Improving Human–Semantic Web Interaction: The Rhizomer Experience Roberto García and Rosa Gil The methodology for developing usable and accessible Abstract—Rhizomer is a generic semantic metadata browser application is called User-Centred Design (UCD). It is based and editor. It is based on “classic” HTML paradigms (table, on an iterative development process based on a detailed study form, etc.) in order to provide an easy and accessible user of the users’ needs, the tasks they carry on in order to meet experience. For end-users, it looks like the web pages they are used to deal with. However, the semantics that lay behind are them and the context in which they are performed [3]. used to improve the user experience through an AJAX-enhanced UCD, as other software development processes, starts with web interface. The user interface is fed by a metadata storage the requirements gathering phase. However, the focus is abstraction layer, which offers a SPARQL endpoint and placed specially on users. First of all, it is important to know implements the DESCRIBE SPARQL primitive in order to who the users are. Then, the following step is to identify the improve the usability of the resulting metadata fragments. The tasks they are going to perform. metadata component is combined with a wiki engine that helps managing the content intended for human consumption. The development process continues with the common Altogether, both components build a simple yet powerful phases, i.e. design, implementation and deployment. Despite platform for Semantic Web portals. these similarities, the focus continues to be placed on the user. In order to keep user needs present during the whole Index Terms— human factors, Semantic Web, user interfaces, development process, the previous phases are complemented web portal with two additional ones that are performed in parallel, for each process iteration, prototyping and evaluation. Prototypes are created from the beginning, for instance I. INTRODUCTION paper prototypes [4], which do not require any T HE SEMANTIC WEB has been around for some time and many people is asking why it has not taken off as quickly as the World Wide Web did [1]. implementation, or simple applications with limited functionality. All of them are used to evaluate the system with users so One of the main problems is that it is not reaching the end- their requirements are taken into account and contrasted with users, who can give it the required critical mass for the developed system just from the beginning and through all widespread adoption. In this sense, one of the main the development process iterations. impediments is that users find Semantic Web applications The evaluation is performed in a controlled environment, very hard to use, i.e. they lack usability. This is true even for usually a usability laboratory, where specialised software researchers and practitioners working in the Semantic Web applications are used to record and analyse the whole field [2]. interaction, i.e. screen capture, key strokes, mouse clicks, user Human-Computer Interaction (HCI) is a multidisciplinary video record and voice, etc. effort to improve the human-computer interface. The focus is The User-Centred Design methodology has been employed placed on user, i.e. to take into account user needs from the in order to develop a usable and accessible Semantic Web beginning and through all the development process, and the metadata browser and editor. The process is sketched in objective is to get usable and accessible products. Section II. Then, in Section III, the solution for metadata Usability is defined as the extent to which a product can be browsing is detailed and, in Section IV, it is shown how used by specified users to achieve specified goals with metadata edition has been faced. Finally, Section V presents effectiveness, efficiency and satisfaction in a specified context the conclusions and the future work directions. of use. Accessibility is used to describe the degree to which a system is usable by as many people as possible without II. USER-CENTRED DESIGN modification and it specifically focuses on people with Following the UCD methodology, the first step has been to disabilities. identify the target users and their tasks. A generic study of Semantic Web users and tasks by Lisa Battle has been used Roberto García is with the GRIHO Human-Computer Interaction Research [5]. Three broad user groups are proposed, together with the Group at the Universitat de Lleida, Lleida, E-25001 Spain (phone: +34 973- 702-740; fax: +34 973-702-702; e-mail: rgarcia@diei.udl.es). main categories of tasks they would perform in order to use Rosa Gil is with the GRIHO Human-Computer Interaction Research Group the Semantic Web, as it is shown in Table 1. at the Universitat de Lleida, Lleida, E-25001 Spain (e-mail: rgil@diei.udl.es). Table 1. Some proposed user groups and task types for the Semantic Web [5] Rhizomer is combined with an easy to use and simple content User Group Task Types management system based on a wiki engine [6]. The wiki 1. End users Information seeking tasks provides the means to create and maintain information objects Information synthesis tasks intended for human consumption. In this case, HTML is better Action-oriented tasks suited so it is used instead of semantic metadata. Information