A hybrid dependency-based approach for Urdu sentiment analysis.

Area of Science:

• Natural Language Processing (NLP)
• Computational Linguistics
• Artificial Intelligence

Background:

• Social media generates vast amounts of unstructured data daily.
• Existing sentiment analysis (SA) methods often rely on inefficient word co-occurrence frequencies.
• A need exists for more accurate SA, especially for low-resource languages like Urdu.

Purpose of the Study:

• To present a novel framework for concept-level sentiment analysis (SA) in Urdu.
• To enhance the accuracy of SA by moving beyond word-level analysis.
• To leverage both rule-based and deep learning approaches for improved sentiment detection.

Main Methods:

• A comprehensive Urdu dataset was created from YouTube reviews and talks.
• The framework integrates predefined language rules with Deep Neural Networks (DNN).
• It employs a hybrid approach: rule-based sentiment flow from words to concepts, with DNN fallback for sentence classification.

Main Results:

• The proposed concept-level SA framework achieved 6-7% improvement over state-of-the-art methods.
• It outperformed traditional approaches like LSTM, CNN, SVM, LR, and MLP on the Urdu dataset.
• The hybrid rule-based and DNN model demonstrated superior polarity detection accuracy.

Conclusions:

• The research introduces an effective concept-level sentiment analysis framework for Urdu.
• Combining language rules and DNN offers a significant advancement in Urdu SA.
• The framework shows promise for analyzing sentiment in diverse Urdu social media content.