AI for business intelligence

• Main Author: Chong, Yow Lim
• Other Authors: Erik Cambria
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/166313
• Institution: Nanyang Technological University

Owing to the rapid advancements of the Internet, consumers can communicate with others around the world to share their opinions and thoughts on different parts of a business or company, ranging from the products offered to the reputation of the brand. Sentiment analysis on consumers’ reviews is key to understanding consumers’ perception of and feelings towards businesses. In this project, several different models were investigated to perform aspect-based sentiment analysis on restaurant reviews. The aim was to create a model pipeline to not only extract the aspect categories mentioned in each review, but also to obtain the sentiment directed towards each aspect category. This allows restaurants to identify areas of strengths and weaknesses, enabling them to take measures to improve their image and business in customers’ eyes. The project began with sourcing for a publicly available dataset for aspect-based sentiment analysis. Data pre-processing was done before the data was fed to various models for training. Performance metrics like macro F1-score and accuracy were used for evaluation. This project explored several models, including support vector machines, neural networks like gated recurrent units and convolutional neural networks, and transformer-based models like BERT, among others. A case study was done using the best performing models to demonstrate an application of this project. The results showed that the transformer-based models outperformed other models in all metrics. At the same time, the neural network models showed relatively poor performance, owing to several gaps or limitations of the project. Possible solutions to address these limitations could be implemented in future work. Ultimately, this study provides useful insights and recommendations for restaurants or businesses to utilise sentiment analysis for their business needs. It is recommended that businesses use pre-trained language models like BERT for sentiment analysis, due to its ease of use, power, and versatility.