XU Wenyi

ongoing PhD project
Team : LFI
https://lip6.fr/Wenyi.Xu

Supervision : Christophe MARSALA

Co-supervision : CHRISTOPHE Benoit

Modeling and exploiting the knowledge base of Web of Things

The concept Web of things (WOT) is gradually becoming a reality as the result of development of network and hardware technologies. Nowadays, there is an increasing number of objects that can be used in predesigned applications. The world is thus more tightly connected, various objects can share their information as well as being triggered through a Web-like structure. However, even if the heterogeneous objects have the ability to be connected to the Web, they cannot be used in different applications unless there is a common model so that their heterogeneity can be described and understood.
In this thesis, we want to provide a common model to describe those heterogeneous objects and use them to solve user’s problems. Users can have various requests, either to find a particular object, or to fulfill some tasks. We highlight thus two research directions. The first step is to model those heterogeneous objects and related concepts in WOT, and the next step is to use this model to fulfill user’s requests.
Thus, we first study the existing technologies, applications and domains where the WOT can be applied. We compare the existing description models in this domain and find their insufficiency to be applied in the WOT. We then propose a new semantic model for WOT using ontology-based approach which composes three main components. This model can describe both static information and the dynamic changes of WOT.
To enable searching objects within our proposed model, we study different similarity measures in order to propose objects to users when there exists no exact match. We further propose a hybrid similarity measure that considers the semantic, the feature as well as the instance property values in matching two objects. This measure also enables user to assign different weights to adjust their context needs.
Furthermore, we propose a personalized recommendation approach. This approach assumes that there exists at least one object that can fulfill user’s request alone. We first design a questionnaire and analyze different user’s needs in searching connected objects. We then propose a fuzzy approach that takes user profile, their location information, the similarity results as well as their expectations in recommending objects.
Later, to overcome the fact that there may not always exist a single object that can fulfill user’s request, we further propose automatic composition methods to search for a chain of composable objects at run time to fulfill user’s request. We apply our proposed approaches in several experiments to evaluate its performance and the results shows our methods have several advantages: efficient, flexible, adaptable to different contexts, etc. In the end, we conclude our research and list some perspectives and future work.

Defence : 01/16/2015

Jury members :

Jacky Montmain,Professeur, Ecole des Mines d'Alès [Rapporteur]
Marek Reformat,Professeur, University of Alberta [Rapporteur]
Bernd Amann,Professeur,LIP6 - UPMC
Matthieu Boussard, Industriel, ALBLF
Benoit Christophe, Industriel, ALBLF
Christophe Marsala,Professeur, LIP6 - UPMC
Nicolas Sabouret,Professeur, LIMSI-CNRS

Departure date : 01/31/2015

2013-2016 Publications