time needed for a novelty search and accurate results in opposition/invalidity searches. And most importantly, finding those crucial pieces of prior art that are hiding in the vast amounts of patent data, but can make your case. More on this in the following parts of this blog series!

Do you still lose precious time defining keywords, synonyms and queries? Do you want to cut the time needed for prior art search? Book a 15-minute demo now and get hands-on with the tool that’ll make your workflow much more efficient.
‍
Source of the cited study: Kong, N., Dulleck, U., Sun, Sh., Vajjala, S. and Jaffe, Adam B., Linguistic Metrics for Patent Disclosure: Evidence from University Versus Corporate Patents (2020). CESifo Working Paper No. 8571, Available at SSRN: https://ssrn.com/abstract=3702123