Twitter Sentiment Analysis in Urdu Language

Abstract

It’s a natural human ability to understand emotions and analyze situations. But the effectiveness with which machines can be trained to demonstrate the same human phenomenon becomes an important question to be explored. Sentiment Analysis (SA) is also considered as sentiment classification in which text which is a document, posts, reviews or tweet is categorized either negative or positive. To know the opinions of people, twitter is considered a rich source platform.SA is being studied for a long time, but in the Urdu language, it is a new research area. This thesis uses a deep learning approach for classifying sentiments in the Urdu Language. Dataset available in this language is not sufficient. In this research, Urdu Tweets are extracted. And manually labeled as negative or positive. Convolution Neural Network CNN is used for classification. As CNN does not directly work on textual data, we use word2vec for converting textual data into vectors. We train word2vec on Urdu language data. Accuracy of 74 % is achieved by using word2vec embedding feature with CNN.