Discovering Deep Knowledge from Relational Data by Attribute-Value Association

Discovering Attribute-Value Association (AVA) is of fundamental importance in knowledge discovery. Market Basket Analysis is an archetypical application. However, most existing algorithms rely only on frequency counts directly obtained from data at the surface and thus cannot reveal deeper knowledge, i.e. the AVAs governed by hidden factors inherent in the data. This paper proposes a new method, called Attribute-Value Association Algorithm (AVAA), which can i) discover statistically significant associations at the Attribute-Value level from relational dataset, ii) disentangle associations to unveil different AVAs corresponding to different hidden factors. The performance of AVAA is validated via experiments on both synthetic and real-world datasets. AVAA demonstrated better identification rate when comparing with Frequent Pattern Mining algorithms, particularly when noise was present. Keywords—Segmentation/Clustering/Association; Explorative and visual data mining