Collaborative filtering based on nonnegative/binary matrix factorization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Collaborative filtering relies on user-item similarities from rating data, often incomplete.
• Matrix factorization techniques like Nonnegative Matrix Factorization (NMF) predict scores for unrated items.
• Nonnegative/binary matrix factorization (NBMF) extends NMF but is typically used for dense data.

Purpose of the Study:

• To adapt NBMF for sparse data in collaborative filtering.
• To enhance recommendation prediction accuracy.
• To improve computational efficiency.

Main Methods:

• A modified NBMF algorithm was developed for sparse rating matrices.
• Unrated entries in the rating matrix were masked to improve predictions.
• A low-latency Ising machine was utilized for NBMF computation.

Main Results:

• The modified NBMF algorithm demonstrated enhanced prediction accuracy for sparse data.
• The use of an Ising machine significantly reduced computation time.
• The proposed method is beneficial for collaborative filtering applications.

Conclusions:

• Modified NBMF with masked entries and Ising machine offers an effective solution for sparse collaborative filtering.
• This approach improves both accuracy and speed compared to traditional methods.
• The study highlights the potential of NBMF for real-world recommendation systems.