Type-2 Fuzzy Decision Making Methods And The Application In Personalized Recommendation

Recently, the researches on big data driven management and decision making have become a hot topic in management science area. Due to the complexity of the decision environment and decision behavior, how to achieve the user’s personalized recommendation based on user’s behavior preference and knowledge discovery and how to combine the multiple attribute decision making (MADM) with the fuzzy-based personalized recommender system in a proper way are very meaningful topics both in theory and practice. This thesis will mainly take the advantages of type-2 fuzzy sets (T2FSs), especially the powerful ability in computing with words and semantic information processing, to solve some decision making problems in personalized recommender systems. The objectives of this thesis will focus on type-2 fuzzy aggregation and decision making, with some related techniques of granular computing (GrC) and type-2 fuzzy multiple attribute decision making, to study on the personalized recommendation model and provide some new ideas and solutions for the problems under complex, dynamic and incomplete information and big data environments for the personalized recommender problem. The main research results are listed as follows:(1) Based on information fusion theory, we develop an interval type-2 fuzzy Maclaurin symmetric mean operator and its dual form. An interesting property of parameter monotonic is studied in details. Compared with the existing interval type-2 fuzzy aggregation operators, the prominent character of our proposed operator it that it can capture the multiple interrelationships among the arguments, and also can deal with such a problem in a flexible way. Based on which, a new method to handle interval type-2 fuzzy decision making problem is developed. Moreover, this new method is applied to the peer review problem in the recommender system of Sciencepaper online. These new aggregation methods not only provide a good way to deal with such a difficult problem of type-2 fuzzy aggregation in theory, but also provide a basis for further study the decision making problem with criteria interaction.(2) Based on multiple attribute decision making theory, we follow our research from three aspects to study interval type-2 fuzzy multiple attribute decision making:ranking method, utility model and optimization model, respectively. Firstly, aiming at the most difficult problem oftype-2 fuzzy sets:Ranking. We propose a new combined ranking method based on three elementary means. The mathematical proof indicates that the proposed ranking method is not only satisfies total order (linear order), but also satisfies admissible order. Based on this desirable property, we further propose an interval type-2 fuzzy combined ranking method with the help of ordered weighted average operator. Secondly, based on behavial decision theory, in virtue of the flexible three parameter (FTP) utility function, we develop some FTP-based interval type-2 OWA operators, and also propose an interval type-2 fuzzy clustering multiple attribute decision making method to deal with large-scale complex decision making problems. Furthermore, we combine the Prospect theory and the VIKOR methods to develop an interval type-2 fuzzy behaviour decision making method based on dynamic reference point, and apply this method to high-tech risk evaluation recommender system. Finally, we extend the classic LINMAP method in multiple objective optimizations under the interval type-2 fuzzy environment, study the decision maker’s preference extraction method and construct some optimization decision models with incomplete attribute weight information, these models are successful applied to mobile phone selection in the recommender system. Therefore, these new methods not only enrich the foundation of type-2 fuzzy decision making, but also extend the scope of application of type-2 fuzzy decision making theory in personalized business recommendation.(3) Based on granular computing theory, we study on the sparsity problem of the scoring matrix in personalized recommendation. Regarding granular computing as an entry-point, we establish a collaborative optimization model based on the coverage and specificity criteria. The corresponding intelligent optimization algorithm is designed to solve this problem. To some extent, the new methods can overcome the limitation of high complexity brought by the two popular methods:matrix factorization and variational optimization, and provide some necessary research solutions to solve the bottleneck problem of personalized business recommendation system.(4) Based on type-2 fuzzy decision making theory, we study the high order fuzzy-based personalized recommendation model. Motivated by the idea of information granular modeling and two popular multiple attribute decision making methods, we propose a multiple attribute hybrid recommendation algorithm with the aid of collaborative filtering and content-based methods. To solve the problem of parameter setting, we further study on the parameter setting method with personalized, differentiated features.Applications are shown for all approaches proposed in this thesis. The related studies can further enrich and improve type-2 fuzzy decision making theory and methods in theoretically, and provide some new tools and methods for e-commerce personalized recommendation in practical applications.