WebEduRank: an Educational Ranking Principle of Web Resources for Teaching

The Internet is a well-established source of information, and it enables services of many kinds. Over the years, many educational systems proposed online systems for the delivery of courses. Lately, educational services on the web have signi cantly enlarged the opportunity for students and instructors to search for educational materials supporting their educational needs. When we focus on instructors, we note that seeking teaching resources on the web is a widespread practice, and some online systems aim to support instructors in this complex and delicate task. In fact, quality of teaching is, without doubt, an important factor for learning, and an important aspect of teaching is the delivery of quality learning materials to students. At the intersection of Information Retrieval (IR) and Recommender Systems (RS) in education, remarkable results have been achieved in the reuse and recom- mendation of educational resources. Back to 2002, we nd a formalisation of a Learning Object (LO) as an object for enabling the reuse of digital materials for learning. The LO provides some educational characteristics of the content called metadata. This proposal was a breaking-through for nally implementing the shar- ing and reuse of educational materials with the application of IR and RS methods. The literature for RS in education spins around the original concept of Learning Object. However, the notion of LO presents many issues for the structuring of the metadata. In particular, the LO alone is unable to provide su cient data for ac- curately performing recommendations and retrieval of resources. The result is that scholars in the eld propose some modi cations to the structure of the metadata with the production of isolated case-studies about small repositories, sometimes not even publicly available. Today, the world-wide-web is mature enough from the educational perspective. So much so that this thesis will use it as the dataset of teaching resources. Certainly, the retrieval from the web of an educational resource is more complicated than from repositories of LOs. The main di culties are the absence of educational metadata among the web pages, and the huge size of the web. Nevertheless, we should highlight that most of the RS and IR systems in education do not bene t from the current standard of metadata of LO. We nd that most of these research work uncovers the signi cant limitations of LO metadata, and so proposals emerge that modify metadata frameworks of LO for performing a recommendation based on particular educational aspects or conditions. We can see the size of the web as a challenge but also as an excellent oppor- tunity. Transferring IR methods from local repositories of LO to the web faces many problems and most of the methods will not work properly. We believe that the size of the Internet is an excellent leverage point because it provides diversity. This feature is critical for educational resources to meet the diverse education characteristics and users' needs. Therefore, this research addresses the problem of using web pages as potential resources for teaching. We will propose a ranking principle that produces relev- ance positions for web pages which re ects their compliance to a teaching context. We start our research working on the identi cation of those educational attributes of a teaching context that actually carry information useful for rating web pages for teaching. Then, our focus moves towards the design of an educational ranking principle based on our de nition of Teaching Context. We call our ranking principle WebEduRank. WebEduRank is a ranking principle for rating web pages focusing on the teaching side of education. For testing the progress of our WebEduRank, we evaluate the accuracy performance of WebEduRank against Google's ranking, BM25F and TFIDF. To conduct this evaluation, we build a dataset of ratings of web pages in educational contexts. We collect these data via an online survey where instructors create a teaching context of their interest and evaluate the use- fulness of some web pages for teaching in such context. With these data, we rst show that we can use the Instructor Pro le (IP) for building a query for a more accurate retrieval of teaching resources. We compare the performance of BM25F and TFIDF interrogated with a user query and with the IP-informed query. The results of the paired t-tests con rm an improvement of the accuracy of the rankings according to ve accuracy measures. Finally, we evaluate WebEduRank with the same data and methodology, and we also compare the accuracy of its rankings with the relevance placements by Google as well as the IP-informed versions of BM25F and TFIDF. Strong positive results leave no doubt about the better rankings pro- duced by WebEduRank, especially when compared with the rankings of Google. Each of the ve measures report a better ranking from WebEduRank, con rming the possibility of ranking web pages for teaching using our WebEduRank.