Summary
Convert document to carne-value data and use the results to conduct your processing. In this video we integrate AI Document Analyzer with OCI Document Understanding V2.0 to identify documentary types and apply a specific extraction model to each.
Generative extraction allows defining in JSON every key, its data type and a description of the field. That's how it's adapted to different formats and to multilingual content in the languages supported by the service.
In the shown flow, the first stage analyzes the first page to distinguish minutes, contracts and prospectuses. The model returns a label. a condition of workflow relates to the original file and chooses the branch and the extractor. Then, the dedicated model processes the pages and gets their keys and values.
We reviewed the JSON of the generative model at OCI Object Storage, AI Document Analyzer installation with Oracle Autonomous Database, model creation and workflow design. Finally, we have been looking at the executions and monitoring module to check the classification and results of each document.
Start
Following the introduction, the carpeting-value of OCI Document Understanding V2.0 and its support for multilingual content are presented. The demonstration starts with a JSON definition of the fields and then integrates models into AI Document Analyzer to choose how each document will be treated.
Generative model V2.0 configuration
A document understanding project select the V2.0 and JSON file previously uploaded to OCI Object Storage. A name is assigned and model production begins. The following section details what information this extraction scheme contains.
Schema JSON: keys, types and descriptions
The JSON defines each key, its type and the description that guides the extraction. Once the model has been prepared, an output bucket and an PDF from the console are selected. The demonstration compares the definition of the fields with the results obtained, before passing to configure the application.
AI Document Analyzer installation and configuration
The installation uses input and output buckets and an instance of Oracle Autonomous Database with the application user. The Setup sets up the access with the walk and tests the connection. The OCI and storage configuration are then completed. changes should be saved to complete access to the platform.
Models for classification and carnation extraction - value
The model module first creates the one that will return the documentary type from its definition JSON. Three specific extractors are then prepared: minutes, contracts and prospectuses. The tour shows how to load, revise and edit the archives and how to consult the state production before using models in a workflow.
Creation of documentary workflow
A workflow with name and description is created and the starting nodes, document loading, document Understanding, conditions and completion are placed. The canvas prepares an initial common stage of classification and branches that will apply models other than loaded documents.
Ranked from first page
The first document understanding node uses the classification model and is set up to analyze only the first page. That stage returns the documentary type to be used to decide the branch of the flow. It does not currently conduct the specific extraction of all fields of the document.
Conditions and routing by document type
Each condition compares the name of the file with the loaded document and checks the label returned by the model. The branches are built for minutes, contracts and prospectuses, each with its extractor. In these nodes, the restriction to a single page is disabled: the model contributes the classification and condition of the workflow determines what extraction is run.
Performance of workflow and review of results
The flow is run with a batch of ten documents and following the order of loading, sorting, filters and extraction. At the end, the results table allows you to search for a file and highlight it to check how far it went. The output of the nodes helps to link the result with their document of origin.
Monitoring and monitoring by document
The Monitoring module allows you to filter and revise a specific document, download and check its classification. A document of risk factors identified as a package leaflet is found in the example. That result explain why his extractor of that category was assigned.
Extraction by page as per model assigned
The document listed as a package leaflet is processed with its specific model and the rules apply to each page. The closure relates this nail-value output to the workflow branches: the classification selects the treatment and the extractor gets structured information about the documentary type.