A recommender system for e-learning

Abstract

Users are finding increasingly difficult to find relevant Content from the Web. Therefore, users are spending a considerable amount of time searching for content that matches their interests. In addition, the content found might not meet the expectations of the user. Saving a user time in searching for resources which are of a good quality in respect to him/her is thus desired. The advent of online social networks has made the tasks of finding relevant material (links to web pages) easier. This is because users are able to form so called friends in the network. This provides for the foundation of various mechanisms for recommending relevant material, or resources. Based on the assumption that people choose with whom to be friends on the network in a wise manner, various algorithms can be formulated. The context of the project is e-Learning. It is crucial that users are recommended the right resources because they depend on these to meliorate their knowledge about an area. The main aim of the project is to provide for a recommender system which gives recommendations to users with a set of resources /users that match their interests. These have three main flavours: i. recommending more friends to users ii. recommending resources to users iii. recommending experts in the social network Ultimately, the end-user would be interested in resources (i.e. links to web-pages). However, recommending friends to users also makes sense in that the manual browsing of friends' pages from social networks also leads users to resources. In addition, functions such as chatting and messaging the social networks provided may also aid the user to come across relevant material on the WWW. For example, a user may suggest a resource to another user through a chat functionality. Thus, recommending friends to a user may aid the latter in finding relevant resources. The algorithms involved in this research work on data which represents a social network. This data consists of a number of users who are associated with tags that they have used to define resources that were previously viewed. Ideally, these tags would be utilized for linking the social network to an ontology, a rich conceptualization of the domain represented by concepts and relationships, in order to improve content-based search. Thus, recommending resources would require navigating the ontology and selecting the resources which match the user's interest by identifying the similarity between concepts through relationships. Furthermore, the utilization of user ratings integrates the notion of friends and user's interests as the suggested resources will be based on the similar users' opinions. This would aid the user in deciding which resource/s from the recommendations list to view. In addition, a user may need not be satisfied with the results that the algorithms return. For example if a resource recommendation algorithm returned two resources in a prioritized manner and the end user would have rather swapped their prioritization, the built system is able to take this into account for future recommendations. Furthermore, it could be the case that a user has already seen a resource which is recommended to him/her by the system. To cater for such a scenario, the built system allows for manual ratings of resources. This entails storing such manual ratings in a database and providing for a mechanism which combines these with the scores the recommendation algorithms return for resources. The same ideas also apply to user (i.e. friend) recommendation. In the case where a user would like to further his/her knowledge concerning a particular domain, recommending resources on the basis of similarity might not be sufficient since the level of knowledge contained in these resources might not meet the user's expectations. Thus, recommending specialists to users is essential. In turn, through features such as chatting in available social networks, a specialist may guide a user.