=Paper=
{{Paper
|id=None
|storemode=property
|title=A Comparison of Usage Data Formats for Recommendations in TEL
|pdfUrl=https://ceur-ws.org/Vol-896/paper9.pdf
|volume=Vol-896
|dblpUrl=https://dblp.org/rec/conf/ectel/NiemannSW12
}}
==A Comparison of Usage Data Formats for Recommendations in TEL==
https://ceur-ws.org/Vol-896/paper9.pdf
An Overview of Usage Data Formats for
Recommendations in TEL
Katja Niemann, Maren Scheffel, Martin Wolpers
Fraunhofer FIT, Schloss Birlinghoven, 53754 Sankt Augustin, Germany
{katja.niemann, maren.scheffel,
martin.wolpers}@fit.fraunhofer.de
Abstract. Recently, a number of usage data representations have emerged that
enable the representation of user activities across system and application
boundaries. Based on these user activity data, systems can adapt to the users
and provide personalized information. A lot of usage data representation for-
mats are already successfully used in real world applications. However, de-
pendent on the purpose, the formats show different advantages and disad-
vantages one must consider when choosing a format for a system. In this paper,
we will present the four most commonly used data representations, namely
Contextualized Attention Metadata, Activity Streams, Learning Registry Par-
adata and NSDL to alleviate the selection of a suitable format.
Keywords: usage data formats, technology enhanced learning
1 Introduction
Attention or Usage Metadata represent the activities of users and their usage of da-
ta objects in specific applications. Aggregating and analysing the usage data provides
the basis for advanced user support systems, e.g. learning recommendation or self-
reflection support. Furthermore, usage data can be employed for annotating data ob-
jects with information about their users and usages, thereby rendering possible object
classifications according to use frequency, use contexts and user groups [1], [2] [3].
Particularly in the domain of learning analytics (see [4], [5] and [6] for more in-
formation on learning analytics) and educational data mining, usage data provide the
basis for learning support systems. For example, based on an analysis of usage data,
irregularities of learning behaviour of students can be identified [7] and the results of
corrective activities by the teacher can be monitored. Another example of the success-
ful application of analysing usage data in learning settings is the reflection and com-
parison of learning activities among students of a learning group. Here, by playing
back their learning activities, students compare themselves with their fellow students
and identify how to improve their learning activities. A further example of the suc-
cessful use of usage data are personalized recommender systems, e.g. in the domain
of learning (see [8] for more details on recent learning recommendation systems).
RecSysTEL 2012 95
Recently, a number of data representation formats for usage data have emerged. In
contrast to simple logging files, these representations focus on the activities of users
and not on those of a system. In this paper, we will present the most prominent exam-
ples, namely Contextualized Attention Metadata, Activity Streams, Learning Registry
Paradata and NSDL Paradata.
2 Usage Data Formats
2.1 Contextualized Attention Metadata
The CAM scheme [9] was defined as an extension of Attention.XML [10] which is
an early approach to capturing and storing attention metadata for single users. In the
current CAM version1, the focus has moved from the user and the data object to the
event itself. This is due to the insight that not every event has a fixed set of attributes.
Fig. 1. CAM scheme
Additionally, only the basic information about an event is stored, e.g. the event
type and the time stamp. All other information, e.g. metadata describing users or doc-
uments involved in the event, are linked. In this way, each entity and also each ses-
sion can be described in a different and suitable way and no information is duplicated.
Fig. 1 shows the complete CAM scheme. The main element of a CAM instance is
the event entry which comprises its id, the event type, the timestamp, and a sharing
level reference. Examples for event types are “send“, “update” or “select”. The shar-
ing level reference points to a description of the specific sharing level which describes
the privacy related issues of the event. Depending on the event, various entities with
different roles can be involved, e.g. when sending an e-mail, there is a person with the
role sender, at least one person with the role receiver and a document with the role e-
mail. Each event can be conducted in a session. A session can, for example, be the
time between booting and shutting down a computer or the time between the login
and logout of a user in a portal.
The current CAM scheme does not have fixed bindings so far. The information can
be stored in XML, RDF, JSON or in a relational database, depending on the purpose
of the data collection.
1
https://sites.google.com/site/camschema/
RecSysTEL 2012 96
2.2 Activity Streams
The Activity Streams specification [11] defines a format for single activities car-
ried out by users. An Activity Stream is a collection of one or more individual activi-
ties. Usually, activities are serialized using JSON.
