Unlocking Understanding: A Novel Pipeline for Automated Concept Map Extraction from Text

Anonymous

Abstract: Concept maps are graphs of entities and their relations that can foster students' understanding of texts. However, manually constructing them is a challenging task. To overcome this, automatic concept map extraction methods have emerged, typically using a pipeline approach to extract entities and their relations. Yet, existing methods face limitations in scalability, scarcity of data and non open-access architectures. To bridge these gaps, we introduce a novel, modularized and open-source pipeline for concept map extraction, using semantic and sub-symbolic techniques. To address scalability, we integrate a summarization step over the input documents and an importance ranking step to make relation extraction more efficient. To tackle data scarcity, we fine-tune a sequence-to-sequence neural model with limited annotated examples. Our approach achieves state-of-the-art performance on METEOR metrics, particularly crucial for concept maps, given the focus on semantic similarity of this metric, and state-of-the-art precision for ROUGE-2. This contribution advances automated concept map extraction, opening doors to wider applications supporting learning and knowledge access.

Paper Type: long

Research Area: Summarization

Contribution Types: Model analysis & interpretability, Publicly available software and/or pre-trained models

Languages Studied: English