Tibetan Aspect Extraction based on syllable feature

The paper proposes a new approach to evaluation target extraction for Tibetan news texts, which integrates Tibetan syllable components and character features to address segmentation errors propagation issues caused by previous models utilizing word embedding. The model used Bi-LSTM for feature extraction after feature fusion in the encoder and CRF(Conditional Random Field) for target output. The experimental results showed that the proposed model performed better than previous models