SV: a visualization mechanism for ontologies of records based on SVG graphics

Abstract

This paper describes SV, a visualization mechanism used to explore digital collections represented as hierarchical structures called ontologies of records. These ontologies are XML les constructed using OAI-PMH records and a clustering algorithm. SV is composed by a web interface and SVG graphics. Through the interface, users can recognize the organization of the collection and access to metadata of documents.

Introduction

called semantic view (SV). The name also refers to the rst two letters of \Support Vector Graphics". SV o ers an interactive view to allow users to explore the content of a federated collection.

The paper is organized as follows. Section 2 describes the features of an ontology of records. Section 3 includes related work. Section 4 and 5 explains the design and implementation of SV, respectively. Experimental results are described in Section 6. Finally, Section 7 includes conclusions and suggests future directions of our work. 2

What is an ontology of records

An ontology of records is a hierarchical structure of clusters of OAI-PMH records that provides an unambiguous interpretation of its elements. Its construction is based on the Frequent Itemset Hierarchical Clustering algorithm [ 8 ]. This structure organizes a collection of documents, this has concept-term relationships useful for keyword based searches. An ontology of records is stored as a well formed XML le that is validated against an XML Schema. An ontology of records has the following features[ 9 ]: 1. Documents are clustered by similarity 2. Clusters in the k -level have labels of k -terms 3. All the records of a cluster share the terms of its label 3

Related work

This section describes some systems that have been used to visualize collections of documents. Proat et al. [ 4 ] use 3D trees to visualize documents organized according to the Library of Congress Classi cation (LCC). Documents are clustered in seven subsets. The interface has controls to rotate or zoom the nodes of trees. The leaf nodes contain metadata of documents.

Geroimenko et al. [ 5 ] have proposed the Generalizad Document Object Model tree Interface (G-DOM-Tree interface) to visualize metadata from XML DOM (Document Object Model) documents. The model displays a hierarchy of labels, this is very similar to the visualization that browsers o er of XML Schema. The interface is implemented as a Java applet or a Flash lm.

Fluit et al. [ 6 ] describe Spectacle, this mechanism uses lightweight ontologies to represent classes of similar objects and their relationships. The navigation can be done by using hypertext or \cluster maps". A cluster map visualizes the objects and their classes.

At last, Sanchez et al. [ 7 ] use a star eld grid to visualize documents from several collections. Documents are stored as OAI-PMH records. The axis of the grid represent attributes of the collections that can be chosen by users. Small polygons are associated with the type of document and di erent colors are used to distinguish the collections.

Desing of SV

The design of SV is addressed to reach the following objectives: { Construct a visualization mechanism with semantic features that allow users to explore a collection of documents { Represent the organization of a collection of documents { Retrieve the metadata and the content of a determined document

In order to reach these objectives, we have used the levels of knowledge proposed by [ 2 ] in the design of SV. We want to uses CORTUPP as a test bed, this is a collection represented as an ontology of records5. 1. Level 1: Organization of the metadata. Metadata is organized in the ontology of records. Content information is stored in dc:title, dc:subject and dc:description elements. 2. Level 2: Organization of the information in the documents. Technical reports have a common structure formed by six mandatory chapters: 1)research propose, 2)state of the art, 3)research design, 4)implementation, 5)results and 6)conclusions. This structure is de ned in a Latex template. The BibTex le format is used to manage the bibliography. A technical report is described as a @techreport entry. 3. Level 3: Organization of the information in databases. The technical reports are stored as PDF les in a database that also includes data and counts of users. Documents are accessible through a web interface. 4. Level 4: Organization of the topics treated in the documents. The dc:subject element stores the topic of a document. Keywords of this element belong to the labels of the clusters in the ontology of records. 5. Level 5: Organization of the concepts, terms and relations. This level is also represented in the ontology of records. 5

Implementation of SV

SV is formed by a web interface and SVG graphics6. SVG is a format developed and maintained by the W3C SVG Working Group. This is an XML application used to describe animated or static two dimensional vectorial graphics. The main feature of these graphics is scalability.

