Sentiment Analysis with Term Weighting and Word Vectors

Metin Bilgin¹ and Haldun Köktaş²

¹Department of Computer Engineering, Bursa Uludağ University, Turkey

²Department of Mechatronic Engineering, Bursa Technical University, Turkey

Abstract: It is the sentiment analysis with which it is tried to predict the sentiment being told in the texts in an area where Natural Language Processing (NLP) studies are being frequently used in recent years. In this study sentiment extraction has been made from Turkish texts and performances of methods that are used in text representation have been compared. In the study being conducted, besides Bag of Words (BoW) method which is traditionally used for the representation of texts, Word2Vec, which is word vector algorithm being developed in recent years and Doc2Vec, being document vector algorithm, have been used. For the study 5 different Machine Learning (ML) algorithms have been used to classify the texts being represented in 5 different ways on 3000 pieces of labeled tweets belonging to a telecom company. As a conclusion it was seen that Word2Vec, being among text representation methods and Random Forest, being among ML algorithms were most successful and most applicable ones. It is important as it is the first study with which BoW and word vectors have been compared for sentiment analysis in Turkish texts.

Keywords: Word2vec, Doc2vec, sentiment analysis, machine learning, natural language processing.