Personalization challenges in e-Learning

Online learning is a hot trend in the education industry, as testified by the proliferation of e-learning platforms and MOOCs on a huge variety of topics, from science and technology to foreign language learning and musical instruments playing. In this talk we will present some of the common challenges in the e-learning industry that we are personally facing at Cloud Academy, an on-line continuous training platform for Cloud technologies. First, the nature of recommendation objects is quite diversified. Typically an e-learning platform comprises several content types, such as video lessons, reading material, quizzes, practical exercises (e.g., hands-on laboratories), etc. Also, they can be organized into composite objects, such as whole courses (groups of lessons) or exams (topic-specific quizzes and exercises). Signals are very heterogeneous, too. Users can trigger all sorts of events while interacting with the learning platform, such as time spent on a given lesson video, outcome of a quiz or a test after a specific number of attempts, natural language feedback, etc. Such variety can be an important and comprehensive source of information but it must be normalized and handled within an appropriate common framework. The objectives of recommendation in e-learning differ from the case of more well-established targets. Although enjoyment remains a driving force (e.g., proposing new interesting courses to an old user), the focus is on learning specific skills or topics for specific goals. In this scenario, recommendation seeks to suggest the new best lesson, exercise, or course to acquire the desired skill while keeping the user engaged, assembling a personalized study path in the process. The pace of the learning activity must be tailored on the user's needs and commitments, like deadlines (e.g., the user plans to get a certification by a certain date), specific goals (e.g., objectives of an employee might be defined by the user's manager) or learning prerequisites (e.g., a course can be consumed only after the student has acquired the needed knowledge, either via other more basic courses or pre-existing skills). Consequently, recommendations must take into account the initial user skills and knowledge as well as her learning objectives (e.g., improving the skills of a non-beginner student requires recommending more advanced material). Transparency plays an important role in the e-learning domain as users are willing to continuously monitor and inspect their progress and understand which training resources better fit their learning goals. Finally, the online learning environment can be made even more complex by \"social features\" such as virtual classes of multiple users, a context that recommender systems both can exploit and need to take into account.