Content Assistance and Recommendations in Learning Material
A Folksonomy-based Approach
Benedikt Engelbert
1
, Karsten Morisse
1
and Oliver Vornberger
2
1
Faculty of Eng. and Computer Science, University of Applied Sciences Osnabrueck, Albrechtstr. 30, Osnabrueck, Germany
2
Department of Mathematics and Computer Science, University of Osnabrueck, Albrechtstr. 28, Osnabrueck, Germany
Keywords: Social Tagging, Recommender System, Learning Material.
Abstract: With the variety of Learning Materials (LM) available in Learning Management Systems and the Internet,
the time a student requires to select the most appropriate content increases. Especially the use of the Internet
to find new LM is time consuming and not necessarily successful. A study accomplished at our university
shows, that students mainly look for alternative explanations, content related exercises and examples, which
can be used in addition to the existing LM. In this paper we describe the System Learning Assistance
Osnabrueck (LAOs), which is based on a collaborative tagging approach with the main goals to give content
related assistance for available LM, but also recommend content in further LM e.g. from the Internet.
1 INTRODUCTION
In former times Learning Material (LM) at
universities covered usually lecture notes and
references to the library where further literature was
provided. In the information age the situation is
clearly different. The distribution of LM is
comfortable, since most universities provide a
Learning Management System. Students can access
digital lecture notes easily. Moreover the type of LM
is more manifold e.g. multimedia content like lecture
recordings or YouTube videos enrich the classical
lecture notes. Furthermore, the Internet expands the
available sources to countless. Many websites
provide open educational resources (OER: under CC
or GPL licence) or other LM (without any licence)
for free. For instance using the engine Google to
search for “algorithms” one will find within the first
results various lecture notes, books, and videos
available for free, but also links to other websites
with further LM and OER. A study conducted in a
computer science course at our university shows that
most of the students invest time to find additional
OER or LM on the Internet. Nevertheless, he or she
perceives the provided lecture notes within the
course as the major material to study with (Engelbert
et al. 2013). Another result was, that the quality of
the provided LM within the course plays only a
minor part and no matter what, students search for
additional LM to extend or complement given LM
with new examples, alternative explanations and
exercises. At this point we see the demand to enrich
given LM with additional information and content-
related connections to new LM or OER, to increase
the students proper use and understanding for the
major material. Further, we see a demand to simplify
the process of searching for additional material. It
has been shown that a huge amount of data and
information can lead to disorganization and mental
overload cp. (Agrawal et al. 2015). To overcome this
we developed a system called Learning Assistance
Osnabrueck (LAOs), which provides a process to
enrich genuine LM like text documents or video
material by a collaborative tagging approach. The
system takes advantage of tags in an adapted
folksonomy structure to subdivide LM into related
content areas. Those content areas can be enriched
by assistance information or can be connected to
other LM. The pedagogical use of tags to find or
discuss context in LM has been considered as a
proved method in several e-Learning scenarios (Fu
et al. 2007; Luo & Pang 2010). In the next section
we describe related work. In Section 3 the concept
and implementation of LAOs is described. We will
work out the goals of the system more clearly and
explain how these goals can be obtained. A formal
model to find related content areas in LM and to
calculate a rating within the recommendation
process is described in Section 4. We finish with
results of a first evaluation and a discussion of the
456
Engelbert, B., Morisse, K. and Vornberger, O.
Content Assistance and Recommendations in Learning Material - A Folksonomy-based Approach.
In Proceedings of the 8th International Conference on Computer Supported Education (CSEDU 2016) - Volume 1, pages 456-463
ISBN: 978-989-758-179-3
Copyright
c
2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
given approach and demonstrate further benefits of
the system.
2 RELATED WORK
Our work is related to several topics in the area of
Educational Data Mining (EDM), Recommender
Systems (RS) and Learning Analytics (LA), but also
to the topic of Social Tagging Systems (STS). In 2.1
we’ll give an overview for related work in the fields
of EDM, RS and LA under name of Recommender
Systems. We’ll discuss the topic STS in section 2.2
separately.
