222 / 1971-01-01 00:00:00

Summarize Online Product Reviews With Two Stage Kernel Estimation

clustering,information extraction,product reviews

We design and experiment with an innovative way to automatically generate product features from reviews. We extract opinions from each review, clusters them by their orientation through an unsupervised learning. From these clustered opinions, we estimate the product feature kernel and the weight kernel, and update the word polarity by minimizing the prediction error with supervised learning. Experiments show that the estimated parameters are reasonable and the outputs provide useful product information.