Sentiment Analysis for Urdu News Tweets Using Decision Tree

In the last few years, with rapid growth in use of networking sites such as twitter and Facebook has been increased greatly. This also attracted the researcher to use social networks data for sentiment analysis. Sentiment analysis is also known as opinion mining is the process of finding out the emotion such as positive, negative and neutral from the series of words. In present, on internet huge amount of data has been generated and to extract useful information from data is also become interest for the researchers. Sentiment analysis has been done mostly in English and Chinese languages. In this paper, sentiment classification is done on Urdu news tweets. The proposed methodology consists upon two steps. In first step data preprocessing is done such as removal of hash tag and removal of stop words is done. In second step feature vector is designed. The feature vector is formulated by through the identification of number of positive words, negative words, and presence of negation and use of POS tags. After formulation of feature vector the decision tree is used as classification algorithm. The decision tree classifies the tweet as positive, negative and neutral. The experimental result of the proposed methodology shows significant success in terms of accuracy and sentiment analysis.

I. Introduction

Fast increase in internet users along with growing power of online review sites and social media has given beginning to Sentiment analysis or Opinion mining. Sentiment analysis is a field of text mining to examine the views of people about specific topic, product, and comments on post of social media networks and reviews site which also raised challenges in opinion mining [1] . Sentiment analysis applications are powerful and broad. One of the challenges in sentiment analysis is to tackle different language such as English, Urdu and Arabic according to their orientation [2] . Opinion extraction is different in different language like English, Chinese, and Arabic etc. One of challenging task in sentiment analysis is to handle each language according to its orientation. Sentiment analysis has been done mostly in English and Chinese languages now present; research has been carried on sentiment analysis of Urdu language. The national language of Pakistan is Urdu. More than 200 million people spoke Urdu. Urdu language is type of combination of many other languages that is Hindi, Turkish, Persian and Arabic. The combination of this language contributed to the complex morphological structure to Urdu [4] . Sentiment analysis goal is to determine the orientations of subject into positive and negative. Most of sentiment analysis approaches are used to handle English text. But these approaches are not used to handle dialects of south Asia language like Urdu, Persian and other languages. All of these languages have different morphology, script and grammatical rules [5] . Major concern related to sentiment analysis in Urdu language is that very limited NLP has done on Urdu. Due to very limited research there is no such Lexicon is available for Urdu language. Urdu is a bi-directional language with an Arabic-based orthography. Bi-directional means that it is very common in Urdu to see an English word written in Latin-based characters.