<![CDATA[The use of search engines as a learning resource in elementary schools increases the risk of exposure to inappropriate content, necessitating a fast and accurate filtering mechanism from the very beginning of the search. This study aims to design and evaluate a Bloom filter-based pre-query filtering system to detect risky keywords as early as the first 1–2 characters of the input. The evaluation was conducted using a query dataset representing the search patterns of elementary school students in grades 1–5 with variations in Bloom filter size (1024, 2048, 4096 bits) and the number of hash functions (7, 10, 15). Experimental results show that the 1024-bit configuration with 7 hash functions yields an average filtering time of 15 ms, a false positive rate (FPR) of 1%, and a false negative rate (FNR) of 0%, thereby meeting real-time response requirements. The 2048–4096-bit configurations reduce the FPR to 0% but increase latency to 18–25 ms. These findings demonstrate a measurable trade-off between latency and filtering accuracy. This study empirically contributes to showing that Bloom filters are effective as a low-latency initial filtering mechanism. The proposed system has the potential to serve as the foundation for developing safer and more responsive educational search engines for elementary school students]]>
Copyrights © 2026