sharing tasks Therefore, Rhizomer is used just for semantic metadata 2. Content curators Content update tasks management while the wiki is used for HTML content. Content distribution tasks Semantic metadata might be used to describe resources, among them the wiki documents but also other documents and 3. Ontologists Ontology update tasks resources. Ontology creation & mapping The Rhizomik web site 1 is based on Rhizomer and a wiki engine and constitutes an example of how they can be used to For UCD it is important to clearly identify the target user. produce a semantic portal for research purposes. In our case, the objective is to develop a Semantic Web instance metadata application so the target is end users. The end users profile characterises common users with no III. SEMANTIC METADATA BROWSING knowledge about the Semantic Web. They are used to web Usability guides how end-user interaction with the applications so the resulting application must look like the Semantic Web is faced by Rhizomer. This interaction is kind of appliances they are used to. commonly viewed, when talking about the Semantic Web, just However, following the current tendency that the Web is from the Semantic Web application towards the user, i.e. what not just a single way communication medium like wikis and can be called Semantic Web browsing, which is detailed in blogs demonstrate, we do not keep a separate group of tasks this section. for the content curators’ user profile. This area is being extensively explored in the Semantic Consequently, although we acknowledge the existence of Web area. Many approaches are based on a graph paradigm, users specialised in content update and distribution tasks, we i.e. the user interacts with a nodes and links graphical consider that our target end users should have the possibility representation. As it has been pointed out, this is not the best to edit metadata. Therefore, we consider the following kinds choice as it is not natural to force the user to interact with of tasks: semantic data through the same model that is used for its − Information-seeking tasks, e.g. to look for a restaurant representation [7]. near the theater that will still be open when the movie is over. The graph model might be useful for the user in some very − Information-synthesis tasks, e.g. to organise the agenda specific scenario, e.g. in order to get a quick view of how of a conference attendant. some data is distributed, but even this is not usually the case − Action-Oriented Tasks, e.g. to build a personalized for Semantic Web data. portal to manage research tasks. Consequently, the latest developments for semantic − Information-Sharing Tasks, e.g. to share pictures with metadata browsing are based on different paradigms. Almost friends and family. in all cases, they are based on a browsing paradigm that can − Content update tasks, e.g. to add new books to a be called the “Subject-centric Approach”. This approach is catalog of published books and edit the metadata of described in Section B. previously added ones. The “Subject-centric Approach” has some usability Following this user profile and tasks selection, the design of problems that have motivated us a slightly modified approach. a metadata browsing and edition component called Rhizomer Our proposal is detailed in section C and it is employed in continued. The Rhizomer project constitutes a technological Rhizomer in order to construct a more user-friendly framework that can be used to build up semantic web portals. experience when browsing the Semantic Web. The Rhizomer Rhizomer manages RDF metadata in a user-friendly way. It browsing approach is based on a simple algorithm, detailed in facilitates not just the common Semantic Web to end-user sections D and E, and it is compared to the subject-centric one interaction provided by semantic web browsers, it also in section F. provides the reverse interaction path: end-users can create, There are other browsing approaches like the table edit and remove semantic metadata. paradigm used in mSpace [8]. However, this table-based The whole interaction is performed through a “classical” system must be configured in order to operate with a concrete HTML interface in a usable way, which minimises user efforts set of data. Consequently, it does not constitute a generic and maximises the benefits they obtain from their Semantic paradigm for semantic metadata browsing. Web experiences. The overall intention is to minimize the gap Finally, the unidirectional experience provided by Semantic among computers and human beings in the context of Web browsers is not enough. The other direction must be Semantic Web. considered and the user should be able to create, edit and The semantic metadata management part provided by 1 Rhizomik initiative, http://rhizomik.net remove semantic metadata in a usable way. This part is The view contains all the triples that have the resource as explained in Section IV. subject and it is shown as a HTML table. The table header shows the identifier of the described resource, i.e. the subject A. World Wide Web browsing paradigm for all triples in the metadata fragment. The following table In the Web, the browsing paradigm is based on navigating rows contain two columns. The first one is for each triple web pages and links, which constitute its basic