Summarizing long-form narratives-such as books, movies, and TV scripts-requires capturing intricate plotlines, character interactions, and thematic coherence, a task that remains challenging for existing LLMs.We introduce NEXUSSUM, a multi-agent LLM framework for narrative summarization that processes long-form text through a structured, sequential pipeline-without requiring fine-tuning.Our approach introduces two key innovations: (1) Dialogue-to-Description Transformation: A narrative-specific preprocessing method that standardizes character dialogue and descriptive text into a unified format, improving coherence.(2) Hierarchical Multi-LLM Summarization: A structured summarization pipeline that optimizes chunk processing and controls output length for accurate, high-quality summaries.Our method establishes a new state-of-the-art in narrative summarization, achieving up to a 30.0%improvement in BERTScore (F1) across books, movies, and TV scripts.These results demonstrate the effectiveness of multiagent LLMs in handling long-form content, offering a scalable approach for structured summarization in diverse storytelling domains.

Large Language Models (LLMs) have demonstrated near-human performance in summarization tasks based on traditional metrics such as ROUGE and BERTScore.However, these metrics do not adequately capture critical aspects of summarization quality, such as factual accuracy, particularly for long narratives (>100K tokens).Recent advances, such as LLM-as-a-Judge, address the limitations of metrics based on lexical similarity but still exhibit factual inconsistencies, especially in understanding character relationships and states.In this work, we introduce NARRATIVEFACTSCORE (NFS), the first "Agent-as-a-Judge" framework that evaluates and refines factuality in narrative summarization.By leveraging a Character Knowledge Graph (CKG) extracted from input narrative, NARRATIVEFACTSCORE evaluates the factuality and provides actionable guidance for refinement, such as identifying missing or erroneous facts.Our experimental results demonstrate that constructing the CKG enables reasoning with 1/3 of the factuality computation used in the prior approach, and achieve three times higher correlation with human judgments.Furthermore, refinement with actionable guidance improves the quality of the summary. 1

In this paper, we introduce an efficient method for the joint linear-programming (LP) decoding of low-density parity-check (LDPC) codes and finite-state channels (FSCs). In particular, we extend the approach of iterative approximate LP decoding proposed by Vontobel and Koetter and further analyzed by Burshtein. By taking advantage of the dual-domain structure of the joint decoding LP, we obtain a convergent iterative algorithm for joint decoding. Our intent was to find an algorithm very similar to BCJR-based turbo equalization (TE) and it seems that we have succeeded in this respect. The result is an iterative solver for the joint decoding LP whose complexity is similar to TE but whose performance is similar to joint LP decoding. The main advantage of this decoder is that it appears to provide the predictability of joint LP decoding with the computational complexity of TE.

In this paper, the linear programming (LP) decoder for binary linear codes, introduced by Feldman, et al. is extended to joint-decoding of binary-input finite-state channels. In particular, we provide a rigorous definition of LP joint-decoding pseudo-codewords (JD-PCWs) that enables evaluation of the pairwise error probability between codewords and JD-PCWs. This leads naturally to a provable upper bound on decoder failure probability. If the channel is a finite-state intersymbol interference channel, then the LP joint decoder also has the maximum-likelihood (ML) certificate property and all integer valued solutions are codewords. In this case, the performance loss relative to ML decoding can be explained completely by fractional valued JD-PCWs.