itec - European Schoolnet

iTEC Project
Title: ITEC-D10_2_V1-1 041102012.Docx
• Multi-attribute content preference modelling: they interpret and model multi-attribute
descriptions provided by users on an element to use them to recommend elements most
adapted to their preferences.
• Multi-attribute content search and filtering: users are allowed to specify their
preferences for certain attributes through search processes. From the data obtained, they
recommend elements satisfying the search and/or filtering criteria most adapted to users’
preferences.
• Multi-criteria rating-based preference elicitation: user preferences are collected from
the element ratings according to several criteria. Recommendations to specific users are
based on their own ratings and other users’ ratings.
According to this classification, and after analysing the behaviour of classical recommenders, we
could say that most of them may be studied as multicriteria recommendation systems; mainly in
the case of knowledge-based and content-based systems due to the way they model users and
elements.
1.2.2. iTEC SDE Recommendation System
As discussed above, the SDE recommendation process is organized in three clearly separated
stages: pre-processing, filtering and sorting of results according to the computed relevance. No
matter this last stage being the most important, we will also discuss the processes taking place
during the other two stages, as they provide the result candidates. Therefore, we will briefly
introduce in Section 1.2.2.1 the initial pre-processing and filtering stages, and we will devote
Section 1.2.2.2 to discussing the recommendation algorithm implemented by the SDE according to
the multicriteria model already introduced. Finally, we will present in Section 1.2.2.3 the factors
considered to compute the relevance of Tool, Person and Event resources, which were collected
as a result of the third iteration of Control Boards from the contributions of iTEC WP
representatives, as discussed in Appendix II.
1.2.2.1. Initial stages
From the input data used to obtain a sorted list of recommended resources, i.e. a requirement from
a Learning Activity and a context for its implementation (i.e. LARGContext), the SDE performs a
semantic pre-processing to generate a semantic graph combining that information with additional
related data in the knowledge base. In this way, we generate an “extended requirement”
simultaneously modelling the Learning Activity’s requirement, the LARG Context and other
properties extracted from the knowledge base. 10 This pre-processing phase isolates the
information needed to query the rest of the data in the knowledge base to reduce both the spatial
and temporal complexity of the final recommendation phase.
Once the input parameters have been consolidated, the filtering phase is devoted to reducing the
number of resources in the knowledge base to those potentially relevant to perform a given activity.
This restriction of resources improves performance, as the volume of data to be processed during
the sorting phase is dramatically reduced. From the specification of a given “extended resource”, a
semantic query is constructed and executed to mark those resources whose properties are fitted to
10 For example, the geographical location of a school can be retrieved from the Technical Setting in the
LARG Context through the existing relation between the Technical Setting and the school.
Page 41/96