building property and the second one for the corresponding objects. blocks. Web pages have content intended for human Tabulator [11] also uses an approach similar to the Subject- consumption and links relate web pages. Web pages content centric one. However, metadata about resources, anonymous and links are based on HTML and derived languages. Their or not, is recursively expanded as a tree. Therefore, it is design should have accessibility and usability principles in possible to expand the properties about an anonymous mind. resource in the same place were the anonymous resource is This browsing paradigm cannot be directly applied to the referred. However, this approach leads to deep tree expansion Semantic Web because it is based on a different ground that mix metadata about different resources, which might model. The building block of the Semantic Web is the triple confuse users and make them loose track of the resource that , which combined builds-up is being described. graphs. There are also other frameworks that are not explicitly Semantic Web metadata is primarily intended for machine oriented to RDF browsing and provide heavyweight solutions consumption. However, it is clear that it must be also that can be adapted to navigation, e.g. the SEAL framework accessible for human users in order to facilitate semantic web [12]. Moreover, there are other Semantic Web navigation applications debugging, results presentation, querying, etc. tools that generate non-HTML views, e.g. Haystack [13] However, the WWW browsing paradigm cannot be applied generates text views and [14] graphical representations. to the Semantic Web. Semantic Web metadata can be packed In many cases the shown metadata fragment is augmented in web documents, e.g. a web-accessible XML file; however it with triples that have the described resource as object, i.e. is not practical to use this approach as the basis for Semantic reverse triples that do not have the described resource as Web browsing. source but as destination. First, not all metadata is available as web documents. Additionally, more triple levels can be included, i.e. the Moreover, semantically related metadata might be packed in triples that have the objects of the original triples as subject, different web documents, which reduces the benefits of and thus recursively. However, it is not common to include semantics-enabled metadata browsing. additional triple levels because it can make the fragments too Usually, Semantic Web metadata is available from big and break semantic coherence, i.e. the user looses the databases. In this case, as it can be also the case for web perspective about what is being described. documents, the amount of metadata is too big for human-user Finally, the semantic coherence is also lost when this consumption. approach is used and anonymous resources are involved. Therefore, what is needed is a Semantic Web browsing Anonymous nodes get identified by the context in which they paradigm and support system that can browse metadata appear, i.e. the triples that reference them and the identified coming from different sources through semantically coherent resources appearing in these triples. fragments that facilitate human consumption. Now the However, this context is broken when the metadata question is: how to define fragments? fragments are built. Anonymous nodes get temporal B. Subject-centric Approach identifiers, which are needed in order to build the graph, but The simpler approach to fragment semantic web graphs is they are also used in the user interface as if anonymous nodes to define the fragment as the set of all triples with the same were like other identified resources. subject. This is also semantically coherent because these For example, Table 2 shows a reference to an anonymous triples are those describing the subject resource. resource as the value of the “N” property. An additional This approach is used in almost all Semantic Web browsers. browsing step is necessary to get the triples describing the For instance, Piggy Bank [9] or Brownsauce [10] generate anonymous value and make it all semantically coherent, as it HTML views for a given resource, i.e. a node of the graph. is shown in Table 3. Table 2 shows an example of such kind of view. Table 3. Additional step for anonymous resource metadata Table 2. HTML Table view of a metadata fragment (anonymous item) Rosa Gil Given Rosa EMAIL rgil@diei.udl.es Family Gil FN Rosa Gil Show Referrers N (anonymous item) Consequently, the same approach is applied to anonymous Show Referrers resources. All the triples that have the anonymous resource as subject are shown. Therefore, the metadata describing the Table 4. HTML Table view of a metadata fragment anonymous resource is shown but the identification context is Rosa Gil lost in the presented view. There might be reverse triples to go EMAIL rgil@diei.udl.es to the identified resource that defines the identification context for the anonymous resource. In any case, however, the FN Rosa Gil presented view does not include the whole context so the user N Given Rosa might not be aware of this fact. Family Gil For example, one common examples of this situation is Show Referrers when RDF containers are used. They do not usually get an identifier so the resources for the containers are anonymous. Therefore, when