The Research About Key Algorithms Of Recommendation System For Mobile User

With the development of wireless network and advanced mobile device,personal recommendation service in mobile environment has attracted people’sextensive attention. there are a lot of study about the recommendation applicationscenes ith the property of context-aware, however most of the study focus on therecommendation of single object, there are litter research about therecommendation of several objects while the recommendation application scenes ofseveral commodities also have widely development prospect, for exampleK-commodity of different kind recommendation in electric commerce and kserveplace sequence recommendation, and current recommendation methods are notapplicable because of not considering about the relationship between commoditiesor the uncertainty of some place providing service. So the corresponding methodsabout two above application scenes are proposed in following parts.Firstly, in order to solve the problem of K-commodity of different kindrecommendation, firstly, the formal definition based on formal combinationoptimization is given and one linear programming algorithm is proposed. But it isnot efficient. So the following approximately algorithm-ACR, which is based onpartition, is efficient in the case that the number of product is large, as a trade-off,the precision become low. Extensive experimental results on both real and syntheticdata sets demonstrate efficiency and effectiveness of ACR.Secondly, for the problem of k serveplace sequence recommendation in staticenvironment, firstly, some definitions about k serveplace sequence are given, andwe conclude the problem is multi-objective optimization problem when consideringthe preference of user is not known. So one classic method about multi-objectiveoptimization, skyline method is used to give people result, and lastly krepresentative skylines are returned as the recommendation result because of alarger quality of skyline. But the number of sequence of combination for eachserviceplace is still huge, so in order to improve effiency, one approximate skylinealgorithm is proposed, which is based on probability partition。The experimentresults demonstrate the effiency of proposed algorithm.Finally, for the problem of k serveplace sequence recommendation indynamic environment, we give one incremental method when the probability ofserverplace to offer service is changeable with the time in real life. And in order to improve the efficiency, we combine reverse index with partial order dominant graphto construct multi-level index. Extensive experimental results on synthetic data setsdemonstrate efficiency of proposed index.