On Information Aggregation Models And Application Towards Complex Evaluation Issues

Multi-attribute comprehensive evaluation is an important part in the field of management science. Essentially, it is used to provide a process of judgment and selection for complicated questions. A classical evaluation process refers to a few obvious steps including identifying the evaluation objective, building evaluation attributes, obtaining attribute weights, selecting or constructing aggregation methods, and getting the final result. Among all of the above steps, information aggregation, where incorporate overall information, is one of the most important.So far, there are great many aggregation methods developed in terms of various perspectives, such as attribute importance, location, relation and the closeness to ideal sample, and the like, therefore, the relevant achivements appear to diverse and decentralized. As to this issue, some researchers suggested combination process and presented a stream of combined aggregation methods. However, the combined methods inevitably meet limits when facing with complex situations, which are as follows.(1) Most of existing aggregating methods primarily consider attribute features or the relationship of attributes. However, considering the distributed structure of attributes is also essential in aggregation. What’s more, it isnecessary to solve complicated evaluation questions or highlight special demands. Usually, by analyzing the distribution characteristics of judgments, we may find some hidden valuable information—such as seeking various groups with different opinions, recognizing special ability of inidividuals, identifying outliers and so on.(2) In traditional aggregation processes, such information with signal form or structure can be fused through aggregation method. Nevertheless, with the development of communicating technology, large-scale group participating in evaluation in different time and space becomes more feasible than ever before. Because individual are greatly different in understanding abilities and knowledge levels, the judgment information offered by them might be inconsistent. Therefore, we need to develop a unitary fusion framework for various types of information to guarantee the effectiveness of large-scale group evaluation.Based on the above analysis, this paper proposes two novel types of aggregation methods, namely, the density aggregation model serving the aim of "inner structure analyzing of one-dimentional data" and the generic information aggregation model serving the aim of "information aggregation of multi-dimensional structure". The main works and conclusions are shown in the following.(1) After introducing the theory of density aggregation operator briefly, we discuss three types of aggregation model such as "linear", "nonlinear", and "harmonic" ones, and then examine and compare the properties of them. In view of this, we further investigate the sensitivity of various operators including arithmetic averaging operator, geometrical averaging operator, harmonic averaging operator and density aggregation operator. As a result, we find that the density operator feature higher sensitivity than others, which can be applied to identify outliers.(2) According to the analysis of density aggregation operators, we apply it to two fields:rewards and punishments management, and information aggregation of wisdom of crowds. Firstly, with regard to the question of rewards and punishments management, we identify advantage and disadvantage of alternatives by classification, and then build the density weights to put rewards and punishments information into use. The validity of its result is addressed based on comparison with the performance of the weighted averaging operator. Furthermore, we extend the density operator to the situation of large-scale group prediction. We come to a conclusion that density operator perform better as the predictions occur under non-rational distribution circumstances, hence it can be used to weaken the effect of outliers.(3) Concerning the problem of information fusion of multi-dimensional structure, we develop a novel type of aggregation method, generic structural information aggregation model. The whole process, including the establishment of information fusion framework, the transformation of blended evaluation data, the algorithm of information fusion framework, and the final normative solving method, is detailedly addressed in this paper.(4) To improve the application performance of generic structural information aggregation model, we further discuss the simplified solving algorithm of information fusion framework, and some numerical examples are given to illustrate the effectiveness of this algorithm. Meanwhile, we apply this generic structural information aggregation model to deal with the issue of government performance evaluation with multi-participants. As expected, it really outputs desirable results.At the end of this paper, we sum up the findings of two types of aggregation models, and point out some remaining issues, which are worthy of being further explored in the future.