@mastersthesis {
	title = {Time-dependent recommender systems for the prediction of appropriate learning objects},
	author = {Krauß, Christopher},
	editor = {Hauswirth, Manfred},
	abstract = {This dissertation deals with adaptive learning technologies which aim to optimize Technology Enhanced Learning (TEL) offerings to fit the individual learner’s needs. Thereby, Recommender Systems play a key role in supporting the user’s decision process for items of interest. This works very well for e-commerce and Video on Demand services. However, it is found to be the case that these traditional Recommender Systems cannot be directly transferred to TEL as the recommendation of course items follows a particular educational paradigm. The special conditions of this paradigm are first investigated and then taken into account for the realization of new algorithms.In order to allow a broad interoperability of a Recommender System with other technical components, a set of open standards and specifications results in a reference architecture for such an adaptive learning environment. Based on the realized architecture, activity data have been collected from students using course materials available online – the courses themselves comprising face-to-face lectures backed up by digital representations of the presented contents, blended learning settings as well as online-only courses. The courses provided access to the course materials via a novel Learning Companion Application. This app also presents learning recommendations to make the content selection more efficient and effective.
Thereby, this work indicates that an educational Recommender System should not be evaluated using standard evaluation frameworks that utilize, for instance, a classical n-fold cross-validation. For this reason, a time-dependent evaluation framework is defined to investigate the precision of the Top-N learning recommendations at various points in time. Moreover, a new measure is introduced to determine the Mean Absolute Timeliness Deviation between an item recommendation and the time when it is actually accessed by the user. Subsequently, four major techniques for Recommender Systems are realized and applied to the collected data, evaluated with the time-dependent evaluation framework and successively optimized. As a reference implementation, a traditional Collaborative Filtering algorithm is developed and extended to incorporate time information. The results are then compared to the results of other time sensitive algorithms: an Item-based Collaborative Filtering approach which has previously been applied to TEL and a new learning path generator which incorporates a set of contextual information. Finally, a novel time-weighted Knowledge-based Filtering algorithm is presented and exhaustively analyzed. The evaluation results indicate that time-dependent filtering which incorporates multi-contextual activity data can produce the most precise recommendations.},
	year = {2018},
	month = {06/2018},
	language = {German},
	school = {Technische Universität Berlin},
	pages = {1-301},
	address = {Berlin, Germany},
	country = {Germany},
	doi = {http://dx.doi.org/10.14279/depositonce-7119},
	url = {https://depositonce.tu-berlin.de//handle/11303/7957 10.14279/depositonce-7119},
	refereed = {does not apply},
	keywords = {adaptive learning, adaptives Lernen, intelligent tutoring systems, intelligente Tutorensysteme, technologie-gestütztes Lernen, technology enhanced learning, time-aware recommender systems, time-dependent evaluation techniques, zeit-abhängige Evaluationstechniken, zeit-sensitive Empfehlungssysteme},
}