How Your Bot Learns
Conversations vs. Training
Your bot's knowledge base is called a "corpus". You can configure your bot so it builds its corpus in two basic ways: conversations, training, or both.
If your bot is configured to learn from conversations, you can converse with it and over time it will learn which responses are best. If you trust the public to converse meaningfully with your bot, you can even configure it to learn from its conversations with the public.
Training your bot allows you to have more control over the responses it gives to other people. It is also a faster, more efficient way to train your bot than simply conversing with it. More information on training your bot
When your bot is first created, its corpus is completely empty. If you opted to let it learn from its conversations, it will attempt to use the phrases in its corpus to respond to user queries. Initially its responses will not meaningfully match the queries you ask it, but over time it will begin to understand how to respond. For example, the first time the bot asks "How are you?" and you respond with "Great, thanks!", the bot will learn that "Great, thanks!" is a meaningful response (and in fact the only meaningful response) to "How are you?". Subsequently, when the bot asks "How are you?", if you provide it with other responses (e.g. "Awesome!") it will learn those phrases as well. So, when somebody asks the bot "How are you?", it selects the most frequently given response that it has seen so far and responds with that.
In order for a bot to recognize a phrase asked by the user, that phrase does not have to be asked verbatim. For example, if the bot was trained to understand the phrases "How are you?", "How are ya?" and "How r u?", it will most likely understand that "How're you" is simply a variation of one of those other phrases, and thus be able to respond accordingly. This "fuzzy" matching is done by BotMark's proprietary algorithms.