ACTIVE ONLINE LEARNING FOR SOCIAL MEDIA ANALYSIS TO SUPPORT CRISIS MANAGEMENT

Abstract

The rapid growth of social media platforms has
transformed them into critical sources of real-time
information during emergencies and crisis
situations. Millions of users actively share updates,
images, and opinions, which can provide valuable
insights for authorities if analyzed effectively.
However, the massive volume, unstructured nature,
and presence of irrelevant or misleading content
make it challenging to extract meaningful crisisrelated
information. Traditional systems relying on
manual monitoring or static models fail to handle
real-time data streams efficiently. To address these
challenges, this project proposes an intelligent
system based on Active Online Multiple Prototype
Classification (AOMPC) for real-time crisis
detection and analysis. The system continuously
processes streaming social media data, classifies
posts into relevant crisis categories, and adapts
dynamically through online learning. It
incorporates active learning strategies that request
labeling only for uncertain data, thereby reducing
manual annotation effort while improving accuracy.
The proposed system integrates Natural Language
Processing (NLP), machine learning, and rolebased
access mechanisms to provide a scalable and
secure platform. Administrators can monitor
suspicious users, while investigators can analyze
event-specific data for informed decision-making.
The system is implemented using Python, Django,
and MySQL, ensuring flexibility and scalability. By
enabling real-time crisis identification and filtering
of relevant information, the proposed solution
enhances situational awareness, reduces response
time, and supports efficient crisis management
operations.