Sentiment analysis in multi-scenarios: Using evolution strategies for optimization

With the developing of blog, mircoblog and social networking sites, researchers and practitioners have paid more attentions to how to accurately get useful positive/negative evaluation information from those web opinion articles which have different writing styles. This is so-called opinion mining and sentiment analysis. However, very few sentiment analysis studies focused on multi-scenarios problem. This study collected some Chinese sentences from one movie blog at Taiwan, and conducted an experiment to infer those authors´ sentiment. We chose two different scenarios, general and horror. After collecting the inferences of readers, we applied evolutionary computing strategies to optimize the tables of basic emotion weights in two different scenarios. The findings indicate that our approach would have better correct rates than past research which considered only one general scenario. Through these findings, we can understand what are people concerned in different scenarios.