Using Atlas Taxonomy to refine content in IKS and refine responses in AI Assistant

*See all Atlas AI & IKS articles here*

A key differentiator in the Atlas Intelligent Knowledge Platform is the taxonomy (terms/tags) that provide a consistent string of context and understanding throughout your content no matter where it is placed.

This enables you to effectively organize your content and information into groups and categories.

This article will explain how you can leverage this feature through the AI tooling.

• How to leverage Terms & Taxonomy in Atlas AI
• In the AI Assistant (chatbot)...
• In the Intelligent Knowledge Studio (IKS)...
• Summary

How to leverage Terms & Taxonomy in Atlas AI

For the Atlas AI tooling, these are the supported metadata terms (in Atlas terminology):

1. Information Type
2. Location
3. Department
4. Entity
5. Activity
6. Subject

In the AI Assistant (chatbot)...

For more information on how to interact with the AI Assistant, please see this article.

A person can use the “tag” icon or the hash (#) key to access these term sets to limit the scope of how the AI responds and/or query against. Pressing the hash key brings up a new fly-out menu where you can view all of the tags which are in scope for the Knowledge Collection (KC) you are in.

You can then use the heading here to find terms from a specific term set, for example, Location.

Utilizing a tag this way can help direct the AI to more specific content i.e. it is only looking at people who have been assigned the tag Location = 'London Office'.

We can do a similar search but utilizing the Subject = Intellectual Property Law

Notes:

1. The All tab shows all terms across the six term sets in alphabetical order irrespective of term set.
2. The current order of term sets in the menu is fixed.
3. Only the parent node for each term displays underneath the term value, not its full path.
4. A term set can be used for multiple purposes, i.e., Language, Office location and Jurisdiction could all be in Location under those headers. Only collections that contains mixed content that has been tagged with all these values could potentially be confusing.   
5. The current term set headers are fixed to the Atlas headers.
6. Typing after the hash automatically filters the list based on what has been typed.
7. The type ahead after the hash finds any character in the header.
8. The type ahead does not search across synonyms or translations.
9. The terms returned only relate to the content in the collection. This is the same behaviour as search refiners (filters), so it is not possible to accidently select a value that is not relevant to the collection in use.
10. The items in the collection itself can also be used to refine the query, this means the title (filename) of each item is useful metadata as well.
11. Selecting the collection defines the content that will be used.
12. It is possible to use multiple terms in a prompt using the hash by applying the AND operator between them, e.g., Term 1 AND Term 2 AND...
13. It is possible to add multiple files with the slash to a prompt by applying the OR operator between them, e.g., (File 1 OR File 2...)
14. It is possible to add both terms and files by applying the same operator logic, (Term 1 AND Term 2...) OR (File 1 OR File 2...)

In the Intelligent Knowledge Studio (IKS)...

Within the IKS tooling - the back-end management area to set-up AI 'Chats', we can leverage the taxonomy in a similar way, i.e. to direct the AI to only look at content which has certain tags assigned to it.

For more information on how to set-up a KC within the IKS, please see this article.

When you set-up a KC and point towards the specific areas of content you want to bring in, by default the KC is going to look at everything within that specific area. Below shows the selection of the workspace 'AM - Global Gourmet Supplies', and the 'Documents' library and the 'Site Pages' library have been selected

Once confirmed, the choices look like the below, where the library or list is displayed in blue, and the site is displayed in grey beneath. You can select up to 1000 libraries and/or lists from the areas you have permissions to.

However, in this particular scenario, I do not want ALL the documents and ALL the pages, I only want some of them.

You can 'query' in KQL (SharePoint Search Keyword Query Language) to ensure that only documents which meet the query criteria are brought into the AI index for use.

In the below screenshot the KQL query filter we have applied is Subject = Areas of Law.

This will ensure that only Documents and Pages which have the Subject tag 'Areas of Law' is indexed, giving you control to only bring through certain documents. It could be a certain Subject, it could be a certain Information Type (such as Case Study, or, Policy'). Utilizing the KQL Query enables the KC Creator or IKS administrator to specify more granular types of content from across different areas.

You can of course use multiple queries with the subtraction '-' sign, and operators of AND or OR.

You can also press the 'test query' button to ensure your KQL Query is functioning and the expected amount of content is being returned to you. Below we are removing any content across both Document libraries or Pages in 2 workspaces where the subject = Areas of Law

This can be important as you only want to index the content which is relevant to your KC (AI Scope of Content). You do not want to bring in irrelevant material.

So if all of the material you want to bring into the KC and Index into the AI is in one or more document library, but is not ALL of that in the document library, you can refine accordingly without having to move or duplicate the content.

Summary

Being able to further refine content is important for a number of reasons. Providing more control over which content is indexed and which content should be asked questions of will improve the accuracy and quality of the AI responses.

For more information on the best ways to implement this best practice across your organization please get in contact with your designated Atlas representative today.