SVD-CNN: A Convolutional Neural Network Model with Orthogonal Constraints Based on SVD for Context-Aware Citation...

Area of Science:

• Computer Science
• Information Science
• Artificial Intelligence

Background:

• Context-aware citation recommendation aids researchers in scientific writing.
• Existing neural network methods struggle with overcorrelation in weight matrices, hindering semantic similarity.
• Accurate citation prediction is crucial for scientific literature development.

Purpose of the Study:

• To propose a novel context-aware citation recommendation approach.
• To improve the orthogonality of the weight matrix in citation recommendation models.
• To explore more accurate citation patterns and enhance link prediction.

Main Methods:

• Developed a novel neural network-based approach for context-aware citation recommendation.
• Focused on improving weight matrix orthogonality to enhance semantic similarity.
• Investigated the interactional features of reference patterns for link prediction.

Main Results:

• The proposed model demonstrated improved orthogonality of the weight matrix.
• Quantitative analysis showed that reference patterns significantly affect link prediction.
• Experiments on CiteSeer datasets confirmed the model's superiority over baseline methods in all metrics.

Conclusions:

• The novel approach effectively addresses overcorrelation issues in citation recommendation.
• The method enhances the discovery of accurate citation patterns.
• This work offers a superior solution for context-aware citation recommendation.