Skip to Main content Skip to Navigation
Theses

Annotation et recherche contextuelle des documents multimédias socio-personnels

Abstract : The overall objective of this thesis is to exploit a user centric means of representation, acquisition, enrichment and exploitation of multimedia document metadata. To achieve this goal, we proposed an annotation model, called SeMAT with a new vision of the snapshot context. We proposed the usage of external semantic resources (e.g. GeoNames ,, Wikipedia , etc.) to enrich the annotations automatically from the snapshot contextual elements. To accentuate the annotations semantic aspect, we modeled the concept of ‘social profile’ with Semantic web tools by focusing, in particular, on social relationships and a reasoning mechanism to infer a non-explicit social relationship. The proposed model, called SocialSphere is aimed to exploit a way to personalize the annotations to the viewer. Examples can be user’s objects (e.g. home, work) or user’s social dimensions (e.g. my mother, my husband's cousin). In this context, we proposed an algorithm, called SQO to suggest social dimensions describing actors in multimedia documents according to the viewer’s social profile. For suggesting event annotations, we have reused user experience and the experience of the users in his social network by producing association rules. In the last part, we addressed the problem of pattern matching between query and social graph. We proposed to steer the problem of pattern matching to a sub-graph isomorphism problem. We proposed an algorithm, called h-Pruning, for partial sub-graph isomorphism to ensure a close matching between nodes of the two graphs: motive (representing the request) and the social one. For implementation, we realized a prototype having two components: mobile and web. The mobile component aims to capture the snapshot contextual elements. As for the web component, it is dedicated to the assistance of the user during his socio-personnel multimedia document annotation or socio-personnel multimedia document consultation. The evaluation have proven: (i) the effectiveness of our exploitation of social graph approach in terms of execution time, (ii) the effectiveness of our event suggestion approach (we proved our hypothesis by demonstrating the existence of co-occurrence between the spatio-temporal context and events), (iii) the effectiveness of our social dimension suggestion approach in terms of execution time.
Document type :
Theses
Complete list of metadatas

Cited literature [141 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00668689
Contributor : Abes Star :  Contact
Submitted on : Friday, February 10, 2012 - 10:57:19 AM
Last modification on : Wednesday, July 8, 2020 - 12:42:08 PM
Long-term archiving on: : Friday, May 11, 2012 - 2:35:14 AM

File

these.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00668689, version 1

Citation

Sonia Lajmi. Annotation et recherche contextuelle des documents multimédias socio-personnels. Autre [cs.OH]. INSA de Lyon, 2011. Français. ⟨NNT : 2011ISAL0025⟩. ⟨tel-00668689⟩

Share

Metrics

Record views

863

Files downloads

3397