AbstractThe assessment of academic journals is becoming more and more critical to many scientific activities. Such assessment can have a significant influence upon many issues ranging from publishing articles in academic journals to determining which universities are qualified to undertake major research projects in a country. Recently, journal ranking has been introduced as an official research assessment tool in many countries. Both peer review and indicator based methods have been exploited for this task in an effort to provide quantitative tools for ranking, evaluating and comparing academic journals. A number of data-driven evaluation measures have been developed based on citations, downloads and other statistical aspects of journals, which have the advantages of being more objective, while consuming less time and finance compared with assessments which based on peer review by human experts.
In this thesis, several intelligent system based methods for journal ranking are presented. The proposed approaches mainly utilise data-driven techniques including: 1) clustering algorithms, which are able to detect groups of academic journals that have similar indicator scores; 2) fuzzy aggregations, which provide more flexible and reliable aggregation of impact indicators than the use of Euclidean and Manhattan distances in journal ranking; and 3) clustering ensembles, by which linguistic variables are introduced to support interpretive clustering of journals. In addition, the mathematical properties of Ordered Weighed Averaging (OWA) aggregation of fuzzy relations are exploited to enhance their application in clustering. Also, link-based fuzzy clustering ensembles are proposed to improve the accuracy and robustness of fuzzy clustering. A method for expediting fuzzy clustering ensembles is introduced to reduce the effort in dealing with the growth of data volumes. Systematic experimental results demonstrated that these methods are not only flexible and interpretable, but also accurate in capturing and reflecting the impact of academic journals. The proposed approaches lead to a number of further developments. These include: selecting and grouping journal impact indicators; learning the weighting vector of OWA aggregation of fuzzy relations; applying fuzzy similarities to explore the boundary region of rough sets; and using the link-based fuzzy consensus function to support re-sampling based clustering ensemble. The proposed approaches for journal ranking can also be employed to solve clustering or ranking problems in applications other than journal assessment
|Date of Award||27 Aug 2015|
|Supervisor||Qiang Shen (Supervisor) & Changjing Shang (Supervisor)|