Document classification using deep neural network with different word embedding techniques

Abstract

Document classification has played a major role in many fields like information retrieval, data mining, etc. where machine learning and deep learning models can be applied. But, before applying any model for classification, textual data must be converted into a numerical measure, where word embedding can help. The selection of appropriate word embedding techniques plays a vital role in classification. So, we analysed the classification performance by widely used deep learning models long short-term memory (LSTM) and convolution neural network (CNN) with various word embedding techniques on five benchmark datasets. The pre-processed dataset is converted into vector representation using a word embedding techniques TF-IDF, Word2Vec, and Doc2Vec. The output is given to the LSTM and CNN classifier and documents are classified as per their context. The CNN classifier with Doc2Vec word embedding technique achieves almost 12% more accuracy as compared to other word embedding techniques on all the datasets.