Marginal relevance selain digunakan untuk information retrieval bisa juga digunakan untuk peringkasan dokumen. Namun pada penelitian sebelumnya ringkasan menggunakan maximum marginal relevance (MMR) menghasilkan akurasi yang masih rendah. Selain itu terkadang ringkasan yang dihasilkan terlalu panjang. Idealnya sebuah ringkasan mempunyai panjang ringkasan tidak lebih dari setengah teks asli. Untuk mendapatkan hasil ringkasan yang ideal, pada penelitian ini terdapat tiga proses utama, yaitu proses preprocessing, meliputi case folding, filtering, tokenizing kalimat, tokenizing kata, removal stopword, stemming, tf-idf, dan cosine similarity. Kemudian proses pengurangan kandidat kalimat ringkasan dengan pengklasifikasian menjadi dua kelas menggunakan metode k-nearest neighbors (KNN). Dan selanjutnya proses pembentukan ringkasan menggunakan metode maximum marginal relevance (MMR). Berdasarkan hasil pengujian menggunakan data sebanyak 35 berita online berbahasa Indonesia, disimpulkan bahwa ringkasan menggunakan KNN dan MMR lebih baik namun belum signifikan dibandingkan dengan ringkasan yang menggunakan MMR saja. Dengan rata-rata selisih precision 9%, f-measure 1%, dan akurasi 2%. Namun ringkasan menggunakan KNN dan MMR masih kurang baik karena rata-rata selisih recall sebesar 5% dibandingkan dengan ringkasan MMR saja.

Marginal relevance not only used for information retrieval can also be used to summaries document. However, previous research has summaries using the maximum marginal relevance (MMR) produces accuracy is still low. Also sometimes the summaries produced too long. Ideally, a summary in length summary of no more than half of the original text. To get the ideal results summary, in this study, there are three main processes, namely the process of preprocessing, includes case folding, filtering, tokenizing sentence, tokenizing word, stopword removal, stemming, tf-idf and cosine similarity. Then the process of reduction of sentence summary candidate with classification into two classes using the k-nearest neighbors (KNN). And further the process of establishing a summary using maximum marginal relevance (MMR). Based on test results using the data as much as 35 Indonesian language online news, it was concluded that the summary using KNN and MMR is better but still not significant compared with the use of MMR summary only. With an average difference of 9% precision, f-measure 1%, and an accuracy of 2%. However summaries using KNN and MMR is still not good because the average difference in recall by 5% compared to MMR summary only.

https://elib.unikom.ac.id/gdl.php?mod=browse&op=read&id=jbptunikompp-gdl-devidwiyan-35904