Entropy-enhanced context-aware event prediction based on ontology and external knowledge
Keywords: Event prediction, Event Graph, Ontology, Graph Entropy, Context Aware, External Knowledge, Graph Neural Networks
TL;DR: We propose an event prediction model based on GNNs, which uses entropy to solve the heterogeneity of event graph and introduces event ontology to describe event context, and we verify the effectiveness on our constructed maritime emergency dataset.
Abstract: Predicting impending events is an attractive task of natural language processing (NLP), which plays an important role in many fields such as emergency management. Current predominant event prediction method is to construct event graphs and learn event representations through graph neural networks (GNNs), which mainly utilizes the semantic and structural information of events to obtain subsequent events, but ignoring the context of event. Meanwhile, these methods does not address the issue of heterogeneity of nodes and edges in event networks. Last, the lack of high-quality event datasets is also a challenge for event prediction. In response to the above issues, this paper proposes the Entropy-enhanced Context-aware Ontology-based model (ECO), which introduces the Entropy calculation module to learn the heterogeneity of nodes and edges, thereby better learning event representations. Furthermore, external knowledge is introduced to the event graph to enhance the semantic information of events during the prediction. Finally, we design a context-aware event ontology for maritime emergency management, and construct a real-world dataset, Maritime Emergency Events Dataset (MEED), to verify our prediction method. Experiments on node classification and link prediction show effectiveness and practicability of our proposed model in realistic scenarios.
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1618
Loading