From 33dcd511456a9e6f2d2fc21a19bd206e4d0a16bb Mon Sep 17 00:00:00 2001 From: Yao Wang Date: Thu, 31 Jul 2025 09:29:26 +0900 Subject: [PATCH] papaer addition: A Retrieval-augmented Complex Story Generation Framework with A Forest of Evidence --- README.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index a15840d..fe0b632 100644 --- a/README.md +++ b/README.md @@ -8,6 +8,7 @@ Given the rapid evolution of this field, we will continue to update the reposito - Evaluation Campaigns - Open-source Projects - Workshops & Tutorials +- Papers ### [Survey Papers]() @@ -67,4 +68,5 @@ Given the rapid evolution of this field, we will continue to update the reposito - Highlight: The workshop of [BREV-RAG@SIGIR-AP 2025](http://sakailab.com/brev-rag/) focuses on the viewpoint of evaluation. ### [Papers]() -TBA +- A Retrieval-augmented Complex Story Generation Framework with A Forest of Evidence ([paper](https://arxiv.org/pdf/2310.05388) + - Highlight: This paper presents GROVE, a retrieval-augmented story generation framework designed to help large language models (LLMs) generate complex and credible narratives. The framework consists of three stages: (1) building a retrieval repository of human-written stories based on target control conditions (e.g., plot, mood, genre, subject); (2) constructing an evidence forest through an iterative “asking-why” prompting process to uncover missing or ambiguous background details; and (3) story rewriting by incorporating selected evidence chains to enrich and refine the story. Experimental results show that GROVE outperforms baselines such as ICL, CoT, and Story-S on both human and automatic evaluations, especially in plot complexity and creativity. The approach demonstrates generalizability even on smaller models like Alpaca-Plus-7B, and includes novel metrics and a detailed ablation study to support its findings.