You might notice that your application does not detect topics or entities correctly in incoming emails or chat messages. Common causes of issues with topic or entity detection include misconfiguration or insufficient model training. Learn about the solutions to these problems.
Cause: Topic or entity model does not have enough training data
Your topic or entity model might lack training data on the pattern that you are expecting.
- Retrain the model using appropriate training data.
- To train a topic model, see Creating machine learning topic models.
- To train an entity model, see Building machine learning entity extraction models.
Cause: Email parser is not detecting the body properly
In the case of an email bot, the email parser might not be trained properly to distinguish between the disclaimer, signature, and email body. As a result, the email parser wrongly identifies the email body as the signature or disclaimer. Typically, the text prediction associated with the email channel is configured not to perform topic detection on the signature or disclaimer, which is why you are not getting the expected results.
- Run a test on the email parser model, using the email text for which
the topic is not detected.For more information, see Testing an email parser.
- Retrain the email parser.For more information, see Training an email parser.
Cause: Wrong ruleset
Your system might contain multiple versions of a text prediction and associated models in different rulesets. The application uses the prediction and model versions in the ruleset that it can access through its own application stack. These versions might not be the versions that you updated, which is why you are not getting the expected results.
- Ensure that the text prediction and the models that you want to use are in a ruleset that belongs to your application stack.