Social networking is nowadays a major source of new information in the world. Microblogging sites like Twitter have millions of active users (320 million active users on Twitter on the 30th September 2015) who share their opinions in real time, generating huge amounts of data. These data are, in most cases, available to any network user. The opinions of Twitter users have become something that companies and other organisations study to see whether or not their users like the products or services they offer. One way to assess opinions on Twitter is classifying the sentiment of the tweets as positive or negative. However, this process is usually done at a coarse grain level and the tweets are classified as positive or negative. However, tweets can be partially positive and negative at the same time, referring to different entities. As a result, general approaches usually classify these tweets as “neutral”. In this paper, we propose a semantic analysis of tweets, using Natural Language Processing to classify the sentiment with regards to the entities mentioned in each tweet. We offer a combination of Big Data tools (under the Apache Hadoop framework) and sentiment analysis using RDF graphs supporting the study of the tweet’s lexicon. This work has been empirically validated using a sporting event, the 2014 Phillips 66 Big 12 Men’s Basketball Championship. The experimental results show a clear correlation between the predicted sentiments with specific events during the championship.