Twitter-sentiment analysis of Moroccan diabetic using Fuzzy C-means SMOTE and deep neural network

Abstract

Effectively managing diabetes as a lifestyle condition involves fostering awareness, and social media is a powerful tool for this purpose. Analyzing the content of tweets on platforms like Twitter can greatly inform health communication strategies aimed at raising awareness about diabetes within the Moroccan community. Unfortunately, the corpus of tweets is imbalanced and the feature extraction leads to data sets with a very high dimension which affects the quality of sentiment analysis. This study focused on analyzing the content, sentiment, and reach of tweets specifically related to diabetes in Morocco. The proposed strategy processes in five steps: (a) data collection from Twitter platforms and manual labilization, (b) feature extraction using TF-IDF technique, (c) dimension reduction using deep neural network, (d) data balancing using Fuzzy C-Means SMOTE, and (e) tweets classification using five well-known classifiers. The proposed approach was compared with the classic system, which works directly on very large, unbalanced tweets. In terms of recall, precision, F1-score, and CPU time, the proposed system can perform highly accurate sentiment analysis in a reasonable CPU time.