A sentence-centered account of text summarization, this work can be applied to any language. Characteristics of the Indonesian language are irrelevant to the proposed approach and are not discussed. Neither is the suitability of the proposed approach to the Indonesian language.

The paper’s contribution is that it exploits an abstract extraction method that reaches an accuracy close to generative summary. It first converts the sentences into sentence vectors, and then calculates the similarity between the sentences, clusters them, and extracts and sorts the selected sentences from the clusters. The first step adopts the Google word2Vec model. It is not clear, however, what method is used to calculate the sentence similarity, except for predefining the similarity threshold. Further, clustering determines “the sentence with the largest amount of information” to be the core sentence of the cluster. This sentence is extracted to be included in the summary.

The experiments were carried out on an especially constituted corpus of Indonesian texts. Tests were performed against ROUGE-1, ROUGE-2, and ROUGE-3 as measurement indices, and the results compared with the results of six popular state-of-the-art text summarization algorithms. The comparison clearly shows the proposed method’s superior performance.

This well-described paper, though lacking some technical details, includes a quite substantial list of related work. It is a good read for scholars and practitioners of text summarization in general, regardless of the language.