Personalized Web Recommendation Via Collaborative Filtering

Recommender Systems, which provide users with personalized information services, are becoming increasingly indispensable nowadays, since they focus on solving the information overload problem due to the development of Web. Because of their important commercial values and great research challenges, recommender systems have been widely applied on industrial community and extensively studied on academic community.Typically, recommender systems are based on collaborative filtering, which predicts the interest of an active user by the rating information from his similar users or items. The objective of this thesis is to study on collaborative filtering to improve the development of personalized Web recommendation.In this thesis, we first propose a new similarity measurement for collaborative filtering to improve personalized product recommendation. Similarity measurement that denotes the similar extent between two users (or two items) is critical to collaborative filtering. Traditional similarity measurement approaches for collaborative filtering can be strongly improved. Our new similarity measurement, named Jaccard Uniform Operator Distance (JacUOD), is more effective than traditional similarity measurement approaches. JacUOD aims at unifying similarity comparison for vectors in different multidimensional vector spaces. Compared with traditional similarity measurement approaches, JacUOD considers the length of vectors and properly handles dimension-number difference for different vector spaces. The experimental results based on the well-known MovieLens datasets show that JacUOD outperforms traditional similarity measurement approaches.With the increasing of amount of Web services on the Internet, personalized Web service selection and recommendation is becoming more and more important. Therefore, we propose a novel collaborative filtering approach, called Normal Recovery Collaborative Filtering, for personalized Web service recommendation. To evaluate the Web service recommendation performance of our approach, we conduct large-scale real-world experiments, involving5,825real-world Web services in73countries and339service users in30countries. To the best of our knowledge, our experiment is the largest scale experiment in the field of service computing, improving over the previous record by a factor of100. The experimental results show that our approach achieves better accuracy than competing approaches.Last but not least, we propose the user-group item-based collaborative filtering approach, called HUCUI, for personalized Open API recommendation in clouds. In recent years, there are more and more Open APIs available on the Internet that can be invoked by independent users for their innovative applications. With the development of cloud computing, the number of Open APIs in clouds is also increasing. With the number increasing of functionally-equivalent Open APIs in the clouds, Open API recommendation is becoming more and more important. For personalized Open API recommendation in clouds, we propose the user-group item-based collaborative filtering approach HUCUI, inspired by the idea that item similarity can be measured based on users with similar preference. Our approach classifies similar users into a user-group, and employs item-based collaborative filtering within the user-group. It has good scalability and prediction accuracy. Extensively experimental results based on real-world QoS values of Open APIs demonstrate the effectiveness of our approach.