Autoregressive models for session-based recommendations using set expansion.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Session-based recommendation systems personalize user experiences by analyzing short-term preferences.
• Existing sequence-based models struggle with ambiguous or unreliable interaction orders.
• Novel approaches are needed to enhance recommendation accuracy in diverse real-world scenarios.

Purpose of the Study:

• To propose a novel session-based recommendation model, Deep Set Session-based Recommendation (DSETRec), that utilizes a set-based approach.
• To overcome the limitations of sequence-dependent models by treating user interactions as unordered sets.
• To capture item coupling and co-occurrence patterns for improved recommendation performance.

Main Methods:

• Developed Deep Set Session-based Recommendation (DSETRec), a model independent of interaction sequence.
• Implemented DSETRec using a deep autoregressive framework for item prediction and reconstruction.
• Conceptualized session data as unordered sets to capture item relationships.

Main Results:

• DSETRec demonstrated superior performance compared to state-of-the-art baselines on benchmark datasets.
• Achieved significant improvements of 13.2% in P@20 and 11.85% in MRR@20 over sequence-based variants on the Yoochoose dataset.
• Showcased effective generalization across both short and long user sessions.

Conclusions:

• The set-based approach robustly captures unordered interaction patterns, enhancing recommendation systems.
• DSETRec offers a more flexible and generalized solution for session-based recommendations, especially when sequential data is noisy or absent.
• This research lays groundwork for developing advanced recommendation systems adaptable to varied session dynamics.