Self-supervised graph neural network with pre-training generative learning for recommendation systems.

Area of Science:

• Computer Science
• Artificial Intelligence
• Law & Technology

Background:

• Current manual or random case assignment systems in procuratorates lead to unbalanced case distribution, impacting judicial fairness and efficiency.
• Real-world data often exhibits a power-law distribution in prosecutor-case category relationships, posing challenges for traditional assignment methods.

Purpose of the Study:

• To propose an intelligent case assignment system that ensures rational and balanced distribution of cases.
• To address the limitations of existing case assignment methods by leveraging graph neural networks and self-supervised learning.

Main Methods:

• Developed an end-to-end Self-supervised Graph neural network model with Pre-training Generative learning for Recommendation (SGPGRec).
• Designed three auxiliary self-supervised tasks utilizing prosecutor-case category interaction graphs and data distribution to capture node features and correlations.
• Constructed a graph neural network recommendation model incorporating power-law data characteristics for improved data representation.

Main Results:

• SGPGRec effectively captures self-supervised signals from intra-node features and inter-node correlations.
• The model generates improved data representations through pre-training, enhancing recommendation accuracy.
• Extensive experiments on real-world datasets demonstrated the effectiveness of SGPGRec compared to baseline methods.

Conclusions:

• The proposed SGPGRec model offers an effective solution for rational case assignment in procuratorates.
• This approach enhances judicial fairness and efficiency by mitigating unbalanced case distribution.
• The study supports the development of intelligent systems for optimized case management.