Target free or target specific: a simple and effective zero sample position detection comparative learning method

PaperWeekly original ·  author |  Liang Bin 、 Chen Zixiao

Company |  Harbin Institute of Technology （ Shenzhen ）

Research direction |  Sentiment analysis 、 Position detection

Paper information ：

Bin Liang, Zixiao Chen, Lin Gui, Yulan He, Min Yang, Ruifeng Xu*. Zero-Shot Stance Detection via Contrastive Learning, The 31th Web Conference (TheWebConf 2022), Apr. 2022.

Address of thesis ：

http://www.hitsz-hlt.com/paper/Zero-Shot-Stance-Detection-via-Contrastive-Learning-WWW2022.pdf

Code address ：

https://github.com/HITSZ-HLT/PT-HCL

In this work , We propose a very simple and effective contrastive learning framework for this task . In particular , We cleverly use a proxy task （pretext task） To assist unknown targets （target） Position detection . The purpose of the agent task is to generate an additional label for each training sample （label）, The position expression of the training sample is marked （stance expression） It's not about the goal （target-invariant） Or specific to the target （target-specific）.