Adobe Lightroom is an example of a domain-speci c information management tool that has adopted many of the practices that have become well-established within the research community such as automatic metadata extraction, fast keyword-based retrieval and querybased collections for the management of images. It also supports the work ow of professional photographers. We will present a web-based system that transfers some of these advanced concepts to the realm of research publications. The new application introduces sophisticated facilities for classifying publications in di erent dimensions based on combined notions of agging, tagging, ratings and colours in addition to manual organisation and sorting in collections. Retrieval and discovery in a potentially large publication corpus is achieved by providing e cient browsing techniques based on a faceted search interface and user-driven categorisation.
Adobe Lightroom1 is an example of a domain-speci c information management
tool for digital photographs. Many professional photographers use it to browse
and manage large collections of digital photographs as well as performing
postproduction activities in an e cient way [
We aim to transfer the ideas and concepts introduced by state-of-the-art information management tools for digital media to other domains and di erent types of personal information. As a rst step, we have developed a system for the management of research publications in the form of PDF documents and their associated metadata. Our system provides a faceted search interface similar to Adobe Lightroom, but adapted for e ectively browsing research publications by taking into account both available metadata speci c to publications as well as user-de ned criteria. We will show how the system can support researchers in tasks and work ows related to scienti c publishing, for example, when carrying out a literature review in order to collect and manage related work.
We will start with an overview of related work in Sect. 2, followed by a presentation of the use case that motivates the system and supported work ows in Sect. 3. In Sect. 4, we will outline the interaction with the system and describe the key elements of the user interface. Section 5 gives an overview of the implementation and our demonstration is summarised in Sect. 6. 2
As well as supporting advanced means of organising and searching for data,
Adobe Lightroom [
Other systems designed to improve information retrieval through the use of
semantic data include Haystack [
While studies proclaim tagging as a viable substitute for le-based document
management [
We will illustrate the Lightroom information management paradigm at the core of our system by considering the example of managing scienti c publications for a literature review. Researchers typically gather and compile their reading lists of relevant publications from various sources. They then read and classify them according to relevance and other attributes based on personal preferences in order to select the set of most relevant papers for a speci c work. Given that a literature review may involve a huge number of publications, a publication management tool that provides a exible and lightweight approach to e cient categorisation as well as fast searching and retrieval can help achieve that task more e ciently. By adopting the Lightroom model, we have designed the system illustrated in Fig. 1 that supports the management and classi cation of research publications along multiple degrees of freedom.
Figure 1 shows a screenshot of the application as we have used it to gather related work for this particular paper. The publications managed by the system have been imported from existing BibTeX les so that the system has access to the publication metadata such as title, author, booktitle, year and location. The tool o ers a faceted search bar at the top where the bibliography can be ltered according to various attributes. For example, we can quickly lter and display only those publications that have been published at CAISE in previous years, and then continue to lter by authors as well as a speci c year or location.
In the centre, all publications that meet the search criteria are displayed as boxes, with the title at the top followed by an excerpt of the abstract and additional metadata. In our example, we have organised the publications into three collections. The collection on the left-hand side contains all publications that are part of the reading list compiled for this paper. The collection in the middle is a hand-picked selection of publications that are related to the topic of faceted browsing and the collection on the right contains papers about existing personal information management tools. These collections were created directly in the browser simply by dragging and dropping publication entries from one collection to another. Note that the faceted search bar on top lters the content of all collections, which is useful when searching across multiple publication libraries.
All publications currently displayed can also be directly manipulated using the in-place editing controls for colouring and rating at the bottom of each box. In our example, we have used these features to colour important publications in red and rate them according to relevance, as can be seen by the stars at the bottom of each entry. Making combined use of multiple and orthogonal classi cation dimensions allowed us to classify the relevant publications and quickly choose the ones to be referenced in this paper. 4
The most important step in designing the proposed system was to decide which concepts from Lightroom are suitable for the management of research publications and how they would translate to the new domain and di erent metadata. Figure 2 highlights the key features of our search interface that align it with the use of metadata and faceted search in Lightroom, but it also illustrates the changes we had to make when moving from digital photographs to research publications.
