Abstract:

When learning, the learner deals with the choice of relevant documents corresponding to his / her educational needs, which naturally leads him to go through manual and tedious searches. And in a Learning management system (LMS) environment that sometimes offers hundreds or thousands of documents, this search becomes endless or without interesting results. The research complexity increase when the learner is confronted with a search for documents in a multi-language environment.

The objective of our contribution is to provide a Big Data Solution. Our solution uses a recommendation system of the most relevant Learning Object (LO) which best meet the needs of learners. The solution is given in the form of a parallel and distributed computing algorithm using the techniques of Data Mining. The solution supports 16 languages including Arabic.

In a first step we have developed a model profile of the relevant object. This model profile is constructed from learning object metadata (LOM) attributes which constitute important indicators of selection in our algorithm. These attributes include context, difficulty, interactivity Level learning, Resource Type ,semantic density and language .In the second step, we have developed a Text Mining algorithm that takes into account the diversity of languages and uses the power of the Parallel and Distributed Processing of a computer cluster.

In summary, our solution consists of six main steps:
(1) creation of the requested model profile using LOM attribute values of the LO being consulted or using the values proposed by the learner,
(2) execution of a First selection to capture a collection of LO which matches exactly with the model profile constructed during step 1. This collection is selected from a Big Data store ,
(3) creating a filtering request based on a list of keywords constructed from the currently viewed LO,
(4) calculation of the scores for the LOs which had succeeded the First selection and had been filtered by the filtering request . These scores are calculated using similarity measurement algorithms (TF-IDF…),
(5) Sorting of the LO according to the scores calculated in the previous step,
(6) Return Top N LO which had the highest scores to our learners.

The implementation and the test of our solution allowed us to provide our learners with the desired LO in their language of learning and in Near Real Time. This contribution allows us to consider to empower our recommendation system with the help of translator processes provided by NLP (Natural Language Processing) techniques.

Keywords:

Big Data, Data Mining, Recommendation System, Near Real Time Processing, Lom, Lms, Distributed Processing, Parallel Processing, Text Mining.