Configuring Auto-attribution for Pega GenAI Knowledge Buddy

In this video, you explore how to configure a prediction model, how auto-attribution applies to content or specific chunks within that content, and how to make a Knowledge Buddy apply auto-filtering when it provides an answer.

You must have access to Prediction Studio to configure prediction models. Within the Knowledge Buddy application, by default, only users with the KnowledgeBuddy:Admins access group have the required access privileges.

After you log in to Prediction Studio, click Models.

To create a new model, click New in the upper-right corner. If you want to create a completely new model or update an existing model, you must have an open ruleset to save it into.

To update a model, in the Model column, click the name of the model you want to update – in this case, Content attributions.

Click Update, then click Update again.

In the Topics column, click the topic you want to update. If you want to add a new topic, you click Add topic. However, in this example you are making changes to an existing topic, specifically Region, so click Region to expand the list. Like adding a new topic, if you want to add more regions, click Manage > Add child.

Click United states.

Within the topic, you create the taxonomy structure according to your business needs. In our example, notice the Should words field that contains the following terms: United States, USA and America. Any time an author includes one of the Should words in content, Knowledge Buddy attributes that content to the United States region.

To add more Should words, click the Should words field after the last should word, then type the new Should word and press enter on your keyboard. Then, in the top-right corner click Save.

You can set up a model to handle multiple attributions or you can configure multiple models, each specialized for a single purpose. Make that decision according to your business needs and best practices.

Let us return to the Knowledge Buddy Portal.

When you create a collection, click Advanced settings to configure auto-attribution for that collection.

In the Content level attribution list, select the Model that that particular collection will use for content level attribution. You can only select one model for each attribution type.

Content level attribution handles auto-attribution for whole content articles created on the Pega Knowledge Buddy portal. Chunk-level attribution does the same, but for each individual Chunk from within content articles. Lastly, Auto-filtering attribution is used to search for specific content based on the attributes of that content.

With Auto-attribution configured correctly, when an Author or other user creates content in the Pega Knowledge Buddy portal, the prediction model will detect any of the Should words and apply the appropriate label and Value to the global attributes of that content.

In this example, if you click Global attributes after you submit the content for ingestion, you can see a Label of Region, a Value of USA and, in the Generated column, the indicator Auto which confirms that this content was auto-attributed.

If you click Chunks, you can see how the chunk-level attribution works: in the Attributes column, the prediction model has correctly attributed Region:USA to that individual chunk.

Keep in mind that global attributes must apply to the entirety of a content article. If each chunk in your content is attributed to a different region, which you can see on the Chunks tab, then the Global attributes tab will not show any auto attributes.

Now let us look at how auto-filtering works.

Auto-filtering applies to how buddies answer questions. When you create a Knowledge Buddy, during the Prompt stage, select the Enable Auto filtering to user request checkbox.

With the checkbox selected, when a user asks a question in which the prediction model detects one of the Should words, the Knowledge Buddy will only use articles or chunks that are appropriately attributed to build an answer. The Buddy ignores any chunks that don't have the correct attribute, even if an article contains multiple chunks.

Auto-attribution, through the use of prediction models, automatically labels any content you create and helps your organization remain better organized. With auto-filtering applied, it also helps refine the answers that a Buddy provides to users.

To summarize, in this topic you learned how to use prediction models to automatically categorize content based on your organization's unique taxonomy and business requirements. The system's ability to handle both content-level and chunk-level attribution ensures comprehensive coverage, whether you are dealing with broad topics that span entire articles or specific details contained within individual content chunks.