Spatiotemporal‐based sentiment analysis on tweets for risk assessment of event using deep learning approach

TLDR

Social media is essential for capturing real‑time public emotions during disasters, and sentiment analysis extracts patterns from user feedback. The study aims to analyze tweets about a specific disaster in a particular location over time to assess risk. An LSTM network with word embeddings generates keywords, which the RASA algorithm uses to classify tweets and compute sentiment scores per location, and the model is benchmarked against several state‑of‑the‑art classifiers in binary and multiclass settings. RASA achieves a 1 % improvement over XGBoost in binary classification and about 30 % better performance in multiclass tasks, enabling governments to implement preventive measures.

Abstract

Summary Social media plays a vital role in analyzing the actual emotions of people after and during a disaster. Sentiment analysis is a method to detect a pattern from the emotions and feedback of the user. The main objective of the proposed work is to perform sentiment analysis on the tweets on a specific disaster context for a particular location at different intervals of time. LSTM network with word embedding algorithm is used to derive keywords based on the history of tweets and the context of the tweets. The proposed algorithm risk assessment sentiment analysis (RASA) uses the keywords generated from the network to classify the tweets and sentiment score for each location is identified. The model is validated with various state‐of‐art algorithms, namely, support vector machine, Naive‐Bayes, maximum entropy, logistic regression, random forest, XGBoost, stochastic gradient descent, and convolution neural networks in 2‐fold scenario: one for binary class and the other multiclass with three target classes. The results infer that the proposed RASA performs better in a binary class scenario with an increase of 1% when compared with XGBoost and 30% in multiclass scenario on an average when compared with all the other techniques. The model helps the government to take preventive measures to manage the posteffect of the disaster event in a location.