Search

This study aims to determine effectivenes Simplified Cirlce Arc method used at the school level to prevent discrimination against student grades. This study National Examination data on mathematics subjects from of the Center for Educational Assessment in DKI Jakarta and Tangerang regions. Using the Rasch Model analysis, data obtained for 2135 in the DKI Jakarta (X) and 2271 in the Tangerang area (Y). The data was obtained after conducting Rasch analysis with Mean Square Outfit (MNSQ) of 0.5

Detection of differential item functioning (DIF) is needed in the development of tests to obtain useful items. The Mantel-Haenszel method and standardization are tools for DIF detection based on classical theory assumptions. The study was conducted to highlight the sensitivity and accuracy between the Mantel-Haenszel method and the standardization method in DIF detection. Simulation design (a) test participants consisted of 1000 responses in both the reference and focus groups, (b) the size of the proportion of DIF (0.1; 0.25; 0.50; and 0.75), and (c) the length of the multiple choice test with 40 choices the answer. Research shows that the Mantel-Haenszel method has the same sensitivity as the standardization method in DIF proportions of 10% and 25%, however, when the ratio of DIF proportions above 25% the standardization method is less sensitive, and conversely the sensitivity of the Mantel-Haenszel method increases. The standardization method has higher accuracy than the Mantel-Haenszel method in the DIF proportion of 10%, however, when the size of the DIF proportion above 10% the accuracy of the standardization method decreases, the accuracy of the Mantel-Haenszel method is higher than the standardization method. Thus, if the ratio of DIF is detected by the standardization method of (≤0.10), then the results of the standardization method are preferred as a reference. Conversely, if the proportion of DIF detected by the standardization method is (≥0.10), then the result of the Mantel-Haenszel method is chosen as a reference.

This research was intended to create Indonesian Text Spelling Correction system with the capability to handle and make correction to both kind of spelling errors, non-word and real-word errors. Existing spelling correction system was analyzed and made some adjustment and modifications to boost its accuracy. The proposed spelling correction system is built with Damerau-Levenshtein Distance that used in existing spelling correction system along with the adjustment and modifications. The result that achieved by the system that uses by existing spelling correction with the word level accuracy of 40.6% and an average processing speed of 18.4 ms per sentence while the result that achieved by the system that uses Damerau-Levenshtein Distance and Recurrent Neural Network with the word level accuracy of 21.3% and an average processing speed of 29.21 ms per sentence. The result of retest text that achieved by the system that uses Damerau-Levenshtein Distance and Recurrent Neural Network with the word level accuracy of 74%. Tujuan dari penelitian ini adalah untuk membuat sistem pengoreksian ejaan teks Bahasa Indonesia, yang memiliki kemampuan untuk menangani dan memperbaiki kesalahan ejaan, baik kesalahan kata tidak sah maupun kesalahan kata sah. Sistem koreksi ejaan yang sudah ada dianalisis kembali dan dilakukan beberapa penyesuaian dan koreksi untuk meningkatkan akurasi. Sistem koreksi ejaan yang diusulkan dibuat dengan metode Damerau-Levenshtein, yang digunakan dengan penyesuaian dan koreksi dalam sistem koreksi ejaan yang sudah ada. Pencapaian yang dicapai oleh sistem koreksi ejaan yang sudah ada menghasilkan akurasi kata sebesar 40,6% dan kecepatan pemrosesan rata-rata 18,4 milidetik per kalimat dibandingkan hasil yang dicapai oleh sistem yang menggunakan Damerau-Levenshtein Distance dan Recurrent Neural Network Akurasi menghasilkan akurasi kata sebesar 21,3% dan kecepatan pemrosesan rata-rata adalah 29,21 milidetik per kalimat. Hasil pengujian ulang teks yang dicapai oleh sistem menggunakan Damerau-Levenshtein Distance dan Recurrent Neural Network menunjukkan akurasi kata sebesar dari 74%.