Probabilistic relational models learning from graph databases

Marwa El Abri 1
1 DUKe - Data User Knowledge
LINA - Laboratoire d'Informatique de Nantes Atlantique : UMR 6241
Abstract : Historically, Probabilistic Graphical Models (PGMs) are a solution for learning from uncertain and flat data, also called propositional data or attributevalue representations. In the early 2000s, great interest was addressed to the processing of relational data which includes a large number of objects participating in different relations. Probabilistic Relational Models (PRMs) present an extension of PGMs to the relational context. With the rise of the internet, numerous technological innovations and web applications are driving the dramatic increase of various and complex data. Consequently, Big Data has emerged. Several types of data stores have been created to manage this new data, including the graph databases. Recently there has been an increasing interest in graph databases to model objects and interactions. However, all PRMs structure learning use wellstructured data that are stored in relational databases. Graph databases are unstructured and schema-free data stores. Edges between nodes can have various signatures. Since, relationships that do not correspond to an ER model could be depicted in the database instance. These relationships are considered as exceptions. In this thesis, we are interested by this type of data stores. Also, we study two kinds of PRMs namely, Direct Acyclic Probabilistic Entity Relationship (DAPER) and Markov Logic Networks (MLNs). We propose two significant contributions. First, an approach to learn DAPERs from partially structured graph databases. A second approach consists to benefit from first-order logic to learn DAPERs using MLN framework to take into account the exceptions that are dropped during DAPER learning. We are conducting experimental studies to compare our proposed methods with existing approaches.
Document type :
Theses
Complete list of metadatas

Cited literature [104 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01901255
Contributor : Marwa El Abri <>
Submitted on : Monday, October 22, 2018 - 6:34:45 PM
Last modification on : Monday, November 5, 2018 - 9:40:28 AM
Long-term archiving on : Wednesday, January 23, 2019 - 6:00:26 PM

File

2018-Thèse-MarwaElAbri.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01901255, version 1

Collections

Citation

Marwa El Abri. Probabilistic relational models learning from graph databases. Computer Science [cs]. Université de Nantes, 2018. English. ⟨tel-01901255⟩

Share

Metrics

Record views

112

Files downloads

490