Loading...

Development of a Deep Learning and Natural Language Processing Based Method in Order to Extract Risky Clauses of Construction Contracts

Kazemi, Mohammad Hossein | 2022

171 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55113 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Alvanchi, Amin
  7. Abstract:
  8. One of the most significant factors for the on-time and successful implementation of construction projects is contract management. Proper management of construction contracts and assessment of potential risks in the bidding process and before its signing have a significant impact on preventing or reducing the occurrence of claims and disputes between the contract parties at various stages of the project. In this research, using the latest deep learning (DL) and natural language processing (NLP) state-of-the-art methods, and various deep neural networks (DNN) architectures a model has been developed for extracting Persian contract risk-prone clauses. In addition, this study provides a comprehensive comparison between various NLP methods in both the stages of data preprocessing and the training of deep neural networks. Studies have shown that, given the small size of the existing corpus, the best approach to the embedding layer is to use the pre-trained model of fastText instead of training the Word2Vec model on all data from scratch. In addition, among all the classical DNNs used, the best results were related to the model based on long short-term memory (LSTM) with an accuracy of 85% and an F1-score of 84%. After this, the results of the bidirectional encoder representation from transformers (BERT) based model have shown the best results with 85% accuracy and 84% F1-score. The following three models are gated recurrent unit (GRU), bidirectional LSTM (BiLSTM), and convolutional neural network (CNN) in case of accuracy and F1-score.


  9. Keywords:
  10. Natural Language Processing ; Deep Learning ; Bidirectional Encoder Representations from Transformers (BERT)Model ; Machine Learning ; Transformer Learning ; Risky Clauses ; Contract Risks

 Digital Object List

 Bookmark

No TOC