the metadata for the resource described with To conclude the presentation of the Rhizomer approach, it container values is shown, the set of container values are not is important to take into consideration the presence of cycles shown together with the resource to which they are associated. in the metadata graph and to avoid considering triples already In order to solve the inconveniences of the Subject-centric added to a fragment. This can be easily implemented as it is approach we propose the Rhizomer Approach that is detailed shown in the algorithm in the next section. in the next section. D. Algorithm C. Rhizomer Approach The browsing fragments are built from a set of selected The Rhizomer Approach for Semantic Web browsing is resources. For each resource, all the triples where the resource also based on fragmenting the metadata graph in a subject- plays the subject role are selected. Then, for each object of the centric way. However, a metadata fragment generated by selected triples, if it is an anonymous resource, all the triples Rhizomer is more than just the considered resource and all the where the anonymous resource is the subject are also selected. triples in which it participates as the subject. Then, and recursively, the same procedure is applied to the The set of triples for a subject-centric is enlarged with all new object anonymous resources. The algorithm for building the metadata that depends on the selected subject for its the fragment for a given resource is detailed in Table 5. identification. Therefore, the graph is traversed starting from Table 5. Rhizomer algorithm for fragments browsing the resource acting as the subject through all possible paths TripleSet buildFragment(Resource r, Triples f) until identified resources or literals, which include all the { TripleSet rTriples = TriplesWithSubject(r) intermediate anonymous resources. All the traversed triples fragment += rTriples constitute a Rhizomer metadata fragment and it is what is for each Triple t in rTriples such that NotInSet(t,fragment) shown at a browsing step. Figure 1 shows a simple RDF graph { where four fragments can be identified. fragment.add(t) Fragment 1 if anonymous(t.object) { Fragment 3 TripleSet anonTriples = buildFragment(o, fragment) fragment.add(anonTriples) Fragment 2 } } Fragment 4 return fragment } Identified Resource Anonymous Resource E. Multi-language Support Literal When a metadata fragment is rendered as HTML or other Figure 1. RDF graph and four Rhizomer fragments format, the triple URIs are replaced with their corresponding As in the case of the subject-centric approach, the resources labels if available. If there is not a defined label for the URI, it appearing as the object of the terminal triples can be browsed is shortened to its fragment identifier, the string after the ‘#’, through new navigation steps that generate new metadata or to the substring after the last ‘/’. fragments describing the resources asked for detail. This has This is done in order to improve readability. An additional been already shown in Table 3, which shows a browsing step improvement, which is not common in existent Semantic Web initiated from the Table 2 fragment. browsers, is multi-language support. The Rhizomer algorithm, However, Rhizomer shows all the metadata in the same in addition to the triples that compose the fragment, selects all identification context together. This way, a greater level of the triples that define labels for all the involved resources, i.e. semantic coherence is maintained and the user experience is subjects, predicates and objects. improved. Anonymous resources do no break down the These labels use to be annotated with a language attribute. metadata describing a resource and they are shown together When the fragment is rendered and a preferred language with the resources that contextualise them, as it is shown in specified, the preferred language labels are selected if they are Table 4. available. If not, the default language has been set to English so the labels with the ‘en’ value for their language attribute of browsing from the user. In any case, all browsing are selected when available. If there is not any label with the approaches will suffer this problem because it corresponds to preferred or default language attribute, a label without the description of the resource describing the whole country language attribute is used. and all triples in this fragment are directly attached to it. Finally, the maximum fragment size for the Music Brainz U2 F. Comparative Study discography is 170 triples. As in the previous case, it As it has been previously shown, the subject-centric corresponds to the U2 concept that directly links to all its approach is the common method for building fragments for discography. Consequently, other browsing approaches will metadata browsing. It has been also shown how the Rhizomer also require a lot of browsing from the user. approach works and how it builds more coherent fragments by As it has been pointed out, the inclusion of the anonymous incorporating all the metadata identified by the described resources together with the resource that identifies them is not resource, i.e. the anonymous resources related to it. common in the existing Semantic Web browsers. However, It might seem that the Rhizomer