Studies On Multi-perspective Recommendation Strategies For Multi-type Users

In recent years,as the coming of information era,the information on the Internet increases explosively,and the Internet has become inseperable from people's life.This dissertation mainly focuses on how to mine information to satisfy users' needs from mass information,and making users take advantage of the information to provide better services.In this dissertation,aiming at different application backgrounds and requirements,multiple recommendation strategies are proposed from different service perspectives for multiple types of users.Specifically,for personal user's information search requirement and social network interaction requirement,and enterprise user's multiple target user mining requirements on the purpose of product promotion,multi-perspective recommendation strategies are presented for multiple types of users.In this dissertation,recommendation strategies are deeply studied for personal user and enterprise user's different requirements.For personal users,there are two basic needs of Internet interaction:information search need and social network interaction need.The former is searching information from mass Internet information,and the latter is acquiring information by following different friends and interacting with them on social network platforms.For these two needs,in this dissertation correlated query path recommendation strategy on multi-domain information integration is proposed to recommend query paths to personal users,and single node overlapping community search strategy on social network is proposed to recommend relevant users which personal users are interested in.For enterprise user,marketing need is an important target user mining requirement for enterprise,and this dissertation mainly focuses on two marketing needs:mining the target users who are most interested in enterprise's products and mining the target users who have the greatest influence on information diffusion.Therefore,in this dissertation multiple node overlapping community search strategy on social network is proposed to recommend interested target users to enterprise users,and hot spread node selection strategy is proposed to recommend the most influential target users to enterprise users.Specifically,the main contributions in this dissertation are as follows:(1)For personal user's information search need,a multi-domain correlated query path recommendation strategy is proposed based on entity class model.This strategy is aimed at recommending other correlated domains to a user according to the user's initial query domain,and these correlated domains are organized in series together to make one correlated query path.First,each domain is abstracted as entity class,from the angle of entity class,a directed-weighted domain correlation graph is constructed through mining correlations between domains by multiple evaluating elements.Then,according to the initial query domain,the correlation graph is utilized to select correlated query paths related with the initial query domain by Random Walk method.(2)For personal user's social network interaction need,based on the application background of recommending friends or related information for personal users,an efficient single node overlapping community search strategy is proposed.This strategy is aimed at searching a user's all overlapping communities he or she belongs to according to the single user node,and it supports online query,and only search the social network in a local way,so it is flexible and light-weight,and can support dynamic network as well.The strategy includes exact method and approximate methods,boundary node constraint is utilized to improve the efficiency of the exact method,and boundary node and node degree are utilized as conditions to adjust the efficiency and accuracy of the approximate methods.(3)For enterprise user's marketing need,based on the application background of recommending target users to enterprise which are interested in enterprise's products,a multiple node overlapping community search strategy is proposed.This strategy is aimed at searching a group of users' overlapping communities that these users belong to according to multiple user nodes.Simply adopting single node overlapping community search approach iteratively may produce many duplicated computations and will affect the efficiency.Thus,in this dissertation a multiple node overlapping community search framework is introduced,and an exact method of multiple node overlapping community search is presented,and also a series of heuristic strategies to adjust the efficiency and accuracy of multiple node overlapping community search are proposed.(4)For enterprise user's marketing need,based on the application background of recommending the most influential users of information diffusion to enterprise,a hot spread node selection approach is proposed based on overlapping community search.First,an iterative promotion model is proposed,which selects influence maximized nodes step by step according to users' behavior feedback,and make the social network platforms play the controller role during information diffusion process,in order to select the most valuable promotion targets from users for advertisement clients.Then,a method to measure the influence of nodes is presented,which is based on overlapping community structure,and through this influence measure method a hot spread node selection approach is proposed based on overlapping community structure,the exact methods of this approach include a basic method and an optimized method.Because the exact solution is an NP-hard problem,the time complexity is very high,and the applicability is not very good,thus an approximate method is proposed based on a scoring mechanism,this method can select hot spread nodes fast and effectively.