2.1 Recommender Systems
RS in general are software tools and techniques
providing suggestions for items to be of interest for a
user (Ricci et al. 2011). RS are also common in the
area of e-Learning and is such an important topic,
where hundreds of papers have been published
(Manouselis et al. 2013). Dealing with content-
related recommendations or assistance is a smaller
domain. Possible approaches vary strongly regarding
to its aims and techniques, which have various pros
and cons. We will briefly discuss those approaches
related to ours. Similar to our approach is the idea of
recommending Learning Objects (LO), where a LO
is the smallest reasonable learning unit. LOs can be
recommended on the basis of a user profile, where
the user profile contains the current state of students
knowledge (Singh & Khanna 2014). LOs are
convenient to cover a certain context and can be
properly assigned to a current state of students
learning progress. Therefore recommendation can be
generated rather easily. The main disadvantage of
LOs is the necessary amount of metadata to make
LOs recommendable (Niemann 2015). Lecturers
may not invest the time to generate them. There are
several articles that apply the use of tags in a RS for
learning. In (Mohsin 2010; Yu & Li 2009) systems
are presented where students can add tags to
websites or LM to describe them more precisely. In
(Mohsin 2010) the tags are used in the
recommendation process to find similar websites to
study with. The approach in (Yu & Li 2009) takes
advantage of tags to help students organizing
documents. (Broisin et al. 2010) presents another
tag-based system to recommend websites. The
approach analyses the tag activities of users and tries
to find similar user groups within the system. On
this basis the system can derive website
recommendation within a user group. The systems in
(Mohsin 2010; Yu & Li 2009; Broisin et al. 2010)
maintain the approach to describe LM on the basis
of tags more precisely. The main downside of those
systems is that the tags refer to an entire document
and specific content within a document cannot be
recommended. The work in (Purwitasari et al. 2011)
however focuses on the idea that students can add
tags to a specific content within a single document,
where the system provides tag recommendations to
help students to find the correct context. Thus the
system recommendations contain context related
information, but cannot recommend the content
itself. In (Machardy & Pardos 2015) a framework to
evaluate the relevance of video resources in MOOC
scenarios is presented. The authors consider the use
of Bayesian Knowledge Tracing (BKT) to trace user
behaviour and derive resource relevance. The use of
implicit behaviour tracking is a reasonable method
and is also considered in our work. However the
approach is suitable for smaller learning resources
like LOs. To recommend resources in dependency of
a learning path is another common approach (Pan &
Hawryszkiewycz 2004). The idea is to design a
learning path depending on a course curriculum and
connect the learning path to suitable LM. Related to
the problem mentioned for LOs already, preparation
time to construct a learning path is time consuming
for the teacher. An entirely semantic approach can
be found in (Heim et al. 2009), where the system
finds similarities in text based resources. Therefore
multimedia documents like audio or video
documents cannot be considered.
2.2 Social Tagging System
In Social Tagging Systems (STS) users can add
freely chosen tags to categorize resources. A
folksonomy is the underlying structure of a STS and
describes the users, tags, and resources, and the user-
based assignment of tags to resources (Hotho, R
Jäschke, et al. 2006). We see the mapping of users,
tags and resources in a folksonomy as the most
promising structure to derive the information we
need to reach our objectives (cp. Section 3.1). The
work from Hotho & Jäschke reflects the idea, that a
resource tagged by important users gets important
itself. The main goals are to search for resources, but
also to apply a ranking which of the resources are
the most important ones. For folksonomies there has
been made some research to the use in e-Learning
scenarios. The system in (Dahl & Vossen 2008)
makes use of a folksonomy in a metadata repository
to easily navigate between learning resources. A
similar approach is presented in (Anjorin et al.
Content Assistance and Recommendations in Learning Material - A Folksonomy-based Approach
457
2011). On the basis of a folksonomy structure the
systems predicts a ranked list of important resources.
STS have been established to make a user-based
classification of resources and therefore seems a
promising technique to categorize not just resources
but also content within resources. We see a lack to
combine RS with a folksonomy structure to make
content related recommendations and give content
related assistance in LM. In Section 3 we will
describe such a system, however we modified the
common use of a STS with it’s free shaped tags to a
more restrictive approach.
