Contextual sentiment analysis for social media genres

摘要: The lexicon-based approaches to opinion mining involve the extraction of term polarities from sentiment lexicons and aggregation such scores predict overall a piece text. It is typically preferred where labelled data difficult obtain or algorithm robustness across different domains essential. A major challenge for this approach accounting semantic gap between prior terms captured by lexicon terms' in specific context (contextual polarity). This further exacerbated fact that term's contextual polarity also depends on genres which it appears. In paper, we introduce SmartSA, classification system social media integrates strategies capture two perspectives: interaction with their textual neighbourhood (local context) text genre (global context). We an hybridise general purpose lexicon, SentiWordNet, genre-specific vocabulary sentiment. Evaluation results diverse show our account local global contexts significantly improve classification, are complementary combination. Our performed better than state-of-the-art media, SentiStrength.