The LitMiner application provides users with a suite of tools for searching the biomedical literature and for manipulating the results. Figure shows the left-hand frame of the interface. The functionality of LitMiner is grouped into five sections:
• The Search tab (highlighted in Figure ) links to PubMed and is the main method of bringing textual information into LitMiner for manipulation.
• The Text Clustering tool clusters the results into a hierarchical subject list based on keywords extracted from the titles and abstracts of the articles.
• The Document Folder enables users to save the results of their searches locally and to share these document sets with other users.
• The Gene Set enables users to compare the co-occurrence of genes in the literature, and to visualize the relationships between genes using a network graph.
• The Thesaurus Manager allows users to fine-tune the aliases that are used to refer to genes in searching.
The LitMiner application is written entirely in Java. Although not web-based, the architecture follows a client-server model. While the client performs much of the processing, all interactions with the databases are routed through the server. Note that this paper does not discuss the gene-alias matching which is one of the facets of the LitMiner application. For a discussion of this technique, the reader is referred to [1
]. This paper will focus on the provision of access to articles in LitMiner and how the immediacy of this access within LitMiner has changed how users interact with the digital library. While each of the five tools in the application serves a separate purpose, they each provide an "articles" panel (see Figure ). This enables users to retrieve articles as part of their manipulation and filtering of information (for example interacting with a graph or clustering documents). Rather than making the digital library a distinct entity within LitMiner access that the digital library provides is integrated into each facet of the application.
Articles table. The results of the search.
Records retrieved by a search are displayed in a table that lists article titles. Those records that are accessible online have a small icon associated with them (see Figure ). A list of articles which are available online was compiled separately and the URL of each, along with an identifier of the article's location, was recorded in a database. Three ways of identifying accessible articles were used:
1. The PubMed record contains a DOI (digital object identifier) in the MEDLINE "AID" field. When the user clicks on the "accessible online" icon, the LitMiner application launches a web browser and loads a website that resolves DOIs [18
]. If successfully resolved, the document (typically in pdf format) is displayed.
2. The article is held in PubMed Central, and is likely an OpenAccess article. It is therefore freely accessible. When the user clicks on the "accessible online" icon, the LitMiner application launches a web browser and loads PubMed's article resolver [19
] with the article's PMID number as the final argument in the http call.
3. The PubMed record is associated with a journal to which CISTI is subscribed. An OpenURL link resolver was used to translate the bibliographic data of an article into a URL (typically of the publisher's website) that gives access to the PDF file of that article.
In this way, the application is able to provide access to articles from many different sources (be it a publisher's server, PubMed Central, or a scholarly society's repository of articles) in an integrated way. The resulting access to the desired document is seamless for the user, whether it is a subscription-only article available by virtue of CISTI's licences with various publishers or an article freely available on the Web. In effect, the result of pre-compiling the location of articles known to be accessible is to extend the breadth of the collection.
From the list of results, double-clicking on any given row will cause the bibliographic details of that article to be displayed in another panel. From this article display panel several options are available to users. If the user chooses to view an online document, a Web browser is automatically launched and the associated file (typically a PDF document located on PubMed Central or on a publisher's Web site) is displayed.
Those documents which are not identified as being accessible online can be ordered via CISTI's document delivery service. A button in the article detail display (see Figure ) sends the bibliographic information for the request to CISTI's Blank Web Order form. This feature is convenient for the user, as they are not required to open a Web browser, navigate to the appropriate order form within CISTI's digital library, and manually type the bibliographic information relating to the article of interest into the Blank Web order form.
Figure 3 Article detail display. The abstract of an article is shown in the article detail display. Note the buttons at the bottom of the display allowing the user to view the full text of the article, to view the record in PubMed, or, if those links fail, to (more ...)
At CISTI the physical journal is located and scanned. The (now electronic) document is then delivered to the user. If CISTI has the journal on hand, the time required to fill such an order can be very short, often the same day. This makes the backfiles of print subscriptions as accessible as the most recent, born-digital articles.
Each order, whether online or through CISTI, is logged to the LitMiner server. This enables an analysis of user ordering patterns. At a high level, the overall number of orders indicates the utility of the service. More detailed study of the journals, authors and subjects most frequently ordered could lead to improved library services.
Between February 2004 and February 2005, three biologists who belong to the core user group and who volunteered their ordering data ordered 360 documents through LitMiner (see Table ). Of these, 246 (68%) were electronic documents which were freely available online (i.e.: OpenAccess journals, articles in PubMed Central, and articles with a Document Object Identifier in their MEDLINE record). In addition, 114 (32%) articles were ordered from CISTI's collection via LitMiner.
The number of documents ordered by three biologists. The orders are broken down by year and by ordering method.
In registering for Loansome Doc ordering, the user has notified PubMed that CISTI is their library. Upon locating an article of interest in PubMed, a user can easily order it from CISTI by sending the article's bibliographic information to the Loansome Doc service. In this way, Loansome Doc is a customized ordering service that is integrated into the PubMed search interface.