The Search Tuner and Pega GenAI™

In this example we learn how you can use the Search Tuner to fine tune the search results of your customers and CSRs, and we see an example of how generative AI can help article creators through assisted authoring.

But first, let us see how the Search works. Let us assume you are searching for a commonly used phrase, for example "I have lost my credit card". The search engine analyzes the search string, meaning the phrase that you are searching for, and extracts one or more keywords from it.

The keywords are then checked against the following predictive models in Prediction Studio:

1. Skipwords, which are ignored during the search.
2. Categories, which link the keyword to a certain category in your taxonomy structure.
3. Tags, which link the keyword to certain tags that are assigned to your knowledge articles.
4. Replacement words, which the predictive model replaces.

There is one additional predictive model that impacts search results. This model helps you by assigning tags (through the auto-tagging feature) to your help articles during their creation.

If any of the extracted keywords link to a category or a tag, the search then goes through all articles containing the same category or tag and returns them as search results. Extracted categories are shown in green squares, and article tags in grey squares.

In our example, you notice that if you search for "Somehow I have misplaced my credit card" the system extracts tags and categories linked to the credit card keyword, but it does not bring up any articles tagged with cancel card, which it did during the previous search. This means that if a customer uses "misplaced" instead of "lost" during their search, they are not directed to the appropriate articles.

So how can you make sure that your customers have a better search experience?

You can approach the situation with two separate solutions. Firstly, you can link the keywords to one or more categories in your taxonomy structure. Secondly, you can link the keywords to one or more article tags. It is possible, but not mandatory, to apply both solutions. As part of this business example we link the keyword to a search tag, however you decide which solution is the best one according to your business needs.

Using Prediction studio, locate the predictive model kmsearch_tags and update it. After the update, the search now extracts the tag or tags that you linked to the keyword, in our example cancel card, and boosting articles that have that tag to the top of the search results. If you create a new topic in kmsearch_tags make sure to first create the equivalent topic in kmarticle_tags, and regenerate article tags. Otherwise, the tag is not assigned to any article, and therefore it cannot be found in any search.

Let us return to the Search Tuner and see how updating the predictive model impacts the search results. Observe that article tags related to cancelling cards are now extracted from keywords in the search string.

You can further refine search results by applying boosts in the Search Configuration page, based on one or more article attributes.

While going through the search results however, you notice one article where the formatting does not look quite right. By clicking Review content, then Edit content, you can directly change the content body and fix the issue in formatting. The content source is available for you to inspect.

Through the AI options button, you access assisted authoring with the help of Pega Gen AI. This provides you and other content authors with an easy way to reword the title, abstract, or in this example re-format the article without changing the text. If you are satisfied with the result, you can save the changes to the article, and preview how it looks to a reader.

You have now reached the end of this video.