Prompt-Based Multi-Interest Learning Method for Sequential Recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Sequential recommendation systems predict user behavior based on historical interactions.
• Multi-interest learning methods aim to capture diverse user preferences from interaction sequences.
• Existing methods often overlook the distinct learning objectives of interest extraction and aggregation, and neglect the dispersion of user interactions.

Purpose of the Study:

• To propose a novel prompt-based multi-interest learning method (PoMRec) for sequential recommendation.
• To address limitations in existing methods regarding the handling of user interaction data and learning objectives.
• To improve the accuracy of next-item prediction by comprehensively learning user interests.

Main Methods:

• Introduced a prompt-based approach to adapt user interactions for multi-interest extractors and aggregators.
• Utilized both mean and variance embeddings of user interactions for comprehensive interest learning.
• Evaluated the proposed method on three public benchmark datasets.

Main Results:

• The proposed PoMRec method significantly outperforms existing state-of-the-art multi-interest learning approaches.
• Experimental results demonstrate the effectiveness of prompt-based adaptation and the use of mean/variance embeddings.
• PoMRec achieves superior performance in predicting user interests and subsequent interactions.

Conclusions:

• Prompt-based multi-interest learning offers a promising direction for advancing sequential recommendation.
• Considering both centrality and dispersion of user interactions, along with tailored learning objectives, is crucial for effective interest modeling.
• PoMRec provides a robust framework for capturing multifaceted user interests in sequential recommendation tasks.