SPARC-HOI

High-Occupancy Itemset (HOI) mining is a critical pattern mining paradigm that measures the
average or total occupancy contribution of an itemset within transactions, preventing the discovery of
long, sparse itemsets with low actual density. However, existing HOI mining algorithms suffer from
massive memory usage due to pointer-heavy prefix trees or heavy bitmap intersection overheads.
In this paper, we present SPARC-HOI, a candidate-free high-occupancy itemset miner designed
for a hardware-efficient execution model: flat primitive arrays, tight monotone upper bounds, static
arena allocation, and cache-linear projection. The algorithm supports both average and summed
occupancy semantics and completely avoids bitmap intersections. We prove the anti-monotonicity of
the Reciprocal Residual Occupancy Boundary (RROB) and the Summed Occupancy Boundary over
suffix projections. Experimental results show that SPARC-HOI achieves up to a 4× speedup and a
680× reduction in Peak RAM usage compared to state-of-the-art baselines.