Mining and Visualizing Cost and Schedule Risks from News Articles with NLP and Network Analysis

Cost overruns and schedule delays in US transit projects have been of growing concern for years. Nevertheless, the data availability and sample size have restricted quantitative analysis toward investigating the risks leading to overruns. Innovative data sources and collection methods need to be identified in addition to traditional surveys and case studies. News articles report on issues and risk events leading to overruns as projects progress but have not yet been explored in the construction domain. The difficulty lies in data compilation and analysis. To fill this gap, the paper tested combinations of different natural language processing (NLP) and machine learning methods to automatically identify risk narratives from news articles. The risk sentences are classified into 5 categories and 26 subcategories through a content analysis approach. Then the risks are ranked and analyzed using an appropriate co-occurrence network. The research demonstrates the possibility of integrating NLP and network analysis for exploring publicly available textual documents to explain project performance issues. The approach serves as a baseline for future studies to develop more intelligent models to examine a wide range of media data and other textual reports in the construction domain.