Using the Classification Analysis Tool (CAT)

A rich test interface is provided to help users understand how a particular model or rulebase template design can influence the classification server. This is the new default classification tool replacing the Classification Server Test Interface (Note this is still accessible through CAT).

The Classification Analysis Tool is a web based application and is accessible by using a modern browser and entering the URL, for example:

http://servername:5059

The following screen should be presented:

Navigating CAT

On the left hand navigation bar, there are 3 main navigation components in CAT which enable users to send requests to the classification server with an additional link to access the old Classification Server Test Interface.

Analyze - CAT default landing page. Starts the document classification workflow.

Test Interface - The Test Interface is the revamped, CSTI. It’s a lightweight graphical tool and tends to be faster than CAT.

Sandbox - Input and test custom rules and examine the classified output.

Submit XML - Simple flow to send XML request to Classification Server and view the XML response.

Rulenet - Allows you to make publish sets active or inactive. It displays the state of all publish sets currently loaded into CCS, so you can see which publish set is being used to classify the content.

Statistics - CCS keeps a record of what it has classified.

Submit a classification

Actions :

1. Add document source
2. Optional - Add metadata fields
3. Optional - Adjust classification settings
4. Send for classification

Specify the document

To test the classification, a document will be required. There are 4 ways to supply this classification source:

• [File tab] - The file from local disc can be dragged and dropped into this panel. Optionally clicking the drag and drop zone will open local file browser.
• [URL tab] - The document URL can be provided.
• [Manual tab] - The document can be typed in manually or pasted using the text area.
• [XML tab] - Custom xml request can be provided. Clicking [Generate XML] button will fill the text area with the xml request containing current form options. Important: custom xml request will be sent only when the XML tab is active

Add metadata to classification test run

The default metadata field is title. Additional metadata fields can be added by clicking [Add Metadata field] button.

A list of suggested metadata tags is as follows:

• Abstract
• Account
• Address
• Application Version
• Attachments
• Author
• Authorization
• Backup Date
• Base File Location
• Bill To
• Blind Copy
• Carbon Copy
• Category
• Character Count
• Checked By
• Client
• Company
• Completed Date
• Count Bytes
• Count Characters with Spaces
• Count Lines
• Count MM Clips
• Count Notes
• Count Paragraphs
• Count Slides
• Count Slides Hidden
• Creation Date
• Department
• Destination
• Disposition
• Division
• Document Comment
• Document Number
• Document Type
• Editor
• Forward To
• Group
• Heading Pairs
• Keyword
• Language
• Last Print Date
• Last Saved By
• Last Saved Date
• Links Dirty
• Mail Stop
• Manager
• Matter
• Minutes Edited
• Office
• Operator
• Owner
• Page Count
• Presentation Format
• Project
• Publisher
• Purpose
• Received From
• Recorded By
• Recorded Date
• Reference
• Revision Date
• Revision Notes
• Revision Number
• Scale Crop
• Secondary Author
• Section
• Security
• Source
• Status
• Subject
• Title
• Title of Parts
• Typist
• Version Date
• Version Notes
• Version Number
• Word Count

Classification settings
Adjustments to the classification request will yield varying results according to the settings and threshold. Clicking [Options] button will open the form below, where options can be edited.

• Article type (selection) - Setting “Article type” to “Multiple article” will result in a document divided by Classification Server into one or more articles.
• Language (selection)
• Threshold (value between 1-100)
• Result clustering (selection)
• Clustering threshold (value between 1-100)

Classification Results

Examine the classification results

The resulting screen should look like the screen above. The main body is occupied by the classified document output as processed and returned by Classification Server. The left panel contains by default the document level categories list used to select and highlight the evidence in the document body, by default all evidences contributing to the scoring of the categories will be highlighted.

If “Multiple article” option was chosen for “Article type” in classification options the document will be divided by Classification Server into one or more articles. This will show has some additional features like article list and article navigation.

There are numerous options available for classification examination. They fall into several categories described below.

Classification output options

Results (tab)
The main content of this panel contains the results from the classification test, this is a list of document level categories. For large of results, CAT offers 2 methods for filtering result sets:

• Search-As-You-Type filter - enter text in the input field to filter down the results in the table
• Drop down filter by class - to select which class to show to inspect a particular set of results

Categories list can be sorted clicking the column header.

Clicking any category row on the list will result in highlighting the evidence in the document panel.

Finder (tab)
This tab offers a search-as-you-type for a text in the document. When the word is typed in the search results are highlighted in the document and also the number of results is displayed below the search window. There are the buttons to navigate between the results.

