Evaluating classification schemes for second screen interactions

We analyze the performance of classification schemes on information collected from social conversation posted in Twitter among audiences of a popular US based TV show. In this research, we consider entropy as a measure of information exchange in a group conversation that is related to social, temporal, and second screen device features. The group conversations are identified by hashtags present in tweets where the number of members in the group interacting is at least two. We apply different classification schemes to more than 4,900 groups identified from 318,000 tweets. The result shows that 5-nn algorithm outperformed the other procedures in terms of misclassification error to identify the informed groups.