Article Preview
TopIntroduction
With the development of internet technology, there are vast amounts of information resources presented on the web. For example, there are hundreds of millions of goods in Taobao, an e-commerce site, and the number of global web sites reaches more than 100 million. Information overload is the major characteristic and challenge of internet. Traditional search engines provide same result for different users, which reduce the utilization of information service platform with the rapid growth of information (Liu & Zhou, 2009). For online education, normal users can hardly get learning material that they really need from the large quantity of resource with the growth of the learning materials. Recommender system provides a proven and effective way to solve the problem of information overload, where recommendations can help users to evaluate unknown items. The recommending processes can provide people with resource according to their interest by predicting users' personal preference, which liberates people from searching vast amounts of information resource. At present, various kinds of recommender systems and algorithms have played an important role on the Internet, such as recommending products at Amazon.com, Taobao.com and jd.com, recommending videos at Youku.com and Youtube.com, recommending books and movies at douban.com.
Propelled by application requirement, there are some popular methods have been proposed, such as collaborative filtering recommendations, content-based recommendations, hybrid recommendations (Liu & Zhou, 2009) and graph-based recommendations. Recommending based on collaborative filtering is the earliest and most popular recommender system. The collaborative filtering method predicts the preference of users according to their similar neighbors which could be discovered by historical evaluation data. The collaborative filtering method has the advantages of adaptability and extensive applicability. However they also show obvious shortage in recommending on the condition of cold start and sparse data for depending on the historical evaluation data. Content-based recommendation structures profiles for users and items, the method does not needs users' personal behaviors to generate nearest neighbors. As a result, content-based recommendation works well in the conditions of new users and new items, and achieves recommendation on sparse data. However the profiles in Content-based method are usually generated by machine learning, and require frequently updating which need high computational cost.
Recommendation systems play a great role in the development of distance learning technology (Sikka & Dhankhar, 2012). Distance learning can be understood as a learning process when the learners are separated by time and distance, which require online courses and examinations for interaction. In order to improve the quality of this kind of education, the most important consideration is how to discover the students' learning behavior patterns based on timely assessment of learning outcomes. Various data mining techniques have been used for assisting tutors in distance learning to analyze the information available in the form of data generated by users (Jain, 2012; Wang, 2011), however the value of the data is far more than that. The developed recommendation systems focus on predicting learners’ behavior and their final performance and supporting the realization of efficient personalized learning. After collecting the learning characteristics of the learners, the preference model of these leaners will be built to predict the best learning path in the distance learning system. In this case, the recommendation system provide an excellent mean for interaction between learners and tutors. With the assisting of the recommendation systems, people will prevent repetitiveness from working or repeated mistake in online learning, rapidly understand and master the knowledge.