Work

Conversational Transcription and Knowledge Graph Representation

Year: 2026

Type: conference-paper

Abstract: In this paper, we investigate the construction of knowledge graphs (KGs) from conversational transcripts, such as those generated during online meetings (e.g., Zoom) or multispeaker dialogs. We specifically address the question of representation, examining whether incorporating the conversational structure, including speaker information and dialog organization, influences the resulting knowledge graph and improves its performance. In particular, we study the impact of chat based representations on the quality of KGs for question answering tasks. Knowledge graphs are automatically extracted using different large language models (LLMs). The evaluation is conducted on a benchmark dataset derived from the first season of the Friends TV series, which includes conversational transcripts and associated question-answer pairs. Our experimental results highlight the importance of conversational structure in knowledge graph construction and demonstrate its impact on downstream question answering performance.

Source: IEEE Annual International Computer Software and Applications Conference (COMPSAC)

DOI: 10.1109/compsac69091.2026.00266

Authors: Karima Boutalbi, Antoine Lévêque, Olivier Le Van, Rafika Boutalbi

Institutions: Cegedim (France), École Pour l'Informatique et les Techniques Avancées, Aix-Marseille Université

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W7203833201

FWCI:

Cites: 14

Cited by:

Related to: 10