Structural Correspondence Learning for Cross-Lingual Sentiment Classification with One-to-Many Mappings