Research On Application Of Web Usage Mining In Personalized Recommendations

With the rapid development of the Internet, the web has been increasinglypenetrating into people’s daily lives, a number of new networking products such asmobile Web, social networking, Internet, Twitter and so on, are changing the traditionalway of receiving information. At the same time, Network data is growing at anunprecedented rate, and people have entered the era of big data. Facing such a hugeamount of data, how to access to information they need has become urgent issue need tobe addressed. Therefore, personalized service technologies is receiving increasinglywide attention.Web server log file records information such as the time that user surfing theInternet, programs that the focus window runs, and page access. Through associatedrules or decision-making tree modeling analysis to this process information, we canobjectively reflect users’ preference to software using and page access mode andinternal relationship between softwares and web pages, recommend softwares and webpages to users, and also make divisions to those users who have similar acts,gainingclass groups through poly class, then recommend right friends to users with somesimilar properties. These methods can provide reference to some web sites or softwareproviders for their improvements in site structures, personalized softwarerecommendations, making friends, or finding potential customers in e-commercestrategy.This article researches the personalized software recommendations and friendrecommendation data based on data mining of Web log. First, this thesis described thebackground meaning of research, present situation of personalized service research andproblems facing it; Second, systematically described the definition and process of datamining and data mining algorithm used in this thesis, and brought forward improving method of Apriori algorithm based on algorithm theories; Then, made a modelinganalysis to the cleaned log data with the help of Spss Clementine data mining tools; andestablished association rules and decision tree models of software-personalizedrecommendation; Finally, this thesis brought forward a kind of friends recommendationproposal based on user’s property information and th software using preferences.Innovations of this thesis are mainly reflected in the following aspects:(1) This thesis,mainly analyzed Apriori algorithm in association rules,andimproved the shortcomings of Apriori algorithm as it has to again scan the database tojudge whether the frequent item sets candidate is the real frequent item sets after itgenerates frequent set. And it also realized the goal of scan only once the database whengenerate frequent item sets through the introduction of prime number decompositionmethod,improved the algorithm efficiency.(2) Put forward the browser market share analysis method based on user-focuswindow, truly reflected users’situation of using browsers, and made an analysis to thechanges of browser market share from May to July.(3) Based on the advantages of research in Web data mining technology, this thesismade models for the cleaned data set with the help of Spass Clementine data miningtools, found the inherent links between software patterns and targetedly recommendedsoftware based on user properties, providing theoretical basis for personalizedrecommendation services of360Software Manager software.(4) In view of the shortcomings in the existing recommendation systems, thisthesis put forward algorithm design method about friend recommendation to usersbased on their attributes and preferences. The programme calculated the ultimatesimilarity of user inclusion property similarity and preferences similarity with the helpof K-Means and cosine similarity clustering. Recommendation of friends with similarbehavior and properties can be realized through similarity.