Managing data within a research unit is not a trivial task due to the high number of entities to deal with: projects, researchers, publications, attended events, etc. When all these data are exposed on a public website, the need to have it updated is fundamental to avoid getting an incorrect impression of the group's performance. As research centres websites are usually quite static, external documents are generated by managers, resulting in data redundancy and out-of-date records. In this paper, we show our e orts to manage all these data using Labman, a web framework that deals with all the data, links entities and publishes them as Linked Open Data, allowing to get insightful information about the group's productivity using visual analytics and interactive charts.
Managing metadata e ectively within a research unit is an ambitious goal, as information systems need to deal with the relationships among the entities that form the organization's data model: projects, publications, researchers and project managers, topics, etc. Most research groups expose their data using a well known Content Management System (CMS) such as Joomla!1, WordPress2 or Drupal3. Nonetheless, in order to extract valuable knowledge from all those data, external tools are needed to perform data analysis techniques. Exporting data in easy to handle formats from the CMS's databases usually leads to the creation of external documents which store data that will later be analysed.
This common situation has the following drawbacks: external documents (e.g., CSV, spreadsheets, text les, etc.) cause data redundancy, resulting in data quality, completeness and updating issues. This gets worse when investigators have their own personal pages (outside the system) where they show the achievements of their researching careers, funding data is managed by the accounting department and so on. When data needs to be updated in di erent
systems, the expected outcome is that at some point data is going to be outdated somewhere, thus leading to errors when trying to get the whole picture of a research unit's performance.
Therefore, we present our e orts towards managing our research group's data,
avoiding redundancy, improving quality and sharing the data in a standardized
and interoperable way. Labman (Laboratory Management) is presented as a
tool to manage all these data, publishing them as Linked Open Data. Linked
Data allows to uniquely identify each entity instance with an URI, encouraging
the creation of relationships among instances in order to discover patterns and
insights in a dataset. Labman is a web application developed in Python using
Django4, and is Open Sourced on its Github's repository page5, where it can be
downloaded and contributed to. Labman is developed to substitute a previous
Joomla! plugin developed at the research unit to publish publication data as
RDF [
This paper is structure as follows: First, we discuss similar e orts in section 2. Next, section 3 elaborates on the bene ts of publishing information as Linked Data. Section 4 exhibits how patterns and knowledge can be extracted thanks to visualization techniques. Finally, conclusions and future work are addressed in section 5. 2
Even though some plugins have been developed to publish data stored within CMS systems as RDF (Resource Description Framework) les and RDFa metadata6, they lack the ability to both make it accesible through a SPARQL endpoint (not allowing complex queries from external entities) and the advantages of publishing them following the Linked Data principles.
Research metadata visualization has also been studied by works such as [
The e orts of iMinds Multimedia Lab [
ResearchGate7 is a social networking site for scientist and researchers to share their work, providing metrics ands statistics to show their performance.
The focus is set on individuals to promote their work, whereas our proposal focuses on providing information on a research unit level basis.
Linked Universities, according to the de nition on their website8 \is an alliance of european universities engaged into exposing their public data as linked data". Specially focused on sharing educational data (e.g., courses, educational materials, teachers information, etc.), it also promotes the publishing of research and publication-related data. Linked Universities highlights the needs to have common shared vocabularies and tools to allow interoperability among how people access information about di erent institutions. Labman takes the recommendations from this alliance at its own core to avoid loosing the bene ts provided by shared standards and vocabularies.
Finally, VIVO9 is a huge project that provides an Open Source semantic web application to enable the discovery of researchers across institutions. VIVO allows any university to manage their data, and publish it using the VIVO ontology. VIVO is specially used among American universities. 3
Linked Data (LD) is a series of principles and best practices to publish data in
a structured way, encouraged by Tim Berners-Lee and the W3C [
The implicit linkage between resources in LD also allows to interconnect resources among them, e.g., a research project with the descriptions of people working on it and the related articles published as the outcomes of the study. Although this feature can also be achieved through plain relational database models, LD allows to connect references to external datasets, so complex queries can be performed in SPARQL, avoiding the potential headaches of joining consecutive SQL sentences and the need of having all the data in our system.
