The document discusses using latent Dirichlet allocation (LDA) to classify reviews into topics and recommend restaurants. LDA reduced the feature dimension by over 99% and the computation time by over 90% compared to bag-of-words modeling. The topics identified include Japanese, Mexican, brunch, bar atmosphere, pizza, Indian, Asian, fast food, sweets, BBQ, and service. The recommendations are evaluated using a normalized distance-based performance measure to minimize error compared to actual rankings.