Searching the Web of Things

Searching the ‘Web-of-Things’ Revealing ambient intelligence using the Semantic Web Benoit Christophe – benoit.christophe@alcatel-lucent.com Bell Labs Research – Alcatel-Lucent Bell Labs France

Overview The ‘Web of Things’ Why do we study it? How do we define and implement it? And… for which usage? Prototype developed Searching the ‘Web of Things’ Facts and questions raised Search strategy: Modeling, Interlinking, Reasoning First implementations and results Conclusion and future works

The ‘Web of Things’ Proliferation of devices [1] Ericsson.com, white paper. More than 50 billions connected devices, 2011

The ‘Web of Things’ At the convergence of (at least) 3 facts… Internet of Things Identification (EPC, RFID) Network of sensors The Web Openness, simple, application agnostic Contributor-based (Web 2.0 ecosystem) Pervasive computing Put computing capabilities everywhere

The ‘Web of Things’ Our approach 1/2 A smart space Federating smart spaces Interconnecting smart spaces [2,3]

A real-world object (RWO) ‘ smart’ home A virtual object (VO) The ‘Web of Things’ Our approach 2/2 ‘ Virtualizing’ connected objects of each smart space [2,3] REST API

The ‘Web of Things’ For which usages? + = New web applications Novel interactions

The ‘Web of Things’ Prototype developed ‘ smart’ home ‘ smart’ office

Searching the ‘Web of Things’ In a near future… Billions of connected objects None of them sharing common data model Accessed by anybody @ anytime… (mobile subscription keep on rising [4])

Searching the ‘Web of Things’ Three axes to investigate Model establishment Handling object specificities… and human perception as well Allowing reasoning Web based! Data models cross understanding SAT-based algorithms or using machine learning Search strategy development Understanding (predicting) the context a search is performed Then triggering the most appropriate algorithms (accuracy vs. fastness)

Searching the ‘Web of Things’ What it may allow Ex: emulating a multifunction copier through a composition of connected objects Model: Some models are shared (i.e. objects have states, functionalities, etc…) While some other are not (i.e. domain based vocabularies to represent structures) F(AnythingVisual) = Picture G(VirtualDocument) = PhysicalSheet H(Sheet) = Sheet “ Document” is a type of “AnythingVisual”; “ Sheet” is a type of “Document”; “ Picture” is a type of “VirtualDocument” “ Sheet” is defined by “PhysicalSheet” F(Sheet) = Picture G(Picture) = Sheet Map: Deduce: H = G o F + =

Searching the ‘Web of Things’ Establishing models - design Connected objects have a physical existence Live & die, become unavailable (idle or for the exclusive use of someone) May move across different smart spaces Are geo-localized May put requirement on the user relative positioning (e.g. sheet of paper shown to the webcam lens) Are owned by someone, shared between people Smart spaces Are reconfigurable (e.g. adding new objects) Can contain other spaces (e.g. a coffee shop in a mall) People Acquire, possess, abandon objects Have a social life (friends, family, etc.) evolving over time

Searching the ‘Web of Things’ Establishing models - design Object description coverage Functionalities offered Functional behavior (state-based machine) Functional requirements (e.g. involves user action, etc.) Ownership and access rights through common definition of people Indoor location description Data produced In terms of semantic web technologies Use of a customized OWL-S process (covering three first points) Use of FOAF (fourth point) Indoor geo-location ontology with links to GeoNames (fifth point) Interlinking all above models to define the whole VO description

Searching the ‘Web of Things’ Establishing Models – overall picture [5] vo-fsm represents finite state machine of a Web-enabled object vo-structures allows object to tell how are structured the data it uses or generates vo-capability allows to map a task with a set of functionalities vo-location represents geographical areas and how they are relatively localized vo-core interlinks all different models and form the description model of Web-enabled objects Description file of a connected object instantiates ‘vo-core’ model

Searching the ‘Web of Things’ Cross data models realization Describing structures of connected objects using the same vocabulary is impossible: It requires agreements with all device manufacturers it requires data models to be changed to comply with such agreed model Giving the opportunity to let providers use their own terms is the key…

Searching the ‘Web of Things’ Cross data models realization BUT! This approach has drawbacks How to know if some objects can be composed based on their I/Os? How to know that objects can be used in Web applications based on their I/Os?

