大群体偏好下的信息融合与共识理论及其在推荐系统中的应用研究

Large group decision making refers to the selection of the best option from a set of alternatives according to the preferences of a large number of people. Compared to small group decision making, the participators in a large group decision making problem are no longer limited to some special experts and the decision information is collected from different places in different time. It is a more appropriate way for the Internet + environment. In this project, based on the limitations of the current literature, we summarized some critical problems that need to be further studied in the field of large group decision making: information fusion and aggregation, incomplete information processing, and consensus reaching process. (1) The theory of information fusion is presented considering heterogeneous information including hesitate information. Based on the development of probability distribution, we provide a new tool to deal with the hesitate information. (2) An approach utilizing the statistical analysis is proposed to manage the incomplete information. In this approach, some estimate criteria are deigned to identify if the missing information can be estimated or not. Then by use of the individual preferences, subgroups’ information, and other groups’ information, a new method to estimate the missing information is given. (3) Some consensus models are proposed to assist the consensus reaching processes under a large group decision-making environment. Taking into account some constraint conditions (e.g., cost, time), interactive or automatic consensus models can be developed. Some manipulation behaviors between the individuals or the sub-groups are analyzed, and correspondingly some control strategies are designed and discussed. (4) The developed theory of large group decision making is applied to improve the effectiveness of some group recommendation systems which become increasing popular in business applications. The results of this study aim to provide new insights for solving the problems under a large group decision making setting.

大群体决策呈现出决策参与人不再局限于某领域专家、决策信息来自异时异地等新特点，是一种更适应于互联网+时代下的决策方式。本项目分析已有研究的局限性，提炼出大群体偏好信息下的信息融合与集结、不完全信息处理及共识一致性等关键问题。（1）提出大群体环境下基于犹豫信息的信息融合理论。发展可能性分布理论，为包含犹豫信息的偏好信息融合提供新的工具。（2）提出基于统计分析的不完全信息估计方法。设计缺失信息的可估计判别准则，利用个体信息、子群体信息、其他子群体信息，给出估计缺失信息的新方法。（3）建立大群体环境下的共识决策模型。考虑共识过程中的约束条件（如成本、时间），设计交互式或自动化共识模型；分析个体或子群体的策略操纵行为，探讨控制策略。（4）拓展大群体决策理论到具有广泛商业应用前景的群体推荐系领域，改进其推荐有效性。研究结果旨在为解决大群体偏好信息下的决策问题提供借鉴，推动现有理论研究的深化。

大群体决策呈现出决策参与人不再局限于某领域专家、决策信息来自异时异地等新特点，是一种更适应于互联网+时代下的决策方式。本项目分析已有研究的局限性，主要研究内容围绕大群体环境下的偏好信息表达与信息融合、个体一致性及共识一致性、决策方法及应用等方面展开。取得的主要研究结果有：（1）提出了大群体决策中基于动态聚类与共识的决策过程。由于个体意见随着决策过程会发生改变，已有研究采用的固定聚类的观点并不能很好的反映现实决策情形。项目提出基于k-均值聚类方法对大群体进行动态聚类，利用可能性分布判断矩阵进行信息融合。（2）提出了一系列基于优化或迭代的个体一致性改进与共识达成模型，解决了大群体决策中个体理性与群体理性的不相容问题。鉴于现有的相关优化模型大都未考虑解的唯一性，提出多阶段个体一致性与群体共识达成的优化模型。对判断矩阵的次序一致性的满足条件进行了理论分析，导出并证明了满足次序一致性对应的约束条件。（3）发展了大群体决策中的决策方法。提出了适用于大群体决策背景的两种多属性决策方法：TOPSIS方法和VIKOR方法。建立了社会网络环境下基于二阶交互的意见动态演化模型，提出控制意见演化的三种策略。运用比较分析与模拟分析等方法对所提出的理论、模型和方法进行了验证。历时四年的研究，项目组共录用或发表相关的论文18篇，其中ESI高被引论文2篇，在科学出版社出版专著1本。根据FMS管理科学高质量期刊目录，所发论文中有2篇属于A级或T1级，有10篇属于B级。这些论文成果已得到国内外一些著名学者的引用，产生了较大的学术影响。项目培养了硕博士生10余名。项目成果发展了大群体决策的基础理论与方法，在决策支持系统、网络数据挖掘、商业推荐系统中具有一定的应用前景，能够为解决群体决策问题提供思路和借鉴。