Abstract: This paper proposed a method of topic extracting with user information protection in Web learning resources.We added the HMM model of user information protect stated into user information protection layer. The model can exit topic extracting procedure automatically once judging the invalid state of user information protection to prevent violations of user information. This paper evaluated HMM model of user information protection with the real data sets of 227 user's query browsing behavior, user link, the user configuration information in four study sites. The experimental data show that for 500 random massages test, the correct ration of the model judgment on the user information protection states is 94%, the incorrect ration of that on false security messages is 0.04. According to the user ratings data of fourlearning sites in the 120 days, the average rate of increase of user rating reaches to 23. 23% after the system is used.

Key words: User information protection, Hidden Markov model,Dynamic aggregation,Semantic level sharing