| 摘要: | 研究期間：10108~10207;In the last decade, the problem of getting a consensus group ranking from all users’ ranking data has received increasing attention due to its widespread applications. The group ranking problem is to construct coherent aggregated results from preference data provided by decision makers. Traditionally, the output of the group ranking problem can be classified into two types: ranking ordered lists and consensus lists. Unfortunately, these two traditional outputs suffer from their own weaknesses. For ranking ordered lists, most previous approaches pay close attention to minimize the total disagreement between multiple input rankings, to ultimately obtain an overall ranking list which represents the achieved consensus. They neglect the fact that user opinions may be discordant and have no consensus, in which we are still enforced to get a complete ranking result. For maximum consensus sequences, there may have many resulting consensus sequences which need to be checked, and thus lead to information overloading, tedious, and not easy to understand. As a result, users are difficult to grasp the whole relationship among items. To overcome these weaknesses, in this three-year project we will propose a new type of group knowledge to represent the coherent aggregated results of users’ preferences. We apply the clustering concept into the group ranking problem and generate an ordering list of segments containing a set of similarly preferred items, called consensus ordered segments, as the output. Accordingly, algorithms will be developed to discover consensus ordered segments from users’ ranking lists. In the first year, we will design a global search GA algorithm for mining consensus ordered segments. In the second year, a local search greedy algorithm with incremental capability to compute objective value will be proposed. Finally, in the last year we will propose an integrated approach, which combines the global search algorithm with the local search algorithm, to solve the consensus ordered segments mining problem. And we will apply PSO+GA to guide the global search procedure and use the greedy method + incremental operations to do the detailed search. The advantages of our approach are that (1) the knowledge is built based on users' consensuses, (2) the knowledge representation can be understood easily and intuitively, and (3) the relationships between items can be easily seen. These advantages indicate that our approach can improve the weaknesses of previous approach in mining group knowledge. |
