Social Web applications such as "Flickr", "Youtube" and "Slideshare" o er a vast body of multimedial knowledge, discoverable through the appropriate search interfaces and API's. This extensive information source, however, is largely unstructured and the available metadata is typically limited to title, tags and description for a resource. On the other hand, Linked Web Data is both structured and well described through a variety of metadata. Combining those sources opens promising direction for knowledge discovery and, at the same time, new challenges for collaborative searching in various Technology-Enchanced Learning Scenarios. In this paper, we explore how to support (collaborative) search in such scenarios through an initial analysis of the Web data landscape and introduce early results from e orts on exploiting Linked Data techniques to solve critical issues in this context.
Learning traditionally has been perceived as a process taking place in some form
of gated communities, where the class-room is increasingly being supplemented
by Technology-Enhanced Learning (TEL) environments [
The open educational resources (OER) movement already partially
considered the liberalization of education and successfully generated an unprecedented
range of standards and content, including examples such as MIT Open
Courseware1 (OCW), GLOBE2 (Global Learning Objects Brokered Exchange) or the more
recent MOOC (Massively Open Online Courses) movement. Still, these
initiatives tend to perceive learning as a process somewhat disconnected from
noneducational activities, leading to a limited scope and take-up [
Web-based knowledge acquisition is characterised by ne-grain and highly
diverse knowledge items, ranging, for instance, from Wikipedia articles and
YouTube videos to scholarly papers and semi-didactic material, such as
usergenerated slidesets. In this context, non-educational information Web resources,
in particular user-generated content, gained similar importance as dedicated
educational material. This development has been fundamentally driven by
technological advancements, such as the emergence of social media together with a
Web of Data which allowed an unprecedented body of knowledge to be reused
and shared across the Web. Technological drivers of this evolution are, for
instance, open APIs and interfaces, and more recently, the Linked Data (LD)
[
While the emerging new forms of (informal) learning require a new set of
techniques and skills, prevalent Web Data present themselves also as an
unprecedented resource for deriving and detecting new pattern and theories about
Web-based informal learning activities. Web Science [
This has recently led to a growing inter-disciplinary community of researchers
from data engineering, Semantic Web, education and social sciences, converging
in disciplines such as Learning Analytics [
In this paper, we argue that Web Science and Linked Data provide resources and methods for (a) analysing, detecting and documenting the ongoing paradigm shift as well as the relevant data landscape and (b) acting as facilitator by resolving interoperability issues which naturally arise when adopting the broad, diverse and less restricted view sketched above. To underline the emerging Websupported practices in learning, we rst introduce results from a study analysing (social) Web information and data for its educational relevance, scope and diversity. We then propose early research results, emerging from the LinkedUp project7, which aim at applying and exploiting LD (principles) together with data analytics approaches to facilitate new forms of Web-based learning by adopting a broader view on educationally relevant resources. 2 (Social) Web Data for Learning: Exploratory
Investigations
Data and their analytics facilitate education on a variety of levels, for instance,
by providing (a) knowledge and information resources as input to learning
processes, (b) social and personal data which enable the investigation and detection
of emerging patterns, (c) social and personal data for use in (learning analytics)
and recommender system scenarios. In particular with respect to (a), it is
assumed that Web data and content, in particular social media and user-generated
content, provide valuable input for knowledge acquisition and informal learning
[
M1: R (% of total)
Thus, this historical data of LearnWeb provides useful information to estimate the educational relevance of particular Web data sources. Our primary research question addresses the perceived educational relevance of Web resources and the diversity of sources, as an indicator for the educational relevance of social media and Web data in general.
Bing
Flickr ipernity
SlideShare
Vimeo
YouTube
For this purpose, for each Web source Si we investigate 3 measures as indicator for educational relevance. (M1) is the proportion (%) of the overall repurposed resources R retrieved from Si, (M2) the ratio of repurposed resources R' from Si per 1K submitted queries, and (M3) the ratio of high-rated, repurposed resources R' from Si per 1K queries. With respect to M3, we consider a resource as high-rated if its average user-assigned score is equal or above the threshold T (here T=2,5; rating scale integers 1..5). Results are depicted in Figure 1 and Table 1. Note, though typical queries, e. g. "astronomia aurora boreale", were de ned with the intention of nding educational material for a certain topic, mostly these appear to not contain any indicator of the learning intention.
Additionally, users were able to manually add links to useful informal
learning resources (category "Links" in Table 1). A breakdown of the most frequently
provided domains con rmed results from Table 1 by revealing YouTube as most
frequent source (416), while the remaining domains varied heavily (Figure 2),
including a range of museums, media broadcasters or educational sources.
