Korea Deep Learning Renews 'DEEP Agen,' Ushering in an Era of Unlimited Documents
TECHWORLD ·
✦ AI Summary
Korea Deep Learning announced a renewal of its document AI solution, "DEEP Agen (DEEP Agen)."
The key is removing restrictions on document types and allowing customers to directly configure changes in new templates, extraction fields, and business standards.
DEEP Agen handles document classification, information extraction, and review, while DEEP Ops supports adding new documents, changing extraction conditions, quality evaluation, and version management.
Korea Deep Learning announced a renewal of its document AI solution, "DEEP Agen (DEEP Agen)." The core of the renewal is the removal of restrictions on document types. The move is aimed at eliminating limits on usage stemming from fixed document templates established at the time of deployment, and at freeing document AI use from the templates defined when the system was built.
In the renewed DEEP Agen, client companies can directly change settings when a new template appears. They can also directly change settings when extraction fields change, and when business standards change. These setting changes allow the scope of use to expand.
Korea Deep Learning combined its document AI operations function, "DEEP Ops (DEEP Ops)," with "DEEP Agen." The roles are divided between processing functions and operational changes. DEEP Agen handles document classification, information extraction, and review, while DEEP Ops supports the addition of new documents, changes to extraction conditions, quality evaluation, and version management.
Korea Deep Learning said that with conventional document AI, adding new templates or business tasks often required separate development, verification, and redeployment. It added that DEEP Ops allows customers to carry out those steps themselves. Through this structure, Korea Deep Learning said it plans to reduce costs associated with repeated deployments and shorten implementation timelines.
The company has expanded automation at the document input stage, from document classification and separation to information extraction and personal data protection.
It can process multiple documents mixed in a single file, such as business registration certificates, contracts, and bankbook copies. It automatically identifies document types and determines the start and end points of each document. Based on those determinations, it can separate documents and process documents that were not pre-registered.
It can extract the required information based on document structure and content, and when processing tables, it preserves row and column structures. It also provides confidence scores for each cell. Personal information such as resident registration numbers and contact details can be masked from the extraction stage depending on settings, and user-defined masking fields can also be configured.
To enable direct responses when business standards change, the company has made it possible to modify extraction fields and conditions in natural language in DEEP Ops. The results of those modifications can be checked immediately. It also provides a function to apply settings before and after changes to the same evaluation data and compare accuracy.
It can also manage setting-change histories, and version management is available. User-specific permission management is also possible, allowing it to respond to continuously changing tasks such as insurance review standards, financial verification items, and public-sector forms.
Korea Deep Learning CEO Kim Ji-hyun said the goal of introducing DEEP Agen and DEEP Ops is to turn the process of redevelopment and verification that occurs whenever new documents or business standards are added into a day-to-day operational area that customers can handle directly. She added that the company will move beyond OCR accuracy and build a document intelligence environment that can continue expanding without restrictions on document types.
Source: TECHWORLD · Lee Gwang-jae
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407045
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.