An Adoptive Learning Process for Social Media Text data Analysis for Disaster Management

Abstract

Social media is a source of low cost and publically available news and information. Social media users are contributing the information during the disaster events are valuable for timely disaster response. Therefore, mining help related data in early time of social media content is a valuable to understand the disasters situation. In this paper, we proposed an adoptive learning process to analyze the social media data to recognize the disaster event and it’s intensity by using unsupervised learning and sentiment analysis. First, the labeled data is used to train a modified Fuzzy C Means clustering. Additionally, the centroids are updated with incorporating reveled new keywords. The improved centroids are useful for identifying help related social media contents. Next, the sentiment score are used to identify the negative sentiments for identifying the severity of the disaster. In order to classify the text according to their sentiments we utilize the SVM and ANN classifiers. Finally, the results of the proposed model is measured and compared with the baseline model. The results revel that the proposed adoptive learning technique is efficient and enhancing the accuracy with the time.