SCI 3.0: A Web-based Schema Curation Interface for Graphical Event Representations

TL;DR

The paper addresses the challenge of representing complex real-world events as structured, hierarchical schemas for NLP and information extraction. It introduces Schema Curation Interface 3.0 (SCI 3.0), a web-based tool built with React, Flask, and Cytoscape that enables real-time graph-based editing of events, entities, relations, and outlinks, bridging automatic schema induction with human curation. The workflow combines an automatic induction stage using hierarchicalschema2023 and GPT-3 with manual refinement via SCI 3.0, followed by a schema-guided prediction loop using the RESIN pipeline to improve coverage. The work demonstrates significant improvements in schema coverage through iterative curation, discusses potential expansions to AMR/UMR and sub-schema libraries, and highlights practical impacts for information extraction, data mining, and policy-relevant analytics.

Abstract

To understand the complexity of global events, one must navigate a web of interwoven sub-events, identifying those most impactful elements within the larger, abstract macro-event framework at play. This concept can be extended to the field of natural language processing (NLP) through the creation of structured event schemas which can serve as representations of these abstract events. Central to our approach is the Schema Curation Interface 3.0 (SCI 3.0), a web application that facilitates real-time editing of event schema properties within a generated graph e.g., adding, removing, or editing sub-events, entities, and relations directly through an interface.

Figures (6)

• Figure 1: Snapshot of the event schema graph in the SCI Viewer. Sub-events are linked as children of the chapter events and feature participants, with their roles described by the argument labels along the dashed directed edges.
• Figure 2: Options displayed when any chapter or primitive event node is right-clicked.
• Figure 3: Prompt for creating an entity, along with its required fields.
• Figure 4: Overview of the rendered schema graph structure. Dark blue diamond elements represent chapter events. Light blue diamonds indicate optional chapter events. Similarly, the the yellow ovals in this image represent primitive events. The expanded primitive event includes participants, their role names, and relation to one another.
• Figure 5: Example window for editing an event node. Allows users to signify event importance, edit event properties, and indicate optionality.