Scientific Article Clustering Using String Similarity Concept

Abstract

Scientific articles are one of the results of a study or research. Scientific articles from a study are disseminated through publications in a journal and a seminar. Journal or seminar committee classifies articles into several research topic clusters. To cluster articles, sometimes the journal committee reads the contents of the article to cluster it manually. The manual process is ineffective and inefficient in terms of processing time. From these problems, a system was created to help classify articles using a computer system based on the contents of the article. The application of article clustering was built using the word frequency method adopted from the term frequency of TF-IDF and using cosine similarity to cluster scientific articles according to the research topics. The processes in clustering scientific articles include calculating the number of words used, searching for words that often appear, eliminating stop words based on the Tala's stop word list, then from words that often appear are made into the new article’s keywords. From the keywords obtained, then the level of similarity is calculated with the keywords used in the research topic using cosine similarity. The research topics used are based on the body of knowledge in computer science and information system based on ACM (Association for Computing Machinery) & IEEE (Institute of Electrical and Electronics Engineers). The system was built in the form of a web application using the PHP programming language and MySQL database with a responsive web design technique. The results of this research are that articles entered into the system could be generated into the new keywords by listing the words that appeared a lot. From the keywords obtained, the articles were successfully clustered using cosine similarity. The level of clarity from the descriptions of the research topics affected the results of clustering. The more clear and complete the description is given, the better the results of the clustering were obtained. From 102 article titles of KNS&I (Konferensi Nasional Sistem & Informatika) 2017 that were entered, 71.6% of articles were clustered correctly. From these data, the largest research publications were obtained in the "information management" cluster.