Abstract:Query-Focused Meeting Summarization (QFMS) aims to generate a summary of a given meeting transcript conditioned upon a query. The main challenges for QFMS are the long input text length and sparse query-relevant information in the meeting transcript. In this paper, we propose a knowledge-enhanced two-stage framework called Knowledge-Aware Summarizer (KAS) to tackle the challenges. In the first stage, we introduce knowledge-aware scores to improve the query-relevant segment extraction. In the second stage, we incorporate query-relevant knowledge in the summary generation. Experimental results on the QMSum dataset show that our approach achieves state-of-the-art performance. Further analysis proves the competency of our methods in generating relevant and faithful summaries.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are two main challenges in **Query - Focused Meeting Summarization (QFMS)**: 1. **Long input text length**: Meeting transcripts are usually very long, and current deep - learning models are unable to encode such long texts at one time. Even for some models that can handle long - text input (such as Longformer proposed by Beltagy et al.), their computational complexity is also very high, which makes it difficult to process long meeting transcripts. 2. **Sparse query - related information**: In meeting transcripts, the parts of information related to the query are sparsely distributed, which means that most of the meeting transcript content is noise information for a specific query. Therefore, the model needs to effectively reduce the influence of these noise information to generate more accurate summaries. To solve these problems, the authors propose a knowledge - enhanced two - stage framework, called **Knowledge - Aware Summarizer (KAS)**: - **First stage**: Improve the extraction of query - related paragraphs by introducing knowledge - aware scoring. In this stage, OpenIE is used to extract knowledge triples from text paragraphs, and the knowledge - aware score of each paragraph is calculated through L2 normalization. Then, combined with the semantic search score (calculated using Multi - QA MPNet) to rank the paragraphs, and select the top k paragraphs with the highest ranking. - **Second stage**: Incorporate query - related knowledge in the summary generation process. In this stage, the FiD - BART model is used, taking the query, the selected paragraphs and the extracted knowledge as input to generate the final query - focused meeting summary. The experimental results show that this method achieves state - of - the - art performance on the QMSum dataset, and further analysis and human evaluation also prove the advantages of this method in generating fluent, relevant and faithful summaries.

Improving Query-Focused Meeting Summarization with Query-Relevant Knowledge

Query-Utterance Attention with Joint modeling for Query-Focused Meeting Summarization

QMSum: A New Benchmark for Query-based Multi-domain Meeting Summarization

Query-focused Abstractive Summarization Via Question-answering Model

Learning to Rank Utterances for Query-Focused Meeting Summarization

Exploring Neural Models for Query-Focused Summarization

Beyond Relevant Documents: A Knowledge-Intensive Approach for Query-Focused Summarization using Large Language Models

SDbQfSum: Query‐focused summarization framework based on diversity and text semantic analysis

QFMTS: Generating Query-Focused Summaries over Multi-Table Inputs

QuerySum: A Multi-Document Query-Focused Summarization Dataset Augmented with Similar Query Clusters

A Knowledge Graph-Based Abstractive Model Integrating Semantic and Structural Information for Summarizing Chinese Meetings

Dialogue acts enhanced extract–abstract framework for meeting summarization

QontSum: On Contrasting Salient Content for Query-focused Summarization

KATSum: Knowledge-aware Abstractive Text Summarization

Fact-Driven Abstractive Summarization by Utilizing Multi-Granular Multi-Relational Knowledge

Investigating Consistency in Query-Based Meeting Summarization: A Comparative Study of Different Embedding Methods

KEMoS: A knowledge-enhanced multi-modal summarizing framework for Chinese online meetings

QTSumm: Query-Focused Summarization over Tabular Data

IDEAL: Leveraging Infinite and Dynamic Characterizations of Large Language Models for Query-focused Summarization

Query-Focused Summarization by Combining Topic Model and Affinity Propagation

Domain Adaptation and Summary Distillation for Unsupervised Query Focused Summarization