Effective Negation Handling Approach for Sentiment Classification using synsets in the WordNet lexical database

Good quality of data preprocessing will have a greater impact on improving the performance of the classification models. Negation is one of the most important features in textual data. Only one negation word can change the polarity of the whole sentence. It is often seen that words like 'not', 'non' or suffixes like 'n't' are removed during noise removal thereby leading to blunders in Sentiment Classification. Effective Feature extraction is the cornerstone of effective Sentiment Analysis and Negation handling is simply essential for this purpose. In this paper, an effective function for handling negations based on First Sentiment Word (FSW) antonymy in the WordNet has been implemented on a set of IMDB movie reviews. The function for Negation Handling created for this paper increased the accuracy of Sentiment Classification by 4-8%. Experiments done in this paper show that improving the quality of the data gives higher results than implementing different state-of-the-art methods like n-grams and even deep learning methods like Word Embeddings, especially when used in an industry setting, where there is a need of quick deployments and changes with cost effectiveness and resource management.