Ensemble Sentiment Model: Bagging with Linear Discriminant Analysis (BLDA)
Dimple Tiwari, Bharti Nagpal · International Conference on Computing for Sustainable Global Development · 2021
Rapid development in information technology and social media platforms increased the demand for sentiment analysis and modelling. Online reviews contain information about customer choice towards a particular product or a service. Existing approaches of sentiment analysis worked on a single learner or predefined ensemble learner to classify the sentiments. This work proposed a bagging ensemble approach with Linear Discriminant Analysis (BLDA) model for analysing the restaurant reviews. The Linear Discriminant Analysis (LDA) was chosen as a base classifier for each random subset in the bagging meta-estimator model. Our solution mainly focuses on enhancement, standardisation, preprocessing, and performance of sentiment analysis with topic-modelling and ensemble learning. Topic-Modelling is used to find the context of a customer regarding a service. The BLDA model's effectiveness has tested on restaurant reviews with various measurement units as Recall, Precision, F1-Score and ROC-AUC curve. The comparative study with Gaussian Naive Bayes (GaussianNB) and K-Nearest Neighbor (KNN) present the proposed model's best performance.