Description of Invention
The practise of examining people's attitudes on a variety of topics is known as sentiment analysis, sometimes known as opinion mining. This includes analysing products, services, events, and problems. Consumers and businesses both depend on online reviews. Reviews are becoming increasingly significant, particularly in the E-commerce sector. Bidirectional Encoder Representations from Transformers (BERT), is used in this study to analyse the sentiment in reviews of Malaysia's Lazada and Shopee products. The methodology involves using dataset from Lazada and Shopee Malaysia where, 7020 customer reviews were collected. The BERT Model workflow involved pre-processing, fine-tuning, validation and test stages on English and Malay datasets. The training dataset contained a total of 5040 reviews for training, while both validation and test datasets contained a total of 1080 reviews respectively. In order to determine the effectiveness and accuracy of the BERT Model, a confusion matrix was generated. The achieved results exhibit substantial accuracy, with the English model achieving 90.73% and the Malay model 84.15% accuracy. The BERT sentiment analysis models show promising results, with high accuracy and balanced performance across various sentiment categories. Further optimizations and evaluations could potentially improve model performance and real-world applicability.