Voice of the beneficiary

Sentiment and topic analysis with human review

Thematic classification and sentiment and trend analysis with accuracy and linguistic-bias measurement, a model and source log, and human review for low-confidence cases. Sentiment output is not used to make a high-impact individual decision without review. CX-REQ-009 CX-REQ-021 CX-PRIV-004 ENT-CX-049

Review list

Confidence below 0.70 or a random sample 5%

Topic trend

Last 8 weeks · Number of negative comments

Model card

Model log ENT-CX-049
Model
Arabic sentiment classifier · v3.2
Provider
Paseer AI Center · Organization key
Validation sample
1,200 comments reviewed by a human
Overall accuracy
87%

Accuracy by dialect

Linguistic bias measurement

Confusion matrix

Actual (rows) × Predicted
PositiveNeutralNegative
Positive412316
Neutral4419852
Negative419434
Most common errors: neutral classified as negative (sarcasm and dialect)