Sequential Recommendation Based on Multivariate Hawkes Process Embedding With Attention.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Recommender systems are crucial for managing information overload in big data.
• Temporal and sequential information can enhance recommendation performance but are challenging to fully exploit.
• Existing methods struggle to accurately model complex user sequential patterns and preferences.

Purpose of the Study:

• To propose a novel sequential recommendation model, multivariate Hawkes process embedding with attention (MHPE-a).
• To accurately model users' temporal interaction sequences and preferences.
• To improve the quality of recommendations and user experience by leveraging both long-term and short-term user preferences.

Main Methods:

• Developed MHPE-a, combining a multivariate Hawkes process with an attention mechanism.
• Utilized the multivariate Hawkes process to model users' sequential patterns in temporal interaction data.
• Employed an attention mechanism to adaptively leverage users' long-term and short-term preferences for accurate predictions.

Main Results:

• MHPE-a accurately models users' sequential patterns.
• The model effectively captures both long-term and short-term user preferences.
• Experiments on lastfm and gowalla datasets demonstrate superior performance compared to state-of-the-art baselines.

Conclusions:

• MHPE-a offers a powerful approach for sequential recommendation by integrating temporal dynamics and attention.
• The model enhances recommendation accuracy and user satisfaction.
• This work advances the field of sequential recommendation systems in the context of big data.