Research On Mining And Recommendation Algorithm Based On Mobile User Behaviors

With the development of mobile communication technology,mobile Internet online services have become an indispensable part of people's daily life.From the mobile users' points of view,it has become more difficult to obtain satisfactory information from the mass information provided by mobile Internet online services,which are referred to as the information overload problem.For Internet service providers,addressing the information overloading problem is one of the keys to operation cost reduction and to raise service efficiency and user satisfaction.Recommender system can accurately identify the satisfactory information for mobile users from mass information,and has been proven to be a successful technology to solve the information overloading problem.Meanwhile,when mobile users are provided with services from Internet service provider,the latter also receive a great many of user behavior information.By analyzing these behaviors,Internet service provider are able to predict the behavioral habits of the mobile users,so as to improve the provision of customized services,raise the user satisfaction level and draw more users.As a consequence,the recommender systems based on mobile user information and the analysis on mobile user behaviors have become one of the most popular aspects in academia.By analyzing mobile user behaviors,three fields are focused in this dissertation: 1)the personal recommender algorithm based on mobile user browsing behaviors;2)the group recommender algorithm based on mobile user rating data;and 3)the mining algorithm based on mobile user attributions.This dissertation has carried out in-depth studies on mining and recommender algorithms based on mobile user behaviors and achieved innovation results as follows:1)A recommender algorithm based on mobile user browsing behaviors is proposed.Due to the frequent occurrences of temporary Internet browsing behaviors made by mobile users,the user browsing behaviors are featured with great uncertainty,which results in moderate recommendation performances of traditional algorithm.For addressing this shortage,a novel recommender model is proposed to predict the future browsing activities of the mobile users and take them as the bases to recommend contents to the mobile users.By analyzing browsing times of a mobile user in everyday during a period of time,the recommender model can obtain the probability that the interesting duration of mobile users for the Internet topic is greater than or equal to minimal interesting duration.Then,neighbor information of the mobile user and concerning subject is applied to obtain interesting level of Internet information for the mobile users.Finally,according to interesting level of the mobile user for Internet information,the most appealing top K topics of Internet information are provided to the mobile user.In experiments,this dissertation uses Internet browsing behavior data of real mobile users to verify the effectiveness.The experimental results show that Internet information provided by the proposed recommendation model in this dissertation is more popular than that provided by the traditional recommendation algorithm.2)A group recommender system based on rating of mobile users is proposed.Due to the shared or different interests among mobile users in groups,it is difficult for the traditional group recommendation algorithm to predict items that can meet the requirements of all mobile users in a group.For addressing this shortage,a novel random group recommendation model is proposed to appeal top K items to all the mobile users in a group.By analyzing rating of all the group membership for the common items,the recommendation model applies the average strategy to abstract the group as a virtual user.And then,a personal recommender algorithm is applied to suggest top K items to the virtual user.And preference model and clustering algorithm based on multiclass are applied to optimize recommendation result of the group recommendation model.In this dissertation,two real public datasets of user's rating are applied to verify recommendation result of the group recommendation model.The experimental results show that the items provided by the proposed group recommendation model are more popular than those provided by traditional group recommendation model.3)A mining algorithm for uncertain frequent attribution sets is proposed based on differential privacy.Under the condition when mobile user privare information is not leaked,it is hard for existing mining algorithm to collectively perform precise mining of user behavior habit information from the uncertain user attribute dataset.For addressing this shortage,a novel uncertain frequent attribution sets of mobile users mining algorithm based on differential privacy is proposed.Based on traditional algorithm of uncertain frequent itemset mining,the sparse vector algorithm and Laplace mechanism in differential privacy are applied to guarantee mining result satisfy differential privacy.And then it is proved that the proposed mining algorihthm can guarantee differential privacy in theory.Finally,two public datasets are applied to verify the efficiency of the proposed mining algorithm.The experimental results show that precision and security of the proposed mining algorithm is better than traditional mining algorithm.