Summary of Large, Small or Both: a Novel Data Augmentation Framework Based on Language Models For Debiasing Opinion Summarization, by Yanyue Zhang et al.
Large, Small or Both: A Novel Data Augmentation Framework Based on Language Models for Debiasing Opinion Summarization
by Yanyue Zhang, Pengfei Li, Yilong Lai, Deyu Zhou, Yulan He
First submitted to arxiv on: 12 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 A novel approach to debiasing opinion summarization is proposed in this paper, which tackles the sentiment bias inherent in existing datasets. The authors recognize that current methods are reluctant to generate negative summaries given positive input texts, resulting in an over-reliance on specific frameworks. To address this issue, a data augmentation framework is introduced that combines large and small language models to balance the emotional distribution of the dataset. This framework involves synthesizing negative reviews by rewriting positive text using a large language model, followed by training a disentangle reconstruction model on generated data. The approach can effectively alleviate emotional bias while being more economically viable than relying solely on large models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about fixing a problem in how computers summarize opinions. Right now, most summary methods are too positive and won’t generate negative summaries even if they’re given negative text to work with. To fix this, the authors came up with a new way to make more data for training these algorithms. They use big language models to rewrite positive text into negative reviews, then train a special model on that new data. This approach can help make summary methods less biased and more accurate. |
Keywords
» Artificial intelligence » Data augmentation » Large language model » Summarization
