[8pla.net]

What choices does a chatbot have when it matches several responses?

1. Randomly choose a response.
2. Rank responses by the most words matched?

For example:

Code

VISITOR: Hello, can I have a computer?

CHAT: PERHAPS YOU DON'T WANT TO*  
RANK: 2
WITH: CAN I

CHAT: DO COMPUTERS WORRY YOU?  
RANK: 1
WITH: COMPUTER

CHAT: HOW DO YOU DO? PLEASE STATE YOUR PROBLEM.  
RANK: 1
WITH: HELLO

This implementation is in progress.  It lets you choose: Ranked or Random,
and remembers the choice you made.  However,  if there is a tie,  it chooses Random.

Ranked is more accurate, but it chooses the best response repeatedly which is a drawback.
Random is less accurate, but it varies the responses, which is a benefit.

http://chatbot.altervista.org/implementation

[ranch vermin]

You need more contextual response.

[8pla.net]

So the question is, What is a context?

"A context includes a unique identifier for each conversation with a user, as well as a counter that is incremented with each turn of the conversation. If we don’t preserve the context,  each round of input appeared to be the start of a new conversation. We can fix that by saving the context and sending it back to the Conversation service each time." said Watson Cloud Developer Vidyasagar Machupalli.

https://www.ibm.com/blogs/bluemix/2017/03/importance-context-chatbot-conversation/

Thanks for your reply, Ranch!

[LOCKSUIT]

Oh I know what you want here, so you have several matching parts, some more matched than others and some equaling matching (don't call this rank nor reward), and some more loved then others (call this rank/reward).

Well, if it was full match and it had still 2 responses then it would say its favorite "go to the hospital" instead of its hated response "don't go to the hospital" because for this AI it has learnt that saying that as answer is a GOOD thing right (duh lol).

But if it matches a 2-word part and one of its responses has a high rank/love whilst it also match another part by 10 words but with lesser loved responses, then, it matches to that one as it is related/in context, and so for this one it then uses best loved answer. If there's 2 matching parts each of 2 words, then it says the response with the highest reward. If 2 2-word matching parts and both have 2 as loved responses then random pick or reason the part and the response to reply to and back with.

However 1 problem. If "Can I have a computer?" doesn't match the full thing and use best loved response, and instead has 2 small matches cus no know the big whole sentence yet cus baby, then it has TWO matches it finds "[Can I] have a [computer] ?" and so then what it does in this case is answers both input matches, it'll answer the 1st match using best loved response "Can I what?" or "it depends", then it'll answer the next match using best loved response.

So there we go. Answer all biggest matching text parts using best loved answer for each biggest match. If you have 2 equally loved responses, say both if not costly else pick one randomly or *that didn't say last time.

[8pla.net]

Yes LOCKSUIT!

I just got started with a simple form of contextual responses.

Hello.
HOW DO YOU DO? PLEASE STATE YOUR PROBLEM.

Whats up?
WHAT ELSE COMES TO MIND WHEN YOU ASK THAT?

I lost my train of thought.
YOU JUST SAID: WHAT'S UP?

CONTEXT:
I LOST MY TRAIN OF THOUGHT
WHAT'S UP?
HELLO

I think it is working.  Now, I have to extend it further... Any ideas?

Thanks for your reply.