NERVIS: An Interactive System for Graph-Based Exploration and Editing of Named Entities

TL;DR

NERVIS tackles the challenge of navigating large, entity-rich text corpora by presenting a graph-based interactive visualization that integrates documents, entity mentions, and entities. The system introduces a three-tier data model, a five-stage visualization pipeline with flexible view and data filtering, and in-system editing, enabling human-centered sensemaking and iterative refinement. Key contributions include the document–mention–entity model with cross-document relations, progressive visualization, and a web-based, extensible interface demonstrated through expert evaluation in the digital humanities. This approach enables flexible exploration, detailed inspection, and downstream analysis for historical and contemporary text collections, facilitating more reliable and interpretable entity analysis.

Abstract

We present an interactive visualization system for exploring named entities and their relationships across document collections. The system is designed around a graph-based representation that integrates three types of nodes: documents, entity mentions, and entities. Connections capture two key relationship types: (i) identical entities across contexts, and (ii) co-locations of mentions within documents. Multiple coordinated views enable users to examine entity occurrences, discover clusters of related mentions, and explore higher-level entity group relationships. To support flexible and iterative exploration, the interface offers fuzzy views with approximate connections, as well as tools for interactively editing the graph by adding or removing links, entities, and mentions, as well as editing entity terms. Additional interaction features include filtering, mini-map navigation, and export options to JSON or image formats for downstream analysis and reporting. This approach contributes to human-centered exploration of entity-rich text data by combining graph visualization, interactive refinement, and adaptable perspectives on relationships.

Figures (6)

• Figure 1: Data model diagrams for text-based named entity visualizations from displaCy Honnibal2020 (left), simple document-entity model (middle), and proposed document-mention-entity model (right).
• Figure 2: NERVIS visualization pipeline. Orange and gray nodes indicate interactable and uninteractable stages, respectively. Green nodes indicate intermediary data representations.
• Figure 3: A comparison of Document–Mention–Entity view (left) and Document–Entity view (right).
• Figure 4: Node (top) and edge (bottom) filtering algorithm flowchart.
• Figure 5: Visual structures used to depict different node types. The top row shows a document node. The middle row illustrates mention nodes (left) and entity nodes (right) across different entity classes. The bottom row depicts mention and entity nodes using an entity class-based color scheme. From left to right the individual node pictograms depict person, location, organization and miscellaneous entity classes.