Event Grounding via Semantic Parsing and Knowledge Graph Anchoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI systems struggle to effectively ground events in free-texts to event-centric Knowledge Graphs (KGs), particularly due to challenges in event representation and sparsity, which limits their ability to perform narrative reasoning.
Innovation Solution
The EvenGround framework addresses these challenges by employing a semantic parsing-based event extraction method with an event normalization module, and a multi-level event abstraction approach. It grounds abstracted events to KGs and uses a Graph Neural Network (GNN)-based model for reasoning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If implicit modeling with language models is used to capture event knowledge, then event understanding capability is improved, but transparency and explainability of reasoning deteriorates
Solution Approach 1:
The patent introduces an event-centric knowledge graph as an intermediary between the language model and the reasoning process. The KG explicitly stores event templates, arguments, and relationships, serving as a mediator that makes the implicit knowledge in LMs explicit and interpretable while maintaining the model's event understanding capabilities
2Loss of information
If explicit modeling with knowledge graphs is used to organize world knowledge, then transparency and explainability is improved, but the ability to generalize to free-texts deteriorates
Solution Approach 1:
The patent employs dynamic event template selection and adaptive argument matching that can adjust to different text formats. The system dynamically selects relevant event templates from the KG based on the input text characteristics and adapts the matching process to handle both structured and free-text inputs effectively
Solution Approach 2:
The patent changes the parameters of event representation by using flexible argument roles and variable event template structures. This allows the same KG framework to handle different text formats by adjusting how events are parsed, represented, and matched to templates
3Stability of the object's composition
If event-centric knowledge graphs are used to represent events, then event knowledge organization is improved, but event sparsity and representation challenges worsen
Solution Approach 1:
The patent designs universal event templates that can represent multiple specific event types through parameter variation. A single event template structure serves multiple functions by accommodating different events with varying arguments, reducing sparsity through generalization while maintaining organizational stability
Solution Approach 2:
The patent discards overly specific event representations that suffer from sparsity and recovers them through template-based generalization. By abstracting specific events into template instances, the system reduces sparsity while preserving essential event knowledge organization
Data Source
AI summary
An AI-based event grounding system is provided. The event grounding system includes an input device, an output device, a graphic processing unit (GPU); and a processor. The processor connects the input device, the output device, and the GPU. The event grounding system receives a free-text through the input device, and the processor perform an event grounding to the free-text through the GPU. The GPU performs event acquisition from the free-text using semantic parsing and acquires a plurality of verb-centric events, and performs event abstraction and acquires a plurality of abstract events, and grounds the abstract events to a plurality of anchor events of an event-centric KG, and reasons a subgraph through a reasoning model, and the subgraph includes the abstract events and the anchor events. The GPU generates a prediction based on the reasoning and provide the prediction through the processor and the output device.