Efficient Personalized Recommendation of Mobile Web Content Using an EM-Based Clustering Method
Ming Xuan He, Alvin Chin, Enhong Chen, Jilei Tian · 2014
Many applications recommend personalized content to users based on their interests. However, personalized recommendation is time and memory consuming especially for commercial systems that have huge numbers of users, requests and big data that require complex computation. Since users do not totally have unique interests, we can cluster similar users then recommend the same items to users belonging in the same cluster. Even though clustering-based recommendations are efficient, the recommendation items to users may not be accurate. We present an Expectation-Maximization (EM) based personalized recommendation method for selecting the appropriate items efficiently and accurately. We use a browser log dataset to compare our method with personalized, k-means, and EM-based recommendations according to average rank, novelty, diversity, and time performance. Results show that based on average rank, novelty and diversity, our proposed method performs close to that of personalized, however it is less efficient than k-means. Since k-means has the worst average rank, novelty and diversity, our method is the best overall.