Improved ensemble based deep learning approach for sarcastic opinion classification

• Published in: Multimedia tools and applications Vol. 83; no. 13; pp. 38267 - 38289
• Main Authors: Maheswari, S. Uma, Dhenakaran, S. S.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2024; Springer Nature B.V
• Subjects: Algorithms; Classification; Clustering; Computer Communication Networks; Computer Science; Customers; Data analysis; Data Structures and Information Theory; Datasets; Deep learning; Electronic commerce; Machine learning; Multimedia Information Systems; Natural language processing; Outliers (statistics); Python; Special Purpose and Application-Based Systems; Bi-LSTM; Agglomerative; Hybrid model; Sarcastic sentiment; Natural Language Processing; TF-IDF
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-16891-9

Sarcastic tweets or reviews confuse the public those who interested to buy the products from online and prior know about the customer’s opinion about the products. E-Commerce business people also come across this problem to understand the customer’s sarcastic opinion. To remove this problem and to help the online customers also to help the business people to know the customer’s opinion is sarcastic or non sarcastic this research work carried out identify the sarcastic opinion. In many existing methods for detecting sarcastic statements commonly followed the part of speech (POS) -Tag, N-Gram methods for extract the features, as well as represented the features by Term Frequency-Inverse Document Frequency (TF-IDF) and binary methods, and for feature selection utilized chi squared, information gain methods. Other than the above mentioned methods authors didn’t use the features such as Likes count, ReTweet count, Replies count from Tweets. Outlier from the features can be identified and detected; typically the outlier removed features will provide the best performance. Not but least, many researchers followed imbalanced dataset for sarcastic detection. These limitations are identified by this proposed work. The novelty of this proposed work is followed the hybrid techniques such as agglomerative clustering method for outlier detection, and outlier removed dataset fed as input to stacked auto encoder (SAE), finally classification and prediction completed by Ensemble of Logistic Regression, Random forest and Bi-Directional Long Short Term Memory(Bi-LSTM). Existing work doesn’t follow the outlier removal method with Deep Learning (DL) method for sarcastic opinion detection. The proposed work is completed with aid of Natural Language Processing (NLP), Python Libraries and Tweets about E-Commerce Amazon products. Proposed model obtained 99.3% accuracy in sarcasm prediction, this outperforming existing sarcasm detection algorithm. This result suggests that the proposed model was successful in detecting and analyzing sarcastic sentiment.