Symptom
- The NLP Classification functionality does not return the expected sentiment when emails include closing courtesy phrases (for example, “Thank you”).
- Emails containing negative content followed by a closing courtesy remark can be incorrectly classified as slightly positive (“a bit positive”).
Environment
SAP Service Cloud Version 2
Reproducing the Issue
- Send an email with clearly negative content followed by a positive closing courtesy phrase (for example, “Thank you”) to the inbound channel that creates cases.
- Allow the system to create a case from the email.
- Open the case and navigate to the Emails tab.
- Observe that the sentiment shown by NLP Classification may be slightly positive despite the negative content.
Cause
The sentiment model treats phrases like “Thank you” as strong positive signals that can outweigh preceding negative content, especially in short texts; this is a known limitation of standard AI sentiment classification.
Resolution
The sentiment score reflects the overall emotional tone of the text as assessed by the underlying AI language model. Standard AI sentiment models are trained on large volumes of natural language text data where phrases like "Thank you", "Best regards", or their equivalents in other languages are statistically strong positive signals. When such phrases appear they can shift the overall score toward positive or neutral when overall text body itself is small. This is a known characteristic (limitation) of industry-standard AI sentiment classification. Alternate reliable approach is with GenAI (but cost is many fold) .
See Also
Keywords
NLP Classification, sentiment analysis, email sign-off, closing courtesy, thank you, a bit positive, misclassification, short text bias, Service Cloud V2, machine learning sentiment, English, Spanish, negative email, sentiment skew, case emails , KBA , CEC-CRM-ML , Machine Learning for SAP Sales/Service Cloud , Problem
SAP Knowledge Base Article - Public