Jan Joshua Velasco, Daniel Hans Tan, Christine Peña, Ken Gorro
Data gathering in a large archipelago is a slow process. This study aims to look into the validity of text in social media posts as an alternative to the manual data gathering technique of collecting from the source to reduce overhead costs. Collecting text from social media platforms Facebook and Twitter have been done and natural language processing techniques such as web mining and text classification have been applied to see if the data is viable as an alternative source and if there is correlation between these posts to reports from the Bureau of Fire Protection (BFP). Results after 10-fold classification and cross validation show that accuracy in classifying is at 77.6%. Only fires with high severity have been completely reported in social media at 100%, while low-severity fires are reported at an average of 17%. Social media posts also have a positive correlation to the fire reports from the BFP at 72.8%. This means that social media posts can be an alternative to manual data gathering for high severity fires, but not for fires with low severity. © 2017 Association for Computing Machinery.
University of San Carlos, Department of Computer and Information Science, Philippines
