Knowledge Search Example: Simple Search

Following is an example of a simple search through Search Widget and Luma Virtual agent (Luma Supported Channel). In this example, the end-user is looking for information on resetting their Gmail account password. The search results in a list of relevant knowledge artifacts.

Knowledge Search in Luma Knowledge is based on the following:

Tenant Configurations

Luma Knowledge uses the following tenant configurations to identify Artifacts for a user query. For example let us consider the following configurations:

Relevancy Score

The Relevancy score of an Artifact determines how relevant is the Knowledge available in the Artifact to the user’s question. The relevancy score is automatically calculated by the system based on the availability of terms in the search phrase, term frequency, and other parameters. Luma Knowledge uses the score to filter artifacts that are not relevant to the user’s query.

In our example, the user searches for “How to reset account password.“

1. Luma Knowledge identifies 3 matching articles and calculates respective relevancy scores:

   • A1: “Change or reset your Gmail password” with a Relevancy score of 12.78

   • A2: “Reset a user's password using Azure Active Directory” and a Relevancy score of 11.42

   • A3: “Send password reset email for an Enterprise ID user” and a Relevancy score of 11.14

   • A4: “Create a new account in Azure Active Directory” and a Relevancy score of 5

2. Based on the tenant configuration Score Range Percentage, Luma Knowledge filters the Artifacts. All the Artifacts that fall in the Relevancy Score range, i.e., between the highest relevancy score for the matching artifacts (12.78) and the calculated Relevancy Threshold (7.66), are considered relevant Artifacts.
   Relevancy Threshold is calculated as the Score Range Percentage (tenant configuration) of the highest relevancy score (60% of 12.78 = 7.66).

   The search result is displayed as:

   Open

   

Confidence Band

The Confidence Score of an Artifact indicates how confident the system is in the identified artifact to help the user with their inquiry. This is calculated based on the metadata generated from the user’s query and metadata of the Artifact. The query’s metadata is matched against the Artifact's metadata available in Knowledge Base. If at least one phrase/value in the metadata type is matched, the respective score is awarded to the Artifact as the Confidence score. For more details on calculating Confidence score, refer to Calculating Confidence score.

In the current example,

1. User searches for “How to reset account password.“

2. Luma Knowledge calculates the confidence score for the matched Artifacts (identified as per the relevancy score):

Now, based on the following tenant configurations Upper Confidence Score Threshold value for LOW Confidence level and Upper Confidence Score Threshold value for MODERATE Confidence level , the matched Artifacts are classified into Confidence Bands. The Artifacts are classified as:

Now based on the Confidence Band, Artifacts are displayed to the end-user. As per our configuration, the Artifacts in High and Moderate bands should be displayed in the search result and Artifacts in Low Confidence Score should be ignored.

Exclude Articles with Low confidence from search results is set to No
Exclude Articles with Moderate confidence from search results is set to Yes

Open

The final Artifacts displayed on the Search Widget are:

Additional Factors for search through Luma Virtual Agent

When the Artifact is searched through the Luma Virtual Agent or any of the Luma support Channels, the following factors play a role in the presentation of Artifacts:

User Intent

User intent is derived from the search phrase. It indicates if the user wants to view Knowledge or execute the predefined service.

For example:
“How to order a laptop” indicates that the user wants to view Knowledge Artifacts on how they can reset the password.
“Create a Service Request” indicates that the user wants help with a service that can reset the password.

For the search “How to reset account password“, user intent or pre-qualification is KNOWLEDGE.

Services on Luma VA

The user’s search request is also used to search predefined services available in Luma VA. Based on the Luma tenant configurations, the Virtual Agent identifies matching services and calculates the skill(s) confidence level. Services are presented to the users along with the Knowledge Artifacts if matching services with High or Moderate Confidence Levels are found. For more information on identifying Skills in Luma VA, refer to NLP Settings.

In the current example, let us consider that no pre-defined services are available in Luma VA. So, the Services’ confidence level is LOW.

Result Presentation preference in Luma VA

Based on the above factors, the system recommends if the end-user should be presented with Knowledge, predefined Service, or both. For our search, the system recommends displaying Knowledge Followed by Skills. For more information on rules for Knowledge and Services display, refer to Integration with Luma Knowledge.

However, based on your organization’s Presentation Preference, the Luma VA administrator can customize the way Knowledge and Services are delivered to the end-user. You can configure Luma VA to always deliver Knowledge followed by Skills, Knowledge & Service together or honor the DE recommendation. The configuration overrides the DE recommendation. For more information, Knowledge & Services Settings.

In the current example, let us consider that the Presentation preference for your Luma tenant is Default. This means the DE recommendation is honored.

Based on the above factors, Luma VA should present Knowledge Artifacts followed by Identified Skills to the end-user. Since there are no matching skills, so only Knowledge Artifacts will be presented.

In the Luma Virtual Agent, the Artifacts are displayed as shown below: