Applying AI Suggestions
The AI system in Codeit consists of a set of layers.
When these layers run, they produce a set of suggested codes for each input item.
After each layer runs, the suggestions produced are applied by the "apply codes" process.
This page details the rules in place within the "apply codes" process.
Codes are only applied to an item if it appears the AI has predicted all the codes needed for that item.
The intention here is to only apply codes when the item is completely coded.
This option is used when the Autocoding mode is set to "Complete Coding".
Any codes that meet the minimum confidence level, are applied to to the item.
This option is used when the Autocoding mode is set to "Partial Coding".
Special consideration needs to be applied when dealing with exclusive (single coded) items.
The rules applied are as follows:
- If the item already has an exclusive code applied, then the AI will not apply any codes
- If the AI generates a mix of exclusive and non-exclusive suggestions, then the non-exclusive suggestions only are applied
- If the AI generates a set of exclusive only codes then the item with the highest confidence will be applied. In the event of a tie, the item that appears first (by ordinal) in the codeframe will be selected
Existing Coded Responses
When codes are already applied to a response, the following rules apply:
- All the Text Matching and Text Rules will be applied (match against codeframe labels, match against existing previous responses, text match rules with regular expressions etc). New codes can be added to already coded responses with those layers.
- The Machine Learning is using existing coded responses as training examples. New codes can be added to already coded responses with the Machine Learning layer but they usually match existing codes.