A Framework for Media Adaptation Using the Web and the ...€¦ · Web 2 S ebastien Laborie and Antoine Zimmermann Media Adaptation using the Web and the Semantic Web. Introduction

A Framework for Media AdaptationUsing the Web and the Semantic Web

Sebastien Laborie and Antoine Zimmermann

SMAP’07

1 Sebastien Laborie and Antoine Zimmermann Media Adaptation using the Web and the Semantic Web

Introduction Motivations

Information Anytime Anywhere

Mona Lisa Image

PNG format

560 x 864

Web




Mona Lisa Image

PNG format

560 x 864

Web




Mona Lisa Image

PNG format

560 x 864

Web

Profile

Format = {JPG,BMP,3GP}Max resolution = 400 x 600




Mona Lisa Image

PNG format

560 x 864

Web

Mona Lisa Image

JPG format

386 x 600




Movie trailer

AVI format

Web

No VideosMovie poster

No VideosNo Images

Movie synopsis




Movie trailer

AVI format

Web

No Videos

Movie poster

No VideosNo Images

Movie synopsis




Movie trailer

AVI format

Web

No Videos

Movie poster

No VideosNo Images

Movie synopsis




Movie trailer

AVI format

Web

No Videos

Movie poster

No VideosNo Images

Movie synopsis




Movie trailer

AVI format

Web

No Videos

Movie poster

No VideosNo Images

Movie synopsis


Introduction Our contribution

Our adaptation approach

Approach Solutions

Find alternatives on the web.

Use information retrievaltechniques.

They must convey the samemeaning (or at least be as muchsimilar as the original).

Identify proximity betweenmetadata.

Select the ones which satisfy aprofile.

Use filtering techniques.

Automatize this process!




Approach Solutions

Find alternatives on the web. Use information retrievaltechniques.









Approach Solutions










Approach Solutions










Approach Solutions








Introduction Contribution

Outline

1 Software Architecture

2 Simulating with current technologies

3 Improving the process with Semantic Web technologies


Software Architecture


World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

a b c d

Adaptation component

d ′out

mi ,{d0,...,dj}

m

m1

m2

d1 height: “864”width: “560”format: PNG

d2 title: “Mona Lisa”author: “Da Vinci”

d3 date: “17/12/2007”title: “Mona Lisa”format: PNG

d4 height: “600”width: “386”format: JPG





World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2









World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2






a©: Description association




World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2






mi ,d′i

m

m1

m2

d′height: “864”width: “560”format: PNGtitle: “Mona Lisa”author: “Da Vinci”

d′1 date: “17/12/2007”

title: “Mona Lisa”format: PNG

d′2 height: “600”

width: “386”format: JPGtitle: “Mona Lisa”author: “Da Vinci”

b©: Description aggregation




World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2






mi ,d′i

m

m1

m2


d′1 date: “17/12/2007”

title: “Mona Lisa”format: PNG



Sim(d′k ,d′

l )

0.76

0.31

0.9

c©: Description similarity




World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

profile

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2






mi ,d′i

m

m2




Sim(d′k ,d′

l )

0.9

d©: Description selection




World Wide Web

Descriptions

d3

d4

d5

d1

d2

Media items

m1m2

m3

m

profile

a b c d


d ′out

mi ,{d0,...,dj}

m

m1

m2






mi ,d′i

m

m2




Sim(d′k ,d′

l )

0.9


Simulating with current technologies


http://ex.com/mona.png

<div>

<img

src="http://ex.com/mona.png"

alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>

http://ex.com/mona.png,

“La Gioconda Mona Lisa”

a©b©c© Similarities: String comparison

Vector comparison. . .

d© Filtering / PreferencesWeb search engines partially implement a©, b©, c© and d©





<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>










<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>










<div>

<img


alt="La Gioconda"/>

...

</div>

http://ex.com/mona.png, {“La Gioconda”}

<div>

<img


alt="Mona Lisa"/>

...

</div>



a©

b©c© Similarities: String comparison







<div>

<img


alt="La Gioconda"/>

...

</div>

http://ex.com/mona.png, {“La Gioconda”}

<div>

<img


alt="Mona Lisa"/>

...

</div>



a©








<div>

<img


alt="La Gioconda"/>

...

