A Hybrid Recommender System Based on Autoencoder and Latent Feature Analysis.

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Recommender systems (RS) are crucial for extracting insights from big data.
• Identifying users' latent preferences is a key challenge in RS implementation.
• Latent Feature Analysis (LFA) and deep neural networks (DNNs) are popular but have limitations.

Purpose of the Study:

• To propose a novel hybrid recommendation model, AutoLFA, that overcomes limitations of individual LFA and DNN-based approaches.
• To enhance recommendation performance by integrating the strengths of Autoencoder and LFA models.

Main Methods:

• Developed AutoLFA, a hybrid model combining Autoencoder and LFA.
• Constructed separate recommendation models using Autoencoder and LFA in distinct metric spaces.
• Integrated these models using a self-adaptive weighting strategy.

Main Results:

• Evaluated AutoLFA on five real-world recommendation datasets.
• Demonstrated significantly superior recommendation performance compared to seven state-of-the-art models.
• Validated the effectiveness of the hybrid approach in diverse scenarios.

Conclusions:

• The proposed AutoLFA model effectively leverages the complementary strengths of Autoencoder and LFA.
• Hybridization with adaptive weighting offers a robust solution for improving recommender systems.
• AutoLFA provides a significant advancement in personalized recommendation technology.