Dual-Tower Counterfactual Session-Aware Recommender System.

Area of Science:

• Information Systems
• Artificial Intelligence
• Recommender Systems

Background:

• Modern information systems like e-commerce and streaming services face challenges in managing user preferences for accurate recommendations.
• Existing session-aware recommender systems (SARSs) often suffer from spurious correlations, leading to biased and unreliable predictions.
• Information theory and causal inference are key to addressing uncertainty and improving recommendation robustness.

Purpose of the Study:

• To develop an innovative approach for SARSs that integrates static long-term and dynamic short-term user preferences.
• To mitigate biases and unreliable correlations prevalent in current recommendation models.
• To enhance the robustness and accuracy of next-item prediction using a counterfactual causal framework.

Main Methods:

• A counterfactual causal framework is employed to analyze the causal influence of long-term preferences on next-item selection.
• A dual-tower architecture is utilized, incorporating a novel data augmentation process.
• A self-supervised training strategy is implemented to address inherent data biases and spurious correlations.

Main Results:

• The proposed method effectively distinguishes between genuine causal effects and spurious correlations in user behavior.
• Experimental results demonstrate superior performance compared to existing benchmark methods in session-based recommendation tasks.
• The approach enhances the reliability and accuracy of predicting user next-item selections.

Conclusions:

• The counterfactual causal framework offers a robust solution for improving session-aware recommender systems.
• Integrating long-term and short-term preferences within this framework leads to more accurate and less biased recommendations.
• This work advances the field of recommender systems by providing a method to overcome common prediction pitfalls.