</div>

http://ex.com/mona.png, {“La Gioconda”,

<div>

<img


alt="Mona Lisa"/>

...

</div>

“Mona Lisa”}



a©








<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>



a©

b©

c© Similarities: String comparison







<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>



a©b©

c© Similarities: String comparison







<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>





d© Filtering / Preferences

Web search engines partially implement a©, b©, c© and d©





<div>

<img


alt="La Gioconda"/>

...

</div>

<div>

<img


alt="Mona Lisa"/>

...

</div>





d© Filtering / Preferences

Web search engines partially implement a©, b©, c© and d©



Improvement

Using current techniques is not enough.

=⇒ Semantic Web technologies can improve this. . .

. . . by adding expressivity:

e.g., “Da Vinci is a Painter”.

. . . by providing deductive capabilities:

e.g., “Mona Lisa was painted by Da Vinci” implies that “Mona Lisa is aPainting”.

. . . by solving heterogeneity problems:

e.g., “La Gioconda same as Mona Lisa”.

This does not change our software architecture!



Improvement

Using current techniques is not enough.

=⇒ Semantic Web technologies can improve this. . .

. . . by adding expressivity:

e.g., “Da Vinci is a Painter”.

. . . by providing deductive capabilities:

e.g., “Mona Lisa was painted by Da Vinci” implies that “Mona Lisa is aPainting”.

. . . by solving heterogeneity problems:

e.g., “La Gioconda same as Mona Lisa”.

This does not change our software architecture!


Improving the process with Semantic Web technologies

Improvement

O1

Painting

Portrait

Abstract

Landscape

Painting ≡∃ hasPainter

hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3d′2

type: Cartoonauthor: “Sato”format: JPG

d4hasTitle: “La Gioconda”represents: ?X?X type Womantype: Portrait

d5

title: “Mona Lisa”author: “Da Vinci”type: MasterPiece

d′3hasTitle: “La Gioconda”

represents: ?X?X type Womantype: Portraittitle: “Mona Lisa”author: “Da Vinci”type: MasterPiece


d2

hasTitle: “Mona Lisa”hasPainter: “Da Vinci”

|= type: Painting


width: “560”format: PNGhasTitle: “Mona Lisa”hasPainter: “Da Vinci”

a©Indexing

b©RDF Merge

c©Similarity

?

?

?

c©Semantic similarity

based on

ontology matching

0.8

0.2

d©Semantic filtering

(e.g., only masterpieces)

0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3d′2



d5





d2


|= type: Painting



a©Indexing RDF

(+OWL ontologies)

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3d′2



d5





d2


|= type: Painting



a©Indexing RDF

(+OWL ontologies)

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2



d5





d2


|= type: Painting



a©Indexing RDF

(+OWL ontologies)

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2



d5





d2


|= type: Painting



a©Indexing RDF

(+OWL ontologies)

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2type: Cartoon

author: “Sato”format: JPG


d5





d2


|= type: Painting



a©Indexing

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2type: Cartoon



d5





d2


|= type: Painting



a©Indexing

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2type: Cartoon



d5





d2


|= type: Painting



a©Indexing

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8



Improvement

O1

Painting

Portrait

Abstract

Landscape


hasTitle

O2

ArtWork

Drawing

MasterPiece

Movie

Cartoon⊥

author

title

v

v

≡

d3

d′2type: Cartoon



d5





d2


|= type: Painting



a©Indexing

b©RDF Merge

c©Similarity

?

?

?


based on

ontology matching

0.8

0.2



0.8


Conclusion

Limitations

§ Limited efficiency for:

private media items;multimedia document.

§ The semantic implementation needs further research:

automatic semantic annotations;merging semantic annotations;semantic similarity.

§ Not implemented yet


Conclusion

Summary

© It is practically implementable;

© The approach is flexible:

no need to implement it for each new format;it encompasses cross-media adaptation.

© It is potentially an efficient solution:

fast (everything is pre-computed);use information profusion on the Web.


Questions

Questions ?

Thank you for your attention !

[email protected]

[email protected]


Documents

A Framework for Media Adaptation Using the Web and the ...€¦ · Web 2 S ebastien Laborie and Antoine Zimmermann Media Adaptation using the Web and the Semantic Web. Introduction