Observing the Performance of the TextRank Algorithm on Automatic Text Summarization for Bahasa Indonesia.

The research about automatic text summarization is common in English text. According to the previous study, automatic text summarization in Bahasa Indonesia is still challenging due to research in this area, especially the research which discusses TextRank algorithm performance, which is still meagerly. Accordingly, this research observes the performance of the TextRank algorithm to summarize the text in Bahasa Indonesia. The TextRank algorithm summarizes a text by sorting out the essential words and relevant sentences regardless of the source language. This algorithm uses a vertex to represent a word. The similarity measurement process will calculate the overlapping words (the same word between two vertices). These overlapping words are represented by the edge, which connects the vertices. Thus, the text forms a graph. This research focuses on the similarity measurement process to determine relevant sentences in a text. As the similarity measurement is critical for the summarization result, this research switches the original process to the Levenshtein Distance algorithm and observes its performance. This research uses the human-produced summarized text by the expert in Bahasa Indonesia linguistics to evaluate the result. The evaluation method is conducted by using ROUGE-1 and ROUGE-2. The result shows that the average of ROUGE-1 and ROUGE-2 for the TextRank algorithm is 0.439 and 0.3186, respectively. Meanwhile, the modified TextRank obtains 0.3999 and 0.2805, respectively. Both of the algorithms have not shown satisfactory results as expected. [ABSTRACT FROM AUTHOR]