by Vagelis H.
In a previous article, we discussed how a chatbot can leverage a document collection to respond to user questions. The simpler method is to return a document snippet that is relevant to the question, while a harder method is Question-Answering (QA), which tries to generate the actual answer to the question.
SmartBot360 is at the forefront of such document-driven chatbot capabilities, which enable the creation of effective customer service chatbots. In this article, we discuss the ways that SmartBot360 offers to enable document-driven AI in a chatbot.
For a live demo, check out https://smartbot360.com/demo/docai, which demonstrates the various ways to query documents in SmartBot360.
The first two ways return a relevant snippet, while the third attempts to directly answer the question.
1) Document Search: This method allows a chatbot to search the pages of one or more web sites. The bot admin first specifies a set of URLs (e.g. smartbot360.com/documentation) to be indexed, and then inserts a Document Search box in the flow diagram of a bot. The bot then asks the user to input a question and returns the most relevant snippets, along with links to the documents that contain them, as shown below.
2) Google Programmable Search integration: This method is similar to the first one, except that here the documents are indexed by Google and not by SmartBot360. Specifically, the admin must first create a Google Programmable Search Engine, and then link it to the SmartBot360 account. Then, the admin inserts a Google Programmable Search box in the flow diagrams, which works similar to the Document Search box. A disadvantage of this method is that it may not be HIPAA-compliant, because Google Programmable Search Engine is not HIPAA-compliant.
3) Question Answering: This is the most advanced method, as the actual answer is searched, using AI. The documents are indexed in the same way as in Document Search. Then, a QA box is inserted to the bot flow diagram. The bot then asks the user to input a question, and it returns the best answer to it, along with the link to the document where this answer was found, as shown below.