approach makes the we have found a similar approach in the DBin project [15]. fragments too big for human user consumption. A statistical This project defines the term called RDFN (RDF study has been performed with different sets of Semantic Web Neighbourhood) that is based on a similar approach. The main metadata. The results of this analysis are shown in Table 6. difference is that this approach is specially tailored for As it can be seen from the statistical analysis, there is a producing digital signatures of pieces of a RDF graph and that slight increase in the size of the metadata fragments when the it does not consider triples directionality. Rhizomer approach is used. It is greater when there is an Therefore, it is not suited for Semantic Web browsing but it intensive use of anonymous resources, for instance in shows that the best way of segmenting a graph is using an ontologies because class restrictions do not use to be approach based on a subject-centric approach plus the identified or in common real world metadata because we do anonymous resources identified by the subject. not use to give identifiers to every resource. Another much more similar approach is Concise Bounded Table 6. Average fragment size in triples Descriptions (CBD) [16] enriched with back links, i.e. links Subject pointing back to the subject being described. Metadata Set Set Size Rhizomer Centric G. HTML Metadata Rendering Musicbrainz 2 8121 6.6 7.7 Once the metadata fragments have been generated, they are CIA World Factbook 3 876 4.3 14.4 shown to the user as an HTML rendering in the web browser. Wine Ontology 4 1839 2.6 7.3 This rendering allows getting a user interface users are comfortable with. It looks like the kind of webs they are used In the case of the Musicbrainz metadata, the difference to interact with. between the subject-centric and Rhizomer approach is smaller In order to produce this rendering, a generic XSL because there are few anonymous resources. Musicbrainz transformation from RDF/XML to HTML has been describes artists, albums, tracks, etc. and all of them must developed. This approach produces consistent results as long have an identifier in order to be easily referred. The only as the metadata fragments to render are generated with the anonymous resources are containers, which are used, for Rhizomer approach and an abbreviated RDF/XML instance, to group all the album tracks. serialisation of them is produced. This slight increase in size of the fragments does not put The abbreviated serialisation produces an XML stream that them out of the human user reach. On the contrary, it keeps all the related triples grouped. Consequently, it is facilitates browsing because fewer steps are required. possible to render them as a set of HTML tables, one for each Moreover, all the metadata related to the same identifier is resource being described, with nested tables corresponding to shown together and this increases its coherence and the the descriptions of the anonymous resources they refer to. usability of the whole system. Resource and property names are rendered as text using the It is also important to consider maximum fragment sizes for appropriate label for the preferred language if available, as it these metadata sets. For the Wine Ontoloty, the maximum has been detailed in Section E. Literals are also rendered as fragment size is 42 triples, which corresponds to the text and, if different language versions are available, the “CavernetSauvignon” class definition. This is a usable size as preferred one is selected. Finally, there are HTML links for all it can be rendered as HTML using the method detailed in resource and property names that allow browsing the metadata Section G in a 370x540 pixels area of a web browser. For the describing them. These links correspond to calls to a SPARQL CIA Factbook metadata about Spain, the maximum fragment endpoint, concretely they are DESCRIBE queries. This kind size is 549. This is a really big fragment, which requires a lot of queries is resolved in a proprietary way in order to generate Rhizomer metadata fragments and to enrich them with the RDF labels that make the multilingual rendering possible. 2 http://musicbrainz.org, U2 discography The resulting HTML pages look like the simple example 3 Factbook for Spain, http://www.daml.org/2003/09/factbook/sp previously shown in Table 4, or the screen capture from the 4 http://www.w3.org/TR/owl-guide/wine.rdf Rhizomik site shown in Figure 2. Moreover, it can be tested metadata using a XSL transformation. Therefore, it is possible from the ReDeFer web page [5] for arbitrary metadata sets. to generate a query form from an example piece of metadata. A related approach to render RDF as HTML is Fresnel [6]. Table 7 shows an example of a form to build semantic queries. Fresnel lenses are specifications about how to render some Table 7. Semantics-enabled query form resources, classes and properties as HTML or other
presentation languages. They allow a great level of has been specified in order to generate a rendering. However, Rhizomer does not allow this level of personalisation but it is …
