Causes of Action Identification in Disputes over Construction Project Contracts Using Hierarchical Learning Based on BERT

Causes of action identification enables appropriate classification of legal cases, which is of substantial significance for case management and resolution. This paper explores the application of artificial intelligence in identifying causes of action in disputes over construction contracts. This study utilizes a dataset comprising public judgment documents and employs the TextRank method to condense judgment documents while maintaining essential information. The classification model incorporates BERT for embedding and the fully connected layer for classification. A hierarchical learning method is proposed to address the performance decrease due to dataset imbalance. The initial dataset is employed to create several training sub-datasets, in which categories with fewer data are merged into a single category to approximate the balance in the data distribution. Each training sub-dataset is used to train a separate classification model. The trained models are then applied sequentially for text classification, and a hash table is used to correlate classification results with their respective causes of action. The effectiveness of this method is validated through 5-fold cross-validation and benchmarked against other methods, demonstrating the superiority of the proposed method in identifying causes of action. In conclusion, this paper presents an innovative method for identifying the cause of action via artificial intelligence.