There is a massive amount of review data on e-commerce platforms, from which product designers can identify product feature keywords and corresponding emotional polarity, which can quickly understand users’ satisfaction with product features. This research uses aspect-based sentiment analysis (ABSA) technology to construct the BERT + Bilstm + CRF model and BERT fine-tuning model, and uses a small amount of review data for model training. Based on the F1 value of the model, a more effective model is selected to automatically extract the product feature keywords and corresponding emotional polarity information evaluated by users from the massive review data, in order to assist designers in quickly understanding users’ satisfaction with each product feature and determining the direction of product iterative design, thereby improving user satisfaction.