Adapting sentiment lexicons to domain-specific social media texts

摘要: Social media has become the largest data source of public opinion. The application sentiment analysis to social texts great potential, but faces challenges because domain heterogeneity. Sentiment orientation words varies by content domain, learning context-specific in domains continues be a major challenge. language poses another challenge since used today differs significantly from that traditional media. To address these challenges, we propose method adapt existing lexicons for domain-specific classification using an unannotated corpus and dictionary. We evaluate our two large developing corpora, containing 743,069 tweets related stock market one million political topics, respectively, five as seeds baselines. results demonstrate usefulness method, showing significant improvement performance. classification.The proposed addresses both domain.We corpora baselines.The evaluation method.