water cycle algorithm

The clustering of large-scale textual data plays a key role in the easy browsing and scrolling of huge documents by organizing their structures. One of its applications is scientific document clustering, which was presented or published in conferences and journals to categorize articles with common topics. The research's purpose is to improve the quality and speed of scientific document clustering and reduce the need for manual parameter settings.

The paper proposes a novel automatic scientific documents clustering method based on the Water Cycle Algorithm (WCA). The proposed method consists of different stages of pre-processing, scientific document representation based on TF-IDF adapted for scientific documents, defining the mechanism of activating and deactivating cluster centers from a certain number of cluster centers in order to create flexibility in the number of scientific document clusters and the WCA to optimize the number of cluster centers and their coordinates.

In this paper, two benchmark datasets, NIPS 2015 and AAAI 2013, are used, which contain information on articles presented at two conferences. Also, the automatic clustering has been implemented based on four meta-heuristic algorithms: Differential Evolution Algorithm, Genetic Algorithm, Artificial Bee Colony Algorithm and Particle Swarm Optimization. Davis-Bodelin index and Chu and Su index were utilized to evaluate the quality of the obtained results. The comparison of the obtained results shows that the proposed method has better quality and efficiency in scientific document clustering and is able to overcome the challenges of unbalanced and large textual data clustering.

Originality/Value:

 In the proposed automatic clustering method, the WCA has been used for the first time. With regard to scientific articles in the same field are presented in a journal or conference, and statistical analysis has a special place to identify keywords of these documents in their clustering quickly, TF-IDF and the mechanism of activation and deactivation of cluster centers have been combined in the proposed scientific documents clustering.