\section{Scenario}
\label{sec:scenario}

\begin{scenario}
To illustrate the problem mentioned in the previous section in a particular context, in the following, we will consider a typical scenario for an early stage researcher (or any kind of researcher) that steps into a new field. The amount of existing publications, and their current growth rate, makes the task of getting familiarized with the relevant literature in a specific domain highly challenging. From an abstract perspective, the familiarization process consists of several stages, as follows:
\begin{inparaenum}[(i)]
  \item the researcher starts from a publication provided by the supervisor;
  \item she reads the publication, thus grasping its claims and associated argumentation;
  \item reaching a decision point, she either decides to look for another publication, or she follows backwards the chain of references, possibly including also publications by the same authors.
\end{inparaenum}
This last step usually involves accessing a search engine (or a publication repository) and typing the names of the authors, or the title of the publication, to be able to retrieve similar results.
\end{scenario}

Each of the above activities has an associated corresponding time component:
\begin{inparaenum}[(i)]
  \item the time assigned to reading the entire publication (or most of it) --- needed to decide whether the publication is of interest or not,
  \item the time associated with finding appropriate links and references between publications,~\footnote{Following all the references of a publication is obviously not a feasible option. Thus, the decision is usually done based on the citation contexts mentioning the references in the originating publication} and
  \item the time associated with searching additional publications, based on different metadata elements, and (manually) filtering the search results. This time increases substantially when an individual is interested in all the publications of a particular author.
\end{inparaenum}

Finally, analyzing the pairs $<$activity, time component$>$ from the metadata perspective, and what it can do to improve the overall process, we can conclude that:
\begin{itemize}
  \item searching and linking related publications as mentioned above, entails the (manual) extraction and use of shallow metadata. Thus, performing automatic extraction of shallow metadata and using it within the linked data cloud, will significantly reduce the time component associated with this activity.
  \item reading the publication to a large extent corresponds to mining for the discourse knowledge items, i.e. for the rhetorical and argumentation elements of the publication, representing its deep metadata. Consequently, extracting automatically such discourse knowledge items from a publication, will provide the user with the opportunity of having a quick glance over the publication's main claims and arguments and thus decrease the time spent on deciding whether the publication is relevant or not.
\end{itemize}

Transposing these elements into engineering goals led us to a three step process, detailed in the following section: \emph{extraction} -- automatic extraction of shallow and deep metadata; \emph{expansion} -- using the extracted metadata within the linked data cloud for cleaning and enriching purposes; \emph{integration} -- embedding the resulted linked metadata within the personal desktop to ensure a smooth search and browse experience, by using the ordinary desktop applications.

As a side remark to the proposed scenario, an interesting parallel can be made with the music domain. Similarly to publications, music items (e.g. music files, tracks, etc) are also acquired and stored by people on their personal desktops, in numbers usually increasing with time. And as well as publications, these can embed (depending on the format) shallow metadata describing them, such as, band, song title, album or genre. Thus, conceptually, the \emph{extraction} --- \emph{expansion} --- \emph{integration} process we propose can be applied also in this domain. In practice, there already exist tools that deal with parts of this process. For example, on the extraction side, there are tools that help users to create or extract ID3 tags embedded into MP3 files or on the expansion side, there exist tools, such as Picard,~\footnote{http://musicbrainz.org/doc/PicardTagger} that clean the metadata based on specialized music metadata repositories (e.g. MusicBrainz). As we shall see in the next section, the result of our work is quite similar to these, but applied on scientific publications.