Similar to Lightroom, we distinguish between metadata and user-driven attributes. Metadata can be extracted directly from the BibTeX source and de ne the publications in terms of various dimensions which can then be used as facets in faceted browsing. Suitable candidates for facets are attributes where the same values appear multiple times and ideally cluster the information space into signi cantly large groups. For each of these facets, the user interface o ers a list of attribute values inferred from the data that can be selected and combined to perform faceted browsing. In terms of research publications, the candidates we have chosen are Authors, Conference/Journal, Locations and Year (top of Fig. 2). Note that the publication title is not a suitable candidate as the title is usually unique and better used in combination with keyword-based retrieval.
On the other hand, user-driven attributes introduce a way for users to classify publications, which allows them to quickly organise, manage and browse their publication corpus. Examples of user-driven attributes in our system include colours, ratings and di erent kinds of agging, e.g. to mark read/unread papers. These user-driven attributes can then also be queried with corresponding toggle buttons and sliders that hide or show matching publications (bottom of Fig. 2).
The main advantage of faceted browsing is that users can only perform selections that actually yield results. Additionally, facets can be re-calculated after every re nement to immediately give the user some feedback about the result size of further selections. All selections and restrictions are performed in realtime and the results are immediately shown to the user. When designing the faceted search interface for the publication domain, it was important to consider the types of queries a user may want to perform. Since publications are often the combined e ort of more than one author, our system provides several search modes for the Authors facet. In the rst mode, several authors can be selected simultaneously, selecting all the publications that have been written by any of the selected authors. The second mode is similar to the rst one, but performs a conjunction of selected authors resulting in all publications that have been written by selected authors collaboratively. The last mode allows publications to be ltered by rst author only.
Another important di erence when moving to research publications relates to how Lightroom visualises the contents of photo collections where it makes extensive use of thumbnails that can be directly created from the pictures. For other types of information, it is typically required to dynamically look up or even create thumbnails derived from metadata, e.g. album art for MP3s or individual frames from movies. While, in our case, the generation of thumbnail images from the PDF is technically possible, it was considered impractical as the scaled down version of a text document might be barely readable and often provides only a snapshot of the rst page. In addition, the two-column layout used by many research publications provides very few visual cues that might help users remember a particular publication based on its thumbnail. That is why we opted for an approach where we create an appropriate representation based on the publication's metadata such as title, authors and abstract. In Fig. 3, an example of our visual representation of a single publication is given. These \thumbnails" are enhanced with small, unobtrusive user interface elements which enable users to directly manipulate some key attributes as mentioned before.
Finally, our system also supports the creation and maintenance of di erent collections of research publications. These can, for example, be used to maintain personal or shared reading lists, individual publication lists as well as all publications of a research group. Our solution allows ad-hoc lists to be quickly created that can be laid out spatially as illustrated in Fig. 4. Using this approach, users are free to arrange the publications according to their preferences since individual publications can be moved, not only within a collection, but also between collections using simple drag-and-drop interactions. 5
The system is based on a rich client web architecture and was created based on the popular jQuery web framework3 in order to support rich and responsive interactions as well as a look-and-feel similar to the original Adobe Lightroom. The server-side components were implemented based on CakePHP4|a PHP web development framework with well-suited abstractions along the MVC design pattern|and are responsible for user account registration and login as well as the storage and retrieval of publication collections. Facet calculation and ltering in the client is based on AJAX and HTML DOM manipulation techniques
that allow us to reduce loading time of publication entries and dynamically show and hide them according to whether they match the search criteria. We have also developed a number of tools for converting BiBTeX to XML and RSS formats, allowing for easy import of existing bibliographies as well as exporting publication lists managed with our system. Moreover, a lightweight version of the system, with only faceted search rather than publication management capabilities enabled, can be integrated with existing web sites, e.g. those of research groups, and allow visitors to quickly lter and browse publication databases. 6
In our demonstration, we will show the publication management system described in this paper. We will provide several example publication databases, including personal and shared reading lists for di erent topics, but we will also give the opportunity for interested parties to explore the novel management facilities using their own bibliographies imported from BibTeX. In this way, visitors will be able to experience a Lightroom-like publication management system for themselves and test whether it enables them to browse and search individual or groups of publications more e ciently.
This work was supported by the SNF under research grant 200021 121847.