KFEA: Fine-Grained Review Analysis Using BERT with Attention: A Categorical and Rating-Based Approach

User reviews contain many key phrases that are crucial for business understanding, but they are often obscured by the sheer volume of reviews. Extracting key phrases from user reviews could help to understand what users are concerned about and provide timely improvement suggestions. Current pattern-based methods for target phrase extraction usually analyze reviews at a coarse-grained level, making the extracted topics unfocused and useless. Hence, in order to address this issue, we proposed a fine-grained analysis approach (KFEA) to extract, cluster, and visualize key phrases from e-commerce reviews. In order to fully utilize the relevant information from comments, KFEA fuses the information like categories and ratings from a large volume of user reviews, and then extracts key phrases with the help of a pre-trained model. A method is also designed to cluster and visualize the extracted key phrases for business understanding. Our evaluation on 6,088 reviews from 6 products shows that KFEA can effectively extract key phrases and perform clustering and visualization. In particular, KFEA achieved an precision of 76.6% and a recall of 81.8% in extracting key phrases from manually annotated data. KFEA’s cross-categories effectiveness is also validated on 16,772 reviews from products like mobile phones, laptops, and furniture.