Re-FRAME the Meeting Summarization SCOPE: Fact-Based Summarization and Personalization via Questions

Meeting summarization with large language models (LLMs) remains error-prone, often producing outputs with hallucinations, omissions, and irrelevancies.We present FRAME, a modular pipeline that reframes summarization as a semantic enrichment task.FRAME extracts and scores salient facts, organizes them thematically, and uses these to enrich an outline into an abstractive summary.To personalize summaries, we introduce SCOPE, a reasonout-loud protocol that has the model build a reasoning trace by answering nine questions before content selection.For evaluation, we propose P-MESA, a multi-dimensional, referencefree evaluation framework to assess if a summary fits a target reader.P-MESA reliably identifies error instances, achieving ≥ 89% balanced accuracy against human annotations and strongly aligns with human severity ratings (ρ ≥ 0.70).On QMSum and FAME, FRAME reduces hallucination and omission by 2 out of 5 points (measured with MESA), while SCOPE improves knowledge fit and goal alignment over prompt-only baselines.Our findings advocate for rethinking summarization to improve control, faithfulness, and personalization 1 .QMSum FAME [EN] GPT 4o Gemini 1.5 pro FRAME GPT-4o GPT 4o Gemini 1.5 pro FRAME GPT-4o MESA (lower is better) Coreference 0 1.22 3 1.58 0 1.64 0 1.45 3 1.570 1.36 Hallucination 3 1.22 4 2.04 1 1.75 4 0.98 4 1.40 1 0.72 Incoherence 4 1.50 4 1.09 3 1.88 4 0.94 4 0.72 3 1.24 Irrelevance 2 1.70 3 1.32 1 1.45 3 1.14 3 1.07 1 1.51 Language 1 1.30 2 1.44 1 1.40 1 1.17 1 1.20 1 1.11 Omission 3 0.40 3 0.38 1 0.16 4 0.16 4 0.31 1 0.00 Repetition 4 1.05 3 0.98 1 1.23 4 0.74 4 0.44 2 0.53 Structure 4 0.90 3 1.70 3 1.24 3 1.57 3 1.53 3 1.46