Emoticon based Sentiment Analysis using Parallel Analytics on Hadoop

Objectives: The major objective of this approach is to provide a sentiment analysis architecture that can operate on streaming big data to provide effective results at tolerable time limits. Method/Analysis: An effective mechanism for analyzing the social networking messages to identify the sentiment levels has been proposed. This is a generic model that can be used for product or organization specific analysis. Further, this method also considers emoticons, which form the integral part of any expressed emotion. The entire process is carried out in Hadoop Architecture using the MapReduce paradigm. Findings: Experiments have been conducted on a Hadoop cluster. Inputs were passed from a client node connected to the cluster. Map Reduce programs were executed in six phases, each phase performing a single task in map and reduce phases. The ROC plot exhibits excellect accuracies with most of the points being clustered in the top left region, some even approaching 100% effectiveness. Even the PR plots exihibits similar efficiency scenario with high positive retrieval rates. Incorporating the emoticons plays a major role in increasing the efficiency of this approach. Novelty/Improvement: This approach uses Hadoop based implementations, involving Map and Reduce operations. Using this approach provides data scalability and improves the efficiency of the results in acceptable time limits.