Conv4Rec: A 1-by-1 Convolutional Autoencoder for User Profiling Through Joint Analysis of Implicit and Explicit...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Traditional recommender systems often struggle to effectively integrate diverse user interaction data.
• Existing models may not fully leverage the nuances of both explicit user ratings and implicit behavioral patterns.

Purpose of the Study:

• To introduce a novel convolutional autoencoder architecture for enhanced user modeling and recommendation.
• To develop a unified model capable of learning from both explicit ratings and implicit feedback simultaneously.
• To improve the accuracy and interpretability of content recommendations.

Main Methods:

• Developed a flexible convolutional autoencoder architecture for user and item interaction modeling.
• Integrated explicit user ratings with implicit feedback from sampling patterns.
• Enabled separate predictions for content consumption probability and high rating likelihood.
• Derived generalization bounds for autoencoders in recommender systems.

Main Results:

• Achieved state-of-the-art performance on both implicit and explicit feedback prediction tasks.
• Demonstrated the model's ability to identify items users might enjoy but not discover naturally.
• Provided new theoretical generalization bounds for autoencoders in recommender systems.
• Showcased improved prediction informativeness and interpretability.

Conclusions:

• The proposed convolutional autoencoder offers significant advancements in user modeling and recommendation.
• Jointly learning from explicit and implicit feedback leads to superior performance and novel insights.
• The model's ability to predict potential enjoyment enhances user experience and content discovery.