Part 6 of 8
4.1 What to do when users ask multiple questions at once?
When a user asks several questions at once or enters a long text, it is usually to address complex topics with subtleties that a single sentence could not summarize. In this case, the chatbot will only react on a few keywords and there is very little chance that the AI will give a relevant answer. It is better to disable AI for this specific case and offer the user other ways to get an answer such as putting him in contact with a human operator in live-chat, by phone or by leaving a message. This temporary deactivation will only be triggered in this case and will reactivate itself right afterwards automatically.
In the advanced settings of your chatbot, there is an option “Disable AI”. It is this feature that allows the user to be directed to a particular sequence when he has entered a text that is too long. You can determine the number of characters beyond which this feature is triggered.
4.2 Autonomous learning
This is called autonomous learning or Machine Learning. This is an extremely complex process that is still limited to the world of research and to the big players in artificial intelligence such as IBM.
Botnation does not have Machine Learning as researchers understand it, but we do have many machine learning algorithms.
With use, your chatbot will determine that certain turns of phrase are likely to match certain responses without you having to add a new AI rule. For the probabilities to be roughly relevant, a large amount of data must be collected.
If you have two rules containing only the Expressions “renting brighton” and “renting london” but a user searches only for the keyword “renting” our algorithms will send him to the most popular sequence, containing at least “renting” in part in one of its Expressions.
In any case, you should regularly check the behavior of your chatbot.
4.3 The AI of a chatbot is never finished
It is important to monitor the use of your users. Only with this data can you correct/educate the ChatBot.
The vocabulary your users will use will probably be quite different from what you had planned. You will have to constantly teach your chatbot new expressions.
To find out which questions have not been answered, use the recommendation algorithm, which we will explain below.
Botnation offers the possibility to A/B test all the content of your ChatBot. We even have an automatic optimization tool.
4.4 The Recommendation Engine
In the AI tab of your bot, in the left column, you will find our recommendation engine. It consists of listing the sentences entered by users that the ChatBot did not understand and prioritizing them according to their occurrence.
You can see the results of the current day, the past month and the current and past years.
This tool allows you toidentify turns of phrase and vocabulary that your chatbot has not understood and will serve as a recommendation for adding new keywords and even new AI rules and content.
Indeed, you will find here requests that you did not think you had to process in your ChatBot.
If you find that the same type of requests are being made regularly, it’s time for you to remedy this by adding content. Whether it’s answering it in the ChatBot or redirecting the user to another online service, this kind of situation must be handled.
This is also where you can see if your users tend to write long messages. If this is the case, it will surely be necessary to disable the AI from a certain number of characters as we have seen previously.
When adding new keywords or phrases, don’t forget the tips we’ve given you. Analyze the phrases entered by your users and translate them for your bot.
Don’t forget to update your Keyword Libraries.
<< Previous part – Table of contents – Next part >>
Download the guide in PDF format