3 SYSTEM GOALS AND
OVERVIEW
In the upcoming section we describe the concept of
LAOs. First, we will work out the main goals of the
systems and how they can be fulfilled, before we
give an overview of the system implementation in
the second part of the section.
3.1 Goals and Requirements
In Section 1 we already stated the problems students
could have when using different LMs. We see the
right selection and the sufficient examination of LM
as the main difficulties for students. Reasons for this
are the huge amount and the variety of LM and
OER, which can be accessed on the Internet. The
limited time a student spends to examine new
material is another main issue. Therefore, we
consider the following goals to help students to get a
better understanding for LM:
• Find content, which is important in the
current state of learning,
• Provide content related assistance,
• Recommend content related additional LM or
OER, to reduce the time a student spends to
search for it.
We further consider the following goals to give
lecturers the possibility to analyse how students use
LM and obtain students feedback on LM content:
• Present content, students have issues to learn
with,
• Present content, which is appropriate for the
students,
• Present statistics on how students uses LM,
• Present LM, which has been used
additionally.
To fulfil these goals we propose the use of a
collaborative tagging approach, where users can add
tags (metadata) to any content within the LM. The
user-generated tags can be seen as an explicit user
feedback to express a student’s opinion on certain
content. In addition we provide implicit user
feedback, which can be derived from the user’s
behaviour (e.g. how long a student used a particular
LM). The implicit feedback helps to generate LM
statistics and to calculate the relevance of the LM. In
section 3.2 we will present both types of user
feedback more precisely. The tag-based approach
leads to the advantage that complex content like
lecture notes or multimedia content can be evaluated
properly. Since tagging based approaches has been
successfully used to classify entire documents, the
technique seems to be promising for a content
related classification cp. (Broisin et al. 2010). We
further assume that the collective intelligence of a
user group helps to identify difficulties or utilities
within LM. The assumption is that if a single student
perceives a LM content to be difficult, important or
helpful others may perceive the same.
3.2 System Implementation
The system can be divided into a tagging- and an
analysing-component. The tagging component
mainly implements the user interface of the system.
The analysing component covers the data analyses
and recommendation process. For the
recommendation process we implemented a method
to extract and rate content from LM. We will give a
detailed explanation of the analysing component in
section 4.The tagging component provides a web-
based tagging feature for text and slide documents,
but also for multimedia documents like video files.
As stated in section 3.1, we distinguish between
explicit and implicit user feedback. In the following
we denote the user feedback as explicit and implicit
tags respectively. With explicit tags, users can
classify LM content according to their opinion by an
explicit user statement. For this we implemented
several tagging features like comment-, rating- or
marker-tools. Furthermore, users can add new LM or
OER to the system or can create an explicit
connection between two content sections. Implicit
tags can be derived from the user behaviour. Implicit
tags help to find important sections in LM, which we
will call content relevance (see section 4.1.2).
Moreover, we make use of implicit tags to create
implicit content connections (cp. section 4.3). An
overview of all available system tags is presented in
Table 1.
CSEDU 2016 - 8th International Conference on Computer Supported Education
458
Table 1: System Tag Overview. E=Explicit, I=Implicit.
Tag Description
Pre-defined Text
Comment (E)
Pre-defined Text Comment (9x positives, 9x negatives)
Quick Tag (E)
Thumbs up/down from a category: Importance, Usefulness,
Understanding, Difficulty
Rating (E) General Rating between 1 and 5 Stars
New Material (E) Add a new Material e.g. from the Internet
Material
Connection (E)
Content related Connection within the same LM or between two
different LM within the system
Marker (E) Marker to mark Content Areas
Page Hop (I) Jump between Pages in a Text Document
Timeline Hop (I) Jump on Timeline in a time-based Document (e.g. Video)
Material Hop (I) Jump between Materials
Residence Time (I) Time a user spend in a certain area of a LM (e.g. text page)
Use Flag (I) Increments the count of LM uses
Different to other tag-based approaches, our
system provides a fixed set of explicit tag types. It is
therefore not foreseen for the user to define
individual tags. However, it is necessary that the
system can interpret the content of a tag properly.