capable of dealing, in a very generic way, with any piece of metadata it encounters. In any case, if a greater level of personalisation is required, the Fresnel lenses rendering As it can be seen in the form, the names of the form fields engine can be integrated in the Rhizomer platform. are URIs coming from RDF schemas and web ontologies. The Consequently, the Rhizomer rendering approach added intention is to make the form fields the building blocks to value is that it is very easy to implement, it just requires an generate triples from forms. The fields’ names are the XSL processor so this work load can be put away from the predicates and their values the objects. The subject, in the case web server and passed to the user web browser. Moreover, it of a query, is not known and so it does no need to be can be managed with AJAX [17] so a greater level of specified. In fact, the resource URI will be the response we interaction through the browser can be achieved. will get from the query. The filled form fields are interpreted as the known terms we use to perform the search. For instance, the form in Table 7 can be filled with the “Corporate news” value for the input field, i.e. Dublin Core title, and the first option of the select field, i.e. music is the subject. In the case of a query form, the fields are interpreted as the triples that build up the query pattern of a SPARQL 7 query. A query like the one shown in Table 8 will be generated by a Javascript from the form values as a result of an “on submit” event. Table 8. SPARQL semantic query for Table 7 form DESCRIBE ?s WHERE { ?s http://purl.org/dc/…/title ?y0; http://purl.org/dc/…/subject ?y1. FILTER regex(?y0, "Corporate news") && ?y1= } Figure 2. Rhizomik interface with Rhizomer zoom If the objective is to edit metadata, then the form fields are interpreted as the building blocks for a set of triples for the IV. METADATA EDITION new metadata. In other words, edit forms are used to edit new or existing metadata. Now, it is necessary to specify the The browsing capabilities already shown provide a simple subject for the triples. This is done using a field named Semantic Web user interface. It is simple because the “rdf:ID” or “rdf:about”. Moreover, there might be also other interaction is limited to selecting the piece of metadata to subjects for the anonymous resources included in the edited show next. However, the user cannot take this interaction metadata fragment. further. The anonymous subjects are defined using hidden form In order to improve user experience, the browsing fields named “rdf:ID” and valued with a temporal identifier capabilities have been enriched with editing and querying just to make possible to build the graph. When all the triples functionalities based on semantics-enabled forms. These are for the anonymous subject have been specified, and another common HTML forms that take advantage of some simple anonymous subject or the main identified subject is to be conventions to make them semantics-enabled. described, a new hidden field with the “rdf:ID” of the subject Moreover, they are automatically generated from RDF of the following triples is introduced. Table 9 shows an example of a metadata edition form. 5 ReDeFer project web page, http://rhizomik.net/redefer 6 Fresnel - Display Vocabulary, http://www.w3.org/2005/04/fresnel-info Table 9. Semantics-enabled edit form
Figure 3. Automatically generated metadata edition form …
V. CONCLUSIONS AND FUTURE WORK If the fields in the Table 9 form are filled with the As it has been shown, Rhizomer uses a common HTML appropriate values, e.g. “http://rhizomik.net/~rosa” for the interface. This approach is usual in Semantic Web browsers non-hidden “rdf:ID” field, when it is submitted, the metadata but it is enhanced by making it more semantically coherent. shown in Table 10 is generated applying a direct conversion The navigation is performed by metadata fragments that are from form fields to triples. fully contextualised in the frame of at least one resource that Table 10. Metadata from Table 9 edit form submit identifies it. Therefore, no additional browsing steps are needed in order vcard:EMAIL to get to the metadata associated to anonymous resources and ; no metadata is presented without an identification context, as vcard:N [ it is the common case when showing the metadata associated vcard:Family "Gil"; vcard:Given "Rosa" ]; to anonymous resources. vcard:FN "Rosa Gil". In addition, in order to the improved browsing experience, the user has a more interactive experience thanks to an Therefore, semantics-enabled forms facilitate a greater level additional set of features. Users can create, edit and remove of interaction with metadata through a Semantic Web semantic metadata as it is browsed. This additional browser. In addition to metadata browsing based on functionality is also available through a common web Rhizomer-like fragments, it is also possible to edit such interface based on HTML forms. fragments or create new ones using the same semantics- The same applies to content, which is also directly editable enabled forms. through the wiki. Therefore, the whole interaction is browser As it has been shown, a direct parallelism can be based and the user does not have to install anything else. established from form fields to triples, so the new or edited The intention is to improve the user experience so all this is metadata can be generated from the user interaction with the currently being tested with real users in the context of a form when it is submitted. Moreover, the reverse way is also usability and accessibility laboratory. Actually, many of the direct, from triples to form fields. Therefore, the edition forms design decisions during the Rhizomer development have been can be generated from existing metadata in order to edit it or based on user test results. generate new metadata based on predefined patterns. For instance, all links in the HTML rendering that allow This two-ways mapping has been implemented in the browsing new metadata fragments are not underlined. Just Rhizomik semantic portal (http://rhizomik.net). The form links that point to new HTML content are underlined. This fields to triples transformation is implemented when the form approach helps user differentiate among metadata and content is submitted using JavaScript. The RDF triples to form browsing links. As they are used to underlined links while transformation is implemented using an XSL transformation. navigating through html documents, the same behaviour is The implementation details are not given here due to space maintained in the HTML rendering. limitations but they can be obtained from the Rhizomik site. Additionally, accessibility is being tested with screen Figure 3 shows a form generated automatically from the RDF readers. The results are quite user friendly because the metadata shown in Table 10 in order to edit it. generated HTML content is based on the available RDF labels and preferred languages and this makes it easier for text-to- speech applications. A part from a more extensive user testing, future work concentrates on incorporating the assisted addition of properties and values depending on the type of the described resource. When the user edits a metadata fragment and tries to add a new property to describe it, a popup window presents a 7 SPARQL Query Language, http://www.w3.org/TR/rdf-sparql-query list of all the available properties, as it is shown in Figure 4. [5] L. Battle, “Leveraging work-centered semantics in human-computer applications,” in Proc. 3rd International Semantic Web User Interaction Workshop, Athens, Georgia, USA, 2006. [6] B. Leuf and W. Cunningham, The wiki way. Addison-Wesley Longman, 2001. [7] D. Karger and m.c. schraefel, “The Pathetic Fallacy of RDF,” in Proc. 3rd International Semantic Web User Interaction Workshop, Athens, Georgia, USA, 2006. [8] m.c. schraefel, M. Karam, S. Zhao, “mSpace: interaction design for user- Figure 4. Assisted metadata creation popup for a SWRC Project resource determined, adaptable domain exploration in hypermedia,” in Proc. Workshop on Adaptive Hypermedia and Adaptive Web Based Systems, First, there are all the properties that are specific to the Nottingham, UK, 2003. resource at hand, i.e. they are restricted to the resource type or [9] D. Huynh, S. Mazzocchi and D. Karger, “Piggy Bank: experience the their domain is one of the resource types. Second, all the Semantic Web inside your web browser,” in Proc. The International properties that are generic, i.e. they have no domain defined or Semantic Web Conference, Springer LNCS 3729, 2005. it is just restricted to any resource. [10] D. Steer, “Brown Sauce: An introduction,” Technical Report HPL-2003- The same kind of assisted metadata creation is planned for 10, Hewlett-Packard Laboratories, Bristol, England, 2003. property values. Altogether may facilitate user interaction [11] T. Berners-Lee, Y. Chen, L. Chilton, D. Connolly, R. Dhanaraj, J. with the underlying ontologies that structure the conceptual Hollenbach, A. Lerer and D. Sheets, “Tabulator: Exploring and Analyzing linked data on the Semantic Web,” in Proc. 3rd International framework where semantic metadata is generated. In other Semantic Web User Interaction Workshop, Athens, Georgia, USA, 2006. words, the user will generate metadata following the [12] J. Hartmann and Y. Sure, “An Infrastructure for Scalable, Reliable restrictions defined by the corresponding ontologies without Semantic Portals,” IEEE Intelligent Systems, vol. 19, num. 3, pp. 58-65, being aware of their existence, at least if the user does not 2004. want to. [13] D. Quan and K.Karger, “How to make a semantic web browser,” in Proc. of the 13th International WWW Conference, NYC, USA, 2004. REFERENCES [14] C. Sayers, “Node-centric RDF Graph Visualization,”. Technical Report HPL-2004-60, Hewlett-Packard Laboratories, Bristol, England, 2004 [1] N. Shadbolt, T. Berners-Lee and W. Hall, “The Semantic Web [15] G. Tummarello, C. Morbidoni, P. Puliti and F. Piazza, “The DBin revisited,”.IEEE Intelligent Systems, vol. 21, num. 3, pp. 96-101, 2006. Semantic Web platform: an overview,” in Proc. Workshop on The Semantic Computing Initiative, Chiba, Japan, 2005. [2] T. Heath, J. Domingue and P. Shabajee, “User interaction and uptake challenges to successfully deploying Semantic Web technologies,” in [16] P. Stickler, “CBD - Concise Bounded Description”, W3C Member Proc. 3rd International Semantic Web User Interaction Workshop, Submission, 3 June 2005. Athens, Georgia, USA, 2006. [17] D. Crane and E Pascarello, Ajax in Action. Manning, 2005. [3] D. Norman and S. Draper, User-centered systems design: new perspectives on human-computer interaction. Lawrence Erlbaum, 1986. [4] C. Snyder, Paper Prototyping: The Fast and Easy Way to Design and Refine User Interfaces. Morgan Kaufmann, 2003