Sentiment Analysis Model on Product Brand Using Social Media Data
TL;DR
The proposed CNN-LSTM model achieved a test accuracy of 95%, outperforming baseline traditional models such as Naïve Bayes, Support Vector Machine, and Decision Tree that utilized single-modality textual data.
Abstract
User-generated content on social media for sentiment analysis is a critical tool for businesses seeking to understand and respond to customer emotions at social media on product brands. Currently, customers’ emotions on social media are expressed using text and images, while existing systems analyze only textual data to understand customer emotions. The study addresses limitations in traditional sentiment analysis methods by incorporating both textual and image features, thereby improving accuracy and reliability in predicting consumer sentiments. The model architecture combined Convolutional Neural Networks (CNN) for image feature extraction and Long Short-Term Memory (LSTM) networks for text sequence modeling. The implementation was carried out using Python programming language with deep learning libraries such as TensorFlow and Keras, and training was conducted using binary cross-entropy loss with the Adam optimizer. Model evaluation was performed using accuracy, precision, recall, F1-score, and confusion matrix metrics. Experimental results on the same dataset indicated that the proposed CNN-LSTM model achieved a test accuracy of 95%, outperforming baseline traditional models such as Naïve Bayes (73%), Support Vector Machine (75%), and Decision Tree (55%) that utilized single-modality textual data.