Summary of Multi-dimensional Optimization For Text Summarization Via Reinforcement Learning, by Sangwon Ryu et al.
Multi-Dimensional Optimization for Text Summarization via Reinforcement Learning
by Sangwon Ryu, Heejin Do, Yunsu Kim, Gary Geunbae Lee, Jungseul Ok
First submitted to arxiv on: 1 Jun 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 The proposed multi-objective reinforcement learning method generates well-balanced summaries across multiple dimensions, including consistency, coherence, relevance, and fluency. The authors introduce two strategies for adaptive learning: MDO_min, which rewards the current lowest dimension score, and MDO_pro, which optimizes multiple dimensions similar to multi-task learning, resolving conflicting gradients across dimensions through gradient projection. A QA-based reward model aligns with human preferences, unlike prior ROUGE-based rewards relying on reference summaries. The approach achieves substantial performance gains compared to baseline models on representative summarization datasets, particularly in the overlooked dimensions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to create summary texts that are good across many different aspects. Right now, most summary methods focus on one specific thing, like making sure the text makes sense or is relevant to what it’s about. But this approach tries to do multiple things at once, which can be hard because they might conflict with each other. To make it work, the authors developed two special techniques: MDO_min and MDO_pro. These help the computer learn how to create summaries that are good in many different ways. The researchers also came up with a new way to measure how well a summary is doing, which is based on questions people would ask about what the text says. |
Keywords
» Artificial intelligence » Multi task » Reinforcement learning » Rouge » Summarization
