Sentiment analysis on Twitter using McDiarmid tree algorithm

In recent years advent of social networking services has created large amounts of data. Microblogging website is a kind of social network in which users share short messages with others. One of the most popular microblogging services is Twitter. Every day millions of people post their opinions and sentiments in this microblog. Due to the large numbers of tweets, finding new approaches to discover and summarize the general overview of a specific topic has become a new challenge. Twitter messages are generated constantly and arrive at high speed and follow data stream model; hence, to predict the sentiment on Twitter we must apply algorithms which can do this in real time and under limited time. Hoeffding tree algorithm is the most popular tool in mining data streams. For this tree algorithm the Hoeffding's bound is utilized to find the smallest amount of instances required in a node to choose a splitting attribute. Replacing the MacDiarmid's bound in Hoeffding tree algorithm, we obtain McDiarmid tree algorithm which is employed in this paper. The accuracy from the McDiarmid tree for sentiment analysis on Twitter is very close to that from the Hoeffding tree; however, the process time of the former has considerably decreased.