FuDL-SM: A Hybrid Fuzzy Deep Learning Framework for Sentiment Analysis of Zoological and Wild Life Conservation based Social Media Data

Background: As microblogging becomes more popular, users produce vast amounts of text expressing different feelings and opinions. Twitter, in particular, attracts millions of users who share brief and informal messages. Sentiment analysis of such posts continues to be a difficult task because of informal language, sarcasm, abbreviations, and ambiguity. Objectives: This study desire to create a new system called FuDL-SM that combines fuzzy logic and deep learning. The goal is to better understand user opinions on Twitter by improving the accuracy of sentiment analysis, especially when dealing with unclear or uncertain expressions in the reviews. Methods: The FuDL_SM model uses deep learning model along with fuzzy membership functions and GloVe word embedding to capture both the meaning of text and the uncertainty in sentiment. Findings: The model was tested on 51,757 wild life reviews which were cleaned and labeled using the VADER sentiment tool. Proposed method achieved 94.28% accuracy, 93.85% precision, 94.12% recall, and 94.03% F1-score, which are better than existing systems. The use of fuzzy logic helped improve accuracy by 10% and made the system better at handling uncertain or mixed sentiments. Novelty: Unlike other methods that use fuzzy logic inside the network layers, this model applies fuzzy inference after the deep learning classification step. This final step helps refine the sentiment predictions and makes the model more dependable. Keywords: FuDL SM, Glove Embedding, Deep Neural Networks, Fuzzy inference system, Wild life conservation, Zoological data