Relevant Feature Extraction to Enhance Aspect-Based Sentiment Classification Using Dependency and Bi-Tagged Features
Neelima S. Ambekar, Anant V. Nimkar · 2023
A growing demand for gleaning insights is observed from a large amount of user-generated content available on the internet. The lack of phrases and the semantic ambiguity of single words both have a negative impact on the reliability of sentiment analysis, and this is particularly true when considering the context of brief texts shared on social media. The proposed research work investigates the contextual knowledge usefulness for relevant feature extraction based on dependency and bi-tagged features (DABTF). The prominent features extracted using Information Gain (IG) value improved the classification accuracy. Based on accuracy, the comparison shows Support Vector Machine (SVM) classifier gives superior performance over Naive Bayes (NB) classifier with value 79.67% for the proposed method. The improvement can be observed in the aspect category classification task for food and service categories with F-measures 92.34% and 89.78% respectively.