Although the investigated samples are rather small, the results of our preliminary
investigation suggest that, in particular social media sources such as YouTube,
SlideShare or Flickr contain a high proportion of material of use for knowledge
acquisition and skill development, while a widespread variety of highly diverse
sources provides additional resources on more speci c topics or domains. We are
currently conducting a more elaborate study based on a wide range of structured
datasets retrieved from the DataHub9. While the DataHub contains 57 datasets
which are explicitly tagged with the term "education", our current studies
consider not explicitly labeled educationally relevant datasets, such as Europeana11,
BBC Programmes [
Facilitating Knowledge Exploration - Towards an educational Knowledge Graph Having investigated on the educational relevance and diversity of Web resources, we argue that Web-based informal learning calls for support beyond the scope of traditional educational technologies, but through aiding learners in exploring Web data and content. 11 http://www.europeana.eu/ 12 http://wordnet.princeton.edu/ 13 http://dbpedia.org
http://kidshealth.org http://www.childrenoftheearth.org http://learnenglishkids.britishcouncil.org http://www.museumofbrands.com
http://www.ducati.it http://www.nationalgeograohic.com
http://www.ecokids.ca http://museo.ferrari.com http://www.slideshare.com http://www.metmuseum.org http://www.deutsches-museum.de http://www.oxfam.org http://www.google.it http://www.mocp.org/ http://www.moma.org/ http://en.cyberdodo.com http://www.bbc.co.uk http://www.flickr.com 5 5 5 5 5 5 5 5 5 6 6 6 7 7 8 9 10 0 5 12
16 15 20
This requires taking into account the diversity of knowledge on the Web and
to make it accessible and digestible from an educational perspective. While the
latter requires the support of learning not by a narrow set of TEL technologies
but by adopting a more general view on Web data, these challenges can only
be addressed by tackling substantial issues, for instance, with respect to quality,
heterogeneity and interoperability of used vocabularies, languages or schemas.
In this context, particularly LD techniques - fundamentally aimed at knowledge
reuse and sharing on the Web - o er potential to act as facilitator. The
opportunities arising from the LD approach are two-fold: (i) LD principles have emerged
as defacto-standard for Web data and knowledge sharing in particular also in
the eld of education [
As part of the LinkedUp project17, we currently conduct a number of data curation activities, aimed at assessing, cataloging and exposing all sorts of Web data of educational relevance (independent of their original intention) where the overall vision entails the creation of an educational knowledge graph which enables learners to explore all forms of suitable Web data and content. This work is two-fold: { Community-oriented data cataloging on the Data Hub: similar to the successful approaches of the Linked Open Data community e ort, a dedicated group ("linked-education"14) is currently being maintained and populated to gather and tag educationally relevant data { Semantic, syntactic and infrastructural alignment : selected datasets will be annotated with a speci c vocabulary which allows data to be exposed in more coherent and accessible ways, aiming to bridge between multiple languages and descriptive approaches.
As part of the latter a general vocabulary ("linked-education" vocabulary15,
adopting VoID16) for the description, cataloging and alignment of
educationally relevant datasets is under development. An initial classi cation of datasets
into educationally relevant categories (such as "educational resource",
"scholarly paper" "video lecture") provides suitability indicators for particular tasks.
While we intend to apply the schema in combination with data interlinking
techniques to provide a learning-oriented view on Web data in general, it is
currently being applied initially to an exemplary dataset17 containing 5,953,623
distinct resources (around 60 million RDF statements) extracted from di erent
sources (Table 2 depicts the sources along with the number of avalable resources).
Sources were selected to re ect the diversity documented above and include
Europeana, BBC Programmes, mEducator Linked Educational Resources [
http://data.linked.education.org/openrdf-sesame/repositories/linked-learning 18 http://acm.rkbexplorer.com/ 19 http://dblp.l3s.de/ In this paper, we have motivated the need for a broader view on Web-based education and in particular educational technologies and provided insights on how Web data in general can facilitate this paradigm shift. In particular Linked Data, o ering a body of structured knowledge and set of techniques, together with traditional collaborative search and data engineering methods can act as a means to integrate learning-related knowledge into a coherent educational graph and support enrichment of the corresponding resources with useful metadata. More recent Microformats and e orts like schema.org20 can further bridge between the structured Web of data and materials on the Web such as web pages, images and videos by providing a common vocabulary enabling search engines to discover unstructured resources.
Our investigation points out promising directions for larger substantial studies to address issues with regard to the diversity of data quality, domain coverage and varying levels of trust, particularly when dealing with distributed data. These represent fundamental obstacles for data consumers in general and apply even more in learning scenarios. Additional and so far under-acknowledged issues arise from the diversity of licensing schemes and policies adopted by various distributed data providers. Hence, highly multidisciplinary communities are required and currently assembled, e. g., by the LinkedUp project, to advance the rede nition of learning on the Web through exploitation of existing Web knowledge and data. 20 http://schema.org: a joint e ort of, for instance, Bing, Yahoo, Google or Yandex to describe common Website markup vocabularies using microdata