Source-linked AI summary
Exploration and Visualization in the Web of Big Linked Data: A Survey of the State of the Art
Nikos Bikakis, Timos Sellis
TL;DR
Very large datasets make exploration and visualization a major challenge, with scalability a vital requirement. This survey reviews system requirements, approaches from database and information visualization, and Web of Data systems; it finds that many Web of Data systems do not address scalability and remain suited to small datasets.
Problem
Large contemporary datasets make manual exploration difficult, while many Web of Data approaches fail to address performance and scalability.
Source: §1 · p. 1Method
The survey describes modern systems’ requirements and challenges, reviews approaches from database and information visualization, and assesses Web of Data systems against those requirements.
Source: §1 · p. 1Results
Most Web of Data systems do not handle performance and scalability, assume all data objects can be displayed, or load whole graphs into memory, restricting them to small datasets or graphs.
Source: §4 · p. 5Takeaways & Limitations
The survey calls for scalability and performance to be vital requirements, including approximation techniques and runtime data retrieval for future Web of Data systems.
Source: §4 · p. 5
Abstract
from arXiv · showhide
Data exploration and visualization systems are of great importance in the Big Data era. Exploring and visualizing very large datasets has become a major research challenge, of which scalability is a vital requirement. In this survey, we describe the major prerequisites and challenges that should be addressed by the modern exploration and visualization systems. Considering these challenges, we present how state-of-the-art approaches from the Database and Information Visualization communities attempt to handle them. Finally, we survey the systems developed by Semantic Web community in the context of the Web of Linked Data, and discuss to which extent these satisfy the contemporary requirements.
1. INTRODUCTION
Data exploration and visualization helps people inspect data, identify patterns, and support sense-making, but contemporary dataset scale makes manual analysis difficult. The paper outlines modern systems’ requirements and challenges, reviews approaches from database and information visualization, and assesses Web of Data systems against those requirements.
1. INTRODUCTION
- Visualization supports data exploration by helping users identify patterns, infer correlations and causalities, and make sense of information.Source: §1 · p. 1
- Large contemporary datasets make exploration and visualization a major research challenge, and scalability is a vital requirement involving storage, access, presentation, and interaction.Most existing Web of Data approaches also fail to address performance and scalability.Source: §1 · p. 1
- The survey describes modern systems’ requirements and challenges, reviews database and information visualization approaches, and assesses Web of Data systems against contemporary requirements.Source: §1 · p. 1
2. CHALLENGES
Modern exploration systems must handle large, dynamic datasets during exploratory work while accommodating diverse tasks and users. This requires scalable data handling and presentation, as well as ways to customize and assist exploration.
Large & Dynamic Data in Exploration-driven Setting
- Traditional systems typically access static, preprocessed data and focus on small datasets that conventional techniques can handle.Source: §2 · p. 1
- Systems must handle billion-object dynamic datasets throughout exploration, including storage, access, presentation, interaction, and operation on machines with limited resources.Online endpoints and streaming data can prevent preprocessing; conventional systems designed for small, preprocessed datasets cannot be adopted for modern settings involving large portions of billion-object datasets.Source: §2 · pp. 1–2
Visual presentation and interaction
- Presenting large datasets risks information overload, so systems need summaries and abstractions that support overview, zoom, filtering, and details on demand.Source: §2 · p. 2
Variety of Tasks & Users
- Scalable exploration must accommodate different user preferences and task requirements, including choices about organization, approximation, and data operations.Systems may also adjust parameters to environmental settings and assist users through visualization recommendations, interest discovery, or explanations of trends and anomalies.Source: §2 · p. 2
3. EXPLORATION & VISUALIZATION SYSTEMS
The survey organizes Web of Data visualization and exploration systems into six categories, spanning browsers, general-purpose frameworks, specialized applications, graph and ontology tools, and libraries. It compares their capabilities and describes representative approaches within each category.
3. EXPLORATION & VISUALIZATION SYSTEMS
- The survey classifies Web of Data systems as browsers and exploratory systems, generic visualization systems, domain-specific tools, graph-based tools, ontology visualizers, and visualization libraries.Source: §3 · p. 2
3.1 Browsers & Exploratory Systems
- Web of Data browsers support resource navigation and representation, while exploratory tools add search, facets, queries, and multiple views.Examples include Tabulator’s maps and timelines and Explorator’s combination of search and facets.Source: §3.1 · pp. 2–3
3.2 Generic Visualization Systems
- Generic visualization systems offer varied data views and operations; some recommend visualizations, while SynopsViz uses hierarchical aggregation and interaction-driven hierarchy construction.The survey compares systems across data types, visualization types, recommendations, preferences, statistics, sampling, aggregation, incremental computation, and runtime disk use.Source: §3.2 · pp. 3–4
3.3 Domain, Vocabulary & Device-specific Visualization Systems
- Domain-, vocabulary-, and device-specific systems address geospatial and time-evolving data, statistical data cubes, and mobile exploration.Examples include faceted geospatial tools, Data Cube chart and map interfaces, and mobile systems that account for small displays and limited resources.Source: §3.3 · p. 4
3.4 Graph-based Visualization Systems
- Graph-based systems support tasks such as relationship discovery, linked-resource browsing, graph summaries, and interactive navigation of RDF graphs.GraphVizdb uses spatial and database techniques for interactive visualization over very large RDF graphs.Source: §3.4 · p. 4
3.5 Ontology Visualization Systems
- Ontology visualizations commonly use node-link layouts, while other approaches use geometric containment or combine representations.CropCircles uses concentric circles for class hierarchies; Knoocks uses nested blocks, and OntoTrix and NodeTrix combine node-link and adjacency-matrix representations.Source: §3.5 · p. 4
3.6 Visualization Libraries
- Visualization libraries let Web pages embed visualizations of Web of Data, including SPARQL results and query outputs.Sgvizler wraps SPARQL results for visualization, while Visualbox supports query building and provides visualization templates.Source: §3.6 · pp. 4–5
4. DISCUSSION
The discussion finds that Web of Data exploration systems generally remain limited in performance and scalability, despite supporting varied data and visualization types. It identifies approximation, memory management, and more sophisticated graph exploration as priorities for future systems.
4. DISCUSSION
- Most Web of Data exploration systems do not address performance and scalability, relying instead on traditional techniques suited to small datasets.Source: §4 · p. 5
- Generic systems support several data and visualization types, while recent systems often recommend visualization techniques based on input-data type.Source: §4 · p. 5
- Only SynopsViz and VizBoard among the surveyed systems adopt visual approximation, leaving most to assume all data objects fit on screen.Source: §4 · p. 5
- Most systems load examined objects into main memory at runtime; SynopsViz instead retrieves data and generates visualizations incrementally from user interaction.This means only part of the dataset needs to be in memory at a time.Source: §4 · p. 5
- Graph-based systems often load the whole graph into main memory, and graph-layout memory demands restrict them to small graphs.Source: §4 · p. 5
- For large graphs, modern WoD systems should consider information-visualization techniques: hierarchical aggregation recursively decomposes graphs into smaller subgraphs forming abstraction layers; edge bundling aggregates edges, and disk-based implementations are another option [22].Source: §4 · p. 5
- Future Web of Data systems should adapt sampling and aggregation, retrieve data dynamically from disk, and consider caching and prefetching.Source: §4 · p. 5
- Beyond visualization recommendations, systems should capture user preferences and assist users throughout large-data exploration and analysis.Source: §4 · p. 6