This would be more difficult if user’s degree of
freedom is too high. Furthermore, students are more
encouraged to use tags, if there is a set of pre-
defined tags available e.g. (Fu et al. 2007).
Technically tags hold a pre-defined non-integer
value between +1 and -1 (tag score). Tags with a
value of +1 indicate a total positive statement; tags
with -1 indicate a total negative statement.
According to the clarity of a statement, the tag holds
a higher (or lower) value (e.g. “The content at this
position is important for the current assignment”
(+0,9); “The content at this position is vague” (-
0,5)). Furthermore, every user is assigned to a non-
integer value between +1 and 0, which we call user
score. A user score depends on how the tags of a
user fit into the amount of tags of a whole group.
Since the system provides an assessment feature, we
also consider adjusting the user score on the basis of
taken assignments. Both – the tag and user score –
are used to evaluate and classify content in LM.
Therefore, users with a higher user score obtain a
higher impact when adding tags to content. We
denote the clustering of tags, which reveals to
certain LM content as content resources. More
precisely the LM is the resource where users assign
tags to and based on the set of tags, the LM will be
divided into content-related resources or sub-
resources. Figure 1 reflects the idea of how the
relation between users (u) and tags (t) and their
values affect the importance of (sub-) resources (r).
To simplify the idea, the figure shows three identical
tags on two sub-resources assigned by four different
users. The size of the circles corresponds to the
respective user-score, tag-score and resource values.
Figure 1: System Implementation Overview.
Figure 1 shows how the value of a sub-resource
relates to a user-score and a tag-score. As already
stated, sub-resources within LAOs are a content
extract or area from a LM. For example it can be a
paragraph in a text document or a timeline section in
a time-based media. To make content extractable,
each tag holds a multimedia coordinate, which can
vary between the different types of LM. In a text
document the coordinate is mapped to X- and Y-
coordinates on a text page. In time-based LM (e.g.
Video) the coordinate is represented as a timestamp.
4 EXTRACT AND RATE
RESOURCES
Section 3 reflects the system concept and
implementation. We proposed the main idea and
defined scores for users and tags. We further stated
how the relation between user-scores, tag-scores and
resources work out. Due to this, we propose in the
following the method to extract content resources
from a LM or an OER for a given set of tags.
Further, we present a formal definition to calculate a
rating for such resources to make it recommendable.
4.1 Extracting Resources
As stated in section 3.2 we extract content from a
LM or an OER by clustering tags. We denote a LM
or an OER as a resource and a clustered set of tags
as a sub-resource. To do so, we make use of the
multimedia coordinates defined in section 3.2.
Overall there are three steps required, which we will
introduce in the following section.
4.1.1 Step 1: Clustering Tags
In the first step we cluster tags to temporary sub-
resources. For this, we consider an ascending
ordered set of tags according to their multimedia
coordinate. In text documents we certainly order
according to the Y-coordinate. Figure 2 (left side)
illustrates a set of tags in a text document. At first,
we calculate the distances between two tags whose
multimedia coordinates are immediately
consecutive. With all the distances between tags we
calculate a mean distance with
Content Assistance and Recommendations in Learning Material - A Folksonomy-based Approach
459
Mean Distance
=
∑
Distances between Tags
Number of Tags
(1)
We define two tags as neighboured if the
distance between both tags is less than the mean
distance. Beginning with the first tag in the ordered
set, we validate if the upcoming two tags are
neighboured to each other. As long as the tags are
neighboured, they can be clustered to a temporary
sub-resource. If two tags are not neighboured, the
clustering for the current temporary sub-resource is
completed. The approach is illustrated in Figure 2
(right side).
Figure 2: Clustering Neighboured Tags.
