Topic Modeling and Text Mining to Improve Gas Related Building Safety
62 million American homes use gas to power their homes. When requesting service, or annual maintenance, reports can be ambiguous, hiding the severity of problems with a particular system, leading to leaks, explosions, which result in injury, death and property damage. The topic modeling algorithm can automatically generates topics (or 'themes') from service notification notes. The results show that topics account for a particular pattern of word use that correlate with patterns of occurrence and observation of risk. Text mining can then used to develop a risk tool to classify incidents as high or low risk. This can help inspectors prioritize different buildings and enforce change when necessary.