Multi-Class Text Classification: Model Comparison and Selection

Abstract

The objective of text classification is to categorize documents into a specific number of predefined categories. We can easily imagine the issue of arranging documents, not by topic, but rather by and large assessment, e.g. deciding if the sentiment of a document is whether positive or negative. While working on a supervised machine learning problem with a defined dataset, there are many classifiers that can be used in text classification. Utilizing dataset of stack overflow questions, answers, and tags as information, we find that standard machine learning systems completely beat human-delivered baselines. These majorly include Naive Bayes Classifier for multinomial models, Linear Support Vector Machine, Logistic Regression, Word to vector (Word2vec) and Logistic Regression, Document to vector (Doc2vc) and logistic regression, Bag of Words (BOW) with Keras. Our paper is a detailed examination and comparison of accuracies among these algorithms.