Data for our AAAI'19, oral paper "Exploiting Coarse-to-Fine Task Transfer for Aspect-level Sentiment Classification".

This dataset acts as highly beneficial source domains to improve the learning of more fine-grained aspect-term level (AT) sentiment analysis. The dataset has three characteristics:

Even with a simple attention-based model for the AT task, our method can achieve the STOA performances by leveraging the knowledge distilled from the AC task.

Each instance behaves as the format below:

inst1: ID1/sentence1/aspect1/label1

inst2: ID1/sentence1/aspect2/label2

inst3: ID1/sentence1/aspect3/label3

....

The sentence containing multiple aspects are arranged together.

0H0FwmPY78v_5u51r2TQrw i did n't dislike the food , but the menu is n't exactly cohesive ... pizza and asian cuisine . FOOD_SELECTION -1

0H0FwmPY78v_5u51r2TQrw i did n't dislike the food , but the menu is n't exactly cohesive ... pizza and asian cuisine . FOOD_FOOD_DISH -1

0H0FwmPY78v_5u51r2TQrw i did n't dislike the food , but the menu is n't exactly cohesive ... pizza and asian cuisine . RESTAURANT_CUSINE -1

0H0FwmPY78v_5u51r2TQrw i did n't dislike the food , but the menu is n't exactly cohesive ... pizza and asian cuisine . FOOD_FOOD 1

positive: 1 neutral: 0 negative: -1

If the data is useful for your research, please be kindly to give us stars and cite our paper as follows:

@article{li2018exploiting,
  title={Exploiting Coarse-to-Fine Task Transfer for Aspect-level Sentiment Classification},
  author={Li, Zheng and Wei, Ying and Zhang, Yu and Zhang, Xiang and Li, Xin and Yang, Qiang},
  conference = {AAAI Conference on Artificial Intelligence},
  year={2019}
}