Korea Intellectual Property Office opens 9.61 million AI training items

The Korea Intellectual Property Office said Wednesday it has opened three new sets of intellectual property records for artificial intelligence training, about 9.61 million items in total, free of charge through its KIPRISPlus service.

The datasets are patent summary learning data, chemistry and metals experiment learning data, and product classification model learning data. The release is intended to spare companies the expense of assembling such material themselves, since intellectual property records are dense with technical terms and complex document structures that are costly to convert into training-ready form.

Patent summary data covers about 16,000 items, pairing original patent specifications with expert-written summaries of the problem to be solved, the main claim and the effect of the invention. Chemistry and metals experiment data, about 3 million items, is drawn from experiment result tables embedded in patent documents as images or text, converted and structured using optical character recognition. Product classification data, about 6.6 million items, links the names of goods and services listed in trademark applications to the similarity classification information used in examination, and can support AI models for automatic product classification, similar-product searches and analysis of market and filing trends.

With the three additions, the office has now opened 22 categories of AI training data, roughly 57 million items. The office said the release is significant because records built and accumulated through administrative work are being converted into training data and returned to the private sector.

"The foundation of artificial intelligence competitiveness depends on how easily high-quality learning data can be secured and used," said Jeong Jae-hwan, director-general of the office's intellectual property information bureau. He said the office would keep identifying and opening high-quality intellectual property data that companies find difficult to build on their own, and support the growth of new services and industries built on that data.

The datasets are available free of charge on the KIPRISPlus website, and inquiries can be made to the KIPRISPlus help desk at 02-6915-1553.