# FRACTAL: Fine-Grained Scoring from Aggregate Text Labels

arXiv: https://arxiv.org/abs/2404.04817 · HTML: https://mlsanity.com/paper/2404.04817/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 28 of its 74 references that are in the map.
- First posted on arXiv 7 Apr 2024.

## Builds on (most cited first, 28 of 28)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Large Language Models are Zero-Shot Reasoners](https://mlsanity.com/paper/2205.11916/) (2022)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Towards a Human-like Open-Domain Chatbot](https://mlsanity.com/paper/2001.09977/) (2020)
- [Fine-Grained Human Feedback Gives Better Rewards for Language Model Training](https://mlsanity.com/paper/2306.01693/) (2023)
- [Teaching language models to support answers with verified quotes](https://mlsanity.com/paper/2203.11147/) (2022)
- [ASQA: Factoid Questions Meet Long-Form Answers](https://mlsanity.com/paper/2204.06092/) (2022)
- [Sparks: Inspiration for Science Writing using Language Models](https://mlsanity.com/paper/2110.07640/) (2021)
- [DocNLI: A Large-scale Dataset for Document-level Natural Language Inference](https://mlsanity.com/paper/2106.09449/) (2021)
- [Weakly-Supervised Action Localization with Expectation-Maximization Multi-Instance Learning](https://mlsanity.com/paper/2004.00163/) (2020)
- [AQuaMuSe: Automatically Generating Datasets for Query-Based Multi-Document Summarization](https://mlsanity.com/paper/2010.12694/) (2020)
- [Deep multi-class learning from label proportions](https://mlsanity.com/paper/1905.12909/) (2019)
- [Learning from Label Proportions by Learning with Label Noise](https://mlsanity.com/paper/2203.02496/) (2022)
- [Co-training for Demographic Classification Using Deep Learning from Label Proportions](https://mlsanity.com/paper/1709.04108/) (2017)
- [Learning from Label Proportions with Generative Adversarial Networks](https://mlsanity.com/paper/1909.02180/) (2019)
- [Multiple Instance Learning Networks for Fine-Grained Sentiment Analysis](https://mlsanity.com/paper/1711.09645/) (2017)
- [Weakly Supervised Classification in High Energy Physics](https://mlsanity.com/paper/1702.00414/) (2017)
- [Generating Summaries with Topic Templates and Structured Convolutional Decoders](https://mlsanity.com/paper/1906.04687/) (2019)
- [Two-stage Training for Learning from Label Proportions](https://mlsanity.com/paper/2105.10635/) (2021)
- [SMART: Sentences as Basic Units for Text Evaluation](https://mlsanity.com/paper/2208.01030/) (2022)
- [Deep Learning from Label Proportions for Emphysema Quantification](https://mlsanity.com/paper/1807.08601/) (2018)
- [Challenges and approaches to privacy preserving post-click conversion prediction](https://mlsanity.com/paper/2201.12666/) (2022)
- [Fine-Grained Relevance Annotations for Multi-Task Document Ranking and Question Answering](https://mlsanity.com/paper/2008.05363/) (2020)
- [Fast learning from label proportions with small bags](https://mlsanity.com/paper/2110.03426/) (2021)
- [Learning under Label Proportions for Text Classification](https://mlsanity.com/paper/2310.11707/) (2023)

## Abstract (arXiv)

Large language models (LLMs) are being increasingly tuned to power complex generation tasks such as writing, fact-seeking, querying and reasoning. Traditionally, human or model feedback for evaluating and further tuning LLM performance has been provided at the response level, enabling faster and more cost-effective assessments. However, recent works (Amplayo et al. [2022], Wu et al. [2023]) indicate that sentence-level labels may provide more accurate and interpretable feedback for LLM optimization. In this work, we introduce methods to disaggregate response-level labels into sentence-level (pseudo-)labels. Our approach leverages multiple instance learning (MIL) and learning from label proportions (LLP) techniques in conjunction with prior information (e.g., document-sentence cosine similarity) to train a specialized model for sentence-level scoring. We also employ techniques which use model predictions to pseudo-label the train-set at the sentence-level for model training to further improve performance. We conduct extensive evaluations of our methods across six datasets and four tasks: retrieval, question answering, summarization, and math reasoning. Our results demonstrate improved performance compared to multiple baselines across most of these tasks. Our work is the first to develop response-level feedback to sentence-level scoring techniques, leveraging sentence-level prior information, along with comprehensive evaluations on multiple tasks as well as end-to-end finetuning evaluation showing performance comparable to a model trained on fine-grained human annotated labels.