Learning Adaptable Approach to Classify Sentiment with Incremental Datasets.
Maibam Debina Devi, Navanath Saharia · Procedia Computer Science · 2020
Identification of sentiment type is considered as a new business tool for the market and its policies, which involves different learning algorithms. Capability to automatically increase performance from experience and learning through data is the concept behind learning algorithms, which often requires large amount of data to train a model. The aim of this article is to find an adaptable classifier with increasing size of dataset. The experiment starts with splitting a dataset of size 10000 samples into multiple sub-set to check performance of different classifiers using term frequency-inverse document frequency (TF-IDF) and count vectorizer (CV) feature extraction techniques. The classifiers are train with sub-set of dataset and their adaptability performance was analyzed. Support vector machine (SVM) and logistic regression (LR) obtained stable and incremental performance with increase in dataset size using CV. Multinomial naive bayes (MNB) outperformed well with TF-IDF. We also analysed of classifiers with respect to performance, feature extraction techniques and data-set size.