Fig. 2. Simplified excerpt of the Activity Streams scheme
Fig. 2 shows the core elements of the Activity Streams scheme. A single activity
must at least contain a description of the entity that performed the activity (actor
property) and the date and time at which the activity was published (published prop-
erty). The Activity Stream Working Group recommends that an activity also contains
a verb, an object, and an id property. The verb identifies the action that the activity
describes (e.g. “accept”, “add”, “dislike” etc.), the object property describes the pri-
mary object of the activity (e.g. the watched movie or the sent e-mail) and the id
property provides a unique identifier for the activity in the form of an absolute IRI
(Internationalized Resource Identifier). The target property is optional and can be
used if indicated by the verb. For instance, in the activity, "John sent an e-mail to
Bill", “Bill” is the target of the activity.
The value of the actor, object, and target property respectively is an Activity
Stream Object. An Activity Stream Object comprises several properties describing the
object and should at least contain an IRI (id property) and a plain-text name for the
object (display property). Additionally, it can contain others such as an object type.
The Activity Base Schema [12] already defines object types to be used with Activity
Streams, e.g. “alert”, “application”, “article”, etc. The object types are further grouped
in six classes, i.e. audio and video objects, binary objects, events, issues, places, tasks.
Depending on the class, objects may contain further properties, e.g. startTime and
attendedBy for Events. Furthermore, any object within an Activity Streams object can
be extended with properties not defined by the core Activity Streams’ specification to
provide as much flexibility as possible.
2.3 Learning Registry Paradata
The Learning Registry Paradata format [13] is basically an extended and altered
version of the Activity Streams JSON format. It was defined to store aggregated us-
age information about resources. The Learning Registry Paradata specification states
explicitly that the Activity Streams format should be used if mainly individual actions
are stored.
RecSysTEL 2012 97
As for the Activity Streams, a basic LR Paradata statement consists of three key el-
ements: actor, verb, and object (see Fig.3). The actor refers to the person or group
that does something and is represented by a string or LR Paradata object (as defined
later). The verb refers to the action that is taken. In its simplest form, it just contains
the action name (e.g. “taught” or “viewed”), but it can also be specified in more de-
tail, which is the main difference of the LR Paradata and the AS scheme. The object
refers to the thing being acted upon using a string or a LR Paradata object.
Fig. 3. Simplified excerpt of the Learning Registry Paradata scheme
A LR Paradata object may contain an id, an objectType, a description of the object,
i.e. an array of keywords, a measure related to the object, a date and a context. Apart
from description and date, each element can be represented by a string or by a JSON
object without pre-defined scheme. The values of the elements depend on the ob-
jectType, which can be e.g. a person, a group, a learning resource, a LMS, etc. Within
the verb, an action is specified that holds the verb’s value (action), additionally, it can
contain any element specified for the LR Paradata object [14], [15].
2.4 NSDL Paradata
The NSDL Paradata format was defined to capture aggregated usage data about a
resource (e.g. “downloaded”, “favourited”, “rated”) which is designated by audience,
subject or education level [16]. In contrast to the other usage data formats presented
so far, this format is not event, but object-centric. Each data object has exactly one
NSDL Paradata record, which is identified by a recordId and must contain the URL
of the resource to which the paradata record applies (usageDataResourceURL). The
most important element is the usageDataSummary, which comprises all available
usage statistics/information about a resource using five different types of values. An
Integer/Float value represents the number of times certain actions have been per-
formed on the resource, e.g. how often it was viewed or downloaded. A String value
is a textual value that has been associated to the resource, e.g. a comment. A Rat-
ingType value is the numerical average that represents the judging of a resource on a
numerical scale, e.g. a rating according to its usability. A VoteType value represents
the number of positive and negative responses to a resource, e.g. good or bad for use
in classroom. A RankType value represents the standing of a resource in a hierarchy,
e.g. best of 2010.
RecSysTEL 2012 98
Besides its type and value, each usageDataSummary element contains the begin-
ning and ending date for the usage data (dateTimeStart, endTimeStart), information
about the audience that conducted the event (“educator”, “student”, “general public”,
the subject of the used resource (e.g. “computing” or “mathematics”) and in which
educational level (edLevel) the resource was used (e.g. “MiddleSchool”, “Grade 7”).
Fig. 4. Simplified excerpt of the NSDL Paradata scheme
For lack of space, these elements are not shown in Fig. 4, but only the elements
that are dependent on the type of the usageDataSummary element. Please see
http://ns.nsdl.org/ncs/comm_para/1.00/records/planets.xml for an extensive example.
3 Conclusion
We reviewed the four most popular usage data representation formats that are be-
ing used in the learning domain in this paper and described their main properties.
Each format has been created with a specific purpose in mind, so one must be clear
about the further applications that will use the collected usage data when choosing the
most suitable format.