Searching the ‘Web of Things’ Cross data models realization This is a problem of similarity establishment

Searching the ‘Web of Things’ Machine learning approach A domain ontology seen as a complete universe [6] data model ( ontology ) A class and its members

Searching the ‘Web of Things’ Machine learning approach Separate universe around a class (e.g. C i ) [6] data model ( ontology ) A class and its members All other model elements

Searching the ‘Web of Things’ Machine learning approach Create representations from groups [6]

Searching the ‘Web of Things’ Machine learning approach Representations can take multiple forms but: Are the cornerstone to find similarities Some examples can be: Represent a class using its elements name Represent a class using its full ontological description (A has some properties, A hasComplementClass B, A hasValue only { α , β , γ }) Custom class description by adding knowledge (facts, rules) based on the domain involved

Searching the ‘Web of Things’ Machine learning approach Once represented, the universe can be trained by a classifier Classify The same way you represent A , then represent B (a concept belonging to another domain) Test with the classifier if B seems to belong to A or Ā Finally do the reverse operation: create Train a classifier with U 2 then test if A belongs to Obtain the joint distribution of A and B: Once joint probability distributions obtained, evaluate the distance between A and B For instance, by using “Jaccard” distance:

Searching the ‘Web of Things’ Data models intertwining prototype

Searching the ‘Web of Things’ Designing search strategies Adapt (predict) search strategy to the requester Humans privilege quick answer and usually comply with approximation Machines can wait but need accurate (i.e. exact) results Based on established ‘context of search’, trigger algorithms: Using probabilistic models for fastness Using graph traversal process for accuracy

Searching the ‘Web of Things’ Predicting type of search

Searching Web-enabled objects Using semantic profiles for searching objects Case of a request coming from an application looking for an object matching accurately some requirements From the incoming discovery request: Extract requirements of the request Translate each requirement to a graph

Searching Web-enabled objects Using semantic profiles for searching objects From a set of Web-enabled objects that belong to a given smart space: Access semantic representation of object’ functionalities Navigate to the ontological definition of their I/Os Extract the graph of each I/O Send stream functionality

Searching Web-enabled objects Do graph comparisons Graph analyzer module Lookup set of object graphs (Who has ‘A’, Who has ‘B’, etc.) Compute matching score Ex: Webcam has ‘B’ and ‘G’ while copier has ‘D’ and ‘I’ matching( α , camera ) = 6/11 matching( α , copier ) = 4/11 2: Lookup graphs 1: load requirement graph Results = {(camera, 55%); (copier,36%)} 3: return classified results Finally, match request requirements with structures uses or generated by an object

Searching the ‘Web of Things’ Conclusion & Remaining works We hope that the Web of Things be able to benefit from the Semantic Web Remaining works: Models: To continue to create links with other ontologies: DBPedia, Dolce, etc. And to check these models deal with OWA issues Cross data models understanding Developing more representations of ontological concepts Evaluating SAT-based algorithms instead of Machine Learning Search strategy: Clustering smart spaces Implementing context of search prediction And obviously… Tools to generate descriptions of connected objects in order to really test and validate our approach (currently, only some tools allowing creation of semantic profiles have been done)

References [1] Ericsson.com, white paper. More than 50 billions connected devices, 2011 [2] B. Christophe et al. The Web of Things vision: Things as a service and interaction patterns . Bell Labs Technical Journal, 16(1):55-62, 2011 [3] M.Boussard et al. Providing user support in web-of-things enabled smart spaces . Proceedings of the 2 nd International Workshop on the Web of Things, 2011 [4] Monitoring the WSIS targets, World Telecommunicaiton/ICT development report, 2010 [5] vo-* models, http://bring-models-to-life.appspot.com/wot/models/index.html [6] A. Doan et al. Ontology Matching: A machine learning approach . Handbook on Ontologies in Information Systems, 2003

Thanks Benoit Christophe Bell Labs Research Alcatel-Lucent Bell Labs France [email_address]

Searching the Web of Things

More Related Content

Similar to Searching the Web of Things (20)

Recently uploaded (20)

Searching the Web of Things

Editor's Notes