Comparative Analysis of NLP Text Embedding Techniques with Neural Network Layered Architecture on Online Movie Reviews

In NLP world, there is a need to convert the text data into numerical form in a smart way of text embedding with the machine learning architecture. In this research, the comparative text embedding methods of Binary Term Frequency, Count Vector, Term Frequency - Inverse Document Frequency, and Word2Vec is used for converting text to meaningful representations of vectors, containing numerical values. In order to analyze the performance of the various text embedding techniques, Neural Network Layered Architecture is designed for movie review’s polarity classification to include input layer, dense layers followed by the ReLU (Rectified Linear Unit) activation layers and Sigmoid activation function to make classifications on the basis of training–testing performance. Word2Vec text embedding scored the highest training and testing accuracy among all the text embedding techniques of Binary Term Frequency, Count Vector, Term Frequency - Inverse Document Frequency, and Word2Vec with 89.75% and 86.94% respectively with \(\pm { }1.0\,\,{\varvec{error}}\) for the online movie reviews.