Navigate Up
Sign In

Crowdsourcing a Recommender System A Case for e-Learning

Publication Date:
 

crowdsourcing 1 .jpg

Written by Mario Mallia Milanes


Mallia Milanes Mario.jpgRecommender Systems are attracting a lot of attention lately.  We can see the reason behind this.  They are useful in nudging us to make that purchase which would have otherwise went unsatisfied or to suggest music or videos we normally like watching.  Companies of the likes of Amazon, eBay and Netflix invest heavily in such systems because they help to increase sales revenue.  So, the aim of a recommender system is twofold; to induce sales, and to reduce information overload.  The techniques behind the Recommender Systems are rooted in information retrieval and filtering. There are two basic approaches to recommender systems, namely, Collaborative Filtering, and Similarity Indexing.

In Collaborative Filtering, the algorithm matches the current customer’s buying patterns to that of others with similar tastes. Recommendations are put forward based on the premise that clients with similar buying patterns would be interested in similar items. So, in this case the idea is centered on building customer profiles based on their behavior within the system.  These profiles are then used to generate lists which may attract, or even help, a user to make choices.

In Similarity Indexing products are tagged to facilitate grouping. Here the benefit is that the system does not actually need to understand what a product is. When a user chooses a product, others like it are put forward as a recommendation based on tagging similarity.

As expected, there are issues with both approaches. The Collaborative Filtering methodology suffers from what is called the cold-start problem.  This happens when the algorithm has very little or no data to go on. This makes user profiling difficult at first. On the other hand, a Similarity Indexing solution is very computationally intensive.  Imagine a typical situation with thousands of users on-line concurrently, each user going through many items available for sale. Here system latency would be a serious issue to contend with.

Extending the Use of Recommender Systems

Recommender systems have been successfully used but as expected their main stay has been in sales-based scenarios.   Naturally I would like to focus of this article away from the technicality of Recommender Systems.  Given that Recommender Systems have proved their worth by keeping customers motivated to purchase, the question posed here is: could Recommender Systems be useful in other domains?  The answer to this question is yes.  I shall argue that applying them to education could reap similar benefits.

Recommender Systems in Education

In today’s world adult learning and skill acquisition update has become somewhat important.  The Internet greatly facilitates this with the instant access to tens of service providers offering a plethora of courses at low cost to anyone who wants to join and learn. Both the caveat, and the challenge, is that of motivating clients to stay-on. The fact that a course has been purchased is only part of the transaction.  Maximum benefit would only be realized when a student has graduated or successfully completes a course.  To achieve this, several factors must be considered, namely:

1.    Access to peers and participation of peers in learning;

2.    The availability of suitable domain experts that can assist learning;

3.    Suitable material that is calibrated in such a way to support the learning process.

In this article we shall argue that Recommender System technology can be used to solve the problem.

Adaptation to e-Learning

For the sake of convenience and economies of scale many e-Learning systems are produced as a one-size-fits-all package.  They suffer from a lack of personalization.  One way to jump over this hurdle is to “cocoon” the student within an automated learning environment that recommends and coaches learners with adequate resources and personalized suggestions.  This would be made up of domain experts, peers, and material to draw on.  Material can be crowdsourced, by having many input points feeding the student with his necessities.

crowdsourcing 2 .jpgThe point of the Recommender System in this setup would be to prevent a cognitive overload by supplying too much in too little time to the user.  

Moreover, the system would have to deal with information relevance apart from its timeliness. 


Hybrid Approach

In our case only the data relevant to the student’s needs and study level is brought up.  This can be done in a similar way as if one is buying items of the Internet. The item in our case being study material.  In addition to the material, the recommender system will have to associate the current user with others who have similar needs or experiences and allow then to interact.  This aspect of the proposed setup will facilitate learning by peer interaction.

Apart from peer association domain experts would also be matched according to the interests of the student. The declared expertise of the domain expert being tagged for association purposes.   Finally, the material needed can be made available through a number of likely sources such as peers, the open Internet, domain experts and also the entity organizing the course.

Crowdsourced material, experts and peers can be pulled together on the fly to create an environment similar to a class, but without the boundaries imposed by space and time.  This would give the student the opportunity to share, experience and scaffold though material till the level of skill is acquired. Teams of agents can be employed to make up the learning system.  A recommender system can also comprise a sub team of agents which cooperate to deliver timely information to the student.

 crowd sourcing.jpg

Figure 1 Top Level Architecture of Proposed System

 

By creating a crowdsourced Recommender System that could adapt to the needs of students individually would put the learner at the centre of learning.  This would help students gain experience as they progress along with their studies and in turn collaborate with others in their learning experience.  An intelligent environment will certainly help with student retention rate and additionally improve skill acquisition. A recommender system is only a small, but important, part of the e-learning eco system.  Information must be media neutral and different elements have to be combined to display the same results by different means that appeal to the user. In this article arguments have been put forward in favour of the use of artificially intelligent techniques to overcome specific shortcomings within e-learning systems. It is strongly believed that the lack of personalization brings about a unfavourable sense of isolation that hinders rather than facilitates the learning process.

The use of a recommender systems based on latest technologies to deliver personalised education material is opportune and suits all requirements and objectives. Such a methodology further assists to alleviate the issue of information overload as specifically targeted educational material will be put forward to the individual learners. The recent developments in technology has enabled recommender systems to move to their next phase whereby networked technologies unleashed resourceful affordances that before were not possible, and that potentially they can take e-Learning to its next generation.


 ​