Modeling and Applying Implicit Dormant Features for Recommendation via Clustering and Deep Factorization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• E-commerce recommender systems (RS) face performance degradation with large datasets.
• Dataset sparsity and the need for effective feature extraction are key challenges.
• Implicit features and user-generated tags can enhance recommendation quality.

Purpose of the Study:

• To improve the performance of e-commerce recommender systems.
• To leverage hierarchical implicit features and customer tag information for better predictions.
• To address data sparsity and cold-start issues in recommendation.

Main Methods:

• Customer clustering based on a customer rating matrix.
• Deep factorization of a weighted nonnegative matrix factorization (WNMF) model.
• Integration of customer tag information to regularize the factorization process.

Main Results:

• Achieved Mean Absolute Error (MAE) of 0.8011 with 60% data and 0.7965 with 80% data.
• Demonstrated MAE rates of 0.8781 and 0.9046 in 50 and 100 customer cold-start scenarios.
• Outperformed baseline models that used individual features (customer, product, or tags).

Conclusions:

• The proposed method effectively enhances recommender system performance.
• Combining hierarchical implicit features with tag information improves rating prediction accuracy.
• The approach shows viability in handling sparse data and cold-start scenarios.