This concept of aggregation is not new to librarians, who have been aggregating content into library collections for centuries. Some librarians are bibliographers who create lists of information sources on various topics of interest. Some of the sources listed in a bibliography may be in the library's local holdings, but some of them may be located elsewhere—perhaps at another library or online on a website. In the traditional analog practice of librarianship, collection management included purchasing books and subscribing to print journals, cataloging them, and arranging them on shelves for patrons to find and use. With the shift from print to digital technology, many of the same principles of collection management are now employed in the aggregation of electronic resources such as databases and e-journals. Libraries are also involved in collecting born-digital content on platforms such as institutional repositories, and many librarians are digitizing special collections and presenting and managing digital libraries.

In all of these activities, librarians define the information that constitutes a collection. This is typically guided by collection development policies that are informed by the mission of the library and the information needs of its patrons. The boundary of such an aggregation is usually established by a librarian as well, in a library's catalog, for example. In other words, what separates a book that belongs to a collection from one that does not?

In most cases, the immediate user of a library collection is assumed to be a person, and librarians have designed their interfaces for people to use. This makes perfect sense in the analog world, where print collections are classified and physically arranged in a building with signs to direct patrons in navigating and using a collection. But in the case of digital libraries, some of a library's collections may also be in electronic formats that are available over a network. The users of digital libraries can be human beings or computer programs.

The problem is that most digital libraries have been provisioned for people, not computer programs, to use. When you boil it down, the primary point of access for a digital library is usually a webpage that presents information along with links to other webpages. The language used to create webpages, Hypertext Markup Language (HTML), was developed for exactly that simple purpose: to mark up and present electronic information to human beings. It is obvious to a person who is accessing a webpage in a digital library that they are looking, for example, at a page in a book that has been digitized. Visual cues such as page numbers at the bottom of the screen or breadcrumb navigation through a table of contents conveys meaning: the semantics of the object being viewed. It doesn't matter if the page numbers are at the bottom left or bottom right of the page, people understand the construct of a “book” and can immediately recognize what it is that they are viewing and how to navigate and use it. People also understand the relationships implied in a book such as those between pages, chapters, works cited from other books, the title page, and the index at the back of the book.

Computer programs cannot understand the semantics of such Web resources unless the resources are exposed and expressed in a way that can be identified and understood by a machine. This is the goal of Object Reuse and Exchange: to provide a standard for identifying aggregations of Web resources and describing the constituents or the boundary of an aggregation.

Librarians invest a great deal of time and care in preparing and assigning metadata to describe digital objects and in presenting and managing objects in digital libraries. There is no question that this process adds considerable value to the digital objects. Unfortunately, most of this value is lost when the user is a computer program and not a human being because a computer program cannot identify and understand how to use the object or any of the valuable information added to it by the librarian. A computer program doesn't know it is accessing a digitized book when the book is being represented in HTML; likewise, it has no way of figuring out the title of the book, its author, and other information about the book that may be obvious to a person who can recognize and read the book's metadata.

The URL (or more generally, the Uniform Resource Identifier or URI) to a Web resource in a digital library typically does not link directly to the digital object itself but to a representation of the object. This representation is usually a “splash page” that is presented for a person, not a computer program, to read and comprehend. To illustrate the point, look at the digital object being represented in figure 1, which is a conference poster that has been deposited into a library's institutional repository.

It is obvious to you, a person, that clicking the Download button will download a copy of the conference poster. But suppose someone wanted to write a program to download all of the conference posters from the collection. The program could retrieve a list of the URIs of every object in the Library Research Publications collection (this piece of metadata could be harvested easily using the OAI-PMH), but the program would probably malfunction when it tried to download the first object because it would receive this splash page instead of a PDF or other file that constituted the actual poster. The program has no way of knowing that there is an additional step (to click the Download button) to download the file. It also can't make much sense of the rest of the information being presented, such as the title of the poster, its authors, the document type (which identifies it as being a poster), the abstract, and the link to a supplementary report that provides context for the poster.

You could make an argument that, with some knowledge of this specific institutional repository and collection, you could write a program that is aware of the link behind the Download button and could accomplish this task. You might even be able to reassemble some of the structured metadata by indexing the page or applying some other heuristics, like those that Google uses for ranking relevant search results. But would this program work with a different digital library that presents different representations of its objects? Chances are it wouldn't work with any precision because the splash pages that it would encounter would be constructed differently. For example, the Download button might be located somewhere else on the page, or instead of a Download button, the title of the object might be a link that the user is expected to click to download the object. There are some other important questions that could be asked. Could such a program be able to differentiate between conference posters and other types of objects in the collection? What if you wanted your program to download and assemble all of the posters or their supplementary files from a particular conference and those files were archived across multiple institutional repositories? What if you wanted to move a set of objects from a digital library to a preservation repository or another digital library platform without losing their semantics?