Graph Representation Learning for Assisting Administrative Penalty Decisions

The application of artificial intelligence opens up a new path for administrative punishment and law enforcement, which is of great significance to the modernization of the country’s governance capacity. Graph representation learning has been widely used in many judicial scenarios. Most existing administrative legal documents for cause determination and penalty decision are made by means of natural language processing. Due to the many representation methods of information elements in the administrative law enforcement documents, the identification of the cause of action and the decision of punishment are difficult to make, which makes the low accuracy and lack of interpretable. In order to solve these problems, this paper constructs a knowledge graph-based information embedding method to effectively embed knowledge graphs into the network, and builds two graph convolutional neural network frameworks based on node classification and graph classification to realise intelligent assisted case determination and penalty decision based on graph representation. The experimental results show that the graph neural network-based framework is a better choice and the results of multi-task classification are significantly better than using only a single task.