Sentimental Analysis of Twitter Data on Hadoop

Data is something without which organizations can never reach any conclusion and cannot extract any particular pattern. These data sets are the sources on which organizations rely while taking important strategic decisions. There are many social platforms on which people around the world are accessing and these platforms are generating a huge amount of data. This data can be differentiated on the basis of their volume, velocity and variety. Organizations term such a huge amount of data as Big Data. These social data sets are of great use for improving business strategies. Nowadays, twitter has become a great social platform for expressing different opinions. This paper focuses on MapReduce-based sentiment analysis of data received through twitter. The data is first cleaned to retain only text, then MapReduce is applied to get the frequency of each word which is then matched with the dictionary created for positive and negative words over Hadoop environment. The results are compared with Naïve Bayes and SVM classifier. It has been observed that time consumed by the proposed system is 45% less than SVM and 38% less than Naïve Bayes. The accuracy in terms of a total number of words detected, positive and negative words, was also observed to be 11%, 16%, 18% respectively in case of SVM and 9%, 13%, 16% respectively in case of Naïve Bayes.