A Predictive Model for Tweet Sentiment Analysis and Classification

摘要

Sentiment analysis over Twitter offers organisations and users a fast and effective way to monitor publics’ feelings towards events especially during crises, hence, motivated much work on twitter data. In this study, predictions on positive, negative and neutral sentiment based on security are analysed. A polarity classification of tweet messages was done with VADER algorithm considering contextual analysis. The analysis was performed by removing stop words in the tweet along with Wordnet lemmatiser for the morphological analysis of words in the features sets. As well as subjected to word sense disambiguation to consider contextual usages of words using a path length corpus based lexicon. Term Frequency (TF) and Term Frequency Inverse Document Frequency (TFIDF) are used as feature extraction from tweets and evaluation of the features reduction was carried out by calculating the accuracy of the predictions on sentiment and tweet messages with Chi-Square to explore the possibly useful features. Finally, validations are done with machine learning models at different sequence to compare the performance between each model.