Performance Analysis Of Machine Learning And Deep Learning Models For Text Classification

Information retrieval and data fields that concentrate on automated lengthy text categorization are becoming more popular due to the quick development of modern science and technology. However, machine learning-based text categorization learning must be thoroughly examined and assessed from an algorithmic, model, assessment, and other perspective. This article classifies texts using the deep learning approach. First, the traditional text classification procedure is explained, which includes pre-processing data, segmenting the text, creating features, summarizing the classification algorithm, and so on. Second, text categorization is done using CNN with multiple convolution kernels. The word vector model takes the role of the conventional bag of words model, and text characteristics are extracted using a multi-dimensional convolution kernel. Deep learning models—primarily convolutional neural networks and multilayer perceptrons—have been more and more popular recently for use with these kinds of datasets. Since there isn't any such integrated study, the purpose of this research is to evaluate how well deep learning models perform on various cancer gene expression datasets. Three deep learning models, two feature selection techniques, and four cancer gene expression datasets have been taken into consideration for this aim. The experiment in this study showed that CNN models performed promisingly in text document classification, but it also raised the following questions that need more, in-depth investigation: 1) How to compare the performance of the suggested models more effectively. Which text-based categorization tasks are best suited for each model? 2) How to better comprehend the facts. Specifically, what was learned by the deep learning model? 3) Adjusting hyperparameters. The process of hyper-parameter tuning is one that requires expertise. 4) Label imbalance: how to modify the sample weight in a loss akin to the bootstrap approach when the data samples are labeled unevenly. 5) On a given data set, what is the simplest neural network structure that can get a certain level of accuracy?