Self-optimization Rule-chain Mining Based on Potential Association Rule Directed Graph
Hongyun Ning, Liu Jin-lan, Zhang De-gan · 2008
This paper presents an ACO-based (ant colony optimization) mining algorithm aiming to discover longer rule-chains directly. Firstly, a potential association rule directed graph (PAGraph) is created, in which, the dynamic heuristics is used to record participant-intensity of edge. Secondly, making use of ant's positive feedback, pheromone on edge that ants passed is adjusted by heuristics so that it could make paths, which have longer rule-chains, have higher selection probability. Meanwhile, a bitwise-AND operation is introduced to compute rule's confidence easily. Finally, the experimental results show the proposed method can sufficiently capture longer rule-chains and it also confirms the robustness of the algorithm.