Did you know that LivePerson has a Conversational AI forum for builders? Check it out here!

In Conversation Builder, a knowledge base is a great tool to answer questions about a variety of topics specific to your bot's area of expertise. One would typically add a knowledge base integration in a fallback dialog to provide simple answers to topics not covered elsewhere in the bot.

The Knowledge Base application in the Conversation Builder platform lets you create and manage knowledge bases and articles, which are then processed by an NLU engine. This allows the bot to leverage intents and entities to recommend the most accurate articles to an inquiring consumer. FAQ bots are driven by a knowledge base full of articles, intelligently delivering the right content to the consumer at the right time.

Access Knowledge Base

To access the Knowledge Base application

  1. On the left sidebar in Conversational Cloud, click the icon.
  2. In the Conversational AI dashboard, click Knowledge Base.

How a knowledge base works

Search modes

When you integrate a knowledge base with a bot via a Knowledge Base integration, you specify a "mode" for the search; this determines the type of search that is performed. Possible modes include:

  • Intents: First, a text-to-text search is performed; the search algorithm checks the consumer's input against the title and intent qualifiers (if used; intent qualifiers are applicable only when using Knowledge Base intents, not Domain intents).

    If a match isn't found by the first search, the Knowledge Base next uses Natural Language Understanding (NLU) algorithms to match the consumer input to articles. If it finds a match to a reasonable degree of certainty, it returns the article. The article attributes against which the NLU engine compares the consumer's input include the title, intents (either Knowledge Base intents or Domain intents), and the positive and negative learnings that the article has acquired.

    If a match isn't found by the NLU search, as a fallback, a final text-to-text search is performed. The search algorithm checks the consumer's input against the title, summary, detail, intent qualifiers (if used; intent qualifiers are applicable only when using Knowledge Base intents, not Domain intents), and tags in the articles.

  • Intents Only: This mode is like the Intents mode except that the final text-to-text search isn't performed as a fallback.

  • All: A text-to-text search is performed. The search algorithm checks the consumer's input against the title, summary, detail, intent qualifiers (if used; applicable to Knowledge Base intents only), and tags in the articles.

Thresholds

When the Knowledge Base uses Natural Language Understanding (NLU) algorithms to evaluate a consumer's input against a knowledge base, it scores the articles based on the confidence level of the match: VERY GOOD, GOOD, FAIR PLUS, FAIR, or POOR. If you're using Knowledge Base intents, the confidence score breakdown for LivePerson NLU v1 is used. If you're using Domain intents, the breakdown for the NLU engine used by the domain is used. For the breakdowns, see here.

When you integrate a knowledge base with a bot via a Knowledge Base integration, you specify the minimum score that a result must have in order to be returned. (The highest performing article with that threshold is returned.) You can select from VERY GOOD, GOOD, or FAIR PLUS. The default value is GOOD. If you downgrade the threshold to FAIR PLUS, be sure to test whether the quality of the results meets your expectations. It's generally recommended to keep the quality above FAIR PLUS.

Knowlege Base intents versus Domain intents

When dealing with a knowledge base, an intent is the question that the consumer has ("What's your return policy?") or a problem the user needs to solve ("I can't remember my password.").

While the article's title represents the intent of the article, you can also specify the intent using alternate formats. You can do this right inside the article by adding what are called "intent qualifiers":

Or, you can define these alternate formats in a domain in Intent Builder. Here, they're called "training phrases":

When you add a knowledge base, you'll specify which configuration approach you're using for the knowlege base:

  • Knowledge Base Intents: Select if this if you're using intents (i.e., intent qualifiers) that reside in the article itself.
  • Domain Intents: Select this if you're using intents from a domain that's defined in Intent Builder.

Languages

When you add a knowledge base, you'll specify its language.

When a text-to-text search is performed and a match is found, the knowledge base's language doesn't play a role. Whatever article is matched is simply returned.

When an NLU search is performed, the knowledge base's language works differently based on how you've defined the knowledge base:

  • If the knowledge base uses Knowledge Base intents, the knowledge base's language is used. In this case, only English and Spanish are supported.

  • If the knowledge base uses Domain intents, the domain's language is used instead. In this case, the languages supported depend on the domain's NLU provider.

Active versus inactive articles

An article is either active or inactive. Active articles are returned in knowledge base searches in Knowledge Base integrations while inactive articles aren't returned.

  • An article is active if its Enable Article setting is turned on. If the article has Valid From and/or Valid To dates (optional, specified in UTC), it is only active during the specified time period. If no dates are specified, an enabled article is always active.
  • An article is inactive if its Enable Article setting is turned off. It's also inactive if the setting is turned on, but the current date and time in UTC falls outside the time period specified by Valid From and/or Valid To.

In the Knowledge Base application, you can easily identify an article's current status:

Using Valid From and Valid To dates

You can specify optional Valid From and/or Valid To dates when you add an article.

For example, you might run a promotion over a holiday that has an associated FAQ, and that FAQ should only be used for a finite period of time. Or, you might have an article that you want to become active on a certain date and remain so indefinitely. You can satisfy requirements like these with Valid From and/or Valid To.