Adaptive Deep Modeling of Users and Items Using Side Information for Recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence

Background:

• Matrix factorization (MF) is a common technique in recommender systems but often overlooks individual user preference diversity.
• A fixed user preference representation in MF leads to inaccurate and repetitive recommendations.

Purpose of the Study:

• To propose a novel adaptive deep latent factor model (ADLFM) that addresses the limitations of traditional MF.
• To enhance recommender systems by accurately modeling individual user preferences and item characteristics.

Main Methods:

• Developed ADLFM, a latent factor model that adaptively learns user preference factors based on specific items.
• Introduced a new user representation method using rated item descriptions, not just ratings.
• Implemented a deep neural network framework with an attention mechanism for adaptive user representation learning.

Main Results:

• ADLFM significantly outperformed existing state-of-the-art baseline methods in extensive experiments on Amazon datasets.
• Further analysis confirmed the substantial contribution of the attention factor in improving the model's performance.

Conclusions:

• ADLFM effectively models individual user diversity, leading to more accurate and personalized recommendations.
• The proposed adaptive approach and attention mechanism are crucial for enhancing recommender system performance.