Urdu Aspect-Category-Opinion-Sentiment (UACOS) Quadruple Extraction: A Transfer Learning Approach

Aspect-based sentiment analysis (ABSA) has gained substantial attention for extracting detailed insights from text, such as product reviews. In this study, we initiate the task of Urdu Aspect Category Opinion Sentiment (UACOS) Quadruple Extraction to improve sentiment analysis for Urdu. This task involves extracting aspect-category-opinion-sentiment quadruples, capturing both explicit and implicit sentiment in Urdu text. We adapt the well-established Aspect-Category-Opinion-Sentiment (ACOS) frame-work to the Urdu language, creating the ACOS dataset tailored to its unique linguistic and cultural characteristics. To process this dataset, we employ transfer learning, adapting models like TAS-BERT-ACOS, JET-ACOS, and Extract-Classify-ACOS, originally designed for English, to perform quadruple extraction in Urdu. Our work addresses key challenges, such as the lack of fully annotated datasets for implicit sentiment analysis and the complexity of classifying multiple sentiment components simultaneously. The results indicate the potential of adapting advanced sentiment analysis models to low resource languages, providing a robust framework for extracting sentiment information from Urdu text and paving the way for future research in underrepresented languages. Among the models evaluated, the Extract-Classify-ACOS method achieved the highest F1 measure on both the restaurant-UACOS and laptop-UACOS datasets, demonstrating the effectiveness of this approach for quadruple identification of aspect categories, and opinion sentences in Urdu.