Multi-view knowledge representation learning for personalized news recommendation

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Retrieval

Background:

• Personalized news recommendation faces challenges due to diverse and dynamic user interests.
• Existing models struggle to integrate candidate news into user interest modeling, limiting accuracy.

Purpose of the Study:

• To propose the Multi-view Knowledge Representation Learning (MKRL) framework for enhanced personalized news recommendation.
• To improve user interest modeling by jointly considering user history and candidate news characteristics.

Main Methods:

• Developed MKRL framework with a multi-view news encoder and candidate-aware attention mechanisms.
• Integrated convolutional neural networks and multi-head attention for contextual information capture.
• Dynamically weighed user interactions and candidate news based on relevance.

Main Results:

• MKRL framework demonstrated superior performance over state-of-the-art baselines.
• Achieved enhanced recommendation accuracy and relevance in experiments.
• Validated effectiveness on three real-world datasets.

Conclusions:

• The MKRL framework effectively captures dynamic user interests for personalized news.
• Incorporating candidate news directly into modeling significantly improves recommendation quality.
• The multi-view approach enables richer and more personalized news recommendations.