Determination of Sentence Similarity Level Using Vector Space Model (VSM) and Word Relationship Weighting for Plagiarism Detection for Indonesian Documents

Plagiarism is an essential issue in writing for both articles and scientific works. One example is the error of taking a reference in a paragraph on someone else's scientific work. Suppose the manual checking process is carried out to determine plagiarism actions. In that case, it will be difficult and lead to the big problems. A computerized system is considered capable of quickly finding similarities in writing articles or scientific works that lead to plagiarism. In this study, detection of similarity in the document that leads to plagiarism is carried out by building an inverted index for each sentence in document in the database by performing data preprocessing then testing one by one for each sentence in the test document using the VSM method with term-frequency (TF) and inverse document frequency (IDF). The results obtained are that VSM with TF-IDF weighting can calculate sentence similarity with a high similarity value of 1 (100%). However, VSM with this weighting does not consider the word's position, then using the proposal of a 1-gram word weighting technique with a comparison process from left-to-right (left-to-right) and right-to-left (right-to-left) between database sentences and sentences. The test shows that there are differences in the similarity results. In the sentence examination time on testing, it is also found that the more the number of examinations, the more time is needed in the process of checking the similarity between sentences.