Deep Non-linear and Unbiased Deep Decisive Pooling Learning–Based Opinion Mining of Customer Review

• Published in: Cognitive computation Vol. 15; no. 2; pp. 765 - 777
• Main Authors: Kuppusamy, Saraswathi, Thangavel, Renukadevi
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Classification; Computation by Abstract Devices; Computational Biology/Bioinformatics; Computational efficiency; Computer Science; Computing time; Customers; Data mining; Datasets; Deep learning; Efficiency; Eigenvectors; Learning; Ontology; Recommender systems; Semantic web; Semantics; Sentiment analysis; Students; Non-linear ReLU-enabled activation function layer; Opinion mining; Unbiased decisive pooling layer; Customer reviews; Skip-gram input layer
• ISSN: 1866-9956; 1866-9964
• DOI: 10.1007/s12559-022-10089-1

Nowadays, opinion mining is becoming a trending research topic. The existing opinion mining detection schemes were used to determine the status of product’s performance, wherein information is disseminated (positive or negative), even though the existing methods did not provide high accuracy and also increase the computational complexity and computation time. To overcome these issues, a deep non-linear and unbiased deep decisive pooling (N-UDDP) learning is proposed in this paper for getting accurate opinion mining of customer review. Initially, the customer reviews are taken from Amazon dataset. Then, the customer reviews are given as input to N-UDDP learning for getting accurate opinion mining of customer review. The proposed N-UDDP learning contains skip-gram input layer, non-linear ReLU-enabled activation function layer, and unbiased decisive pooling layer. The skip-gram input layer is utilized to obtain the computationally efficient features from input customer review. The non-linear ReLU-enabled activation function layer is used to obtain the unique and relevant features through eigenvector generation. Finally, the mining accuracy can be improved via unbiased decisive pooling layer by explicitly incorporating their confidence over the pooling function through decisive rules. Initially, the customer reviews are taken from Amazon dataset. Then, the customer reviews are given as input to N-UDDP learning for getting accurate opinion mining of customer review. The proposed N-UDDP learning contains skip-gram input layer, non-linear ReLU-enabled activation function layer, and unbiased decisive pooling layer. The skip-gram input layer is utilized to obtain the computationally efficient features from input customer review. The non-linear ReLU-enabled activation function layer is used to obtain the unique and relevant features through eigenvector generation. Finally, the mining accuracy can be improved via unbiased decisive pooling layer by explicitly incorporating their confidence over the pooling function through decisive rules. The proposed N-UDDP method attains 10.99%, 44.83%, 17.53%, 11.93%, 11.19%, 53.96%, and 13.285% higher accuracy than the existing methods, like ontology-driven feature engineering, multi-task learning framework, CNN-RNN, NA-DLSTM, MLP, autoencoder, and MV-DNN. The proposed N-UDDP method shows better accuracy by reducing the computational time and overhead.