Shah Khusro et al. (2016) and Chris Anderson name this problem as quite challenging for collaborative filtering. It appears when one user had totally different preferences than others and so cannot be put in any of clustered groups (Khusro, Ali , & Ullah , 2016), (Anderson , 2011). A. Kanimozhi et al. (2014) implemented this problem for e-learning saying that the easiest way to avoid it is to make students and teachers to fulfill their personal or course IDs and so to make them share information (Kanimozhi & Raj, 2014).
Concept 6: Sparsity
YiBo Chen explains that such challenge appears, when there are too many users, most of which gave no feedback to lots of items. Although, author tried to solve this problem with the help of association retrieval technology, he recognized that such approach can lead to information overloud and thus it’s not advisable for usage (Chen, 2011).
Jie Lu found out that sparsity is a corn stone of the personalized e-learning recommendation system and in order to overcome this challenge, recommendation system should be more accurate both with students’ and courses’ information analyzing it in multiple directions (Lu, 2004). Bahram Amini et al. (2011) agree adding that another possible solution for this problem is clustering of the items (Amini, Ibrahim, & Othma, 2011).
Concept 7: Latency Problem
Shah Khusro et al. (2016) pointed out that such challenge exists when new items are added to the database more frequently and in the same time recommendation system gives ratings only to old items, while new ones need to wait. Author warns that popular solution of adding collaborative filtering elements can cause another problem – overspecialization. There are other three possible ways out of this problem: model-based recommendation system, clustering methods and combination of user stereotype with concept-based approach (Khusro, Ali , & Ullah , 2016). Bahram Amini et al. (2011) adds that another approaches for solving latency problem can be K-Nearest Neighbor1, Cosine or Correlation based similarity.
To conclude, the most common challenges which relate to data requirement in e-learning recommendation systems are Cold start problem, Scalability, Data Set Sharing, Challenges, Shilling Attacks, Grey/Black Sheep, Sparsity and Latency Problem. However, various approaches, methods and techniques can be applied to cope with each of these problems
Another side of data needed for recommendation challenge is privacy problem since information required by advisory system can be confidential. Shah Khusro et al. (2016) stated that possible solution for this challenge can be simple usage of collaborative filtering since for this method all data is taken only from users’ ratings and stored in special repositories. In this case, nothing is taken from outside of the recommendation system borders (Khusro, Ali , & Ullah , 2016).
However, Joseph A. Calandrino et al. (2011) strongly disagree with the fact, that collaborative filtering is solution for privacy challenge. In their work, scientists proved on example of e-commerce web-sites (like Hunch, Last.fm, LibraryThing, and Amazon), that while giving recommendations collaborative-based advisory systems takes into consideration all transactions of user, including those, which he would not make known of his own free will (Calandrino , Kilzer , Narayanan, Felten, & Shmatikov, 2011).
John Canny supports this concept, saying that private users’ data can be even sold by one company’s web-site, which is on the stage of bankruptcy and used to give collaborative-based recommendations, to another one. Author sees possible solution for this challenge in giving control to the user about the data: user should be able to manage how, when, with whom and what to share. In order to implement this in reality, John Canny proposes to combine anonymous profile with personalized recommendation (Canny, 2002). It should be pointed out, that this solution is quite effective for e-commerce sphere but not for e-learning, since the profile here cannot be anonymous.
Katrien Verbert et al. (2012) noticed, that lots of researchers tend to think, that privacy problem is not so much common in e-learning environment. However, this approach is wrong and as a prove of this author gives some legislative acts as European Directive on data protection 95/46/EC2 (Verbert, et al., 2012).
A. Kanimozhi et al. (2014) strongly agrees adding, that privacy is one of the major problems in e-learning environment. That is why researchers developed their own e-learning recommendation system with a new and quite important element – Role Based Access Control (RBAC), which gives extra privacy protection, because in the projects there were lots professors, heads of different departments and other important scientists workers enrolled (Kanimozhi & Raj, 2014). RBAC clusters all users in different roles and gets permissions about information based on the following role (Sandhu , Coynek , Feinsteink, & Youmank, 1995). In such way, in e-learning recommendation systems can identify what is permitted to do with teacher’s or student’s data and will not mix these limitation, so student will never have excesses to the materials teacher does and vise verse