Font Size: a A A

Research On Dynamic Recommendation Algorithm For SaaS Multi-tenant

Posted on:2016-09-09Degree:MasterType:Thesis
Country:ChinaCandidate:J H SuFull Text:PDF
GTID:2308330464456854Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Nowadays incloud computing era, Saa S software delivery models based on Internet is hold in esteem, especially amongthe SME group. In the multi-tenancy mode of Saa S environment, multiple users can rent the same serviceand make customization according to the actual need, which not only satisfy the tenants’ diverse needs, but also give full play to the personalized characteristics of services. The personalized demand of the enterprisewhich moving toward the development of information networkin the long tail of software services market is increasingly prominent, and how to provide the required service in a wide variety of resources for the enterprise group has become a hot spot in research.For the services’ recommendation, collaborative filtering can be used which is most widely used of the dynamic recommendation technology. Although CF has achieved great success in the electricity business and other fields, also face the two typical problems: data sparseness and interestdrift. Sparse problem is due to the high-dimensional data which is generated by lacking of information in the preliminary stage of recommendation system and insufficiency new information in the process of system; Interest drift is caused by the changes of hobbies due to personal or others or even the environment. Saa S based on the fluctuant data has dynamicsof online system, thus the sparseness and drift will appear at any time. The two problems interact with each other, as well as the accuracy of the recommendation. To solve the problem, this paper combines multi-tenant application scenarios with personalized recommendation technology, mainly solving data sparseness and interest drift in the recommendation. The specific research work is as follows:1) The relationship between dealing with the sparse dataand the inaccuracyIn the Saa S environment,servicesare recommended to tenants, exploring the sparse data in the recommendation process. There are many literaturesstudy separate data sparseness from cold-start study, this paper argues that cold-start, new users,and the problem are caused by sparse data. The existing solutions mainly are filling and reducing dimension, which are applied to the environment with the data that relatively single, but the multi-tenant service platform environment in Saa S mode is complicated that limited the use of traditional methods.This paperproposes a collaborative recommendation based on LMF, classifyingthe heterogeneous relations of Saa S services, using the LFM model regularizing the single mode and factorizing the dual-modebased on the basis matrix factorization,then extracting the implicit informationto extend relevant data set, whichcan ease the inaccurate problem due to data sparseness in the collaborative filtering algorithm.2) Process of interest driftThere are many factors affect interest, personal preferences, product quality or the public evaluation, etc., but the basal information of existing solutions is relatively single, and interest model is only updated recently or used the memory function simply, which leads toover-fitting and lose diversity. For the tenants’ variable interest, this paper proposed the interest evolution strategy based on time effect which oriented the different levels of granularity, dividing the informationinto explicit and implicit information, then modeling and updating accordingly. The evolutionary algorithm based on explicit information is on the basis of service granularity, combining the different factors that affect interestswith time, establishing TRSVD dynamic model, through the score value to recommend services, then using the combination method to generate the personalized service set; The evolutionary algorithm based on implicit information is on the basis of service’s customizable point granularity, quantitating the behavior of tenants to the evaluation of services, updating short-termand long-term interest, then using genetic algorithm convert recommendation that consists of custom points with small granularity into personalized service set. Finally two results be fused and will get the prediction of the tenant personalized service.3) The experiment and testBecause of the particularity in multi-tenancy mode, the Saa S platform-related data should be collected and classified in the first place, then testing two algorithms separately. For the single mode and the dual-mode which based on LMF,usingregularization and factorization methods separately and exploring the relationship between data sparse degree and recommendation precision; Interest evolutionary strategy based on the time effect update the interests of explicit information as well asthe long-term and short-term interest of implicit information, then test the fusion of these two algorithms. Finally, analyze the experimental results, and compare accuracy and recall of the different kinds of recommendation.According to the above several aspects of research and related experiments, it is concluded that the proposed two kinds of optimization algorithm is effective to some extent make up for the deficiency of the collaborative recommendation: collaborative filtering algorithm based on LMF shows a good improvement of the accuracy under the condition of sparse data; And the interest evolution strategy based on the time effect alleviates drift, also improves the accuracy of recommendation results. The effect of the combination is obvious, reaching to the destination of this study.
Keywords/Search Tags:SaaS, multi-tenant, dynamic recommendation, sparse data, interest drifting
PDF Full Text Request
Related items