Application of advanced text analytics to patent documents

Natural Language Processing is increasingly been applied to a number of different areas including patents analysis, for example for semantic searching and patent landscape summaries. Some patent search tools have strong word analysis capabilities built in, such as Patseer.

Patent_Insights can also apply this technology directly to one or more patent or other documents.

As an example of this, I downloaded a set of 15 patents filed by Telsa for self-driving cars, and applied a couple of text analysis algorithms to the title and abstracts of these documents.

So what did we learn?

Yake algorithm

The Yake algorithm was published in 2018, and is claimed to outperform other text extraction algorithms. I benchmarked this algorithm against some other text analysis algorithms, and it appeared to give the best results. The output of this algorithms is summarised below:

Word pairs

An approach sometimes used in patent searching is to find pairs of words in a patent. We tested this by looking for the most common pairs of nouns in the sentences for the title and abstract text for these 15 patents - the outcome is shown below:

Learnings

These are only examples of what can be learned from text analytics - this approach is highly flexible and can be applied to a range of document types in a variety of ways - please contact us for the details.