Recommendation Systems are central in current applications to help the user find relevant information spread in large amounts of data. Most Recommendation Systems are more effective when huge amounts of user data are available. Educational applications are not popular enough to generate large amount of data. In this context, rule-based Recommendation Systems seem a better solution. Rules can offer specific recommendations with even no usage information. However, large rule-sets are hard to maintain, reengineer, and adapt to user preferences. Meta-rules can generalize a rule-set which provides bases for adaptation. In this chapter, the authors present the benefits of meta-rules, implemented as part of Meta-Mender, a meta-rule based Recommendation System. This is an effective solution to provide a personalized recommendation to the learner, and constitutes a new approach to Recommendation Systems.
TopTheoretical Background
In the field of educational recommendations, there are approaches which use a modified page ranking algorithm for the generation of recommendations (El Helou, Salzmann, Sire, & Gillet, 2009). This algorithm takes into account a number of actors, activities, and resources, as main entities. The user’s activity is used to create a directed graph representing the entities and links between them are generated. After that, a rank is assigned to nodes, and finally, this information is used to generate recommendations for a user query. This work is valuable to us because meta-rules take the user’s activity for the automatic generation of rules, as an input.
In addition, in the field of rule-based recommenders, there are some relevant reports in the literature (Abel, Bittencourt, Henze, Krause, & Vassileva, 2008), which describe a rule based Recommendation System for online discussion forums for the educational online board Comtella-D. Actually, the system is able to call several encapsulated recommenders, collaborative filtering or content-based recommenders, and the rules decide according to the amount and type of user data which recommender should be called. In doing so, the rules define a meta-recommender which is very interesting but tangential to this work, mostly because the obtained rule-set is very simple. The rules just decide which recommender system to pass the control to and the actual recommendation is provided by the attached recommender systems.