SV uses Xerces, this is a Java parser used to extract data from an ontology of records. The classes of SV are built using Java language. In the interface, each document, that is, an OAI-PMH record, is represented with a yellow star in a blue gradient background. The background is divided in ve parts that correspond to the rst levels of the ontology. These levels are divided by lines that form angles of 90 degrees. The distribution of the lines try to re ect an estimation of the amount of documents that can be found in each level. The documents closer to 5 CORTUPP is available at http://server3.uppuebla.edu.mx/cortupp/ 6 http://www.w3.org/svg/ the upper left corner belong to the rst level of the ontology, these documents share one term. The second level shows the documents that share two terms, and then on. The stars have di erent size according to their level, they are bigger at the rst level and smaller at the last one.

The interface of SV is a SVG graphic of 502 per 502 pixels. XML Parser is the Java application used to construct the XML document that contains the interface. XLink is used to create hyperlinks between documents and their metadata. Given a click on a star, users can allow the metadata on the right panel. Figure 1 shows the SV interface where only six documents at the second and third level were included, however SV is designed to support until 500 documents. The colors can be modi ed without requiring compilation because they are stored in a text le. The mechanism is accessible at http://informatica.uppuebla.edu.mx/ visualizacionPI/index.html. We have described SV, a visualization mechanism of federated collections based on ontologies. SV has semantic features represented in the interface such as the location of documents in the ontology and the similarity between documents. Additional semantic information is stored in the metadata attached to each document and in the ontology of records. Through SV interface, users can access to metadata or download a document.

CORTUPP was used as a test bed for SV, however, any collection of OAIPMH records represented as an ontology of records can be visualized. Although the size of an ontology of records can impact the visualization of SV, its design is exible enough to support distinct collections. As future work, we plan to expand SV to show the clusters and their labels. Then, we would like to incorporate tagging and recommendation mechanisms.

References

1. Geroimenko V. , C.C. : Visualizing the Semantic Web . XML-based Internet and Information Visualization . Segunda edicion edn . Volume 1 . Springer, Wokingham, England ( 2003 ) Libro. los primeros cuatro datos son obligatorios .

2. Kruk , S.R. , McDaniel , B. : Semantic Digital Libraries. Springer-Verlag, Berlin, Heidelberg ( 2009 )

3. Gruber , T.: A translation approach to portable ontology speci cation . Knowledge Adquisition 5 ( 2 ) ( 1993 ) 199 { 220

4. C. , P. : Sistema uva: interfaces para visualizacion de grandes colecciones digitales . Tesis de maestr a, Universidad de las Americas Puebla, Santa Catarina Martir S/N, San Andres Cholula, Puebla, Mexico. ( 2002 ) Tesis de maestr a. Los primeros cuatro campos son obligatorios .

5. Geroimenko V. , G.L. : Interactive interfaces for mapping e-commerce ontologies in Visualizing the Semantic Web . XML-based Internet and Information Visualization . Segunda edicion edn . Volume 1 . Springer, Wokingham, England ( 2003 ) Libro. los primeros cuatro datos son obligatorios .

6. Fluit C. , Sabou N. , H.v.F. : Ontology-based information visualization . In: Visualizing the semantic web: XML based internet and information visualization . Volume 1 . Segunda edicion edn. Springer, Wokingham, England ( 2002 )

7. Sanchez J. A. , Quintana M. G. , R.A. : Star- sh: Star elds+ sheye visualization and its application to federated digital libraries . Proceedings of the 3rd Latin American Conference on Human-Computer Interaction (CLIHC 2007 , Nov .) ( 2007 )

8. Fung , B. , Wang , K. , Ester , M. : Hierarchical document clustering using frequent itemsets . In: Proceedings of the Third SIAM International Conference on Data Mining , (SDM'03 , San Francisco, California, May),, San Francisco, CA, USA, SIAM ( 2003 ) 59 { 70

9. Medina , M.A. , Sanchez , J.A. : Ontoair: A method to construct lightweight ontologies from document collections . Mexican International Conference on Computer Science 0 ( 2008 ) 115 { 125