Stance Classification-2

Stance Classification - 2 (SC2): An SC task is used to estimate a politicians' position from her/his utterances. Taking a lesson from the last SC subtask evaluated at the PoliInfo-2, we are planning to revisit it by taking into account the following two aspects. We first redesign the classification task. In the last task, the information source of the classification was assembly minutes as a whole. We found that members of an assembly tend to state their stance on a given topic explicitly at the beginning of their speech. While most of the task participants successfully exploited this to achieve good classification accuracy, it made the task unexpectedly easy. Therefore, in SC2, we focus on a classification of members' opinion about a given topic without any explicit statement on their stance. We then extend the target minutes to several local governments, other than Tokyo, in Japan, e.g., Sapporo and other governments in Hokkaido prefecture.

• Input: 

  • A topic

  • An assembly member's opinion sentences on the given topic without explicit statement of neither approval nor disapproval.

• Output:

  • Relevance (Binary) : Either approval or disapproval on a given topic

• Evaluation:

  • Accuracy  (for automatic evaluation)