Improving Attributed Long-form Question Answering with Intent Awareness

arXiv cs.CL / 3/31/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes that improving an LLM’s “intent awareness” can raise the quality of long-form, knowledge-intensive question answering and report generation.
It introduces structured, tag-based methods to extract implicit intents authors use when writing and citing sources, aiming to better align model reasoning with human document goals.
Experiments show that extracted intents improve zero-shot long-form report generation and also help create higher-quality synthetic data for fine-tuning smaller models.
Results report average gains of +2.9 points for large models and +12.3 points for small models versus baselines across multiple scientific report generation tasks.
The study further finds that intent-aware models make better citation choices and generate reports with substantially improved readability.

Abstract

Large language models (LLMs) are increasingly being used to generate comprehensive, knowledge-intensive reports. However, while these models are trained on diverse academic papers and reports, they are not exposed to the reasoning processes and intents that guide authors in crafting these documents. We hypothesize that enhancing a model's intent awareness can significantly improve the quality of generated long-form reports. We develop and employ structured, tag-based schemes to better elicit underlying implicit intents to write or cite. We demonstrate that these extracted intents enhance both zero-shot generation capabilities in LLMs and enable the creation of high-quality synthetic data for fine-tuning smaller models. Our experiments reveal improved performance across various challenging scientific report generation tasks, with an average improvement of +2.9 and +12.3 absolute points for large and small models over baselines, respectively. Furthermore, our analysis illuminates how intent awareness enhances model citation usage and substantially improves report readability.