Korea Deep Learning Renews 'DeepAgent,' Strengthens Document AI Operations

Korea Deep Learning announced on the 16th that it has completed a renewal of its document AI solution, DeepAgent. The renewal focused on strengthening operational functions to respond to changes in document types and business rules.

The core of the renewal is reducing the need to repeat separate development and redeployment every time a new document form or extraction standard is added. To that end, Korea Deep Learning carried out the DeepAgent renewal by integrating DeepOps, its document AI operations function.

The roles have been structurally separated. DeepAgent handles document classification, information extraction, and review, while DeepOps handles new document registration, extraction condition changes, quality evaluation, and version management.

As a result, customers can add new documents directly in DeepOps and also change extraction items and conditions themselves. This structural division of roles helps customers manage document and business rule changes that occur after deployment on their own.

This function automates the document input process and treats mixed documents such as business registration certificates, contracts, and bankbook copies within a single file as processing targets. The system first identifies the document type and then automatically separates documents by recognizing each document's start and end points.

It then operates in a flow that groups or splits documents again depending on the number of pages or items, as with contracts. Multi-page contracts are grouped as a single document, while multiple contracts in one file are split and processed by case.

In the next step, information can be extracted based on structure and content even for documents that were not pre-registered. During extraction, the row and column structure of tables is preserved, and the confidence level of each cell is also provided. In addition, personal information such as resident registration numbers and contact information can be masked during the extraction stage depending on settings, and the items to be masked can be designated separately.

If business rules change, extraction items and conditions can be modified in natural language in DeepOps. For example, a standard that previously extracted the entire contract period can be changed to extract only the contract start date, and the revised result can be checked immediately.

Through the renewal, Korea Deep Learning made it possible to compare accuracy before and after setting changes, using the same evaluation data to compare results. It also said it can manage configuration histories, version management, and user-specific permissions.

The company said it provides the original text location along with items that require additional review during inspection. It also said that if the person in charge corrects a value, the corresponding result data is accumulated and that the accumulated data can later be used for quality evaluation.

Korea Deep Learning plans to expand the scope of DeepAgent based on this renewal. The expansion target covers document operations across finance, the public sector, and enterprises.

Kim Ji-hyun, CEO of Korea Deep Learning, said the goal is to shift the process of redeveloping and revalidating every time new documents or business rules are added into a daily operating area where customers can respond directly, through DeepAgent and DeepOps. She also said the company aims to build a document intelligence environment that can continue to expand without limits on document type, beyond simple OCR accuracy.

Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407015

References

• https://www.epnc.co.kr/news/articleView.html?idxno=407015