Manouselis N., and Sampson D. (2004) Recommendation of Quality Approaches for the European Quality Observatory. In the Proc. of the 4th IEEE International Conference on Advanced Learning Technologies (ICALT 2004), Joensuu, Finland, August 2004.
Recommendation of Quality Approaches for the European Quality Observatory Nikos Manouselis, Demetrios Sampson Department of Technology Education Advanced e-Services for the Knowledge Society and Digital Systems, Research Unit, University of Piraeus, & Informatics and Telematics Institute, 150, Androutsou Street, Piraeus, Centre for Research and Technology Hellas, GR-18532, Greece 42, Arkadias Street, Athens, GR-15234, Greece e-mail:{nikosm, sampson}@iti.gr Abstract This paper presents a neighborhood-based collaborative filtering algorithm used in the European Quality Observatory (EQO) web portal, in order to provide recommendation hints about e-learning quality approaches to the users of the portal.
1. Introduction Recommender systems were originally defined [1] as “using the opinions of a community of users to help individuals in that community to identify more effectively content of interest from a potentially overwhelming set of choices”. The term has currently acquired a broader connotation [2], describing “any system that produces individualized recommendations as output or has the effect of guiding the user in a personalized way to interesting or useful objects in a large space of possible options”. This paper presents the implementation of a recommendation mechanism that aims to support the users of the European Quality Observatory (EQO) web portal. The recommendation technique engaged is based on the principles of collaborative filtering [1,2]: it takes into consideration previous input from users and exploits similarities between the user profiles, so that to provide recommendation hints to a user searching for quality approaches.
2. Recommendation of e-Learning Approaches in the EQO Web Portal
Quality
The European Quality Observatory (EQO, http://www.eqo.info) is a European initiative funded by the eLearning Programme of the European Commission. Firstly, EQO aimed to develop a common descriptive framework for the categorization of quality approaches (QAs) that are applied in the field of e-learning. Additionally, the EQO project deployed a web-based repository that collects and stores descriptions of elearning quality approaches. In this context, the EQO web portal aims to provide services that will support and facilitate users in identifying interesting quality
approaches. Therefore, the recommendation mechanism should aim to predict how interesting will the description of a QA be for a specific user. Recommendation of QA descriptions is developed using a neighborhood-based collaborative filtering algorithm [3]. Collaborative filtering systems work by collecting ratings from a community of users for items in a given information domain, and matching together people who seem to be having similar preferences. They provide three advantages compared to other approaches [2]: support for filtering items whose content is not easily analyzed by automated processes; the ability to filter items based on subjective user preferences; and the possibility to provide novel and serendipitous recommendations. Nevertheless, other techniques are also studied for the EQO portal [4]. The problem in automated collaborative filtering is to predict how well a user will like an item that he/she has not rated, given a set of previous ratings from a community of users. The recommendation algorithm functions as a prediction engine that collects all the ratings and uses a collaborative filtering algorithm to provide predictions. Two different modes of operation will be developed for the EQO recommendation module, so that two distinct user tasks are supported: • Prediction upon a specific item: how will a user like a specific item viewed? This is the question to be addressed in an ‘Annotation in Context’ task [5], where users are viewing a QA. • Prediction of top-k items: what are the k items that the user will most probably like? This is the question to be addressed in a ‘Find Good Items’ task [5], where users are submitting search queries to the repository. The algorithm used to support the ‘Annotation in Context Task’ is engaged every time a QA description is presented to a user. The recommendation mechanism attempts to predict the rating this QA would get from the user, based on his/her profile. If the item has already been rated by the user, the specific rating value is presented, along with an indication on whether he/she wants to provide a new rating. The way recommendation will be provided to the user is presented in Figure 1.
Manouselis N., and Sampson D. (2004) Recommendation of Quality Approaches for the European Quality Observatory. In the Proc. of the 4th IEEE International Conference on Advanced Learning Technologies (ICALT 2004), Joensuu, Finland, August 2004. Let us present an example scenario. In Table 1, the ratings of three (3) users upon five (5) QA records are presented. When User 3 requests to view QA1 (an ‘Annotation in Context’ task), a prediction of the rating User 3 would have given to this QA record (QA1) is calculated as p3,1=2,63. Therefore, a recommendation of 2,63 (that is, 3 out of 5 stars in the graphical scale used) is presented to the user.
User 1 User 2 User 3
QA1 4 1 ?
QA2
QA3 5
3
QA4 5 2
4
QA5 5 3
QA6 4 3 2
Table 1: Users vs. Items matrix
3. Conclusions Figure 1: Interface design for the Annotation in Context in QA record recommendation
This paper presented how recommendation hints are provided to a user searching for interesting quality approaches in the EQO web portal, by the application of a neighborhood-based collaborative filtering algorithm that has been specialized to support two recommendation tasks. There are several issues that can be furthermore explored, such as the application of different recommendation techniques [4], and the way such services can serve as quality assurance mechanisms in virtual user communities sharing items of content.
4. Acknowledgements The work presented in this paper is partially funded by the eLearning Programme of the European Commission, and more specifically the “EQO: European Quality Observatory” initiative.
5. References
Figure 2: Interface design for the results of Find Good Items in QA record recommendation
The algorithm used to support the ‘Find Good Items’ task is engaged every time an Advanced Search query is submitted to the EQO repository. The recommendation module scans the results of the query, and based on the prediction of the ratings for the retrieved QAs, presents a ranked list of QAs. The way recommendation will be provided to the user is presented in Figure 2.
[1]. Resnick, P., Varian, H. R. “Recommender Systems”, Communications of the ACM, 40 (3), 56-58, 1997. [2]. Burke R. “Hybrid Recommender Systems: Survey and Experiments”, User Modeling and User-Adapted Interaction 12, 331-370, 2002. [3]. Herlocker J.L., Konstan J.A., Borchers A., Riedl J., “An Algorithmic Framework for Performing Collaborative Filtering”, SIGIR’99, Berkeley, CA, USA, 1999. [4]. Manouselis N., Sampson D., “A Multi-criteria Model to Support Automatic Recommendation of eLearning Quality Approaches”, in Proc. of EDMEDIA’04, Lugano, Switcherland, 2004. [5]. Herlocker J.L., Konstan J.A., Terveen L.G., Riedl J.T., “Evaluating Collaborative Filtering Recommender Systems”, ACM Transactions on Information Systems, Vol. 22, No. 1, January 2004.