Patents are complicated publications that contain many data elements that are essential for describing the invention disclosed and how the patent’s claims can be used by the public. There are more than 20 million active patents from more than 200 national patent offices and billions of dollars spent on the filing, maintaining and researching in the patent space. Artificial intelligence is poised to transform the patent space by reducing costs, creating stronger patents and more accurately determining the value of patents in the marketplace.
A patent allows the owner to have a temporary monopoly on an invention in exchange for the public disclosure of how the invention works. The patent application process includes several laborious steps including researching the prior art (earlier public documents related to the inventive idea in the application under examination), determining priority dates, and creating claim language that correctly describes the invention. When the inventor presents the invention to a patent examiner, he is obliged to specify the prior art of which he is aware and make his case as to why his invention is novel and nonobvious in light of the prior art. The examiner then reviews the prior art and does his own search to determine if the patent application’s claims are novel and nonobvious. This human analysis is often constrained by the resources of patent office: the number of examiners, the data sources available and various time constraints.
This is not the only problem in the patent industry. Issues exist in the marketplace relating to its scope. Patent transactions are estimated to exceed $180 billion per year, but currently only involve about 2% of the active patents. The reason that the patent asset class is so illiquid is due to the complexity of the assets themselves (the complex nature of the publication) which requires a resource intensive process to analyze patents. The marketplace also lacks transparency. Business managers struggle to determine the complete set of patents that exist in a particular technology area and the specific aspects of the technology that they cover. In parallel, investors struggle to know if the patent is valid and valuable.
Patent documents themselves contain a lot of data: priority date, inventors, classifications, examiners, agents, owners, claims and claim elements (just to name a few). They are also connected together in a large network (similar to the world wide web). When they are examined they are associated with related prior art documents. They can in turn become prior art for other applications under examination. When this data created during the application process is analyzed by a machine learning engine, many problems facing the examination and valuation of patents can be solved.
Today, many patents searches are done by text searching and, to a lesser extent, by semantic search engines. The problem with such tools is that their broad searching capabilities return a lot of noise along with the useful results. In some cases a search for “hot swappable drive” returns results including machines for making hot water.
IPwe’s tools bring intelligence to this process by using the all of the patent data and the patent network. By modeling the diverse data and information via various networks including patent and non-patent literature, classification hierarchies, and other aspects, each patent is associated in different ways with other patents in the network, forming a multifaceted patent knowledge network. As certain associations between patents have higher weight than others, queries can be tailored against a targeted universe of patent data with higher relevancy for the goals of the searcher. Instead of keyword searches against the full text of the documents, the query can address only the claim elements themselves matched against a high-dimensional representation of the patent specification and claims, thereby pinpointing relevant text snippets from a known similar universe to a specific claim element. This process provides more targeted, accurate results allowing for quicker examination.
For determining the quality of a patent, IPwe’s tools use a similar approach. Using a random walk with restart process on the patent knowledge network, the tools determine how important relevant to its peers each patent is. This process recreates how likely would a searcher, skilled in the technology, find the patent. Extensive research into this approach has shown to work in other fields and successfully identify important web-pages (by performing this process on hyperlinks between webpages), proteins involved in certain biological processes (by performing this process on a protein-protein interaction network), and social influencers (by performing this process on a social network or follower-followee network). When applied to patents, along with other techniques, our tools can provide a measure of a patent’s quality.
IPwe’s tools based on artificial intelligence and machine learning provide a disruption compared to the current text searching techniques -including semantic search- and will help bring transparency to the patent market. Inventors will be able to research their ideas, business managers will be able to able to find patents targeting a specific technology, and investors will ascertain the relative quality of their portfolios through the power of machine learning techniques which efficiently leverage the vast amount of patent data already available. As more data relating to patent acquisition and licensing is fed into the IPwe engine, new transaction opportunities will emerge, transforming patents into a true asset class.