Hamiltonian deep neural network fostered sentiment analysis approach on product reviews
Narahari Ajmeera,P. Kamakshi
DOI: https://doi.org/10.1007/s11760-024-03014-6
2024-02-29
Abstract:In recent times, online shopping has become commonly used method for consumers to make purchases and engage in consumption with the rapid advancement of Internet technology. Enhancing user satisfaction is achievable through sentiment analysis (SA) of the enormous user evaluations found on e-commerce platforms. However, accurately predicting the sentiment orientations of these user reviews remains a challenge due to varying sequence lengths, text arrangements, and intricate logic. Nowadays, sentiment analysis is widely employed to assess customer feedback, which holds great significance in determining a product's success. In the past, people relied on word-of-mouth reviews to judge a product's quality. This practice of sentiment analysis is extensively applied in social media. Natural language processing (NLP) plays a crucial role in deciphering sentiment, also referred to as opinion mining or emotion AI, as it encompasses the collective perception of customers. In this manuscript, a Hamiltonian Deep Neural Networks-based Sentiment Analysis on Product Recommendation System (HDNN-SCOA-SA-PR) is proposed. First, the data are gathered from Amazon Product Reviews dataset. Then the data are pre-processed utilizing adaptive self-guided filtering for space tokenization, Gensim lemmatization, and Snowball stemming. By using Structured Optimal Graph-Based Sparse Feature Extraction, the features are extracted. Extracted features are selected using Single Candidate Optimization Algorithm. Finally, the classification process is done using Hamiltonian deep neural network and classified sentiment analysis as positive, negative, neutral. The proposed HDNN-SCOA-SA-PR method is activated in Python, and the efficiency of the proposed method is analyzed with different metrics, such as accuracy, sensitivity, RoC, precision, error rate, F1-score,computation time. ROC is evaluated and compared to the existing methods, such as sentiment analysis based upon machine learning of online product reviews with term weighting including feature selection (SAPR-FS-ENN), sentiment analysis of product reviews depend upon weighted word embeddings along deep neural networks (SAPR-WWE-DNN), improving sentiment analysis for social media applications utilizing an ensemble deep learning language (ISA-SMA-ECN-PR), respectively.
engineering, electrical & electronic,imaging science & photographic technology