A Fusion Attention Mechanism with Bi-LSTM-Based Sarcasm Detection for Selecting High-Profit Products

Social media has become an invaluable resource for gathering customer feedback in today’s digital era. As technology and the internet evolve, vast quantities of data are generated from numerous sources such as social blogs and websites. Recently, webpages and blogs have emerged as primary channels for obtaining real-time consumer feedback. However, the sheer volume of blogs on the cloud has led to overwhelming information in various forms, including attitudes, opinions, and reviews. For product reviews, it is possible to discern the reviewer’s perspective on specific product characteristics rather than the product as a whole. Nonetheless, sarcastic reviews pose a challenge in accurately determining the sentiment class of the review. This paper proposes a sarcastic review prediction-based sentiment analysis approach to address this issue. The proposed method comprises three stages: pre-processing, feature extraction, and prediction. Initially, reviews are collected and pre-processed using tokenization, stop word removal, and stemming. Following pre-processing, feature extraction is performed, wherein sentimental, punctuation, and TF-IDF features are extracted from each review. These extracted features are then fed into a Bidirectional Long Short-Term Memory (Bi-LSTM) network with an attention mechanism to predict whether a product review is sarcastic or non-sarcastic. The fusion attention mechanism assigns attentive weights to each word, enhancing the prediction accuracy. The performance of the proposed approach is evaluated using various metrics and compared against state-of-the-art techniques. The results demonstrate the effectiveness of the proposed method in accurately predicting sarcastic reviews.