Thanks to the millions of documents that need to be analysed, AI can supplement human intelligence to analyse patent and market data. Using AI will involve a shift from keyword-based searches and conventional Boolean operators for patent discovery to AI-enhanced semantic searches using neural networks for high retrieval efficiency and accuracy.

Assistance provided by AI

The significant growth in the volume of patent and non-patent literature – which is present in multiple formats, languages and sources – to be searched and analysed by patent to identify the most relevant information has made the process of carrying this out manually overwhelming and laborious. AI provides benefits to patent searching in terms of time saved and costs saved, while providing accuracy and assuring quality. Instead of prior art research taking between two and four days, AI can do all this legwork in just a couple of hours, thus giving the searcher a head start when it comes to manual analysis. Semantic-based algorithms can help a searcher to find the most relevant documents regardless of the terminology or language used as they focus on digging out information and performing document retrieval by identifying the meaning in accordance to an appropriate context. AI-based patent search tools are powered by neural network analysis and semantic vectors. Incorporating a language model and a literature knowledge graph will result in more relevant results with less noise or junk and fewer false positives, instead resulting in accurate, complete and deep data for market insights. AI provides a more domain-specific and nuanced understanding of the words that users are typing in along with the user’s query and context.

IBM is offering Watson, an IP advisor that leverages AI for fast patent ingestion, better insights, and analytics. IBM Watson is capable of analysing massive amounts of unstructured data that is input wherein the data is in several languages and is from several sources. Patent software companies are now relying on AI to produce advanced patent tools using the hybrid approach of incorporating the features of conventional patent tools with AI capabilities.

Even patent offices are hunting for best possible AI solutions to handle the ever-increasing volume of the literature and to increase efficiency for examiners. AI and machine learning also analyse the behaviour of the searcher, compare and include competitor and market information with patent data and thus, incorporate market and business information in the analysis. AI-based bots can be used to extract clean and structured data from any website with ease.

AI initiatives at the USPTO and the EPO

The USPTO is utilising AI to help examiners to review pending patent applications by augmenting classification and searches – currently a high priority – with it. The office has also explored using AI for search expansion and refinement, assisting with patent classification tools and locating similar images. The most promising of these AI capabilities have been identified and are being prioritised for inclusion into the search system in order to pilot them with examiners.

Meanwhile the EPO has been developing business solutions using AI and machine learning for patent searches at various degrees of implementation, including the automatic generation of search queries and automatic searches for prior art for patent applications that are to be examined.

AI and machine learning may not answer every question asked by a patent searcher, however, these applications are easy to integrate with conventional manual systems and tools. In addition, they can be used to reduce the patent sample space, giving, for example, the top 50 patents to analyse from an input of a larger set of patents or identifying the top 10 patent prior art references, excluding results have found previously. IP solutions based on AI and machine-learning techniques look certain to be a significant part of the patent industry tools going forward.

Amit Goel, Prashant Singhal, Raj Kishore

Effectual Knowledge Services Pvt Ltd

This article first appeared in IAM. For further information please visit