Sentimental Analysis of Industry 4.0 Perspectives Using a Graph-Based Bi-LSTM CNN Model

Today's modern society mainly depends on Internet for every fundamental task in their life, like sharing thoughts, education, business, and Industry 4.0, etc. Internet has strengthened the base of society digitally. Searching for the reviews and comments for a particular product from the former or present customers has become compulsory for making decision or a purchase, helping them to make a fair deal of the product by view from social media on Industry 4.0. The increased use of data over Internet has led to rise of many e-commerce websites where people can buy things as per their requirement without even stepping out of their house using analysis of text using natural language techniques in Industry 4.0. The graph-based modelling of sentiment analysis classifier needs to divide the textual data into training dataset and testing dataset. First, preprocessing work is performed to improve the data reliability by removing unwanted information and fixing typing errors. To train the models, the whole dataset is used and to measure the classification performance of the categorized models 10-fold validation technique is being used. The paper proposes a combined hybrid model for sentiment analysis using CNN and independent bidirectional LSTM networks to enhance sentiment knowledge in order to address the issues mentioned for sentiment analysis. The proposed CNN model uses global max-pooling for retrieving context information and to downsample the dimensionality. Lastly, to acquire long term dependencies, a distinctive bidirectional LSTM is used. To emphasise each word's learning ability, parts-of-speech (PoS) are tagged in the LSTM layer. In addition, the regularization techniques, batch normalization, and dropout are used to prevent the overfitting issue. The proposed model is compared with a collaborative classifier with six classifiers and each of them predicts the sentiments separately, and the majority class prediction is taken under consideration. The proposed Bi-LSTM CNN model achieves an accuracy of 98.61% along with PoS tagging of the sentiments.