VISDOM Services

Taxonomies and Training Data

The taxonomies and training data page has been replaced in Contract Insight with a version that is connected to VISDOM Services.   This article provides examples for creating and viewing records.  

To access the VISDOM Services taxonomy and training data pages first navigate to the VISDOM configuration pages using the main menu options Manage/Setup > Visdom Configuration and then use the Training Data option located in the side menu.

The following page contains a list of the System defined VISDOM taxonomies as well as any user defined taxonomies that have been migrated.  This page also contains an area for a new object called Virtual Taxonomies.

There are several new taxonomies available in VISDOM Services that were not present in previous versions.  These taxonomies target specific contract types.  New taxonomies and virtual taxonomies will be added regularly and will be automatically be available in Contract Insight applications without the need for updates or patches.

Click on any of the tiles to view the associated clauses.  

User Training Data and System Taxonomies

In the previous version, users could add additional user defined training data to system defined taxonomies.  The system defined and user defined training data would be combined when the application trained the data to generate the statistics used to identify unknown clauses.

This feature has been removed from versions supporting VISDOM Services.  A replacement “Suggest a Clause” feature will be added in the future where user defined data can be submitted and reviewed prior to being applied to the training data.  

The reason for this is that in versions prior to VISDOM Services, all training data, both system and user defined, resided in the clients database.  This meant that each client had a copy of the system defined training data which they could then add to.  

In the VISDOM Services architecture all clients are using the same copy of system defined taxonomies.  Allowing potentially inappropriate training data to enter the repository unchecked could affect the service’s ability to identify clauses which would then affect all clients.  This does not apply to user defined taxonomies.

Efforts have been made to improve and expand the system defined training data.  The goal is to provide training sets which are robust so that clients find no need to provide additional samples of system defined clauses.  Training data will be continually updated and added for all system defined categories.  

System defined taxonomies are clearly marked as read only.  Menu options to edit the data are removed, and efforts to navigate directly to editing pages with a system defined ID will result in an error.

A list of System Defined Taxonomies is available HERE.

A list of Clauses that VISDOM Services can recognize is available HERE.

User Defined Training Data

Taxonomies

Users may create user defined taxonomies and clause categories.  

The minimum requirement for implementing a taxonomy are:

•    The Taxonomy must contain at least two clause categories

•    Each clause category must contain at least two different training data samples

To create a taxonomy, use the Add Taxonomy option on the Taxonomies page

Provide a name a description for the taxonomy and click save.    

Once the record is saved return to the taxonomy list and the newly created taxonomy will appear as a new tile in the Taxonomies area.

Clause Categories

To add clauses to a taxonomy first the select the user defined Taxonomy to edit by clicking on the tile from the Taxonomies page.  This will load the clause list page where you can select the Add Clause option from the side menu.

Enter a name and description for the new clause and click Save.  

Once the record is saved return the Clause Category list page and the new category will be displayed in the grid.  Note the additional Edit, Data and Delete options that are available on user defined data.  

Contract Insight will display a warning message at the top of the page when there are an insufficient number of clause categories.  The minimum requirement is two.  

Training Data

To add training data to a clause, click the Data button in the grid next to the clause you want to update.  This will navigate to the Training Data List page where you can use the Add Training Data option in the side menu.

Enter the training data and click Save.  The training data will be updated and the taxonomy will be retrained.  Navigate back to the training data list and the newly added data will now be in the grid.

Virtual Taxonomies

Virtual Taxonomies are a new feature available with VISDOM Services.  Virtual Taxonomies are a way to combine clauses from several different Taxonomies to improve classification results. Virtual Taxonomies were added to address the issue of misclassifications that can occur when classifying with different taxonomies using multiple passes.  

Misclassification issues can occur when classifying with multiple taxonomies because the classification engine must choose a probability for each category.   This is because VISDOM does not predict if an unknown text is or is not a particular clause, instead it predicts which clause in a given set is most similar to the unknown text.

For example there may be two taxonomies, each containing three clauses

Next we take some text that is most similar to Clause A. and have VISDOM evaluate the text against Taxonomy 1 and then again against Taxonomy 2.   This might produce the following results.

