Guide for classifying sentences into Intents

For your xBot to be well trained, we recommend the following best practices :

Classify at least 5 sentences per intent.
Maximum of 30 sentences classified by intent.
Classify at least 1 sentence in your test set sentences for 5 classified sentences, in order to be able to compare the results after a classification session.
Keep the number of classified sentences roughly identical among all intents.
If several intents are expressed in a sentence, then it is wise to archive it.
Classify relevant distinct sentences. Your xBot needs a clean sentence base to learn correctly. For example, if the sentence “What’s the weather in San Francisco?” is already classified, then it is better to classify this sentence Will it be sunny tomorrow in New York? than this sentence “What’s the weather in New York? . This last sentence is less distant , too close to the first sentence and should not be classified in the same Intent.
Classify only sentences without spelling and grammar errors. Your xBot will be smart enough to deal correctly with typos. For example, classify the phrase “if a part does not fit, can I send it back?” rather than “if a prat is not fiting_, can I send it back?_”.
Only classify sentences that correspond to the response / scenario of this intent. No need to “confuse” your xBot in its training! 😊
Acronyms and abbreviations should be avoided … unless clearly used and understood by your users.
The sentences collected on the “Pending” page (Sentences awaiting classification) are unique, in the sense that any identical phrase (including the same keystrokes, spaces and punctuation) is filtered by our engine, so you will never classify the same phrase twice.
It is better to archive a sentence rather than to delete it. This will allow the unwanted sentence never to appear again as sentence awaiting for classification in the “Pending” page.

Documented/reviewed 27/07/2017 7:12:55 AM