The Method for Generating Recommended Candidates through Prediction of Multi-Criteria Ratings Using CNN-BiLSTM

Multi-criteria ratings improve recommendation accuracy, but asking users to enter a rating per criterion is a burden. So the approach was to infer per-criterion ratings from review text instead.

• CNN-BiLSTM derives per-criterion ratings from reviews, aggregated through linear regression to predict the overall rating
• The learned weights are interpreted as the user’s priorities, and a new score matrix is built for recommendation
• User–item similarity is computed against those priorities to generate recommendation candidates

Evaluated on real TripAdvisor data, outperforming a general SVD-based recommender.

CC BY-NC 3.0