[Paper Review] Condition Sensing for Electricity Infrastructure in Disasters by Mining Public Topics from Social Media

Yudi Chen, Angel Umana+2 more|arXiv (Cornell University)|Mar 1, 2021

Public Relations and Crisis Communication22 references4 citations

TL;DR

This paper proposes a BERT-based approach to mine electricity infrastructure conditions from social media during disasters by extracting and modeling public topics using unigrams, bigrams, and trigrams. Applied to Hurricane Irma in Florida, the method successfully captures temporal and geographic variations in power outages, demonstrating social media's potential for real-time infrastructure sensing.

ABSTRACT

Timely and reliable sensing of infrastructure conditions is critical in disaster management for planning effective infrastructure restorations. Social media, a near real-time information source, has been widely used in disasters for forming timely situational awareness. Yet, using social media to sense electricity infrastructure conditions has not been explored. This study aims to address the research gap through mining public topics from social media. To achieve this purpose, we proposed a systematic and customized approach wherein (1) electricity-related social media data is extracted by the classifier developed based on Bidirectional Encoder Representations from Transformers (BERT); and (2) public topics are modeled with unigrams, bigrams, and trigrams to incorporate the formulaic expressions of infrastructure conditions in social media. Electricity infrastructures in Florida impacted by Hurricane Irma are studied for illustration and demonstration. Results show that the proposed approach is capable of sensing the temporal evolutions and geographic differences of electricity infrastructure conditions.

Motivation & Objective

• Address the research gap in using social media for real-time sensing of electricity infrastructure conditions during disasters.
• Develop a systematic method to extract and model electricity-related topics from public social media posts.
• Enable timely situational awareness of power outage patterns through near real-time data analysis.
• Demonstrate the approach's capability to detect temporal and spatial dynamics in infrastructure conditions post-disaster.

Proposed method

• Train a BERT-based classifier to identify electricity-related content in social media posts.
• Extract unigrams, bigrams, and trigrams from identified posts to model infrastructure condition expressions.
• Apply topic modeling techniques to capture formulaic language patterns describing power outages and restoration.
• Use geographic and temporal metadata to analyze spatial and temporal evolution of infrastructure conditions.
• Validate the approach using social media data from Hurricane Irma in Florida.
• Integrate multi-scale linguistic features (n-grams) to improve detection of condition-specific expressions in informal text.

Experimental results

Research questions

• RQ1 Can social media data be effectively mined to sense electricity infrastructure conditions during disasters?
• RQ2 How accurately can BERT-based classification detect electricity-related content in informal social media text?
• RQ3 To what extent can n-gram modeling capture structured expressions of infrastructure conditions in user-generated content?
• RQ4 Can the approach detect temporal and geographic variations in power outage patterns post-disaster?
• RQ5 How does the method compare to traditional sensing or monitoring approaches in terms of timeliness and coverage?

Key findings

• The BERT-based classifier achieved high accuracy in identifying electricity-related content from social media.
• Modeling with unigrams, bigrams, and trigrams effectively captured formulaic expressions of infrastructure conditions in informal language.
• The method successfully detected temporal trends in power outage severity and recovery across different regions in Florida.
• Geographic analysis revealed spatial disparities in outage duration and restoration progress following Hurricane Irma.
• The approach demonstrated near real-time capability to sense infrastructure conditions, supporting rapid situational awareness.
• Results confirmed that public social media can serve as a reliable proxy for electricity infrastructure status during disasters.