User permissions required: ‘Datasets admin’
Before you start training, you’ll need to choose whether to enable sentiment analysis when creating your dataset. This is an important decision as it will effect how you label each verbatim, as well as the output of the platform's predictions.
If you choose to enable sentiment analysis, every time you apply a label you will need to select whether it has positive or negative sentiment (there is no neutral sentiment).
This does make the labelling process slightly slower, however, for more emotive communications data, it provides a very useful indication of the overall sentiment of each label (i.e. are people happy with X or dissatisfied about Y).
When would you enable sentiment analysis?
Sentiment analysis is most useful for more emotive communications data, such as customer (or employee) feedback reviews and surveys, or support tickets and chats, when you’re trying to gain a sense of customer (or employee) satisfaction (or dissatisfaction) regarding various topics.
Sentiment analysis is not typically recommended for communications data that is generally neutral in tone, such as shared mailboxes for BAU teams interacting with each other or external counterparts (though there can be exceptions). In these kinds of data sources, sentiment is usually only expressed occasionally, but you would need to assign positive or negative sentiment to each label if it was enabled.
For more neutral datasets, it is typically easier to capture sentiment with certain inherently positive or negative labels, such as ‘Frustration’ or ‘Chaser’ as there are far fewer cases where sentiment is explicit.
How to enable sentiment is covered in the previous article here.
Previous: Multilingual sources and datasets | Next: Amend a dataset's settings