Authors: Seyoung Park, Junegak Joung, Harrison Kim
This paper presents a new artificial intelligence method for analyzing online customer reviews to help companies better understand what users think about specific parts of a product. While previous tools could identify general categories like “camera,” they often struggled to distinguish between more detailed “subfeatures,” such as a phone’s front versus rear camera. To solve this, the researchers used a sophisticated language model called SBERT to turn review sentences into numerical data that capture the context and specific meaning of words. They also developed a specialized “loss function” to ensure the AI performs accurately even when some product features are mentioned much less often than others in the data. The system was tested on thousands of real-world reviews for smartphones and headphones from websites like Amazon. The results showed that this new approach was significantly more accurate than older models, achieving high performance scores even for rare or complex customer comments. By using this method, businesses can gain practical, detailed insights from customer feedback to improve product designs in their early stages.
Fig. 1 Conceptual overview for extracting subfeature-level design insights from customer reviews