Chatbots are becoming more and more common in companies and on websites. However, not all chatbots work the same way. The way they work varies from one provider to another, which can confuse novice or experienced users who use them. In this article, we will review the different ways chatbots work.


What is a chatbot?

Resulting from the combination of the English terms “chat” and “robot”, “chatbot” refers to conversational agents. These conversational agents, also called “conversational robots” and sometimes even “virtual assistants”, are generally available online in the form of instant messaging. Their vocation is to answer users’ questions and even to guide them in their steps. Instead of conversing with human beings, users talk to robots. The answers proposed by chatbots are often prepared beforehand by human beings (see Chatbot vocabulary: terms to know).


How can a chatbot answer?

The responses of chatbots depend largely on the type of engine they are based on (see Key concepts of Machine Learning).

Chatbots based on artificial intelligence or Natural Language Understanding (NLU) methods are able to understand the question freely asked by a user within the limits of the competences they have been assigned (see How to choose between NLP and NLU? and Key concepts of artificial intelligence). Thus, they analyze the statement asked by the user in order to find the most appropriate answer to the question.

Chatbots based on less advanced techniques such as NLP (Natural Language Processing) also need to prepare the chatbot’s potential answers beforehand, but the bot will not understand the meaning of the question asked: either the user’s statement matches a question contained in its FAQ (Frequently Asked Questions), or there is no result. These chatbots have a limited number of preconceived answers. To avoid the frustration of no answer, these chatbots generally propose guided discussion threads with buttons. It is therefore not possible to ask a question freely to the bot, you must click on buttons until you get the expected answer.

Although all chatbots are designed to help users find answers to their questions or to automate certain actions, one should not confuse “chatbot” with “search engine”: one should always keep in mind that the knowledge of chatbots is limited to very specific fields of expertise.


How to use a chatbot?

Interactions with chatbots often proceed as follows:

  • Step 1: The chatbot is present on an interface, for example on a website or in an instant messaging service. Usually, the chatbot automatically displays a “welcome message” when a user arrives on the interface.
  • Step 2: Depending on the type of chatbot, the user sends a request to the bot either by freely writing a question or by clicking on one or more buttons successively.
  • Step 3: If the chatbot has an answer, it responds to the user otherwise it displays its “No result” message.
  • Step 4: Depending on the user’s needs, steps 2 and 3 are repeated.

Users can also leave feedbacks about the answers given by the chatbot. They can indicate their degree of satisfaction and whether or not the bot was able to provide them with an answer.


What about Hubi?

Hubi is a metabot based on advanced techniques that allow users to freely ask their questions. Hubi’s knowledge base provides answers to user queries while scenarios automate actions. The scenarios incorporate both open-ended questions (free speech) and closed-ended questions (buttons). Despite the button-guided conversation system, a new feature allows users, if enabled, to write an answer that does not appear in the choices offered in the buttons and to get an answer by exiting the scenario.

Camille is a computational linguist. Following two experiences in Parisian start-ups on named entity recognition and callbots, she recently joined the team at Hub Collab as a chatbot scriptwriter.