Feature Selection Optimization for Identification of Citation Sentences in Scientific Journals

Scientific journal publication is an easy way to disseminate findings or research. But in doing the writing study, researchers often forget the source of the basics of the theory and the results of previous studies that are used as a reference. In some cases, these actions can be considered as a form of plagiarism, they can also damage the integrity of an academy. Therefore, citation writing is very important in research. In this study, researchers tried to detect citation and non-citation sentences contained in the CL-SciSumm dataset in 2018 using classifiers such as SVM. Not only that, classification is also helped by using feature selection to reduce unneeded dimensions. Feature selection used includes TF-IDF, IDF, WordNet and Word2Vet Similarity. In doing classification, it shows that the use of TF-IDF and SVM methods is the best combination of other features, this is evidenced by the high accuracy value that reaches 90,23%, recall reaches 92,76%, and f-measure reaches 88,13%.