In order to enhance the interoperability among usage data analysis tools and usage
data storage silos, our next step will be to provide guidelines on how mappings be-
tween formats can be implemented and what has to be considered. All formats are
open and allow supplemental, not pre-defined elements. Additionally, the specified
vocabularies are not perceived as complete and for instance in a CAM instance, an
entity can be described by any metadata scheme. Thus, no one-size-fits-all mapping
among the formats is possible. In contrast, mapping can only be defined for specific
application scenarios. By providing automatic mapping rules based on specific appli-
cation scenarios, access to usage data collections will be facilitated. Nevertheless,
further work remains to be done in terms of further generalizing the mapping rules so
that the automatic conversion tools become less application scenario dependent.
RecSysTEL 2012 99
References
1. Schuff, D., Turetken, O., D’Arcy, J., and Croson, D. 2007. ‘Managing E-Mail Overload:
Solutions and Future Challenges’, Computer, 40 (2), 31-36.
2. Hauser, J. R., Urban, G. L. Liberali, G., and Braun, M. 2008 ‘Website Morphing’, Market-
ing Science.
3. Adomavicius, G., Tuzhilin, A. 2005. ‘Toward the Next Generation of Recommender Sys-
tems: A Survey of the State-of-the-Art and Possible Extensions’, IEEE Transactions on
Knowledge and Data Engineering, 17 (6), 734-749.
4. Elias, T. 2011. ‘Learning Analytics - Definitions, Processes and Potential’, retrieved July 4,
2012 from http://learninganalytics.net/LearningAnalyticsDefinitionsProcessesPotential.pdf
5. Dawson, S. 2011. ‘Analytics to Literacies: Emergent Learning Analytics to evaluate new
literacies’. Workshop on New Media, New Literacies, and New Forms of Learning, London,
December 2011. Retrieved July 4, 2012 from
http://blogs.ubc.ca/newliteracies/files/2011/12/Dawson.pdf
6. Ferguson, R. 2012. ‘The State of Learning Analytics in 2012: A Review and Future Chal-
lenges’. Technical Report KMI-12-01, Knowledge Media Institute, The Open University,
UK. Retrieved July 4, 2012 from http://kmi.open.ac.uk/publications/techreport/kmi-12-01
7. Scheffel, M., Niemann, K., Leony, D., Pardo, A., Schmitz, H.-C., Wolpers, M., Delgado
Kloos, C. 2012. “Key Action Extraction for Learning Analytics”. Proceedings 7th European
Conference on Technology Enhanced Learning (EC-TEL 2012), Springer.
8. Katrien Verbert, Nikos Manouselis, Xavier Ochoa, Martin Wolpers, Hendrik Drachsler,
Ivana Bosnic, Erik Duval. 2012. "Context-Aware Recommender Systems for Learning: A
Survey and Future Challenges," IEEE Transactions on Learning Technologies, vol. 99, no.
PrePrints, , 2012
9. Schmitz, H.-C., Wolpers, M., Kirschenmann, U., Niemann, K. 2012. ‚Contextualized Atten-
tion Metadata’. Human Attention in Digital Environments, Eds: Claudia Roda, Cambridge
University Press, Cambridge, US, 2012
http://www.cup.es/catalogue/catalogue.asp?isbn=9780521765657
10. Çelik, T. 2005. ‘Attention.xml Technology Overview’, retrieved July 4, 2012 from
http://tantek.com/presentations/2005/01/attentionxml.html.
11. Snell, J., Atkins, M., Norris, W., Messina, C., Wilkinson, M., Dolin, R. 2012. JSON Activity
Streams 1.0, retrieved July 4, 2012 from http://activitystrea.ms/specs/json/1.0/
12. Snell, J., Atkins, M., Recordon, D., Messina, C., Keller, M., Steinberg, A., Dolin, R. 2012.
Activity Base Schema (Draft), retrieved July 4, 2012 from
http://activitystrea.ms/specs/json/schema/activity-schema.html
13. Paradata Specification V1.0. Retrieved July 4, 2012 from
https://docs.google.com/document/d/1IrOYXd3S0FUwNozaEG5tM7Ki4_AZPrBn-
pbyVUz-Bh0/edit?hl=en_US&pli=1#
14. Paradata Cookbook, V 1.0 © Copyright 2011, Learning Registry. CC-BY-3.0. Retrieved
July 4, 2012 from
https://docs.google.com/document/d/1lggCnowWsDgQxrNjYRAgh2KNwKfq-
MV8vLJzRXbAaos/edit?pli=1#
15. Paradata in 20 Minutes or Less, V 1.1 © Copyright 2011, US Department of Education: CC-
BY-3.0. Retrieved July 4, 2012 from
https://docs.google.com/document/d/1QG0lAmJ0ztHJq5DbiTGQj9DnQ8hP0Co0x0fB1Qm
oBco/edit?hl=en_US&pli=1#
16. NSDL Paradata, 2012. Retrieved July 4, 2012 from
https://wiki.ucar.edu/display/nsdldocs/comm_para+%28paradata+-+usage+data%29
RecSysTEL 2012 100