Research Of User Access Prediction Based On Web Log Mining

Internet grows explosively since the beginning of the 21st century; we all experience the advantage of information. However, as the managements and researchers of Internet, we need to make good use of the tremendous data resource which brought by the rapid growth Internet, and mine useful knowledge from that to guide the construction of Internet in order to open a more humanism and more intelligent Internet era.Web log mining mainly studies the activity of web users browse in order to understand user's interests,hobbies and visited habit, so as to provide users with better services. Nevertheless, user access prediction is the core of web log mining, which predicts the next access page or the future access pages according to the history access information and the current access path. We can make use of the prediction result to improve the web server performance, increase the cache utilization and provide users with personal service.This paper analyzes the advantage and disadvantage of existing classical user access prediction algorithms, proposes Markov chain and association rule prediction algorithm (MAPA). This algorithm uses second-order Markov chain to find the pages which users may visit in next step or future, so as to generate the candidate prediction page set, and then corrects the Markov prediction result on forward and reverse perspective according to the two-items association rules, and gets the last prediction page. This algorithm integrates the advantage of Markov chain and association rule well.This paper proposes Markov prediction model with feedback (MPMF), the model creates the history prediction tree (HPT) step by step during the prediction process, saves the history prediction information into HPT, and determines whether the prediction is correct according the user feedback. This model generates the candidate prediction page set by two-order Markov prediction algorithm at first, and then adjusts the prediction algorithm dynamically according to history prediction information, and gets the prediction page at last.Experiments result shows that the accuracy of MAPA and MPMF is good. Theoretical analysis proves these two prediction methods with linear time complexity, so the prediction efficiency is also meet the requirement.