Abstract

Cross-Domain ABSA has proved effective in extracting more descriptive sentiment information from the reviews or feedback as an important step in improving the experience of the customers in electronic commerce. This work examines the performance of ABSA in the cross-domain scenario where models trained on the review data of one domain for example electronics domain are applied to another domain such as fashion domain. Here, we put forward a risk mitigation strategy that builds upon transfer learning and domain adversarial training methodologies to enhance the overall resilience and reliability of sentiment estimations across multiple product domains. The proposed model was tested using data obtained from different e-commerce retailers, such as Amazon, eBay, and Alibaba concerning various categories of products including electronics, fashion, and home appliances. The outcome of experiments show better performance over most of the compared methods of single-domain ABSA and cross-domain approaches. The model offered greater accuracy, recall rate, F1-score, and cross-domain efficiency, which demonstrated the model’s effectiveness and versatility. The consequences for e-business companies are significant. Improved sentiment analysis allows businesses to obtain more specific data about customers’ opinions, correct mistakes when developing products, and adjust their advertising approaches. Also, the use of advanced text analytics to measure and monitor such aspects in different product areas offers a competitive edge and improves product innovation decisions. However, there are several limitations to the study itself, such as the variations that may arise among the domain, and the limited availability of the data. Further work is to be done on more complex and sophisticated methods of domain adaptation, using external resources; the main model itself should also be much faster and more efficient in scalability. This work thus indicates an opportunity for cross-domain ABSA to generate practical insights for enhancing customers' experience within e-commerce.

Statistics