RareSense: Rarity-Aware Similarity Search for Anomaly Retrieval in Transactional Data
A novel framework named RareSense has been developed for conducting similarity searches on sparse set-valued data, with a focus on transactional anomaly detection. This framework identifies minimal rare itemsets as intermediate structures, formulates dependable rare association rules, and translates objects into sparse rare-rule profiles for comparison via weighted Jaccard similarity. The weights for the rules incorporate factors such as inverse support, confidence, lift, structural complexity, and stability, allowing neighborhoods to be defined based on shared rare evidence instead of standard feature overlap. The study illustrates that IDF-weighted Jaccard is a specific case of RareSense. This research can be found on arXiv with the identifier 2607.28879.
Key facts
- RareSense is a rarity-aware similarity framework for sparse transactional anomaly data.
- It mines minimal rare itemsets as intermediate structures.
- It derives reliable rare association rules.
- It maps objects into sparse rare-rule profiles.
- It compares objects using weighted Jaccard similarity.
- Rule weights combine inverse support, confidence, lift, structural complexity, and stability.
- IDF-weighted Jaccard is a restricted singleton case of RareSense.
- The paper is available on arXiv with identifier 2607.28879.
Entities
Institutions
- arXiv