The Missing Path: Analysing Incompleteness in Knowledge Graphs
Authors:
Marie Destandau,
Jean-Daniel Fekete
Abstract:
Knowledge Graphs (KG) allow to merge and connect heterogeneous data despite their differences; they are incomplete by design. Yet, KG data producers need to ensure the best level of completeness, as far as possible. The difficulty is that they have no means to distinguish cases where incomplete entities could and should be fixed. We present a new visualisation tool: The Missing Path, to support th…
▽ More
Knowledge Graphs (KG) allow to merge and connect heterogeneous data despite their differences; they are incomplete by design. Yet, KG data producers need to ensure the best level of completeness, as far as possible. The difficulty is that they have no means to distinguish cases where incomplete entities could and should be fixed. We present a new visualisation tool: The Missing Path, to support them in identifying coherent subsets of entities that can be repaired. It relies on a map, grouping entities according to their incomplete profile. The map is coordinated with histograms and stacked charts to support interactive exploration and analysis; the summary of a subset can be compared with the one of the full collection to reveal its distinctive features. We conduct an iterative design process and evaluation with 9 Wikidata contributors. Participants gain insights and find various strategies to identify coherent subsets to be fixed.
△ Less
Submitted 13 January, 2021; v1 submitted 16 May, 2020;
originally announced May 2020.
Path Outlines: Browsing Path-Based Summaries of Knowledge Graphs
Authors:
Marie Destandau,
Olivier Corby,
Jean-Daniel Fekete,
Alain Giboin
Abstract:
Knowledge Graphs have become a ubiquitous technology powering search engines, recommender systems, connected objects, corporate knowledge management and Open Data. They rely on small units of information named triples that can be combined to form higher level statements across datasets following information needs. But data producers face a problem: reconstituting chains of triples has a high cogni…
▽ More
Knowledge Graphs have become a ubiquitous technology powering search engines, recommender systems, connected objects, corporate knowledge management and Open Data. They rely on small units of information named triples that can be combined to form higher level statements across datasets following information needs. But data producers face a problem: reconstituting chains of triples has a high cognitive cost, which hinders them from gaining meaningful overviews of their own datasets. We introduce path outlines: conceptual objects characterizing sequences of triples with descriptive statistics. We interview 11 data producers to evaluate their interest. We present Path Outlines, a tool to browse path-based summaries, based on coordinated views with 2 novel visualisations. We compare Path Outlines with the current baseline technique in an experiment with 36 participants. We show that it is 3 times faster, leads to better task completion, less errors, that participants prefer it, and find tasks easier with it.
△ Less
Submitted 8 October, 2020; v1 submitted 23 February, 2020;
originally announced February 2020.
