Scientific publications.
-
2019 Ibrahim Dellal
Gestion et exploitation de larges bases de connaissances en présence de données incomplètes et incertaines. (Management and Exploitation of Large and Uncertain Knowledge Bases). University of Poitiers, France,
-
2019 Ibrahim Dellal, Stéphane Jean, Allel Hadjali, Brice Chardin, Mickaël Baron
Query answering over uncertain RDF knowledge bases: explain and obviate unsuccessful query results. Knowl. Inf. Syst. 61(3): 1633-1665 (2019)
-
2018 Ibrahim Dellal, Stéphane Jean, Allel Hadjali, Brice Chardin, Mickaël Baron
Traitement coopératif des requêtes RDF dans le contexte des bases de connaissances incertaines. Document Numérique 21(1-2): 9-35 (2018)
-
2017 Ibrahim Dellal, Stéphane Jean, Allel Hadjali, Brice Chardin, Mickaël Baron
On Addressing the Empty Answer Problem in Uncertain Knowledge Bases.
-
2017: 120-129 Ibrahim Dellal, Stéphane Jean, Allel Hadjali, Brice Chardin, Mickaël Baron
Traitement coopératif des requêtes RDF dans le contexte des bases de connaissances incertaines. INFORSID 2017: 277-292
Project: "QaRS4UKB - Improving Query Results for Uncertain Data".
https://github.com/lias-laboratory/qars4ukb
In today's world, our knowledge graph contains countless facts, and often, each fact has a confidence score telling you how sure we are about that information.
Example 1 (Finance): Imagine a prediction that says, "There's an 85% chance this stock will rise." Here, 85% is our confidence score.
Example 2 (Health): Or consider a medical test that suggests, "There's a 95% probability this patient might be allergic to penicillin." Here, 95% represents our confidence.
When someone interacts with our knowledge graph, they typically want answers that meet a certain confidence level. For example, they might only want facts that are 90% sure or more.
However, there might be times when the knowledge graph can't provide answers that meet the desired confidence, leading to no results. This can be perplexing and inconvenient for the user.
What does our project do?
Rather than merely saying "Sorry, no answers," our project assists users in two significant ways:
-
Understanding the Problem: We explain which aspects of their query made it difficult for us to locate a confident answer. These difficult parts are termed "αMinimal Failing Subqueries" or αMFSs for short.
-
Suggesting Alternatives: We also offer suggestions for similar queries where we can provide confident answers, known as "αMaximal Succeeding Subqueries" or αXSSs.
Technical Bits (for those interested):
We've formulated innovative methods (algorithms) to make this possible:
-
αLBA: This technique identifies the αMFSs and αXSSs for a query based on a specific confidence level (like the 95% in our health example).
-
NLBA, Bottom-Up, Top-Down, and Hybrid: These methods achieve a similar outcome, but they can handle a variety of confidence levels, not just a single one.