Adaptive Word Embedding to Reduce the Dimensionality of the Document to Vector Representation

Abstract

Sentiment Analysis is a methodology of detecting the emotions from the text. It is an application of Natural Language Processing (NLP) methodology. The NLP enables us to know the common day to day language of the people. This will helps to decipher the sentiments of the users and hence explain liking and disliking of the people. The traditional bag-of-words models lack the accuracy of sentiment classifications. The intention of this project is to improve the accuracy of the sentiment classification by employing the concept of dimensionality reduction. Reducing the dimensionality of a large document helps to reduce the computational cost and increase efficiency. Word embedding methods capture the context of a word in a document which helps to reduce the dimensionality of text data. Vector representation of the words using a technique like Word2Vector proves to be very effective in interpreting the meaning and hence the sentiments. The words in the document will be converted into vectors. Each word is assigned a unique value (vectors) such that these vectors represent its context, meaning, and semantics. The resulting word vectors are wont to train machine learning algorithms within the sort of classifiers for sentiment classification. We use the Machine Learning classifier Naive Bayes to analyze the sentiment from the given pre-processed dataset (word vectors). Our experiments on real-world datasets show the improvement in the accuracy of sentiment classification using the word embedding techniques.

Keywords: Dimensionality reduction, Sentiment analysis, Vector representation, Word embedding.

View PDF