Why use intents?
If your knowledge base is an external knowledge base with LivePerson AI or an internal knowledge base, it can make use of Natural Language Understanding or NLU to find the right answer (article) to use to respond to a consumer's message. What's so great about NLU? It's superior to the alternative, which is text-based search. NLU is much more nuanced and leverages AI to identify various attributes of a message: meaning, intent, sentiment, and more. It makes answer retrieval more accurate.
To take advantage of NLU, you'll need to link the articles in the knowledge base to the intents in a domain. So, in summary, using intents is the way you tap into the power of NLU.
Intent Manager offers a set of prebuilt domains. These are designed to get you up and running quickly with intents.
In KnowledgeAI, associate the domain with the knowledge base. Then associate the domain's intents with the knowledge base's articles.
Train the knowledge base (and the underlying intents) until you get the performance you expect.
When NLU is used, the consumer's message is evaluated against the intents that are associated with the articles, and the highest scoring article is returned as the answer.
For more details, see below.
Associate a domain with a knowledge base
You associate a domain with an external knowledge base when you add the knowledge base:
And you can likewise associate a domain with an internal knowledge base when you add the knowledge base:
Associating the domain gives you access to the domain's intents, so you can associate them with the articles. This is the next step in connecting your content to intents.
Associate intents with articles
After you've added a knowledge base that is associated to a domain, configure the articles so that each is linked to the appropriate intent.
You don’t need to link your articles to intents right away, as the Intent field is optional. This is deliberate because it allows you to get started with a knowledge base by adding just the articles first. Then, you can create intents for the content you care about the most, and link those to the relevant articles. This means you can focus on specific content areas in your knowledge base, and manage the content overall with varying levels of effort on your part. The approach gives you flexibility as you maintain the knowledge base over time.
Tune a knowledge base
For information on this, see here.
A knowledge base search is performed using a specified "search mode." The search mode determines the type of search. Possible modes include:
- Intents Only
When the Intents mode is used, an exact match, text-to-text search is performed first. If a match isn't found by the first search, KnowledgeAI next uses Natural Language Understanding (NLU) algorithms to match the consumer input to articles. And if a match isn't found by the NLU search, a final, text-to-text search is performed as a fallback.
Intents Only mode
The Intents Only mode is like the Intents mode (above) except that the final, text-to-text search isn't performed as a fallback.
When the Text mode is used, a text-to-text search is performed:
- With an external knowledge base with LivePerson AI, the search algorithm checks the consumer's input against the title and tags.
- With an internal knowledge base, the search algorithm checks the consumer's input against the title, summary, detail, Knowledge Base intents (intent qualifiers), and tags.
You can improve the quality of your knowledge base answers by linking your articles to intents and performing intent-based searches. However, often this change is introduced gradually, as time and opportunity allow. Typically, Text searches are used when you haven’t yet linked your articles to intents.
Be aware that, when a Text search is performed, if a match for the search phrase is found, the associated confidence score is reduced to FAIR_PLUS if either of the following is true:
- The search phrase has less than three (3) words.
- The search phrase and the stored content don’t have matched words in sequence (with two deviations).
Therefore, if you’re using a Text search, LivePerson recommends that you test and tune the knowledge base using a threshold of FAIR_PLUS. If you aren’t satisfied with the results, increase the threshold to GOOD.
Scoring and 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 the knowledge base is…||Then…|
|an external knowledge base with LivePerson AI||the scoring breakdown for the NLU engine used by the associated domain is used|
|an internal knowledge base with Domain intents||the scoring breakdown for the NLU engine used by the associated domain is used|
|an internal knowledge base with Knowledge Base intents (intent qualifiers)||the scoring breakdown for LivePerson (Legacy) is used|
For these confidence score breakdowns, see here.
When you implement a knowledge base search within a bot via a Knowledge AI interaction, you specify the minimum score that a result must have in order to be 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.