Initial Centroid Selection Optimization for K-Means with Genetic Algorithm to Enhance Clustering of Transcribed Arabic Broadcast News Documents

Abstract

In this research a collection of artificial intelligence techniques are combined together to optimize the process of clustering textual transcripts obtained from audio sources. Since clustering techniques have drawbacks that if not taken care of will produce sub optimal clustering solutions, it’s essential to attempt to optimize the clustering algorithms to avoid sub optimal solutions. As an attempt to overcome this problem, different artificial intelligence techniques are applied to avoid clustering problems. The main objectives of this research is to optimize automatic topic clustering of transcribed speech documents, and investigate the impact of applying genetic algorithm optimization and initial centroid selection optimization (ICSO) in combination with K-means clustering algorithm using Chi-Square similarity measure on the accuracy and the sum of square distances (SSD) of the selected clustering algorithm. The evaluation showed that using ICSO with genetic algorithm and K-means clustering algorithm with Chi-square similarity measure achieved the highest accuracy with the least SSD.