The Internet's widespread use has had a significant impact on electronic commerce. The trend of review-oriented consumption, where consumers rely on customer reviews of a film, is gaining popularity in the market. E-commerce platforms face a significant challenge in accurately interpreting user sentiments from the large volume of customer evaluations. This research suggests a BERT-based ecommerce reviews sentiment analysis algorithm to address the aforementioned issues[3]. Our approach to researching sentiment analysis involves analysing annotated data and labelling entities using the BIO (B-begin, I-inside, Ooutside) data labelling pattern. By utilizing this method, we are able to accurately identify and classify entities within the data, and determine their sentiment. Based on experimental findings on the Taobao cosmetics review datasets, our approach has demonstrated significant improvements in both accuracy rate and F1 score when compared to conventional deep learning methods.