Ex Libris has developed a new service called the bX Recommender that takes advantage of social data to determine related works for any given article. This approach brings Web 2.0 concepts to the realm of scholarly content to provide a new and innovative service.
The vast body of scholarly information presents a challenge for researchers who want to be aware of all relevant articles on a given area of study. A number of techniques can help, like working through the citations in one article, exploring additional works by cited authors or searching using keywords or subject terms. Even when using these techniques, however, it is difficult to locate all related material using traditional citation analysis. The idea that there should be an automated mechanism for discovering related scholarly material must be a fairly common sentiment among academic researchers.
One of the ways to determine relevancy or relatedness involves leveraging data that is collected to understand the way that users interact with resources. A search engine like Google, for example, can improve its relevance rankings through its internal data regarding the links that have been clicked in response to a given query. In a result set where many items seem more or less equivalent according to keyword match criteria, data on the items that users actually select from the list can be an important clue regarding relevancy to that query. Search engines are well positioned to perform this kind of socially guided ranking since they have access to massive quantities of associative data between links and user click-throughs.
The bX Recommender service mines a vast repository of data from link resolvers across multiple institutions. With link servers mediating the process of connecting users to scholarly content, an interesting opportunity emerges to gather user data and apply it to value-added services. In the same way that search engines rely on social data to determine relevancy for Web-based resources, Ex Libris has devised a service that relies on the user data from link resolvers. Many libraries make use of link resolvers to provide access to the ever-increasing body of articles that are represented in their collections of e-journals.
Ex Libris pioneered the genre of link resolvers when it introduced SFX as the first commercial product in this arena. A number of other products have since joined the fray, such as 360 Link from Serials Solutions, the WorldCat Resolver, and LinkSource from EBSCO. Link resolvers have also emerged in the open source arena, including GODOT/CUFS, which was developed at Simon Fraser University in Canada. For libraries that offer large collections of e-journal content, link resolvers have become an almost indispensable component of their technical infrastructure.
The primary purpose of a link resolver is to dynamically calculate links that will take users to the most appropriate copy of an article based on metadata about the article and the profile of subscriptions available from their own library. While the actual link calculations involve a mechanical process, since each operation is invoked on the basis of a request by a user for a particular journal article, the logs of resolvers contain useful data regarding patterns of selections by users. These data reveal relationships among articles based on how they are chosen by a user within a search session. This approach builds on the assumption that multiple articles chosen within the same session by user are related in some way. While the relationships cannot be determined within any given session, aggregating the data across millions of sessions reveals patterns of association between articles.
Ex Libris has created a massive repository of user data to power its bX Recommender service. The architecture of the product involves formatting use log data into a standardized format and aggregating use data within an institution's link server. Data from multiple institutions is then aggregated and harvested using the OAI-PMH protocol into a central repository. These multiple levels of aggregation produce a large collection of user data that can then be processed by analytical software functions as a service provider that responds to requests for related articles.
The concept of bX emerges from research conducted at the Los Alamos National Laboratory beginning in about February 2006 by Herbert Van de Sompel, who led the original development of SFX, and Johan Bollen. [See “An architecture for the aggregation and analysis of scholarly usage data” JCDL 2006). Both currently work at the Los Alamos National Laboratory.
As part of the proof-of-concept for this project and the prototype of the bX Recommender service, Van de Sompel and Bollen aggregated data from the SFX user logs across the digital library repositories at the Los Alamos National Labs and those of the libraries of California State University. Ex Libris has taken advantage of this research to create a commercial service to provide article recommendations.
The company initially announced its work to develop the bX Recommender service in January of 2009. This work involved a group of about 20 libraries as development partners, which included California State University, Tsinghua University Library in Beijing, China, Boston College, Monash University in Australia, and others.
The ability of the service to provide relevant recommendations improves as the size of the repository as user data increases. In most cases, libraries involved with the bX Recommender service will contribute their own link resolver server data to be harvested into the central repository.
One of the key features of bX is the ability to control the recommendations by limiting them to data from a given institution or specified types of institutions. A library might, for example, choose to offer recommendations from only its own user patterns. Another scenario would involve a library in an undergraduate college or university choosing to offer recommendations based on use data from similar institutions.
The bX Recommender does not necessarily have its own discrete interface. Rather, it supports a recommendation feature that can be incorporated into other interfaces. For libraries that use SFX, for example, a related articles feature can be added to the menu of services offered when the SFX button is invoked.
One of the main methods that libraries can use to explore the bX Recommender is an API (applications programming interface).Using an API allows the library to use programming or scripting languages to embed a recommendation feature in any appropriate context within their Web presence. California State University, for example, uses the bX API to integrate bX recommendations into their metasearch environment. The bX API can return results in XML, RSS, or Atom feeds.
Ex Libris offers bX as a hosted service, where the company manages the central repository and manages the process of harvesting use data from participating libraries. While Ex Libris primarily targets libraries that use their own SFX, since it relies on the OpenURL standard followed by all link resolvers, the design of the product accommodates competing products. General availability of the service was announced in May 2009. bX represents the latest in a series of collaborative projects between Herbert Van de Sompel and Ex Libris. Van de Sompel has been involved in developing many of the concepts and protocols that have become important elements in digital library technologies. He played key roles in the creation of the OpenURL protocol, the Open Archives Initiative Protocol for Metadata Harvesting (OAIPMH), and more recently Object Reuse and Exchange (ORE). Ex Libris has been engaged in these efforts and has found opportunities to channel them into commercial products. The company's SFX product derives from Van de Sompel's work at the University of Ghent.
Ex Libris anticipates that other services may also be possible through the bX repository of link server use data. The initial offering of the bX Recommender service may be followed by other services that involve user interactions with scholarly content.