The global patent system is under pressure from a rapid growth in patent application volume and complexity. As patent offices struggle to meet growing workloads, patent examinations can be delayed, impacting efficiency, putting patent quality at risk, and reducing customer satisfaction.
To combat this, patent offices are taking a number of approaches to improving the speed and quality of their patent examinations, including global IP harmonisation, hiring more examiners, and introducing technologies such as AI.
AI is being used to optimise prior art search workflows, patent classifications, machine translation, and other functions. The speed and accuracy with which AI can analyse millions of data sets reduces the time taken for reviews from hours or days to minutes.
Applying AI successfully in patent office workflows
A collaboration between CAS and the National Institute of Industrial Property (INPI) of Brazil yielded significant improvements in prior art searching using an AI-enabled workflow solution:
Up to a 50% reduction in examination times;
Reduced search times for over 75% of applications processed; and
Contributed to an overall reduction of 80% in the office’s patent backlog
Three important insights into how predictive solutions should be structured to optimise patent office workflows are below:
1. Clean, structured data
Using high-quality data to train AI models and to perform searches is crucial for identifying patents with similarities and adjacent patents to produce high-relevancy results.
However, most organisations use data that is unstructured, making it difficult to search and therefore less useful for AI. Publicly available scientific data can also include transcription errors, mislabelled units, and overly complicated patent language – as well as being published in any one of more than 60 languages.
These issues are especially challenging in fields such as chemistry and life sciences, where substances are often described inconsistently across publications and with key words embedded in tables or images, making the information unreadable by most search technologies.
High-quality data can resolve this challenge to improve the relevancy of results returned through AI-enabled prior art searching.
2. Multiple algorithms
Using structured data, multiple algorithms can be trained to find semantic, syntactic, connectedness, and substance similarities.
The INPI Brazil solution integrated four types of algorithms for text-based and substance-based analysis, including deep learning and term frequency-inverse document frequency. Using multiple algorithms allowed the AI to find multiple types of substance similarities all in one multifaceted solution.
Once the first-level algorithms had returned their similarity results, an ensemble algorithm analysed the results and arrived at a single, prioritised list of the publications that were most likely to be in conflict with the target patent for examiners to review.
3. Leveraging human expertise alongside technology
Pairing technology with a team with expertise in areas from data analytics and workflow integration to high-performance computing, cloud computing and scientific searches improves project outcomes.
For INPI Brazil, the CAS IP search team was able to support the examiners’ searches by validating algorithm results during solution development and by performing highly complex searches to augment the office’s capabilities. The combination of human expertise with expediting technology proved to be a successful formula.
Optimising for the future
Patent offices will continue to require approaches that enable them to meet stakeholder expectations while operating under operational and resource constraints. Flexible technological solutions will improve efficiencies and productivity and will be vital investments for the future.
Download the complete whitepaper here.
