codeit provides a number of different tools that use AI (Artificial intelligence) to perform coding tasks so that you can reduce or minimise the amount of coding that needs to be done by a human. One of these methods used is Machine Learning.
What is Machine Learning?
Codeit learns from the coded examples you provide and uses this learning to autocode. This extends the autocoding beyond text or pattern matching, to autocode based on semantic theme.
For example, coding examples like:
"It helps me relax"
"I find it calming"
will enable to Codeit to autocode verbatims like:
"This is great for chilling out"
"It is great for unwinding"
even though these phrases are completely textually dissimilar.
Machine Learning can be run after a first wave or initial batch of coding has been completed - as machine learning needs to learn from how the initial batch of coding was completed by the human coders. As more and more coding is completed, the machine learning can apply more coding. Note that if you have any coding that had already been completed before you started using codeit, you can also import that previously coded data into your project so that the machine learning can be applied from the start.
What if I do not have coded data?
It is recommended to have at least 100 coded verbatims before running the Machine Learning through the Apply AI start menu option.
If you do not have at least 100 coded verbatims setup in your task (either coded manually or imported through the import wizard), a warning will be displayed when running the AI Autocoding. You can choose to ignore this warning and the AI will only use the code labels to match verbatims. This can lead to a drop in performance depending on the labels of your codeframe.
See here for more details about the Apply AI option in the Start Menu.
AI Settings
You can control how much autocoding is done by the AI using the AI Settings, more details on the settings here. Also refer to the Complete & Partial coding guide which explains how codeit allows you to specify the level of "completeness" required before a verbatim can be autocoded.
You can use the coding screens to review the coding done by the AI, a guide for the symbols used in the verbatim screen for the different AI functions is available here.
AI Training data
Using the Wave AI Settings, you can control which Wave is used in the model's training data when re-training a model.
By default, the model is not re-trained automatically, instead the user needs to turn on the "Retrain Rob" setting.
Autocoding new data
Once trained, the Machine Learning model will run automatically after a Data Sync, or an Import on any Wave marked as In Progress, or Verifying.
The AI will also only attempt to code the verbatims marked as Rejected, or not yet checked (e.g without status Finished or Approved).
Finally, by default the AI will not change coded verbatims. You can change this default in the Company Settings.
Supported Languages
A list of the supported languages is available here.