Applying and Comparing Hidden Markov Model and Fuzzy Clustering Algorithms to Web Usage Data for Recommender Systems

As the information on the Web grows, the need of recommender systems to ease user navigations becomes evident. There exist many approaches of learning for Web usage based recommender systems. In this study, we apply and compare some of the methods of usage pattern discovery, like simple k-means clustering algorithm, fuzzy relational subtractive clustering algorithm, fuzzy mean field annealing clustering and hidden Markov model, for recommender systems. We use metrics like prediction strength, hit ratio, precision, prediction ability and F-Score to compare the applied methods on the usage data of the CTI Web site of DePaul University. Fuzzy mean field annealing clustering and hidden Markov model acted better than other methods due to fuzzy nation of human behavior in navigation and extra information utilized in sequence analysis.