article on best practices associated with artificial intelligence will be of great help to you in understanding the definitions and help you understand the best way to manage such a project.
our
1. The principles
1.1 Function
The Intent Manager is the interface from which you drive the iAdvize artificial intelligence to use it in a bot scenario. When configured correctly, this application allows the understanding of the intents of your visitors as well as providing an adequate reply.
1.2 Managing the languages of your project
Since intent detection (also known as NLU: Natural Language Understanding) relies on written language to work, it can only work in one language at a time: that's why the interface starts by presenting to you the list of languages supported by our AI. If your visitors speak several different languages, you will have to work on the intent detection separately for each language.
1.3 Intent detection, a long-term project
Once your artificial intelligence is in place, it must be regularly monitored, trained and improved! For more information, please refer to the bot master's guide and especially to the article dedicated to the good use of artificial intelligence. To make this maintenance easier, our AI extracts every week all the undetected expressions of the last 7 days, filters them and groups them by theme to suggest you adding expressions relevant to your intentions.
2. Intent suggestions: optimize intents with the help of our AI
Intent suggestions are here to save you valuable time when optimizing your intents. Every week our AI engine extracts all the undetected expressions from the past 7 days, filters them, and groups them to suggest only relevant expressions to add to your intents.
2.1 How does it work?
- Every Monday our AI engine extracts the complete list of all undetected expressions from the past week.
- We filter expressions that are very close to each other to suggest just one instance of each expression.
- Our AI then groups expressions together that should go to the same intent.
- Each group is then shown as a suggestion to add to an existing intent or to create a new intent.
- You can now review the expressions in each suggestion and decide if you would like to validate them and if they need to be added to an existing intent or a new intent.
2.2 How to review an intent suggestion?
- Go to the Intent manager and choose the language of your intents.
- If your intelligent bot has been running for at least 2 weeks you should be seeing new intent suggestions at the top of the screen.
- Click on the “review” button next to a suggestion and you will be sent to the suggestion review page. If you want to remove a suggestion from the list, click the “ignore” button and the suggestion will be deleted.
- On the Intent Suggestion review page read through the expressions and click the “validate” button for each expression you would like to add to your intents and “ignore” for each expression you don’t want to add to your intents.
- Once you have reviewed all the expressions, you need to decide if they should go to an existing intent or a new intent.
NB : Be mindful that intents need to be strictly distinct so if you already have an intent with similar expressions choose the option to add to an existing intent.
- If you have chosen to add the expressions to an existing intent, you will see a dialog to select the intent to which you would like to add the expressions. Select the right intent and confirm.
If you have chosen to add the expressions to a new intent, double check the name of the new intent at the top of the screen and click the button to create a new intent.
3. Use collaborative training to improve intent suggestions
If you have Collaborative Training activated your intent suggestions will be of even higher quality. As mentioned above, our AI groups undetected expressions into themes. By using Collaborative Training, your agents help improve the grouping done by AI and remove expressions that shouldn’t be grouped together in the same intent.
6.1 Pre-requisites for the CSV file
Please note the following pre-requisites for the CSV file before making an import:
- The CSV file needs to be lower than 50Mo
- The CSV file has to be with a comma separator
- The use of a semicolon (;) character is forbidden
- The file can’t have empty lines
- The file can’t be empty
- An intent can have a maximum of 2000 expressions (applies in total)
- An expression can have a maximum of 768 characters (applies in total)
- Intent can’t be empty without expressions
- A line can’t miss an intent name
When a CSV file