The application of Centroid linkage hierarchical method and Hill climbing method in comments clustering online discussion forum

Several problems are rised in order to enhance the effectiveness of communication in online discussion. The similarity and repetition of comments in terms of questions in the sentences or text meanings as well as triggers the emerging of miscommunication amongst participants in a forum discussion are investigated. Moreover, some comments seems are ignored or not been touched by other participants and in advance the effective used of forum discussion as knowledge acquisition and sharing can not be achieved. This paper studies the application of Centroid Linkage Hierarchical Method (CLHM) Algorithm and Hill Climbing methods in findings the similarity value of participants comments and clustering based on it. The analysis follows the text mining process including text processing, text transformation, attribute selection and pattern discovery. In order to test the validity and accuracy of both application methods, confusion matrix in euclidean and consine similarity were calculated. As the results, from variety numbers of comments groups, including Bersosial.com in 17 comments, Indowebster.com in 27 comments and Teknojurnal.com in 51 data comments provided the value of well-separated clusters performed. This testing also defined that the alteration of threshold and altitude did not affect the clustering process. From the calculation of F-measure values in confusion matrix explained that consine similarity provided better result that euclidean distance where teknojurnal 0.89, indowebster,com 0.71 and bersosial.com 0.57. This showed that CLHM algorithm and Hill climbing methods are effective approaches and have been successfully applied in comments clustering of online discussion.