Document Classifier

Overview

VisualVault’s Document Classifier feature allows VisualVault implementors to create and train their own machine learning models, which will be used to automatically assign the Document Type index field.

Objective

The Document Classifier leverages an AI learning model to automatically assign the Document Type index field during a bulk upload or bulk scanning process. This process helps prevent human errors associated with incorrect designation, as the AI is trained on large volumes of documents specific to the customer.

When uploading a small batch of documents, a user may assign index fields manually for the batch; for large-scale bulk uploads containing a mixed set of document types, manual batch indexing becomes time-intensive and costly. The document classifier is VisualVault's solution to this issue by automatically sorting and assigning document types accordingly.

Description

Key Functionality

• AI-Powered Classification: Uses an AI learning model to automatically assign the Document Type index field.

• Bulk Upload Support: Operates during bulk upload or bulk scanning processes.

• Error Reduction: Minimizes human errors in document type designation.

• Customer-Specific Training: AI is trained on a large volume of customer-specific documents for improved accuracy.

• Efficiency in Large Uploads: Eliminates the need for time-consuming and costly manual indexing during large-scale uploads.

• Automatic Sorting: Automatically sorts and categorizes mixed document types within a bulk upload.

Quickstart Guide

New to the Document Classifier? Use this guide to learn the basics.

Walkthrough

Prerequisites

• Determine which file types need classification​

  • All files in the customer database of the specified types will be classified.​

  • PDF and PNG are supposed file types.

• Training Data​

  • You must have real-world sample files that have been manually labeled (Document Type value set). ​

  • Training data must be uploaded or imported into the customer database’s VisualVault doc library with the Document Type index field specified (labeled).​

  • Recommend a minimum of 100 sample files per Document Type. This is only a guideline (more is always better)​.

  • Upload documents to a folder in the Document Library. These documents' content will be used for training the classification machine learning model. Supported file types include PDF, DOCX, XLXS, PPT, TXT, and TIFF formats.

• Index Fields​

  • Best Practice is to use the system generated “Document Type” field when configuration classification. ​

  • Create a numeric Index Field named “Confidence” to store the auto-classification confidence level for each Document that is classified.

Create a New Classifier Model

To create a new Classifier Model:

1. Navigate to Process Design Studio, from the Enterprise Tools tab of the Control Panel.

   

2. From the left menu, select Document Classifier (under Analytics). Click New Classifier Model.

   The New Document Classifier Model window will open.

   

Create/Edit Document Classifier Model

In order to edit a Model, go to the Document Classifier screen and click on the item's Edit button.

1. Fill in the Model Name

2. Fill in the Model description

3. Select the Document Library folder(s) where the training data is located.

   1. See Prerequisites.

4. Select the Index Field “Document Type”

5. Select the Input Type

   1. Image and Text: When the layout is consistent for every doc type, then select the “Image and Text” option.

   2. Text: When there may be varying layouts for the same doc type, then select the “Text” input. When unsure, “Text” is the better default choice.

6. Click Save. The Created Classifier should now appear on the list, with its status set to Initializing.

7. Click Train

Evaluate Classification Quality

1. Review the model’s calculated Threshold value.​

   

   1. Threshold value = Minimum Document Confidence Level​

      1. This is a value between 0 and 1 representing a percentage. The threshold should be greater than or equal to the confidence level. If low confidence, a new model with additional training data or accurate Doc Type labeling is required.​

   2. Upload a sample of customer documents that were not part of the training data and confirm the confidence level is >= Threshold value.​

   3. If low Document confidence, a new model with additional training data or accurate Doc Type labeling is required.​

2. Review the Confusion Matrix Chart​

   

   1. Diagonal representation of Document Types is ideal​

   2. Look for the “True Label” to match the “Predicted Label”.​

   3. If multiple mis-matched Doc Types, a new model with additional training data or accurate Doc Type labeling is required.

3. Review the Scatter Plot Chart​

   

   1. Tight cluster of points = high accuracy​

   2. If you cannot “draw a circle” around a Document Type without selecting another Doc Type, a new model with additional training data or accurate Doc Type labeling is required.

Delete Document Classifier Model

In order to delete a Model, go to the Document Classifier screen.

1. Select an item from the list.

2. Click on the Delete icon in one row, or select multiple items on the list and then click on Delete Model(s) button.

User Groups and Permissions

Need help? Please contact your organization’s admininistrator, for troubleshooting. If the issue remains unresolved, VisualVault Support is available to assist.