Catagorizing Expressions

Re: Catagorizing Expressions

Postby SangramMCP » Wed Sep 06, 2017 11:38 pm

Thanks Franzz,

Can you have any reference website, which can guide me how to frame expression, and some varied examples of the same. This will help us a lot in framing expression as well as grouping intents
SangramMCP
 
Posts: 47
Joined: Wed Jul 26, 2017 7:41 pm

Re: Catagorizing Expressions

Postby Sreejith » Thu Sep 07, 2017 4:08 pm

theres no dedicated page for writing all types of expressions sorry... the developer network has tutorials on writing example and template expressions and using entities and context...a expression is a sentence with entities...simple :)
If your experiment needs statistics, you ought to have done a better experiment - Ernest Rutherford
User avatar
Sreejith
Help & Support
 
Posts: 7
Joined: Sat Apr 06, 2013 6:16 pm

Re: Catagorizing Expressions

Postby SangramMCP » Tue Sep 12, 2017 7:10 pm

Hi,

Thanks for the response.
I have 1 more query on top of it?

Lets say, I have a cafeteria related chatbot.

I have added some 5 expressions for the same. which were queried before. Now, a new expression for the same context was typed by user. but the same is not there in the expression. How are we telling the bot, to understand a new expression, which was not categorized earlier. What is the role of ML in this scenario. How to make the bot learn by itself.
SangramMCP
 
Posts: 47
Joined: Wed Jul 26, 2017 7:41 pm

Re: Catagorizing Expressions

Postby Leslie » Tue Sep 12, 2017 10:33 pm

Actually the role would be of Natural Language Processing. What you are asking is that, if you have an expression and the user types something semantically similar then the intent should be given higher confidence score.

Well, this is exactly the sole purpose of the NLP engine of Oscova. That is, to find similarities between developer created expressions and user submitted requests.

Here's an example:

  • I love Pizza -> Your expression
  • I like Pizza -> User request

Both the above sentences are very similar as they reflect the user's likeness towards pizza and may possibly belong to a single intent. However, in reality the 2 expressions aren't exactly the same so the intent score should be slightly less when the user types "I like pizza". This is by design as if similar sentences were given exactly the same confidence score then it would become impossible to compute any sort of differentiation.

From 2.6.0 Oscova always uses ML algorithms and only when you explicitly select Deep processing mode Oscova will forcefully create deeper neural network layers that take a lot of memory and may not always gives extremely higher confidence, as this is the case with any bot architecture. For performance critical applications we recommend using the Standard processing mode.

To enhance the NLP, developers can provide additional information to Oscova. For which we've built in multiple APIs. You can either load WordNet lexical database or load Word Vectors, in Machine Learning Word Vectors are vector representations of words as words cannot be directly fed into Neural Networks.
User avatar
Leslie
Lead Software Architect
 
Posts: 287
Joined: Fri Sep 14, 2012 12:20 pm

Re: Catagorizing Expressions

Postby SangramMCP » Tue Sep 12, 2017 10:39 pm

Hi Leslie,

If it is similar lind of expression, it is ok.
But what about

User: I love food( Its a part of Expression, where it is very generic)
and
User: I love Pizza(Its not a part of expression, I mean Pizza is not a entity).

My bot should understand that Pizza goes under food items. Pizza is just an example, some different thousands of food name can't be added in the expression.

Is that possible
SangramMCP
 
Posts: 47
Joined: Wed Jul 26, 2017 7:41 pm

Re: Catagorizing Expressions

Postby Leslie » Thu Sep 14, 2017 1:15 am

You may notice that I didn't entity-fy the words in my example. Whether it be love, like, food or pizza the degree of word relatedness is measured using similar procedures by the NLP engine.
User avatar
Leslie
Lead Software Architect
 
Posts: 287
Joined: Fri Sep 14, 2012 12:20 pm

Previous

Return to General Discussion

Who is online

Users browsing this forum: No registered users and 2 guests
cron