Abstract:
In recent years, social media and other forms of digital communication have become a trend, encouraging much research on sentiment analysis. Integrating multiple machine ...Show MoreMetadata
Abstract:
In recent years, social media and other forms of digital communication have become a trend, encouraging much research on sentiment analysis. Integrating multiple machine learning and deep learning methods has become a viable solution to improve the reliability of sentiment analysis, especially in outlier opinion mining and cross-domain. The purpose of this study is to develop a model that combines various analytical techniques for mining outlier opinions effectively across different application domains. A comprehensive methodology was explicitly employed, the data was collected from different public sources, and the data preprocessing procedures included text cleaning, tokenization, and embedding using BERT and Word2Vec. The hybrid model architecture was designed with three branches, such that each branch involves different combinations of the following algorithms: SVM, CNN, GBM, and GRU. The performance of the proposed model was assessed based on widely used measures (accuracy, precision, recall, F1 score, AUC-ROC) over multiple datasets and cross-domain settings. The hybrid model performed better than single models in accuracy, precision, recall, and F1-score. The model eliminated out-of-scope opinions, and performance remained stable across different domains. Advanced embeddings and the proposed hybrid architecture helped improve feature representation and image classification. The proposed hybrid sentiment analysis model improves performance over various datasets and domains.
Published in: 2024 30th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)
Date of Conference: 03-05 October 2024
Date Added to IEEE Xplore: 12 November 2024
ISBN Information: