Interoperability of repositories: the simple query interface in ARIADNE.

This article reports on our experiences in providing interoperability between the ARIADNE knowledge pool system (KPS) (Duval, Forte, Cardinaels, Verhoeven, Van Durm, Hendrickx et al., 2001) and several other heterogeneous learning object repositories and referatories.

Introduction

This article gives an overview on the role of the Simple Query Interface (SQI) (Simon, Massart, Van Assche, Ternier, Duval, Brantner et al., 2005) in providing interoperability between metadata repositories. It reports on work that has been done bridging different Learning Object Infrastructures and tries to give an overall picture of the current status. Interoperability has been achieved with two types of systems: learning object referatories and learning object repositories.

A learning object referatory (e.g., EdNA online, MERLOT) maintains metadata about resources that are not managed by the system. It describes content and provides references to learning objects or learning resources. As these systems do not host the content, they are challenged with managing references that are often not persistent. A referatory usually also offers mechanisms to detect changes and trigger updates on metadata, as content on the Web evolves over time without notification.

A learning object repository (LOR) (e.g., the ARIADNE KPS, http://www.ariadne-eu.org) offers besides a metadata repository also a content repository. For this type of system, there is no need to deal with the issues described above, as the system manages the content itself. Administering the content comes with the obligation to provide other content management services like digital rights management, version tracking, making the content accessible, etc.

As our SQI applications only deal with querying metadata, this article will not cover content management and hence, we limit ourselves to metadata repositories only. These are offered by both learning object repositories and learning object referatories. We assume however that all metadata repositories are able to provide a reference. This reference can either resolve directly to the content, or can lead a user to a page, describing how to proceed further if he/she wants to obtain the content.

In this article, our understanding of metadata repository will cover an individual storage layer (relational database, XML store) which can be part of a peer-to-peer network (Nejdl, Wolf, Changtao, Decker, Sintek, Naeve et al., 2002), federated search network, etc. An ARIADNE knowledge pool system is considered as one storage layer, while the federated search layer that will be described below is not. This federated search layer is not part of the ARIADNE Knowledge Pool System, although it is used by some ARIADNE tools.

SQI API 

SQI is a specification created under the auspices of the CEN/ISSS Learning Technologies Workshop. The partners involved in this process all represent very different kinds of metadata repositories. This has led to a query transport specification that is applicable in different contexts and that meets very different architectural requirements. For ARIADNE, the SQI API enables two kinds of interoperability.

* Tool interoperability. Users have more freedom in picking a search engine, to search into the ARIADNE knowledge pool system. From the moment that the API is implemented on a system, its searching functionality becomes available in a whole range of applications that know how to send queries to an SQI enabled search engine. These applications that were formerly very strongly coupled to the ARIADNE knowledge pool, are now reusable on non ARIADNE repositories. Providing such an interface between search components, leaves organisations with more freedom in providing search functionality in their e-learning framework.

* Interoperability between metadata repositories. The SQI API is beneficial for both foreign repositories targeting interoperability with ARIADNE and ARIADNE aiming interoperability with other repositories.

- At the time of writing, the ARIADNE SQI interface is integrated in Edusource (Splash Federated Search, 2004) through an ECL gateway, MERLOT (1997) and Celebrate (Celebrate, n.d). Other implementations in EdNA (2005), Lionshare (2005) and EducaNext (ELENA, n.d.) are ongoing. These repositories benefit from SQI by future integrations of other SQI-enabled repositories becoming available for free.

- Inversely, the ARIADNE search tool can benefit from other repositories that implement SQI because the federated search layer it uses can be extended with other SQI targets. As more metadata repositories provide SQI targets, integrating them is only an issue of configuration provided that they offer queries in the same query language and return results in the same metadata binding. Currently, the ARIADNE community can search through SQI in MERLOT, EdNA, EducaNext, RDN, Pond and Celebrate.

SQI comes in two communication scenarios: an asynchronous scenario and a synchronous scenario. One must implement at least one scenario. In the synchronous scenario, a client (source) sends a query to a repository (target) and actively waits for the target to return results. Note that in this scenario, a program (thread) halts until it receives results. In the asynchronous scenario, clear distinction is made between two actions. After the source has sent a query to a target, it proceeds execution and does not wait for the target. Using a result listener service (implemented on the source), a target can later register its results. This scenario assumes that the target is able to contact the source.

Federated Search Through SQI

Searching beyond the borders of ARIADNE is of great value to the end users. Before, SILO--ARIADNE's search and indexation tool--only delivered results that were indexed by ARIADNE members. With each new repository that is added through SQI, more heterogeneous content becomes available to the ARIADNE community. Both the synchronous and asynchronous interface are useful here. On one hand, a front-end tool requires an easy way to integrate API. On the other hand, the back-end architecture needs to be able to send queries and retrieve searches in a flexible way. SQI has proven to be very useful here.

A programmer or web-designer that wants to integrate a search API into a front-end tool, does not want to deal with the complexities of asynchronous querying, ranking and merging results. Asynchronous requests suffer an additional complexity when dealing with software that runs on desktops. As these are often hidden behind a network address translation (NAT) or firewall, it might not always be possible for a target to initiate contact with the hidden source and hence receiving the results asynchronously. Therefore, both SILO and a search prototype tool only use the synchronous SQI API.

In the back-end, a federated search component acts as a gateway to different SQI enabled repositories. This component offers a synchronous SQI interface to the front-end and is able to issue both synchronous and asynchronous queries in the back-end. The federated search component is able to forward queries to both API's. The asynchronous API has proven to be helpful here, as it can forward a query asynchronously to a number of repositories and use one listener to collect all results. The component can also federate queries to synchronous SQI targets, as we don't want to exclude targets that do not implement the asynchronous interface. In the latter scenario, we need a separate thread for each synchronous target to which the federated search layer connects. This thread first sends the query to the foreign target and next synchronously waits for results.

SQI makes it possible to implement a query service that is either stateful or stateless. In the stateless scenario, an implementation usually re-executes a query when a source polls for additional results. This is not the behaviour we target when providing SQI on top of a federated search target. As re-executing a query might not only be time consuming, but will likely return results in another order. Because of that, a stateful federated search layer was designed, where the query string identifies the state of an ongoing query. The ARIADNE knowledge pool offers a stateless SQI service, as the time necessary to execute a query on this target is negligible.

In this section, we briefly outlined how searches are federated to other repositories. Note that SQI is agnostic about the architecture for interoperability and that we only explained one possible solution for this architecture. In the Edusource network, it is up to the client to decide in which repositories to conduct a search. Using the Edusource Communication Layer (ECL), a layer which can be implemented on top of a repository, a client can query a repository. The client, however, doesn't send its query to an intermediate layer, but to each repository separately which the client can discover through a Universal Description, Discovery and Integration (UDDI) service. As the UDDI registry contains information about the query templates a repository supports, the client itself is able to do the mapping from one query language to another query language. Note that in this context some ECL to SQI gateways are already operational (Splash Federated Search, 2004).

Query Languages

All partners involved in this network, offer different search methods on their repository. At a technical level, searches differ in semantics, encoding and underlying metadata standard.

* The ARIADNE KPS implements two query languages. Users can express advanced queries that are transported between applications serialized as XML messages. The following excerpt exemplifies such an XML document.</p> <pre> <advancedQuery> <And> <queryItem comparison="contains"

path="ariadneDocument/title" value="Java"/> <queryItem comparison="lower"

path="ariadneDocument/publicationDate" value="10/11/1993"/> </And>

</advancedQuery> </pre> <p>As most end users only want to provide search terms to the query application without the need know what the exact semantics of the query is, ARIADNE also defines a metadata independent query languages. Using the latter query language applications can send a list of search terms to the KPS:</p> <pre> <simpleQuery> <item>LOMster</item>

<item>Ternier</item> </simpleQuery> </pre> <p>Note, that it is our intention to enrich these queries with the profile of a user so that better ranking or results can be presented.

* Edusource offers different templates. The most limited template enables clients to only search using search terms. In the most general case an ECL enabled repository can offer full XQuery capabilities.

* The EdNA Online Distributed Search enables Google-like searches. If one uses the HTML API, searches are encoded as a parameter in the URL. The following example illustrates this: a parameter q has as a value several search terms delimited by a '+.'

http://api.edna.edu.au/search.xml?q=relativity+special

Although the EdNA API only offers limited query possibilities, some of the backhand repositories that implement this EdNA API, offer more expressive search functionality.

* Similarly to ARIADNE, MERLOT offers two search syntaxes:

- MERLOT's simple search syntax makes it possible to glue search terms together with boolean operators:

"learning object" OR "learning resource" AND "lor"

- An advanced query syntax defines a proper way of expressing searches that offer richer expressiveness. As illustrated by the following example, one can search into individual metadata fields using this syntax:</p> <pre> <search> <title>LOMster</title> <creatorname>Ternier</creatorname>

</search> </pre> <p>In our first interoperability step, a greatest common denominator for all query languages was offered at the level of the federated search service that forwards queries into the network. As all participating repositories offer a possibility to issue search term based queries, this is the only search functionality that is offered for now.

CONCLUSION AND FURTHER WORK

In this article, we briefly outlined how ARIADNE is connected to and connects to different metadata repositories through SQI. These repositories currently only agree on exchanging search terms. Future interoperability steps will focus on exchanging semantically more expressive searches. This will enable tools and software agents to formulate requests (e.g., based on a user profile) that will better match the needs of an end user.

Nothing has been decided yet at the level of ranking the individual results. Although all metadata repositories already leverage some kind of ranking, the aggregation layer is left with the problem of merging different arbitrarily ranked results. In a next phase, all partners could either agree on the same ranking algorithms or implement different algorithms that can then be selected by the clients.

Currently, in the synchronous scenario, a target decides how long to wait until the first results are sent back. For some clients it is however necessary to provide results just fast (e.g., Google) while in other usage scenario's, time doesn't matter as long as the application provides good results. (e.g., searches into a peer-to-peer network). Note that no changes at the level of the SQI API are necessary to provide this functionality.

References

Celebrate (n.d.). Celebrate. Retreived August 31, 2005 from http://celebrate.eun.org/

Duval, E., Forte, E., Cardinaels, K., Verhoeven, B., Van Durm, R., Hendrickx, K., et al. (2001). The ARIADNE knowledge pool system. Communications of the ACM, 44(5), 73-78.

EdNA Search Centre (2005). Education network Australia search centre. Retreived August 31, 2005 from http://www.edna.edu.au/edna/search

ELENA (n.d.) The ELENA smart space for learning. Retreived August 31, 2005 from http://www.hcd-online.com/ubp

LionShare (2005). Lionshare. Retreived August 31, 2005 from http://lionshare.its.psu.edu/

MERLOT (1997). Multimedia Educational Resource for Learning and Online Teaching federated search. Retreived August 31, 2005 from http://fedsearch.merlot.org/

Nejdl, W., Wolf, B., Changtao, Q., Decker, S., Sintek, M., Naeve, A., et al. (2002). EDUTELLA: A P2P networking infrastructure based on RDF. In Proceedings of The 12th International World Wide Web Conference. Budapest, Hungary.

Simon, B., Massart, D., Van Assche, F., Ternier, S., Duval, E., Brantner, S., et al. (2005). A simple query interface for interoperable learning repositories. In Proceedings of The 14th International World Wide Web Conference. Chiba, Japan.

Splash Federated Search (2004). Splash federated search. Retreived August 31, 2005 from http://pool.iat.sfu.ca/fedsearch/

STEFAAN TERNIER AND ERIK DUVAL

Katholieke Universiteit Leuven, Belgium

[e-mail]

[e-mail]