Understanding Common Perceptions from Online Social Media

TL;DR

This paper develops probabilistic language models to extract collective perceptions from geotagged Twitter posts, demonstrating their application in understanding public opinions and supporting power grid restoration after storms.

Contribution

It introduces a novel probabilistic modeling approach to analyze geotagged social media content for insights into collective perceptions across different geographical areas.

Findings

Models successfully extract perceptions from Twitter data.

Application demonstrates utility in power grid restoration.

Method applicable to various geographic and topical analyses.

Abstract

Modern society habitually uses online social media services to publicly share observations, thoughts, opinions, and beliefs at any time and from any location. These geotagged social media posts may provide aggregate insights into people's perceptions on a bad range of topics across a given geographical area beyond what is currently possible through services such as Yelp and Foursquare. This paper develops probabilistic language models to investigate whether collective, topic-based perceptions within a geographical area can be extracted from the content of geotagged Twitter posts. The capability of the methodology is illustrated using tweets from three areas of different sizes. An application of the approach to support power grid restoration following a storm is presented.