IMPLEMENTATION OF SUPPORT VECTOR MACHINE METHOD IN CLASSIFYING SCHOOL LIBRARY BOOKS WITH COMBINATION OF TF-IDF AND WORD2VEC

Abstract

The development of technology in education is integral to enhancing its quality, such as implementing information technology in school libraries. Searching for books in school libraries is time-consuming due to conventional book classification, lacking organization based on classifications. Therefore, implementing information technology in school libraries is crucial to improve library management effectiveness. An innovative solution optimizing library management involves leveraging artificial intelligence, particularly machine learning. In applying machine learning to library book classification, Support Vector Machine acts as an algorithm understanding patterns and characteristics of book titles, categorizing them into Dewey Decimal Classification (DDC). The dataset comprises 10 classes aligned with DDC. Random data collection follows an 80:20 scale for training and testing data. Data preprocessing is an initial research stage, addressing imbalanced data through oversampling. Testing the SVM algorithm with a linear kernel and C = 1 parameter is conducted three times using different feature extraction methods: TF-IDF alone, Word2Vec alone, and a combination of TF-IDF and Word2Vec. Model performance evaluation employs K-Fold Cross-Validation. After the three objective tests, the most accurate book classification results were obtained using a combination of TF-IDF and Word2Vec feature extraction. It's concluded that SVM's book classification method can be applied, yielding the highest accuracy of 73% with the TF-IDF and Word2Vec feature extraction combination. This outperforms other feature extraction methods, with precision at 83%, recall at 72%, and an F1-Score of 76%.