Chatbots Magazine Become a Writer Intro to Chatbots

Find the best answer for the user question

(mleal) #1

Hi everyone!

I’m currently working on a possible customer support bot for a company. One of its requirements is the ability to support the user to ask questions about some applications (at this moment we’re focusing on FAQ) and then find the most probable answer in that case.

At the moment we are mapping intents+entities of the user input to an entry in the database, but at some point this may not be enough.

I was wondering if anyone here is also dealing with question-answering at this level. If so, how are you designing your solution? What recomendations do you have?

Thank you so much in adance, any kind of opinion/insight will be much appreciated :slight_smile:

(a.pinzonacevedo) #2

I’m experiencing a very similar issue with an upcoming development. I’m curious to know what you’ve been thinking about of if you’ve received any feedback.
We’re thinking of using WordPress to upload the answers and metadata, but recognize that it might now be enough.

( #3


How about using regular expressions. For example:

User, “How long before I get support?



All regular expressions above match.
The best match is the regular expression
with highest word count (6).

You may rank all of them based on their word count.
NOTE: The best response repeating over and over
could come across as unnatural.

(Kath) #4

I have feeling that using regular expressions is just a dead end. Because there are too many versions of the same like in your case it could be:
how long…
how much time…
what time…
which time…
how many minutes…
approx how … (all the previous versions)


(mleal) #5

Regular Expressions are fit enough for some cases, for example, I am using AIML to deal with overall chitchat where it is not very important to me to kno what the user is saying. However, when it comes to deal with user intentions or requests to a specific service (ie, requesting a service of the company or just asking a question about a product), it is more important to know in more detail what does he want. Regular expressions are not enough here, there are a lot of ways the user can express himself to ask for something. If we think about changing the language (from english to portuguese, for example), we would not only have to rewrite all the rules but also have in account the own linguistics expressions of each language. We are trying an artificial inteligence approach to deal with this, trying to do question similarity between the user input and the questions we have stored in a database. I can’t really say how this is going since I am now in a different project, but I hope we’ll know some results soon.

( #6

AIML uses SRAI tags for reductionism, which uses recursion to reduce the stimulus (user input) to a simpler level. Regular Expressions are just a tool, though a super tool, I think. Though they are cryptic, and have a steep learning curve, they are powerful, once you learn them. They don’t replace A.I. but I think it is fair to say that REGEX may extend narrow A.I.

Please know that I support building a smarter Artificial Intelligence. Good luck with your A.I. research.

( #7

Yes, your right Kath. In terms of narrow A.I. that is a common criticism. However, once you get up to 10,000, 40,000 or 300,000 records in the chatbot brain (database), the convincibility of the chatbot sky rockets. In other words, the conversation simulation improves greatly.