A Classifier to Detect Informational vs. Non-Informational Heart Attack Tweets

Social media sites are considered one of the most important sources data in many fields, such as health, education, and politics. While surveys provide explicit answers to specific questions, posts social have same implicitly occurring text. This research aims develop a method for extracting implicit from large tweet collections, demonstrate this an concern: problem heart attacks. The approach is collect tweets containing “heart attack” then select those ones with useful information. Informational which express real attack issues, e.g., “Yesterday morning, my grandfather had while he was walking around garden.” On other hand, there non-informational “Dropped iPhone first time almost attack.” starting point manually classify 7000 either informational (11%) or (89%), thus yielding labeled dataset use devising machine learning classifier that can be applied our collection over 20 million tweets. Tweets were cleaned converted vector representation, suitable fed into different machine-learning algorithms: Deep neural networks, support (SVM), J48 decision tree naïve Bayes. Our experimentation aimed find best algorithm build high-quality classifier. involved splitting dataset, 2/3 used train 1/3 evaluation besides cross-validation methods. deep network (DNN) obtained highest accuracy (95.2%). In addition, it F1-scores (73.6%) (97.4%) classes, respectively.