Structural Correspondence Learning for Cross-lingual Sentiment Classification with One-to-many Mappings