hello,

i’m working on ,y new project which is ” Arabic chat bot” as a customer service.
so fare i’ve crated the XML file and the chatbot.py file using Python and the chat bot works very well.
but now i’m moving toward trying the chat bot in other words “machine learning”.
so the problem is :
1-which algorithm is best to use (neural network or clustering)?
2-is ot best to still make the XML file as the knowledge base or crete an sql DB file?

need you help….!!!

Typically in question answering systems question-answer pairs are required for training the machine learning.  In dialog systems, this translates into chat logs, where the human is proactive and the machine is reactive.  I still think machine learning could be applied to raw AIML; I’m just not sure how to do it exactly.  I don’t know of any published research on applying machine learning directly to raw AIML.  Theoretically, one could do automated question generation, as well as answer generation, on AIML.  AIML lives in trees or graphs; so, this may be related to applying machine learning to other NLP graphs.  I would like to know myself what the similarities and differences are between AIML trees and other kinds of graphs used with machine learning.

I happened upon this fellow who had a neural net learn from Cleverbot‘s response pairs, but so far I am not impressed by any attempt at dialogue using neural nets, including Google’s attempt, which ended up simply parroting a movie script to the letter. Unless that is what you want, I can’t recommend neural nets for conversation.

There has been some work done on generating AIML from text corpora:

http://www.editlib.org/p/43452/article_43452.pdf

This paper describes an algorithm that will generate FAQ-style AIML given a domain and source texts. While not quite “machine learning,” this technique does show how computational linguistics might be applied to AIML.

I would do this with the GNU octave package The Shogun Machine Learning Toolbox® you would need oct2py 3.5.7 and this would provide you with unseen avantages over other methods but this would have a very high learning curve if you do not have the time to learn octave and have a avanced math background you would find this impossible and a total wast of time but you might find this usefull you could also send same data to octave and R also called GNU s and use CRAN machine learning R will do some things better than on some learning like data mining where octave does better on other things like math but R also has a high learning curve but allow you benifts like R could let your chatbot be a fire safety statistal expert with GNU ocatve you could install mathlab and have a better visual link to your machine learning

Chatbot is a real technological leap forward, which allow us to build applications inside the messengers, which means that users don’t need to install one more application. Many routine phone calls such as delivery services and bills paying can be automatized with the help of bots. Now, the time of your project launching matters even more than the complexity and cost of your web or mobile app development. The crucial part here is what tasks your product performs. Chatbots are already in use by many multinational companies, where they successfully perform sales pipeline and maintain the communication process with the customers. Smaller companies tend to adopt this technology as well. Read article “How to create a chatbot for business, step-by-step guide with an estimation” - https://www.cleveroad.com/blog/how-to-create-a-chatbot-for-business—step-by-step-guide-with-an-estimation