Analysis of Public Health Concerns using Two-step Sentiment Classification

Abstract

Our aim is to develop a sentiment analysis tool for public health officials to monitor the spreading epidemics in a certain region and time period. Analyzing the public concerns and emotions about health related matters is an important issue to know the spreading of a disease. In this work, sentiment classification of Twitter messages is focused to measure the Degree of Concern (DOC) of the people about a disease spreading. In order to achieve this goal, the disease related tweets are extracted based on time and geographical location. Then, a novel two-step sentiment classification is applied to identify the personal negative tweets. First, the clue-based algorithm is used to classify the personal tweets from non personal tweets by using subjectivity clues. Next, lexicon-based algorithm and Naïve Bayes classifiers are applied to classify negative and non-negative personal tweets. The personal negative tweets are used to measure Degree of Concern. The Public Health Surveillance System (PHSS) is also developed by using visualization techniques such as maps, graphs and charts to visualize the Degree of Concern (DOC) of the epidemic related twitter data. The visual concern graphs and charts can help health specialists to monitor the progression and peaks of health concerns of people for a disease in particular space and time, so that necessary preventive actions can be taken by public health officials. Negation Handling and Laplacian Smoothing techniques are used with Lexicon Based classifier and Naïve Bayes classifier to improve performance.