The \Open" term in Linked Open Data indicates that is freely available to everyone to use and republish data as they wish, without copyright and patent restrictions. All the information published on Labman is of public domain by default, making it freely consumable through its SPARQL endpoint. However, there is an option to mark a certain's project funding as private. If marked, this nancial information will be used for the generation of top level visualizations
(those which give a full view of the unit's performance), but no funding charts will be rendered for that speci c project and the funding amounts triples will not be generated. 3.1
To encourage the adoption of Labman among the Semantic Web community,
we have used well known vocabularies to describe the data of the di erent
entities in our data model. The Semantic Web for Research Communities (SWRC)
[
Labman stores data both in a relational database and as RDF triples (the relational database is used to increase performance and to allow non-semantic erudits to work with relational dumps). When an instance of any model is saved in Labman, a call is triggered to publish the instance and its attributes as RDF, generating or updating the referenced resource and its associated triples thanks to the rules of mapping speci ed for each model. Those triples are loaded into an Open Link Virtuoso14 instance to be later on accessible through the dedicated SPARQL endpoint15. Semantics can be enabled/disabled on demand for a full deploy of Labman through general settings (useful when installing a local instance of Labman to get a taste of the system and easing the transition from a legacy relational database model). A single management command allows to make a full dump of the relational database and publish it as RDF triples. The list of available extra commands within labman can be consulted through the {help modi er of Django's manage.py command line feature.
To help with publications data adquisition, a Zotero16 parser has been developed in order to extract all publication-related data and import it in Labman's system, publishing it as Linked Open Data using the previous described ontologies. Thanks to Zotero and the browser plugins, all metadata regarding a publication is extracted from well known publication indexing databases such as Web of Science, CiteSeer, Springer, Elsevier and so forth.
As the same authors may appear under slightly di erent names on di erent sites, Labman implements an author alias algorithm to perform term disam10 http://www.morelab.deusto.es/ontologies/swrcfe 11 http://www.foaf-project.org/ 12 http://bibliontology.com/ 13 http://muto.socialtagging.org/core/v1.html 14 http://virtuoso.openlinksw.com/ 15 http://www.morelab.deusto.es/labman/sparql 16 https://www.zotero.org/ biguation and apply the corresponding substitutions. This simple algorithm uses Python's di ib library17 to compare strings (e.g., author full names), taking as input string pairs of all the author names present in Labman, and returning a similarity ratio between them (as shown in table 1. If the ratio is greater than the given threshold, both strings are sent to Labman's administrators for dissambiguation checking. If the match is approved, the incorrect author name from the pair is assigned as an alias of the valid name. A periodic background task unlinks all the referenced triples to the invalid alias, and assigns them to the correct author resource.
The adage \A picture is worth a thousand words" ts perfectly when visualizing research related info, as the huge amounts of data related to projects, funding agencies, organizations, researchers and so forth makes them perfect candidates to be rendered using visual representations, instead of displaying all the information in text form without highlighting underlying connections.
In order to access and interact with the visualizations through any web browser, web graphics libraries such as Google Charts18, d3js19 and sigma.js20 have been used. Charts and graphs are rendered on the screen using JavaScript, with data extracted from Labman using Python.
Due to the interlinked nature of Linked Open Data, most visualizations showing linked entities are rendered as graphs and linked nodes. Visualizations are available on the Charts section21 and on the extended information subsections of the webpage. 4.1
Project managers and principal investigators usually depend on funds to continue their research work. In a transparency e ort, Labman allows to consult how much money is gathered from public entities, as displayed in gure 1. 17 https://docs.python.org/2/library/difflib.html 18 https://developers.google.com/chart/ 19 http://d3js.org/ 20 http://sigmajs.org/ 21 http://www.morelab.deusto.es/labman/charts/
Funds are provided by public administrations and organizations, usually under a named funding call. Labman also takes this information into account and allows to compare di erent calls' performances. For example, the principal investigator can view historical records from european FP7 and spanish INNPACTO funding calls to design the new budget strategy for the forthcoming years. Geographical scopes can be de ned and related to Geoname's22 feature classes, to classify funding call levels according to their e ect area. 4.2
Research would not be possible without the collaborations of di erent researchers working together to generate new knowledge. Being able to detect research networks is a fundamental insight to have always present, together with the communities of practice our unit takes part in and the evolution and the interactions with members of external disciplines.
