Past work on affect analysis of tweets has focused solely on detecting emotions, and ignored questions such as 'who is feeling the emotion (the experiencer)?', 'towards whom is the emotion directed (the stimulus)?' ,and 'what is the purpose behind the tweet?'. We automatically compile a large dataset of tweets pertaining to the 2012 US presidential elections, and annotate them for style, purpose, and various semantic roles of emotion. The annotations give interesting insights such as people tend to convey opposition twice as often as support in electoral tweets. We use the labeled data to develop a classifier for detecting emotions. It obtains an F-score of 55.86 on a 8-way classification task. We show how the stimulus identification task can also be framed as a classification task, circumventing more complicated problems of detecting entity mentions and coreferences. Our supervised classifier outperforms competitive baselines and obtains an F-score of 58.30. We also develop a system that automatically classifies electoral tweets as per their purpose, obtaining an accuracy of 43.56% on an 11-class task and an accuracy of 73.91% on a 3-class task (both accuracies well above competitive baselines).