Behavior Analysis of Twitter Feed using Dymamic Synonym and Abbreviation Mapping Modules on top of a Trained Naïve Bayes Classifier
B. N. Shankar Gowda, Siddharth Mark Joseph, Vibha Lakshmikantha · 2016
The current generation is largely associated with the online social networking services that enable users to exchange information, emotions, and thoughts in their own way. The existing techniques highlight the static nature of supervised learning with respect to behavioral analysis of twitter posts. In this work, in order to analyze the behavior or sentiments of online users, a novel concept of a dynamic synonym mapping and abbreviation mapping is proposed to achieve higher accuracy. A unigram feature extractor model is used for training a Naïve Bayes Classifier to classify the user's tweets dynamically as positive and negative tweets. The paper highlights the importance of Synonym mapping and Abbreviation mapping mathematically, in the context of a Naïve Bayes Classifier and through custom example tweets which contain words that are not present in the feature list. These unknown words are then mapped to words present in the feature list or vocabulary. The process of mapping these words and abbreviations to words in the feature list has been clearly illustrated. Effort has been made to enhance the quality of output by preprocessing the raw tweets to obtain a quality dataset for analysis. Our proposed methodology proves both conceptually and mathematically the increase in accuracy of prediction of the classifier.