Perbandingan Semantic Search Berbasis Sentence-Bert dan TF-IDF pada Pencarian Dokumen Ancaman Siber

Abstract

The increasing number of cyber threat documents has made the information retrieval process increasingly complex because keyword-based methods are unable to understand semantic relationships between documents. This study aims to compare the performance of the Term Frequency–Inverse Document Frequency (TF-IDF) and Sentence Bidirectional Encoder Representations from Transformers (Sentence-BERT) methods in a cyber threat document retrieval system. The research stages include Importing the CyberBERT Dataset, text preprocessing, building a search model, testing using queries, and evaluating the results based on document relevance and response time. The Dataset used consists of 33 documents processed into vector representations using TF-IDF and 384-dimensional embedding using Sentence-BERT. The results show that both methods are able to return relevant documents to the given query, but produce different document sequences due to differences in text representation approaches. The TF-IDF method has a faster response time of 0.0089 seconds, while Sentence-BERT requires 0.0312 seconds. On the other hand, Sentence-BERT is better able to capture semantic relationships, resulting in more contextual searches. Thus, TF-IDF is superior in terms of efficiency, while Sentence-BERT is more effective in understanding the meaning of documents, making it suitable for semantic search systems in the cyber threat domain.