In gure 2, a force directed graph is selected to represent project
collaborations present in the system. When hovering over an element, only the node's
community is visible, allowing to consult who each person is related with. Node's
size is calculated using Eigenvector centrality, a value which increases if the
connection with other central nodes of the graph is relevant, and the color of each
node indicates the community it belongs to. Community belonging is calculated
using modularity, and a di erent color does not mean they do not work for the
same organization, but that their connections make them beloging to a di erent
group of interconnected people. The calculations for generating these graphs are
further explained in [
Figure 3 shows the egonetwork [
Sometimes relations between researchers are not explicit (e.g., two researchers have not worked together in a project or co-authored the same paper, but both of them work in the same knowledge area). In order to identify these relations we have implemented a similarity coe cient using the tags of the papers of each researcher. To ascertain the similarity of one researcher with another we have devised the following formula: coef = jB \ Aj
A j j
Where A is the set of tags belonging to the base researcher and B is the set of tags belonging to the researcher which we want to compare the base researcher with. It must be taken into account that this similarity coe cient is not symmetrical. The reason is the topic similarity for a given researcher is considered within the whole of its tags, without taking into account the whole topics a related researcher works in. This situation is common for novel PhD students with a few published articles, who share all their topics with their advisors (due to their co-authorship), but senior researchers will have a broader set of areas they have worked on because of ther large research trajectory. Together with the identi cation of research networks, knowing which topics those networks and the involved researchers are working in is fundamental to understand the most relevant areas the group is focusing on. Projects and publications are tagged with concepts in Labman, published as dc:subject triples using the muto:Tag ontology. The rst obvious visual representation is to generate weighted lists (also known as tag or word clouds) of the topics used by a researcher. Figure 5 displays the topics used by one of our researchers, being the size of the tag representative of its weight (i.e., the bigger the tag, the more proli c in that area).
Research topic evolutions are also a good indicator to detect which areas
the group is focused on. The historical evolution helps understanding which
topics are no longer hot amongst researchers, and which topics have died to
evolve into new research areas (e.g., from Ubiquitous computing to Internet of
Things to Weareable computing ). ArnetMiner [
Eventually, establishing a robust taxonomy of topics leads to the identi cation of interest groups and expertise hubs around topics, allowing to relate researchers, projects and publications automatically where no previous obvious hints were available to connect them. Describing resources using shared vocabularies and connecting to external knowledge datasets allows to create these links in a global space, opening new doors to knowledge discovery thanks to the use of Linked Data principles. 5
Using dynamic visualizations of research information published as Linked Open Data, helps end-users (i.e., those consulting the information either from the research centre or visitors) to discover real connections between its members and the di erent entities modelled in the system. Visually representing which topics the researchers work on, who can be consulted about a certain area, and the historical collaborations within the group members can be of great help to guide the development of new project proposals, discover non-obvious potentials and address the key entities.
As future work, we will continue working on the generation of new visualizations to provide opportunities to improve the performance and strategic vision of a research centre. There is a strong continuous commitment to connect entities with external datasets, in order to evolve from a four-star to a full Linked Open Data data space, making Labman able to answer complex queries with data not present in our system, including other descriptions to the same resources available on the Web of Data. Finally, a ne detail level when well de ning and describing topics will allow for deeper analysis of data, taking into consideration the evolution of topics through time and how research areas are hierarquically structured. Actually, topics are cleaned and reviewed automatically on a regular basis to improve how resources are tagged. Better data completeness will lead to more enlightening reports, so automatizing even further the data adquisition stage will bene t all users. 6
The research activities described in this paper are partially funded by DeustoTech, Deusto Institute of Technology, a research institute within the University of Deusto and the Basque Government's Universities and Research department, under grant PRE 2013 1 848.