4.1.2 Step 2: Validate Sub-resources
Now in step 2 we have to validate the relevance of
the temporary sub-resources. First we check if the
amount of clustered tags for a sub-resource is big
enough. For our model, the condition in equation 2
needs to be satisfied
Number of Tags for atemp. Sub-Resource
NumberofallTags
N
umber o
f
all temporary Sub-Resource
(2)
Afterwards we need to examine if the sub-
resource is in a relevant area, where an area is part
of a LM or an OER e.g. a text page. We denote an
area as relevant if the residence time (the time all
users spend in this area derived from implicit tag)
and the tag appearance (all tags in this area derived
from all explicit tags) exceeds the respective mean-
score with
Residence Time Mean
∑
ResidenceTimeofeachLMarea
Number of all LM areas
(3)
Tag Appearance Mean
∑
TagsofeachLMarea
Number of all LM areas
(4)
4.1.3 Step 3: Clustering Sub-resources
The third step is to investigate if two sub-resources
can be clustered again. This can be done if sub-
resources are close enough according to their
multimedia coordinates. Clustering two sub-
resources basically means to merge both respective
tag sets. Similar to step one, we first determine the
distance between all available sub-resources in a
certain area. For this, we use the upper/lower tag
coordinate of the respective sub-resource. Using the
mean distance score for all sub-resources with
ResourceMean Distance=
∑
Distance between Sub-Resources
Number of all Sub-Resources
(5)
we examine if two sub-resources can be merged to a
single one. After the third step the sub-resources are
finalized and can be classified. This process will be
described in the upcoming section 4.2.
4.2 Calculate Resource Score
In section 3.2 we already described the relation
between users, tags and (sub)-resources. In the
classifying step we now make use of this approach.
For this, we switch to a more formal description. Let
users , tags and sub-resources are finite sets,
whose elements are called users (u), tags (t) and sub-
resources (r). Let
∆
⊆ be a relation
defined by
∆
,,|∈,∈,∈,
useruassignedtagttosub-resourcerattime∆
(6)
For simplicity we assume that
∆
will always be
considered in the current system state so that we use
the simplified notation for the cumulative
∆
. We
consider the function
:→0;1 to determine
the user-score for user ∈. Further, we consider
the function
:→1;1 to determine the tag
score for any ∈. Let ∈ be a sub-resource in
the system. We use a matrix notation to map the
relation between a sub-resource r and the elements
of the sets and . So let
be
,
,
⋯
,
⋮⋱⋮
,
⋯
,
, with
1
|
|
,1
|
|
, where
,
|
,
,∈∃∈|
(7)
is the number of tags
assigned by user
to a sub-
resource . Let
be a function
:→1;1 to
determine a score value of a sub-resource ∈
which describes a positive or negative statement
strength of a sub-resource defined by
∑∑
∗
∗
,
||
||
∑∑
∗
,
||
||
(8)
4.3 Making Predictions
In section 4.1 we described the extraction process
for sub-resources in LM or OER and in 4.2 we
presented a model to calculate the resource score for
any extracted sub-resource. In the following section
we will describe how the extracted resources and the
respective score can be used to present additional
information in LM. This will be done separately for
students and lecturers.
CSEDU 2016 - 8th International Conference on Computer Supported Education
460
4.3.1 Students
As stated in section 4.2 the resource score of
for any sub-resource ∈ holds a non-integer value
within [-1,+1]. According to a positive or negative
statement, the function
yields a higher
positive resp. negative value. With the tag context
assigned to the sub-resource, it is possible to derive
an assistance statement for any sub-resource and the
related content. Therefore, students get information
for relevant content, which can be used to prove
their existing knowledge. Figure 3 shows an
example of a difficult content within the system.
Furthermore, sub-resources will be connected to new
LM or OER, which has been added to the system by
the user group. E.g. difficult content can be
complemented with LM that helped other students
already (cp. example in Figure 3). Additionally, the
system adds content related connection between the
LM or OER, which is already available in the
system.
Figure 3: Content Information in LAOs.
This is possible by using the explicit material
connection tags from the user group. Further the
system can derive implicit material connections
whilst analysing the material usage of each user. In
Figure 4 the material use over the time is shown.
Each colour presents another LM or OER.
Figure 4: Material Use over the time.
With the switch between two LM and the related
tag context, an implicit material link can be derived.
