Efficiency and generalization of the extraction of negative association rules for the extraction of latent rules

2023 Fiscal Year Final Research Report

• Project/Area Number: 19K12096
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: University of Yamanashi
• Principal Investigator: Iwanuma Koji 山梨大学, 大学院総合研究部, 教授 (30176557)
• Project Period (FY): 2019-04-01 – 2024-03-31
• Keywords: データマイング / 負の相関ルール / 圧縮 / 極小生成子 / アルゴリズム / 一般化 / 飽和集合

Outline of Final Research Achievements

In this study, we investigate an efficient method for extracting negative association rules for the purpose of extracting rule-type knowledge latent in huge data. Compression of a large number of rules is extremely important because it is considered to be essentially a generalization and abstraction of rules. Therefore, we developed several compression principles for negative rule sets, principles for fast extraction of negative rule sets in compressed form, and algorithms for fast execution of these principles. For example, we have theoretically clarified the properties of minimal generators and developed algorithms for their fast extraction. We have also developed a fast on-line extraction algorithm for strongly closed itemsets. Furthermore, we have developed an algorithm for directly enumerating negative rules in a compressed form on an enumeration tree, which is formed on the downward closure property of minimal generators.

Free Research Field

人工知能基礎

Academic Significance and Societal Importance of the Research Achievements

現実の世界を考えるとき，潜在因子も考慮した法則の発見抽出は重要な課題である．しかし，これまでのデータマイニングの研究では殆ど考慮されてこなかった．潜在的ルールの発見抽出は，統計的学習での潜在パラメータの推定問題等とは異なる問題であり，重要である．本研究の成果である負ルール集合の圧縮原理と直接抽出アルゴリズムにより，抽出計算が大幅に効率化・高速化でき，実用レベルの潜在関係規則のマイニングへ近づくことができた．ルール集合の圧縮は内在する共通な現象を発見して一つにまとめる作業と考えられる．これはルールを一般化することに相当し有用である．この圧縮に基づくマイニングはこれまで殆ど研究されていなかった．