Combining Linguistic and Semantic Information in an Emotion-based Model for Polarity and Intensity Classification