Shrinking the Sentiment Analysis for Signed Network Construction

To construct a signed network for online social networks would be very useful for Ad Serving, friend recommendation, and Web spammer detection. In the construction of the signed network for certain online social networks (e.g. microblog), sentiment analysis plays a critical role, and the construction of a signed networks turns out to be a big challenge with the rapid growth of the online social networks's content. Considering a non-trivial quantity of links could be predicted according to other links, we investigate the probability to reduce workload of the sentiment analysis for online social networks' content by sign prediction. Based on the analysis of the redundancy and balance degree of Slashdot, Wikipedia, and Epinions datasets, we propose an algorithm to shrink the sentiment analysis for signed network construction. Our evaluation shows that the shrinking algorithm manage to reduce the workload of sentiment analysis (from 21.6% to 48.7% for different datasets) with sacrifices of prediction accuracy from 6-17%.