Topic-aware sentiment prediction for Chinese ConceptNet

Sentiment dictionary is a valuable resource in sentiment analysis research. Previous works propagate sentiment values from existing high quality dictionary on semantic networks to build wide coverage dictionary efficiently. But this approach suffers from quality degradation during propagation. In this work, we propose a topic-aware propagation method on Chinese ConceptNet to ease the issue. With this approach, every terms will have different sentiment values under different topics. The experimental result shows that the generated topic-aware sentiment dictionary helps improve the performance of polarity classification for texts.