Aspect-based sentiment analysis for restaurant reviews on food items

• المؤلف الرئيسي: Liu, Zhixuan
• مؤلفون آخرون: Hui Siu Cheung
• التنسيق: Final Year Project
• اللغة: English
• منشور في: Nanyang Technological University 2023
• الموضوعات: Engineering::Computer science and engineering
• الوصول للمادة أونلاين: https://hdl.handle.net/10356/166061

This project investigates techniques for performing aspect-based sentiment analysis (ABSA) on restaurant reviews on food items, which may be deployed to the existing FoodHunter system, a food review system for canteens at Nanyang Technological University (NTU). The task is divided into two subtasks: food item extraction and ABSA using the extracted food items. Approaches are investigated and evaluated using a newly created dataset for ABSA on food items, which was adapted from the SemEval 2014 dataset through manual data cleaning and annotations. Two approaches for food item extraction are implemented and evaluated, and the N-gram Dictionary-based method is selected due to its efficient processing time compared to that of Noun-Phrase-T5 method with similar F1-score. For ABSA, two approaches are implemented and compared, and the PyPi ABSA Pipeline method is chosen over BERT for ABSA as Sentence Pair Classification due to its higher weighted average F1-score. Case studies are carried out to clarify the advantages and drawbacks of the models. Besides obtaining a final working model for ABSA, this project also demonstrates the power and efficiency of transfer learning and pre-trained language models in complex Natural Language Processing (NLP) tasks. It also highlights the importance and usefulness of available tools and packages for various tasks.