作者: Benjamin Letham , Cynthia Rudin , David Madigan
DOI:
关键词: Context (language use) 、 Artificial intelligence 、 Cold start 、 Statistical learning theory 、 Sequence 、 Generalization 、 Machine learning 、 Bayesian probability 、 Data mining 、 Association rule learning 、 Event (probability theory) 、 Computer science
摘要: We present a theoretical analysis for prediction algorithms based on association rules. As part of this analysis, we introduce problem which rules are particularly natural, called "sequential event prediction." In sequential prediction, events in sequence revealed one by one, and the goal is to determine will next be revealed. The training set collection past sequences events. An example application predict item placed into customer's online shopping cart, given his/her purchases. context problem, have distinct advantages over classical statistical machine learning methods: they look at correlations subsets co-occurring (items b imply c), can applied natural way, potentially handle "cold start" where small, yield interpretable predictions. work, two that incorporate These used both supervised classification, simple enough possibly understood users, customers, patients, managers, etc. provide generalization guarantees these algorithmic stability from theory. include discussion strict minimum support threshold often rule mining, an "adjusted confidence" measure provides weaker condition has support. paper brings together ideas theory, mining Bayesian analysis.