Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RAG-RLRC-LaySum at BioLaySumm: Integrating Retrieval-Augmented Generation and Readability Control for Layman Summarization of Biomedical Texts

Yuelyu Ji, Zhuochun Li, Rui Meng, Sonish Sivarajkumar, Yanshan Wang, Zeshui Yu, Hui Ji, Yushui Han, Hanyu Zeng, Daqing He

TL;DR

RAG-RLRC-LaySum tackles the problem of making biomedical literature accessible to lay audiences by grounding abstractive summaries with retrieval-augmented generation and steering readability through reinforcement learning. The approach combines a Longformer Encoder-Decoder backbone, Wikipedia-based retrieval, neural passage re-ranking, and PPO-based readability optimization, with LLMs used for paraphrasing and refinement. Across PLOS and eLife, the framework achieves higher relevance and readability while maintaining factuality, outperforming baselines and illustrating the importance of grounding and readability control in lay summarization. The work advances public engagement with biomedical research by offering a practical, extensible method to produce clear, accurate lay explanations of complex science.

Abstract

This paper introduces the RAG-RLRC-LaySum framework, designed to make complex biomedical research understandable to laymen through advanced Natural Language Processing (NLP) techniques. Our Retrieval Augmented Generation (RAG) solution, enhanced by a reranking method, utilizes multiple knowledge sources to ensure the precision and pertinence of lay summaries. Additionally, our Reinforcement Learning for Readability Control (RLRC) strategy improves readability, making scientific content comprehensible to non-specialists. Evaluations using the publicly accessible PLOS and eLife datasets show that our methods surpass Plain Gemini model, demonstrating a 20% increase in readability scores, a 15% improvement in ROUGE-2 relevance scores, and a 10% enhancement in factual accuracy. The RAG-RLRC-LaySum framework effectively democratizes scientific knowledge, enhancing public engagement with biomedical discoveries.

RAG-RLRC-LaySum at BioLaySumm: Integrating Retrieval-Augmented Generation and Readability Control for Layman Summarization of Biomedical Texts

TL;DR

RAG-RLRC-LaySum tackles the problem of making biomedical literature accessible to lay audiences by grounding abstractive summaries with retrieval-augmented generation and steering readability through reinforcement learning. The approach combines a Longformer Encoder-Decoder backbone, Wikipedia-based retrieval, neural passage re-ranking, and PPO-based readability optimization, with LLMs used for paraphrasing and refinement. Across PLOS and eLife, the framework achieves higher relevance and readability while maintaining factuality, outperforming baselines and illustrating the importance of grounding and readability control in lay summarization. The work advances public engagement with biomedical research by offering a practical, extensible method to produce clear, accurate lay explanations of complex science.

Abstract

This paper introduces the RAG-RLRC-LaySum framework, designed to make complex biomedical research understandable to laymen through advanced Natural Language Processing (NLP) techniques. Our Retrieval Augmented Generation (RAG) solution, enhanced by a reranking method, utilizes multiple knowledge sources to ensure the precision and pertinence of lay summaries. Additionally, our Reinforcement Learning for Readability Control (RLRC) strategy improves readability, making scientific content comprehensible to non-specialists. Evaluations using the publicly accessible PLOS and eLife datasets show that our methods surpass Plain Gemini model, demonstrating a 20% increase in readability scores, a 15% improvement in ROUGE-2 relevance scores, and a 10% enhancement in factual accuracy. The RAG-RLRC-LaySum framework effectively democratizes scientific knowledge, enhancing public engagement with biomedical discoveries.
Paper Structure (20 sections, 5 equations, 2 figures, 4 tables)

This paper contains 20 sections, 5 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Retrieval Augmented Generation (RAG)
  4. Reinforcement Learning for Readability Control (RLRC)
  5. Large Language Models
  6. Experimental Settings and Results
  7. Datasets and Evaluation
  8. Performance of Baseline Models
  9. Effect of Neural Re-rankers
  10. Effect of Large Language Models
  11. Effect of RLRC
  12. Related Work
  13. Conclusion and Future Work
  14. Limitations
  15. Finetuning models
  16. ...and 5 more sections

Figures (2)

  • Figure 1: Knowledge Retrieval Augmented, with the trained re-ranker, can provide more relevant knowledge based on the first generation.
  • Figure 2: This figure illustrates the architecture of the proposed RAG-RLRC-LaySum model. During the training phase, we employ the Longformer Encoder-Decoder (LED) model as the backbone beltagy2020longformer. We enhance the model's capabilities through Wikipedia knowledge retrieval during inference. We utilize large language models (LLMs) such as ChatGPT and Gemini to further improve readability and enhance textual clarity by modifying prompts. For controlled text generation, readability scores are utilized to guide the model in generating expected outputs. The outputs of these scores are normalized to ensure text consistency and quality across generated texts.