Summary of Learning with Silver Standard Data For Zero-shot Relation Extraction, by Tianyin Wang et al.
Learning with Silver Standard Data for Zero-shot Relation Extraction
by Tianyin Wang, Jianwei Wang, Ziqian Zeng
First submitted to arxiv on: 25 Nov 2022
Categories
- Main: Computation and Language (cs.CL)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary In this paper, researchers propose a new approach to improve the performance of supervised relation extraction (RE) methods by leveraging large-scale silver standard data. Zero-shot RE methods have already shown promise in converting RE tasks into other NLP tasks and achieving high accuracy without relying on extensive annotation efforts. The proposed method involves detecting a small amount of clean data within the silver standard dataset, fine-tuning a pre-trained model using this selected data, and then inferring relation types using the finetuned model. Additionally, the authors introduce a class-aware clean data detection module that considers class information when selecting clean data. Experimental results demonstrate a significant improvement in zero-shot RE task performance on TACRED and Wiki80 datasets, with an increase of 12% and 11%, respectively. By utilizing extra silver standard data from different distributions, the approach can be further optimized. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding a way to make computers better at understanding relationships between things in text. Right now, computers are really good at this task when they have a lot of labeled training data. But what if we didn’t need that much training data? Researchers came up with an idea: take the labeled data and use it to fine-tune a pre-trained model. This means taking a model that was already trained on some other task, but not specifically for this relationship understanding task. Then, they use the fine-tuned model to make predictions about new text. The results show that this approach can work really well, especially when combined with extra data from different sources. |
Keywords
» Artificial intelligence » Fine tuning » Nlp » Supervised » Zero shot