What we find is that Taxonomy 2 has produced a high probability result even though the text is most similar to Clause A. The reason that Clause D is rated so high is because Clause D is the best choice  within Taxonomy 2. The problem this creates is that it’s difficult to know which Taxonomy is correct, so to get around this we can combine the taxonomies so that all six clauses are evaluated in one pass.

This approach might give us a result that looks more like this:

In this result, it’s clear that Clause A is the best fit.

A good design for Virtual Taxonomies is to try and match the Contract Type being processed.  There are a few system defined Virtual Taxonomies that come preconfigured for Employment Contracts and Confidentiality agreements and Lease Agreements.  More will be added as the training data is improved.

You can create your own Virtual Taxonomies that combine system defined clause categories with user defined categories.  Using Virtual Taxonomies you can also exclude categories that are less likely to be present which can reduce false positives.  For example, you may get better results by excluding the Force Majeure clause when processing Confidentiality Agreements.

Viewing Virtual Taxonomies 

To view the clauses associated with a Virtual Taxonomy, click on a Virtual Taxonomy tile in the Taxonomies page.  System defined Virtual Taxonomies are also documented on the public pages of the Visdom as a Service web site. 

System Defined Virtual Taxonomies are read only.  A reminder of this is displayed at the top of the page.

A list of System Defined Virtual Taxonomies is available HERE.

Creating Virtual Taxonomies

To create a new Virtual Taxonomy, click Add Virtual Taxonomy on the side menu

Provide a name and optional description for the new virtual taxonomy and click Save.

Follow the prompt back to the Taxonomy List to see the new Virtual Taxonomy in the list.

Editing Virtual Taxonomies

Click the tile of the Virtual Taxonomy to edit from the Taxonomies list, then select the Manage Clauses option from the menu

The manage clause list is modeled after the Fields by Type editor and should be familiar to most users.  Use the Add and Remove links to manage the clauses that are associated with the Virtual Taxonomy.

Training

Unlike user defined Taxonomies, user defined Virtual Taxonomies are not retrained automatically when a change occurs.  This is because Virtual Taxonomies may contain system defined categories which will consist of many training records.  The combination of the larger training data and the ease of adding and removing categories from Virtual Taxonomy makes the decision of when to train a better choice for users.  To train the data use the Train Virtual Taxonomy option in the page menu.

New Algorithms

Two new algorithms are now available with Visdom as a Service.  Both were added to address problems that have been encountered in the past that were difficult to solve with existing algorithms.

Data Points

Data Points are specific items that VISDOM can locate or calculate.  As of this writing the Data Points are documented on the VISDOM Services website HERE.

You can map data points to fields in Contract Insight using the Data Point algorithm. Select the Data Point algorithm from the list and the edit the algorithm and choose the data point you would like to use.

One field that can benefit from Data Points is the Contract End Date which has proven to be difficult to capture with current VISDOM methods. VISDOM Services will return the date if it is stated explicitly or more commonly when the length of the agreement is found in the Term clause.  In the latter case the End Date is calculated by extracting the effective date and adding the Term.

VISDOM Services locates the Effective Date in the Preamble

VISDOM Services locates the contract length in the Term clause

Both Effective Date and Term Length are returned as Data Points and an additional value of End Date is calculated and returned as a Data Point.

Contract Insight can now map both the effective and Expiration Date fields to the Data Points with no additional configuration needed.

New Data Points will be added over time and like, taxonomies and categories, will be automatically available to Contract Insight without the need for updates or patches.

Time Spans

A new time span algorithm has been added to improve Visdom’s ability to set the term dropdown.  The values in a time span field can be difficult to map to the text values that are retrieved from clauses. This algorithm converts the potential and located values to a common format so that they may be compared.

Map a dropdown field containing timespans to a VISDOM Data Point in the ISO-8601 timespan format.

The system will convert the potential values of the dropdown to the ISO format.

The extracted value is also converted to the ISO format.  

The system can now compare the two values and select the correct Dropdown item