Star ratings compress a lot of nuance into a single number. This project mines the opinions embedded in product review text — not just sentiment polarity, but the specific attributes reviewers praise or criticize — and tests whether recommendations built on that mined opinion structure outperform recommendations built on ratings alone.

The work was presented at the 2025 POM Annual Meeting as part of a broader research thread on turning unstructured consumer text into structured signal for downstream decision systems.