Contrastive cross-domain sequential recommendation via emphasized intention features.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Cross-domain sequential recommendation aims to predict future user interactions using historical data from multiple domains.
• Existing methods often struggle with noisy representations from undifferentiated cross-domain information and neglect joint user preference modeling.
• This leads to suboptimal performance in personalized recommendation systems.

Purpose of the Study:

• To propose a novel model, C²DREIF, for improved cross-domain sequential recommendation.
• To effectively integrate single-domain and cross-domain information while constraining cross-domain data noise.
• To simultaneously capture users' long-term and short-term preferences for accurate intent extraction.

Main Methods:

• Utilizes Gaussian graph encoders for information representation, constraining correlations and capturing contextual information.
• Employs a Top-down transformer to extract user intents, considering both long-term and short-term preferences.
• Applies entropy regularization to enhance contrastive learning and reduce randomness from negative samples.

Main Results:

• The proposed C²DREIF model effectively constrains cross-domain information, reducing noise during representation generation.
• It accurately captures user intents by jointly considering long-term and short-term preferences across domains.
• Enhanced contrastive learning improves the model's robustness and accuracy.

Conclusions:

• C²DREIF offers a significant advancement in cross-domain sequential recommendation by addressing key limitations of existing approaches.
• The model demonstrates superior performance in personalized preference extraction and prediction.
• Future work could explore further refinements in cross-domain information fusion and user intent modeling.