Caractéristiques
- Types de propriété intellectuelle : Logiciel
- Stade de développement : TRL4 - Validation de la technologie en laboratoire
-
Secteurs d'applications :
Numérique - Réseaux - Télécoms - Systèmes
-
Domaines scientifiques :
SCIENCES ET TECHNOLOGIES DE L'INFORMATION ET DE LA COMMUNICATION
- Mots-clés : AI ; Search engine ; Optimization
Description
Most search engines use one optimized search method (core) with fluctuating performance when requests greatly vary. Search engines can leverage user information to compensate, however, this is not always possible (eg due to GDPR constraints).
• MCS improves document search accuracy by selecting in real time the best search method for each users’ query
• Search results are up to 20% more relevant according to the desired optimization (eg users’ satisfaction, conversion rate…)
• Benchmarked against 5 international reference collections (example below)
• Maximum performance boosts when user queries greatly vary
• Limited performance impact in production
• Do not leverage any user behavior data
Spécifications techniques
|
ERR-IA@20 |
α-nDCG@20 |
NRBP |
|
|
Oracle |
.80 |
.77 |
.81 |
|
MCS |
.77 |
.71 |
.75 |
|
BM25 |
.35 |
.42 |
.31 |
|
SQE |
.40 |
.38 |
.39 |
Avantages concurrentiels
• Search accuracy up to 20%
• Best perf. when users’ queries greatly vary
• Do not rely on users’ personal data
Champs d'application
• Search engines
• Information retrieval systems without personal data
• Recommandation systems
