Oscova Machine Learning

[IranCahtBot]

Dear FranzZ,

I studied Oscova Machine Learning document, but I don’t understand what learning means in this document.
It said, at first we should retrieve the training data as an XML Document and then generate a trained neural network model by Syn Bot Studio and load the trained model into Oscova.
What is the difference between when we use Oscova Machine Learning and when we don’t use Oscova Machine Learning.
Would you please give me a sample Project in this respect?
I appreciate sincerely taking time to respond me.

With regards,
Ebrahimi

[Fantom]

Oscova uses 2 modes for processing a user request. It is either rule based decisions that allow Oscova to fastly (within milliseconds) find similarity between the expressions you wrote and what the user wrote or forcing a deep neural network to learn the same. At the moment Rule mode is superior to RuleAndML mode in performance (as per our unit tests).

When you enable RuleAndML mode Oscova is forced to train a deep learning neural network model to find contextual similarity. The final score generated by this Deep Neural Network model is then calibrated with the Score generated by Rule mode and the intent gets a average score of the two. When you train Oscova neural network model by saving it to XML file and loading it in Syn Bot Studio, the bot studio still uses Syn.Bot framework in background to train the model. So training it in Syn Bot Studio and Syn.Bot framework will have the same result. The only reason we allow saving it to XML file and loading in Bot Studio is that you can see training in real-time in Bot Studio's GUI. Otherwise there is no difference.

ML mode is recommended for carefully designed projects that use very less of @sys.text entity if you are making use of @sys.text then running Oscova in Rule mode should give higher results. From version 2.6.0 Oscova will always be using Machine Learning in background. As both Rule and RuleAndML mode will be obsolete and will be replaced by Standard and Deep processing mode.