| description abstract | Disputes routinely arise in construction projects and significantly affect costs and scheduling. Learning from previous disputes is pivotal for construction contract management. This research focuses on extracting valuable information from government-issued statute that is involved in construction contract dispute, which is underexplored but useful for better construction contract management. The research presented in this study explores and evaluates five typical shallow learning models and four deep learning models for the multilabel text classification task that provide the ability to analyze dispute cases with statute outcomes automatically. Furthermore, model optimizations in some control variables (i.e., model grid search) are conducted to provide constructive model selection suggestions in practical text mining applications. Results show that the text convolution neural network model with 256 filter number and [1,2,3,4] filter size is a suitable backbone architecture for classifying construction dispute cases, which produced the best performance with the P@1(%), P@3(%), P@5(%), NDCG@1(%), NDCG@3(%), and NDCG@5(%) by 65.99, 54.60, 44.32, 65.99, 62.41, and 65.09. In conclusion, the contributions of this research mainly cover the following: (1) exploring and evaluating several multilabel classification models in construction dispute classification tasks and making further model optimizations and (2) the automatic generation of government-issued statutes enabling contract administrators to understand and evaluate the worth of their claims prior to taking it to litigation and therefore put in place strategies to reduce and resolve dispute in construction contract management. | |
