A CNN-LSTM Hybrid Approach for Sentiment Analysis in Online Product Ranking with Probabilistic Linguistic Term Sets

Abstract

In the context of the dynamic digital economy, accurate analysis for online product reviews is required to make it convenient process both for consumers and businesses. This paper proposes a new hybrid method which combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for sentiment analysis based on online product ratings. Current sentiment analysis methods perform inadequately due to their inability to sense complex linguistic pattern and further made worse by noise & ambiguity in textual data. We tackle these challenges by proposing the integration of Probabilistic Linguistic Term Sets (PLTS) with CNN-LSTM framework, which grants a rich understanding of sentiment due to inherent linguistic uncertainty.
This kind of model starts with the CNN layer since it is best for capturing local patterns, and then followed by LSTM which is great as modeling long-term dependencies within a text. For sentiment classification, since PLTS can model the linguistic termsbased on probability distribution over ontology concepts, it enhances reliabil-ity and robustness of the implementation for a better accuracy in analysis. We conduct extensive experiments on benchmark datasets to compare our hybrid CNN-LSTM model with simple ones and obtain state-of-the-art performance in terms of precision, recall, F1-score as shown next. The results demonstrate the promise of this method to improve sentiment analysis accuracy in online product reviews, an important aspect for decision making in e-commerce.