Introduction
When codeit autocodes your data, it will attempt to apply all the relevant codes to each verbatim.
However, there will be occasions when the AI is not able to apply all the codes needed to fully capture all of the themes in a verbatim. For example, if a theme is mentioned that is not included in the codeframe, or if a theme is expressed in an unusual or obscure way.
The autocoding mode in codeit allows you to control whether autocoding should be applied to verbatims where coding would only be partially complete.
Autocoding Mode
This setting in codeit has two options.
| Option | Description |
|---|---|
| Minimise Coding Gaps | Using this option will limit the codeit AI so it will only apply autocoding to a verbatim if it is sufficiently confident it will apply ALL of the necessary codes. Using this option will increase the chance that each coded verbatim is completely coded and therefore does not need any further coding refinement. However, it will reduce the overall amount of autocding that codeit can apply. |
| Maximise Coding Volume | Using this option will allow the codeit AI to apply ANY codes that it thinks are applicable, even if that would leave "gaps" - i.e. themes within a verbatim that are not captured by codes. Using this option will increase the overall amount of coding that codeit can apply, however, some verbatims will need to be checked and further refined. |
Detailed Explanation
Minimise Coding Gaps
The aim of the "Minimise Coding Gaps" mode is ensure that the AI can only apply suggestions from the machine learning if the verbatim is completely coded. To understand this, consider the following example:
Suppose that the codeit AI is presented with the verbatim:
"They give good service, but it was expensive and the staff were rude"
Suppose also that the Machine Learning generates the following codes, in response:
| Suggestion | Text Suggested |
|---|---|
| Code 3: Good Service | they give good service |
| Code 7: Poor Price/Value / Too Expensive | it was expensive |
In this example, the AI has only suggested 2 codes (Code 3 and Code 7), but these do not capture all of the themes expressed in the verbatim.
Therefore, it is likely there is a gap in the coding. In this sense the codes applied are not considered "complete" and the AI has missed a code - i.e. 'the staff were rude'. Using "Minimise Coding Gaps", in this case, would result in none of these codes being applied.
So, you would use "Minimise Coding Gaps" mode if you want to be as sure as possible that applying the Machine Learning suggestions will result in completely coded items with no gaps of this sort. Note, that this sets quite a high bar for the AI and will result in higher "completeness" of autocoding at the expense of a lower number of items autocoded overall.
Maximise Coding
The "Maximise Coding Volume" mode offers an alternative approach - we allow the Machine Learning Layer to apply any codes that meet the minimum "Autocoding Threshold". This increases the possibility of gaps in the coding, but hugely increases the volume of codes that the AI is able to apply.
In the example above, for verbatim "They give good service, but it was expensive and the staff were rude" , the 2 suggested codes would be applied by the AI when using this mode.
The codeit AI will use the Coverage threshold to calculate how much the verbatim should be "covered" to be considered as completely coded by the AI.
See more details about the coverage threshold here.