Sentiment analysis from social media in crisis situations

Abstract

Advent of social media has created an unprecedented environment for people to share their thoughts with the world. These online platforms like facebook, twitter are usually the first resort people turn to in times of crisis to voice their opinions and relay other crucial information. But when it comes to detecting sentiments out of this gigantic pool of opinions, it becomes an arduous task and doing it manually is practically impossible. Hence different methods have been devised to perform automatic polarity classification of textual data. This paper provides a brief overview of different techniques being developed for analyzing social media data, particularly twitter data. We developed a workflow for applying sentiment analysis to a comparatively new domain of natural disasters to detect public emotions in crisis. A base line model is designed and trained on unigram features using Naïve Bayes. The model is further tested on Kashmir floods dataset collected from twitter and an overall accuracy of 67% is achieved. The result provides valuable information which will assist the authorities to strategize their actions with due consideration to public sentiments and hence ameliorate the process of managing such situations.

In the Eye of the Storm: Social Media and Crisis Management

“…Sentiment grammar identifies these and then associates them with relevant targets and the owners of those opinions. Sentiment aggregation combines the scores of each feeling expressed on the networks [97].…”

Building Social Resilience and Inclusion in Disasters: A Survey of Vulnerable Persons' Social Media Use

“…SM provides people with a unique platform to express their thoughts and feelings publicly during times of disasters (Qiu, Xu, Wang, & Gu, 2020). Using data on Twitter referring to the Kashmir floods, Kaur and Kumar (2015) develop a model to assist authorities strategize during disaster based on public sentiments. Kim and Hastak (2018) leveraged data from Facebook in the city of Baton Rouge after the 2016 Louisiana flood to analyse the emergent networks after the flood.…”

Space-Time Surveillance of Negative Emotions after Consecutive Terrorist Attacks in London

“…The rest of the tweets received 0 if they did not express these negative emotions, such as a tweet of the related news. This process is commonly accomplished through sentiment analysis using text-mining techniques, such as supervised or unsupervised machine learning algorithms [32][33][34][35][36][37]. However, these algorithms typically require a large sample dataset to develop classification rules, use the rules to classify the remaining tweets, and then evaluate the performance of these classifiers.…”