Refine (tab)
Refine tab offers options to update the document classification. The options can be modified and re-submitted, the classification request will be resent to Classification Server which will respond with a updated set of results.

Article List (dropdown)
If “Multiple article” option was chosen for “Article type” the document get divided by CS into one or more articles. This dropdown offer a shortcut to navigate between articles in the document area.

View XML (dropdown)
Allows to examine XML request that was sent to Classification Server as well as the XML response.

Eye (button icon)
Shows the document fields and article related metadata. By default the fields are hidden unless results are triggered in the body by selecting from the category results.

Document (button icon)
Shows a modal popup with document level metadata.

Paragraph (button icon)
Toggles paragraph symbol marked in the document body.

Article level options.
When the document is divided by Classification Server into articles there are additional options available:

1. [Article Score and Metadata] button above each article.
   This opens a panel with the list of metadata and a list of categories results for the current article. The categories list is similar to the one for the document. It contains the “Show” button and clicking the button highlights the evidence for the article.
2. Article navigation
   Below the document panel there is a navigation bar that contains [Previous Article] and [Next Article] buttons.

Category examination

• Clicking any category on the list in the left panel will highlight the evidence for the category in the document and show the Evidence Key panel on the right.
• If there is no evidence for the category an alert will be displayed.
• Hovering over highlighted evidence will show an information about current category and annotation.
• Clicking the highlighted evidence will open the rulebase tree panel on the right of Evidence Key panel.

Evidence Key, Term type filter and annotations

Annotations are in principle, the rulebase tree labels and consequently the evidence labels coming from the ontology. They are automatically generated by Publisher, but can be also added by a template editor. Annotations fall into groups generally as direct evidence eg PT’s and NPT’s or contextual evidence eg associative relationships. Non-contributing evidence is the evidence that can be found in the rulebase tree leaves, but is not the active category evidence (was rejected by one of the rules on the way to the category node). Non-annotated contributing evidences are highlighted but will have no annotation (occurs when classifying with entity extraction rulebases).

Evidence Key panel contains a list of annotation groups which is a legend for coloring the annotations in the document. There are a few options available in the panel:

• Evidence navigation with [Prev] and [Next] button to iterate through the highlighted evidence.
• Show evidence tree which will show the category rule in the rulebase tree.
• Filtering highlighted evidence by clicking on annotation group.
• Optional toggler to highlight only contributing evidences or highlight all including non-contributing evidences.

Rulebase Tree

When a piece of highlighted evidence is selected in the document body, we can quickly identify why this rule has been highlighted and show directly where the rule is triggered in the ruletree.

The rulebase tree is opened in a context of selected evidence. This means that initially the rulebase tree opens up to the leaf node where the clicked evidence was found. The rulebase tree menu bar contains following options:

• Toggle display of inactive branches - Shows or hides branches of the tree that were not triggered during the classification process. By default those branches are hidden.
• A drop down list of the following; Collapse/Expand whole tree, Search tree window (marks the tree nodes containing search term) and a rulebase tree node icon key
• Cluster evidence - Contains a list of articles with score and [View] button that allows to jump directly to the article.

Tree node Clicking on the rulebase tree node results in:

• Redrawing the highlighted evidence in document area selected node being presented.
• Showing the Tree node information panel containing more detailed information about the node.
• Links - Some of the nodes are link nodes directing to another node. If the target node is in the same rulebase then the tree will be scrolled to the node and the target node will be selected. If the target node is in another rulebase then the current rulebase tree view is replaced by the rulebase tree containing the target node and the target node will be selected. Also the new rulebase tree will be opened in the context of the evidence that was clicked previously to examine the original rulebase. It is possible to return to the previous rulebase using the breadcrumb links displayed over the rulebase tree root. To follow these links you may use the context menu by “Right-clicking” on the node.

Sandbox

Provide custom rules and the document CAT also features a sandbox where you can test custom rules and accompanying document or text to test classification.

Actions :

1. Enter custom rules in the “Rule base” text area - also feature support for autocompletion of rulebase tags. For more help on rule base syntax there is [Documentation] link provided above the area.
2. Manually enter your classifying text/html or attach a document via the file upload tab.
3. Click [Test Classification]

The resulting screen show the classification result screen. In the document category panel there will be a category PARENT of class SANDBOX and this category will contain custom rules typed in the previous screen.

Submit XML

This tool is primarily aimed at advanced users - allows quick submission of xml request and see the literal response from classification server. Click here for XML DTD of requests to Classification Server.