An Efficient CSPK-FCM Explainable Artificial Intelligence Model on COVID-19 Data to Predict the Emotion Using Topic Modeling

Abstract

Incessant COVID-19 pandemic negatively impacts nations throughout the globe. It is necessary to determine how people react to public health interventions and understand their concerns. Twitter is a social media platform that has emerged as a tool for disseminating information, debating concepts, and reviewing or commenting on global issues. This study applies Explainable Artificial Intelligence (XAI) methods, like Cosine Similarity and Polynomial Kernel-centered Fuzzy C-Means (CSPK-FCM) centered topic modeling and Fuzzy Logic with Improved Long Short-Term Memory (FL-ILSTM) centered Sentiment Analysis to COVID-19 data on Twitter. The proposed model has five major steps: preprocessing, feature extraction, term weighting, topic modeling (clustering), and classification. Twitter comments relating to the COVID-19 pandemic are initially collected from publicly accessible websites. The collected data are then preprocessed to remove irrelevant information, namely, noises. The Feature Extraction phase is then performed by extracting emoticon and non-emoticon features. The extracted feature dataset is scored: the Term Frequency Inverse Document Frequency-Chi-Square (TFIDF-CHI) method is utilized for non-emoticon, and the score for the emoticon is assigned based on a few criteria. For Topic modeling, the TFIDF-CHI scores are provided to the CSPK-FCM clustering algorithm, which groups the most frequently discussed topics throughout COVID-19. FL-ILSTM executes the Sentiment analysis of clustered topics and emoticon features. It has extraordinary performance when compared to other methodologies.