Building Dynamic Knowledge Graphs from Text using Machine Reading Comprehension

Abstract

We propose a neural machine-reading model that constructs dynamic knowledge graphs from procedural text. It builds these graphs recurrently for each step of the described procedure, and uses them to track the evolving states of participant entities. We harness and extend a recently proposed machine reading comprehension (MRC) model to query for entity states, since these states are generally communicated in spans of text and MRC models perform well in extracting entity-centric spans. The explicit, structured, and evolving knowledge graph representations that our model constructs can be used in downstream question answering tasks to improve machine comprehension of text, as we demonstrate empirically. On two comprehension tasks from the recently proposed PROPARA dataset (Dalvi et al., 2018), our model achieves state-of-the-art results. We further show that our model is competitive on the RECIPES dataset (Kiddon et al., 2015), suggesting it may be generally applicable. We present some evidence that the model's knowledge graphs help it to impose commonsense constraints on its predictions.

ICLR 2019 submission

Cited by in corpus (16)

• Representation Learning for Dynamic Graphs: A Survey
• Recent Advances in Natural Language Inference: A Survey of Benchmarks, Resources, and Approaches
• Learning Dynamic Belief Graphs to Generalize on Text-Based Games
• Question Answering is a Format; When is it Useful?
• Enhancing Text-based Reinforcement Learning Agents with Commonsense Knowledge
• Interactive Language Learning by Question Answering
• Building Dynamic Knowledge Graphs from Text-based Games
• A Heterogeneous Graph with Factual, Temporal and Logical Knowledge for Question Answering Over Dynamic Contexts
• Be Consistent! Improving Procedural Text Comprehension using Label Consistency
• Tracking entities in technical procedures -- a new dataset and baselines
• Multilingual Answer Sentence Reranking via Automatically Translated Data
• Playing by the Book: An Interactive Game Approach for Action Graph Extraction from Text
• DyKgChat: Benchmarking Dialogue Generation Grounding on Dynamic Knowledge Graphs
• Knowledge-Empowered Representation Learning for Chinese Medical Reading Comprehension: Task, Model and Resources
• Effective Use of Transformer Networks for Entity Tracking
• An Approach for Process Model Extraction By Multi-Grained Text Classification