HAI THÁC CÁC MẪU PHỔ BIẾN TRÊN CƠ SỞ DỮ LIỆU TRỌNG SỐ ĐỘNG

Abstract

Pattern mining is one of the fundamental tasks in modern data mining. In particular, mining frequent weighted patterns (PWPs) from quantitative databases is an important problem aimed at discovering frequent patterns based on item weights. However, existing studies have largely overlooked the scenario in which item weights vary over time — known as dynamic weighted databases (dWDBs). In many real-world applications, item weights fluctuate to reflect shifts in their significance, such as profit margins or temporal importance (e.g., air conditioners are more popular in summer, and medical masks become critical during respiratory disease outbreaks).

In this paper, we introduce a new problem of mining PWPs from quantitative databases with dynamically changing item weights — referred to as dynamic quantitative databases. To address this problem, we first propose an algorithm named dPWPT, built upon the traditional tidset structure. We then present a second algorithm, dFWNL, which utilizes a novel data structure called dWNList to efficiently mine PWPs from dWDBs. Finally, we conduct extensive experiments on various dWDBs to evaluate and validate the effectiveness of our proposed algorithms.