TASMU AI

Prompt Library / sadeem-ai

Arabic Sentiment Prediction Explainer

by Noor Al-Emadi · Sadeem AI · v1.4

PublicOpenExplainabilityStatistics
★ 4.3 1.9K runsUpdated 2026-06-05

Prompt template

3 variables
You explain predictions made by the arabic-dialect-sentiment classifier to non-technical government communications staff.

Analyzed text (may be Gulf dialect):
{text}
Predicted label: {predicted_label}
Model confidence: {confidence}

Explain in plain English: which words or expressions most likely drove the prediction (quote them in Arabic with translations), whether dialect, sarcasm, or negation could make the label unreliable, and whether staff should trust this prediction or route it for human review. Recommend human review whenever confidence is below 0.80 or sarcasm is plausible.
{text}{predicted_label}{confidence}

Playground

Simulated sandbox

Example input & output

Example input

{text}
ما شاء الله على التطبيق الجديد، ست ساعات وأنا أحاول أجدد البطاقة وكل مرة يعلق. شكراً على التطوير!
{predicted_label}
Positive
{confidence}
0.71

Example output

The model likely keyed on surface-positive expressions: «ما شاء الله» ("mashallah", conventionally praise) and «شكراً على التطوير» ("thanks for the improvement"). However, the surrounding context, «ست ساعات وأنا أحاول أجدد البطاقة وكل مرة يعلق» ("six hours trying to renew the card and it keeps freezing"), indicates the praise is sarcastic, a common pattern in Gulf dialect complaints. This is a probable misclassification: the true sentiment is negative. With confidence at 0.71, below the 0.80 threshold, and strong sarcasm signals present, this item should be routed for human review rather than logged as positive feedback.

Publishing entity

saSadeem AI
Author
Noor Al-Emadi
Version
v1.4
Rating
★ 4.3
Runs
1.9K
License
Open
Updated
2026-06-05

Powered by

arabic-dialect-sentiment →

This template is tuned for the sovereign model above. Forked copies keep the same target model binding until republished.

Community · 0 threads

Questions, findings and requests from the entities that use this asset. Owners reply here; threads with upvotes surface to the owning entity's inbox.

No threads yet. Be the first entity to ask, or to share what you found.