PENGARUH PENAMBAHAN FITUR TOPICS DALAM PERHITUNGAN NILAI KEMIRIPAN REPOSITORI PADA GITHUB

SYAFINA NURUL AIDA, Suprapto, Drs., M.Kom., Dr.

2019 | Skripsi | S1 ILMU KOMPUTER

Pengembangan sistem deteksi kemiripan repositori GitHub dilakukan untuk mendeteksi repositori yang memiliki kemiripan fungsi satu sama lain dengan menghitung kemiripan fitur repositori. Dalam penelitian sebelumnya, fitur readme files, stargazers, dan waktu pemberian star digunakan untuk perhitungan kemiripan repositori. Namun, diperlukan pengembangan lebih lanjut dalam proses perhitungan kemiripan repositori. Ditambahkan fitur topics untuk menghitung kemiripan repositori GitHub. Pada penelitian ini, dilakukan perhitungan besarnya pengaruh fitur topics sebagai tambahan dalam perhitungan kemiripan repositori. Dilakukan perbandingan nilai success rate, confidence, dan presisi dari evaluasi output antara perhitungan kemiripan repositori dengan tiga fitur (readme files, stargazers, dan waktu pemberian star) dan dengan empat fitur (tiga fitur ditambah fitur topics) untuk mengetahui pengaruh penambahan fitur topics. Sebanyak 501 data repositori digunakan sebagai basis data dan sebanyak 20 repositori diantaranya digunakan sebagai kueri untuk menghasilkan output lima repositori teratas yang memiliki nilai kemirpan paling tinggi dibanding kueri. Pada proses evaluasi hasil pengujian, dilibatkan sebanyak empat orang penilai yang memberikan penilaian pada hasil output perhitungan kemiripan dengan nilai kesepakatan Kappa sebesar 0,47. Nilai evaluasi akhir dari semua penilai kemudian ditentukan menggunakan skala Likert. Perhitungan kemiripan repositori dengan penambahan fitur topics mendapatkan nilai success rate, confidence, dan presisi lebih tinggi jika dibandingkan dengan perhitungan kemiripan repositori tanpa fitur topics dengan nilai succes rate pada T=4 sebesar 55%, nilai succes rate pada T=5 sebesar 25%, mean confidence sebesar 2,35, median confidence sebesar 2,00, mean presisi sebesar 0,16, dan median presisi sebesar 0,20.

The development of repository similarity detection system was conducted to detect similarity between repositories by calculating features similarity. In the research that has been done, readme files, stargazers, and timestamp of star addition can be used for repository similarity calculations. However, further development is needed in the process of repository similarity calculation. Topics feature was added to calculating the similarity of repositories. In this study, the influence of topics features to the repository similarity calculation was calculated. In this study, we conduct a comparison of success rate, confidence, and precision value of final evaluation score from the output between two systems, i.e. system with three features (readme files, stargazers, and timestamp) and one with four features (with topics feature) to determine the influence of topics feature. The number of repository data use as database was 501. From the database there were 20 repositories ID used as query and for each of the query in the similarity calculation there were five first highest value of similarity selected as an output. The evaluation process involved four raters who gave an assessment of the output system with Kappa value was 0,47. The final evaluation score was calculated using Likert scale. System with topics feature addition obtained higher success rate, confidence, and precision values than system with only 3 features (without topics feature), with success rate obtained at T=4 was 55%, success rate at T=5 was 25%, mean confidence was 2,35, median confidence was 2,00, mean precision was 2,00, and median precision was 0,20.

Kata Kunci : Perhitungan Nilai Kemiripan, GitHub, Fitur Topik, TF-IDF, Fleiss Kappa, Skala Likert