The recommendation process covers mainly the
notification of the extracted and rated sub-resources.
This is necessary if
,
a student made a contrary
statement, a student stated content as difficult (both
with explicit feedback), a student did not use certain
LM content or a student seems to be confused (both
with implicit feedback).
4.3.2 Lecturers
From the lecturers point of view the system gives
basically an overview on the students’ activities.
There are several benefits that come along. Firstly,
lecturers get information about how students access
given LM or OER. Among statistics how often and
how long they used the different LM, the lecturer
has access to the detailed user behaviour mentioned
in Figure 4. With the using behaviour it is possible
to trace back the students’ activities and to derive
possible weaknesses. It is worth mentioning that the
user’s information is anonymous. Certainly the
lecturer has access to all extracted sub-resources and
calculated statements. Therefore, he or she can judge
how good or bad the student group can work with
the LM. There is also the possibility to access the
new LM and the explicit LM connections, which
have been added by the students. This helps to
review, which LM is used by the students
additionally and to review if the students find correct
context between various LM.
5 EVALUATION
In the following we describe a first evaluation
setting, which has been used to collect data for
recommendation purposes. A proper data collection
is necessary to show the functionality of the
approach presented in section 4 initially. To do so,
we set up LM from Algorithms & Datastructures, a
classical course within computer science programs,
and 35 students are requested to solve problems out
of that area. An exercise sheet with 20 exercises has
been presented. The exercises covered five different
topic areas, where the difficulty level ranged
between easy and slightly difficult. In the exercises
we asked for factual knowledge, however the
students had to solve algorithmic problems. The
exercises were prepared with the help of an
algorithms lecturer. It was a conscious decision to
present exercises from different topic areas in
different difficulty levels for various question types.
Students should work in different areas of the
material to ensure a distributed data collection. Thus
we are able to make a qualified statement about the
implemented approach. Among lecture notes from
algorithm class, we provided an algorithm book and
Content Assistance and Recommendations in Learning Material - A Folksonomy-based Approach
461
for each topic area one video with the length of 5-15
minutes. The students were allowed to use the
Internet. Nevertheless, it was a requirement to add
the used additional LM to the system. We
accomplished the study in a computer lab with a
time limitation of 90 minutes. Each student worked
alone on the exercises. All together the 35 students
worked 2218 minutes with the given LM. We
received 847 explicit and 7104 implicit tags. 43 new
LM were added to the system. The first outcome,
which seems to be evident, is the significance of the
lecture notes. With 1600 minutes students used the
lecture notes clearly the most. Certainly we
motivated this behaviour in section 1 already. On
average the videos were used just 32 minutes. This
is mainly because of the length of the videos. Only
the provided book was barely noticed. In form of an
expert analysis we evaluated the outcome of the
system. It is conspicuous that the system was able to
work out relevant content. For each content area
sufficiently large set of information had been
extracted. The extracted content was necessary to
solve the exercises properly. This applies for the
lecture notes and the videos. In the book the system
extracted some useful information, which has been
seen as an addition anyway. Especially for the more
difficult exercises the negative statements become
more frequent. However, the additional LM
becomes more frequent equally. This is not
surprising, since students are looking for easy or
alternative explanations cp. (Engelbert et al. 2013).
With the given results the functionality of the system
seems promising to achieve the goals proposed in
section 3. Especially the extraction of useful or
difficult content is working well. Also the number of
added LM is high and adequate enough to enrich the
given LM. In a second evaluation step in summer
2016 we will verify if students resemble the same.
For this, we will ask students to evaluate the
extracted content and further LM according to the
proposed exercises.
6 CONCLUSIONS
In this paper we presented the system LAOs. The
main goal is to assist students in the use and retrieval
of LM or OER. We described an approach on the
basis of user assigned tags in LM and the analysis of
the gathered information. Furthermore, we described
a first evaluation setting, which was intended for
collecting data. With an expert analysis we were
able to approve the proper functionality of the
system. The system extracts content according to
given exercises in a useful manner. Also the
implemented functionality for lecturers to analyse
the student’s use with LM satisfies the expectations.
