In this Building with Watson webinar, IBM Watson Conversation Product Manager Chris Desmarais with show you how to build a chatbot using the Conversation service and Discovery Product Manager Joe Scherping will demonstrate how to extend the function of the bot using Discovery’s cognitive search engine so the bot can answer questions beyond the intents incorporated in its original model.

In this case, the “long tail” refers to probability distributions; for example, how often will a certain question be asked by a audience. The common questions, or “short tail” ones, will form a cluster of likely probabilities and as a question becomes less likely to be asked (usually by being more complex in structure), it will show up on a charted curve further away from the common cluster. This demonstration shows you how to incorporate the Discovery service’s ability to smart-search in a directory of documentation into your chatbot so it can find and return answers to questions it wasn’t trained to answer.

The example they provide for a short tail question is “How do I reset my password?”. A question like this is expected and is often built into the modeling parameters of a chatbot. A long tail question could be something like this: “How do I access my account information remotely through my iPhone 6S?”

In the first question, the Conversation service will use reasoning strategies that focus on the language and the context of the question. In the second question, the Discovery service will use strategies that focus on identifying the most appropriate answer.

Resources for you