Evaluasi Algoritma Textrank pada Peringkasan Teks Berbahasa Indonesia

Abstract

Automatic text summaries are commonly used to summarize text using computer program to get a summary of the text. Usually automatic text summary use extractives method because this method retrieves important information from a text without altering or modifying the information. One of the algorithms that can be used to automatic text summarize is TextRank algorithm. The advantage of using TextRank algorithm is it does not require depth knowledge of a language and does not require data training to be able summarize text. This algorithm works with represent the sentence in text into the graph, then calculate the value of each sentences using similarity to determine important sentences, in this research we also use TextRank modification that is using levenshtein distance to calculate summary by comparing similarities between strings by using enter, delete, or replacing string characters. Automatic text summaries using TextRank is done by summarizing 100 Indonesian texts which will be evaluated using ROUGE. ROUGE evaluation works by comparing summary results from TextRank with manual summaries by experts Indonesian linguist. The result obtained by the TextRank algorithm get an average F-Score of 0,439 on ROUGE-1 and 0,3186 on ROUGE-2. While TextRank modification gets an average F-Score 0,3999 on ROUGE-1 and 0,2805 on ROUGE-2. When compared with the results of a summarize English text using TextRank algorithm that gets an average F-Score of 0,4708, the results are not much different. This mean that TextRank algorithm can be used to summarize Indonesian text.