由於學術績效評價的需要，期刊排名問題已經引起各領域研究者的廣泛關注。過去研究主要集中在解決期刊的排名問題，無論是根據主觀態度的專家調查法或是基於客觀的引文評價方法。但是這兩種方法都有各自的優缺點，而且它們通常互補；另一方面，分級排名常用來提供決策制定與獎勵配置，是實務上相當有價值的方法。然而它是一種資源分配與組合最佳化問題，難以直接透過傳統引文分析方法得到結果，因此本研究試圖提出了兩種全新的方法，前者整合主客觀觀點，後者解決分級排名問題。 在本研究中，我們提出兩個演化式PageRank算法，第一個方法採用多目標粒子群最佳化演算法來平衡引文分析和專家意見的分歧。並透過實驗評估排名結果，證明了它的有效性。結果顯示，本研究的方法提升了專家對PageRank期刊排名結果的滿意程度。第二個演算法利用一種樹狀的染色體編碼來表示分級排名，利用此編碼可以有效地將等級分配與聲望值整合與一個染色體中，再透過遺傳演算法來求出基於引文與類別比例設定的最佳的等級分配。實驗也證明此方法可精準的分配等級比例並能有效的保證等級內成員的相似程度。; The journal ranking problem has drawn a great deal of attention from researchers in various fields due to its importance in the evaluation of academic performance. Most previous studies solved the journal ranking problem with either a subjective approach based on expert survey metrics or an objective approach based on citation-based metrics. Since both approaches have their own advantages and disadvantages, and since they are usually complementary, this work proposes a brand new approach that integrates the two previous approaches. In addition, the class-ranking is quite valuable method to provide decision makers with the incentive preparation in practice. However, it is a resource allocation and combinatorial optimization, so it is difficult to get results by the traditional citation analysis method. To this end, we propose the second approach in this study to solve the class-ranking with citation-based data. In this study, we propose two evolutionary PageRank algorithms. The first method uses the Multi-Objective Particle Swarm Optimization to balance citation analysis and expert opinion. Experiments evaluating ranking quality were carried out with citation records and experts’ surveys to show the effectiveness of the proposed method. The results indicate that the proposed method can improve PageRank journal ranking results. The second method uses a tree-based chromosome to represent a class-ranking problem. This encoding can be combining all assigned classes and prestige values in a chromosome effectively. We, also, use the Genetic Algorithm to determine an optimal graded assignment based on the citations and users constraints. Experimental results also proved that this method can be allocation classes precisely, and ensure the similarity between the members of the same class.
