LFG: A Generative Network for Real-Time Recommendation

Junyi Liu

2023-11-25

Abstract:Recommender systems are essential information technologies today, and recommendation algorithms combined with deep learning have become a research hotspot in this field. The recommendation model known as LFM (Latent Factor Model), which captures latent features through matrix factorization and gradient descent to fit user preferences, has given rise to various recommendation algorithms that bring new improvements in recommendation accuracy. However, collaborative filtering recommendation models based on LFM lack flexibility and has shortcomings for real-time recommendations, as they need to redo the matrix factorization and retrain using gradient descent when new users arrive. In response to this, this paper innovatively proposes a Latent Factor Generator (LFG) network, and set the movie recommendation as research theme. The LFG dynamically generates user latent factors through deep neural networks without the need for re-factorization or retrain. Experimental results indicate that the LFG recommendation model outperforms traditional matrix factorization algorithms in recommendation accuracy, providing an effective solution to the challenges of real-time recommendations with LFM.

Information Retrieval

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper primarily attempts to address the issues of insufficient flexibility and high computational cost in traditional matrix factorization-based recommendation systems for real-time recommendations. Specifically:

1. **Challenges of Real-time Recommendation**:
- Traditional recommendation algorithms based on latent factor models (LFM), such as SVD and its variants, require re-performing matrix factorization and gradient descent training when facing new users, leading to high computational costs.
- These algorithms lack the flexibility needed to handle real-time recommendations.

2. **Proposed New Method**:
- The paper proposes a new generative network model—Latent Factor Generator (LFG), which can dynamically generate users' latent factor matrices without the need for re-performing matrix factorization or gradient descent training.
- The LFG model uses deep neural networks to generate user latent factors, thereby achieving flexible real-time recommendations.

3. **Experimental Validation**:
- The paper demonstrates through cross-validation experiments on the MovieLens-100k and MovieLens-1m datasets that the LFG model outperforms traditional SVD and BiasSVD models in terms of basic rating prediction accuracy and real-time recommendation accuracy.

In summary, this paper aims to provide an efficient and accurate real-time recommendation solution through the LFG model, reducing the computational cost of real-time updates in traditional recommendation systems and improving recommendation accuracy.