Incremental Learning based Optimized Sentiment Classification using Hybrid Two-Stage LSTM-SVM Classifier

Abstract

Sentiment analysis is a subtopic of Natural Language Processing (NLP) techniques that involves extracting emotions from unprocessed text. This is commonly used on customer review posts to automatically determine if user / customer sentiments are negative or positive. But quality of these analysis is completely dependent on its quantity of raw data. The conventional classifier-based sentiment prediction is not capable to handle these large datasets. Hence, for an efficient and effective sentiment prediction, deep learning approach is used. The proposed system consists of three main phases, such as 1) Data collection and pre-processing, 2) Count vectorizer and dimensionality reduction is used for feature extraction, 3) Hybrid classifier LSTM-SVM using incremental learning. Initially the input raw data is gathered from the e-commerce sites for product reviews and collected raw is given to pre-processing, which do tokenization, stop word removal, lemmatization for each review text. After pre-processing, features like keywords, length, and word count are extracted and given to feature extraction stage. Then a hybrid classifier using two-stage LSTM and SVM is developed for training the sentimental classes by passing new features and classes for incremental learning. The proposed system is developed using python and it is compared with the state-of-the-art classification techniques. The performance of the proposed system is compared based on performance metrics such as accuracy, precision, recall, sensitivity, specificity etc. The proposed model performed an accuracy of 92% which is better compared to the state-of-the-art existing techniques