Sistem Ekstraksi Informasi Lirik Lagu Daerah Menggunakan Term Frequency-Inverse Document Frequency (TF-IDF) dan Cosine Similarity

Abstract

This study develops an information extraction system for Indonesian regional folk-song lyrics by combining Term Frequency-Inverse Document Frequency (TF-IDF), Cosine Similarity, and Part-of-Speech (POS) Tagging with rule-based extraction. A preliminary survey of 33 respondents indicated limited familiarity with regional songs and difficulty understanding lyric meanings, motivating a web-based system developed using Rapid Application Development (RAD). The system processes lyric queries through case folding, tokenization, stopword removal, stemming, TF-IDF weighting, and Cosine Similarity ranking to retrieve the most relevant song. The selected lyrics are then analyzed to extract nouns, verbs, adjectives, pronouns, predicates, actions, subjects, objects, and related semantic elements. Black-box testing showed that all main functions operated successfully. Retrieval evaluation on ten test queries achieved 100% Top-1 Accuracy, while linguistic extraction against manual annotation produced Precision 0.74, Recall 0.76, and F1-Score 0.75. These results indicate that integrating statistical weighting and linguistic analysis can effectively support cultural documentation and regional-language learning.