Document Understanding

2024 年 5 月

General Release Notes - Document Understanding

5 月 17 日

新增功能

In this latest release, we've made substantial enhancements to the performance of the Generative Extractor, specifically resolving these noted issues:

Deteriorating performance when asking more than 10 questions.
Variations in responses to a particular question whenever a new question is added or an existing one is modified.

Additionally, we've boosted the performance and precision of all our generative AI features, including Generative Extractor, Generative Classifier, and Generative Predict.

5 月 16 日

改进

For improved consistency, we updated the icon for the Upload button from the Build section.

5 月 13 日

发现 API 中的生成式提取程序和分类器

我们很高兴地宣布，发现 API 现已返回生成式提取程序和生成式分类器。您可以使用此 API 中的生成分类器或生成提取程序来处理页面。

有关更多信息，请查看 Document Understanding API 指南中的使用发现 API 部分。

提高通过 API 提取文档数据时提取数据的可信度

A new parameter, boostExtractionConfidence, is now available. This parameter is used to boost the extraction confidence when extracting Document Data with APIs.

For more information, check the Use the Extraction Service section from the Document Understanding APIs user guide.

5 月 9 日

改进

为提高清晰度，我们在创建新项目时更新了每种项目类型的说明。

一般发布说明 - Document Understanding 经典服务

5 月 17 日

新增功能

In this latest release, we've made substantial enhancements to the performance of the Generative Extractor, specifically resolving these noted issues:

Deteriorating performance when asking more than 10 questions.
Variations in responses to a particular question whenever a new question is added or an existing one is modified.

Additionally, we've boosted the performance and precision of all our generative AI features, including Generative Extractor, Generative Classifier, and Generative Predict.

5 月 16 日

错误修复

We fixed a bug in Document Manager for classic projects which was causing differences between the score of the training pipeline and the performance obtained in a running automation. This affects in particular low diversity datasets (with under 10 layouts) in both Automation Cloud™ and Automation Suite. This fix will also be included in the June Cumulative Update for Automation Suite version 2023.10.

To benefit from this fix, the only action you need to take is just to export your dataset again from the Document Manager session and then follow your normal process for training, testing, and deploying your model.

5 月 13 日

发现 API 中的生成式提取程序和分类器

我们很高兴地宣布，发现 API 现已返回生成式提取程序和生成式分类器。您可以使用此 API 中的生成分类器或生成提取程序来处理页面。

有关更多信息，请查看 Document Understanding API 指南中的使用发现 API 部分。

提高通过 API 提取文档数据时提取数据的可信度

A new parameter, boostExtractionConfidence, is now available. This parameter is used to boost the extraction confidence when extracting Document Data with APIs.

For more information, check the Use the Extraction Service section from the Document Understanding APIs user guide.