We assume that the functionality will support
lecturers in getting a better understanding for the
student’s needs and weaknesses regarding to LM.
Nevertheless, it is necessary to show the usefulness
of the system outcome. This has to be proved in an
upcoming evaluation, which focuses on the
validation for recommendations from the student’s
point of view.
REFERENCES
Agrawal, A., Leonard, S. & Paepcke, A., 2015. YouEDU :
Addressing Confusion in MOOC Discussion Forums
by Recommending Instructional Video Clips. In
Proceedings of the 8th International Conference on
Educational Data Mining. pp. 297–304.
Anjorin, M., Rensing, C. & Steinmetz, R., 2011. Towards
ranking in folksonomies for personalized
recommender systems in e-learning. CEUR Workshop
Proceedings, 781(October), pp.22–25.
Broisin, J. et al., 2010. A personalized recommendation
framework based on cam and document annotations.
Procedia Computer Science 1, 1(2), pp.2839–2848.
Dahl, D. & Vossen, G., 2008. Evolution of learning
folksonomies: social tagging in e-learning repositories.
International Journal of Technology Enhanced
Learning, 1(1/2), p.35.
Engelbert, B., Morisse, K. & Vornberger, O., 2013.
Zwischen Nutzung und Nutzen – Die Suche nach
geeigneten Lernmaterialien und deren Mehrwerte im
Kontext einer Informatikveranstaltung. In GMW 2014
- Lernräume gestalten - Bildungskontexte vielfältig
denken. Zürich, pp. 508–519.
Fu, X. et al., 2007. Video Annotation in a Learning
Environment. Proceedings of the American Society for
Information Science and Technology, 43(1), pp.1–22.
Heim, P. et al., 2009. Semantisch unterstütze
Informationsextraktion aus Dokumentenmengen.
Hartmut Wandke; Saskia Kain & Doreen Struve, ed.,
“Mensch und Computer 2009”, pp.415–418.
Hotho, A., Jäschke, R., et al., 2006. Folkrank: A ranking
algorithm for folksonomies. Proc. FGIR, pp.2–5.
Luo, G. & Pang, Y., 2010. Video annotation for enhancing
blended learning of physical education. Artificial
Intelligence and Education (ICAIE), …, pp.761–764.
Machardy, Z. & Pardos, Z.A., 2015. Evaluating The
Relevance of Educational Videos using BKT and Big
Data. In Proceedings of the 8th International
Conference on Educational Data Mining. pp. 424–
427.
Manouselis, N. et al., 2013. Recommender Systems for
Learning, New York, NY: Springer New York.
Mohsin, S.F., 2010. Web based Multimedia
Recommendation System for e-Learning Website.
CSEDU 2016 - 8th International Conference on Computer Supported Education
462
Journal of Advanced Networking and Applications,
223, pp.217–223.
Niemann, K., 2015. Increasing the accessibility of learning
objects by automatic tagging. Proceedings of the Fifth
International Conference on Learning Analytics And
Knowledge - LAK ’15, pp.414–415.
Pan, W. & Hawryszkiewycz, I., 2004. A method of
defining learning processes. Beyond the 21st
ASCILITE Conference, pp.734–742.
Purwitasari, D. et al., 2011. Ontology-based annotation
recommender for learning material using contextual
analysis. In Proceedings of the IETEC’11 Conference.
Kuala Lumpur, Malaysia.
Ricci, F. et al., 2011. Recommender Systems Handbook.
In F. Ricci et al., eds. Recommender Systems
Handbook. Boston, MA: Springer US.
Singh, T. & Khanna, S., 2014. Reinforcement learning
approach towards effective content recommendation in
MOOC environments 1. In IEEE International
Conference on MOOC, Innovation and Technology in
Education (MITE), 2014. pp. 285–289.
Yu, L. & Li, Q., 2009. Personal Media Data Organization
and Retrieval in e-Learning : A Collaborative Tagging
Based Approach. MTDL, pp.1–7.
Content Assistance and Recommendations in Learning Material - A Folksonomy-based Approach
463
