Enhancing sentiment analysis classification for amazon product reviews using CNN- sigTan-Beta activation function

The rapid development of online products paves the way to share customers' opinions on amazon products. Unstructured text reviews and customer feedback are popular resources for customers when making decisions. However, reading through all the evaluations is tiresome, but the volume of customer feedback is enormous. The ability to forecast the precise sentiment polarities of user textual feedback evaluations for a particular entity is still difficult because of phrase length restrictions, textual order variations, and logical complexities. Therefore, an aspect level of analysis is needed, which support the retailers in understanding customer expectation and then modifying the product accordingly. However, many existing machine learning algorithms are available for sentiment detection but fail in accuracy rate. This paper proposes a novel sigTan-Beta Activation Function for Convolution Neural Networks (CNN) to attain remarkable and effective results. First, the sample dataset is pre-processed, and text strings are converted into the vector using Word2Vec, which computes the distance between words and groups them based on similarity. Afterwards, CNN extracts the sensitive features from the data and classifies the product reviews. The proposed model uses the sigTan-Beta Activation Function, which tunes the weight of the neurons to gain accurate performance. The proposed classified as positive or negative classes using the amazon review dataset. The proposed sigTan-Beta Activation Function for Convolution Neural Network (CNN) experiment performs better than existing methods in terms of accuracy, precision and F1-score. Our proposed sigTan-Beta Activation Function for Convolution Neural Network (CNN) achieves 94.5% accuracy to the existing ABO-RF algorithm (89.9%).