Extracting Useful Emergency Information from Social Media: A Method Integrating Machine Learning and Rule-Based...

Area of Science:

• Social Media Analysis
• Emergency Informatics
• Data Mining

Background:

• Social media user-generated content (UGC) is a vital source for emergency information (EI).
• Extracting accurate EI from vast and varied social media data is challenging, particularly with pure machine learning.
• Existing methods struggle with the heterogeneity and volume of UGC.

Purpose of the Study:

• To propose and evaluate a novel machine learning and rule-based integration method (MRIM) for classifying emergency information from social media.
• To compare the performance of MRIM against pure machine learning and rule-based approaches.
• To identify factors influencing the effectiveness of EI extraction from UGC.

Main Methods:

• Developed a hybrid approach integrating machine learning algorithms with rule-based systems (MRIM).
• Conducted comparative experiments using microblog data from the "July 20 heavy rainstorm in Zhengzhou".
• Analyzed the influence of microblog characteristics (word count, address, contact info) and user attention on EI classification.

Main Results:

• The MRIM demonstrated superior performance in classifying emergency information compared to pure machine learning and rule-based methods.
• Microblog characteristics, including specific details like addresses and contact information, significantly impacted classification accuracy.
• User attention emerged as a key determinant in the effectiveness of emergency information extraction.

Conclusions:

• Integrating machine learning and rule-based methods is a feasible and effective strategy for mining emergency information from social media UGC.
• The findings provide practical insights for enhancing emergency information management systems.
• Optimizing EI extraction requires considering both content features and user engagement metrics.