A Scalable Algorithm for Vertical Mining of Quantitative Frequent Patterns

Frequent pattern mining has become popular in big data analytics and knowledge discovery as it discovers sets of items (e.g., merchandise items or events) that co-occur frequently. These frequent patterns are discovered by either horizontally by transaction-centric mining algorithms or vertically by item-centric mining algorithms. Regardless of the mining algorithms used, traditional frequent pattern mining algorithms focus on discovering Boolean frequent patterns, which reveal the presence or absence of specific items within the discovered patterns. However, in many real-life scenarios, the quantities of items within the patterns are crucial. For instance, the quantity of items can significantly impact the profitability of selling the items found in the discovered patterns. A very recent quantitative algorithm called Q-VIPER mined frequent quantitative patterns by representing the big data as a collection of item-centric bitmaps. Each bitmap captures the presence or absence of a transaction containing the item, together with the quantity of that item in each transaction. It then mines quantitative frequent patterns vertically. It works well with small quantity. However, when dealing with large quantity, it generates a large number of sets of candidate quantitative frequent patterns (aka sets of item expressions, or itemexpsets for short). Given that large quantities are not unusual in numerous real-life applications, we design a scalable solution in this paper. The resulting scalable quantitative frequent pattern algorithm called SQ-VIPER significantly reduces the number of candidates to be generated, and thus speeds up the mining process. Evaluation results show that superiority of our SQ-VIPER over the existing Q-VIPER and MQA-M algorithms, which respectively mine quantitative frequent patterns vertically and horizontally.