关键词: 校园一卡通大数据, Transformer, 长短期记忆, 局部异常因子算法, 消费异常检测

Abstract: Most of the existing campus card data mining methods use shallow machine learning and data mining methods such as statistics，clustering，and association rules，ignoring the timing of consumption data and lacking deep expression of data. This paper proposed a Consume2Vec model based on deep neural network，which can deeply mine the timing and correlation among consumption data，and build a consumption anomaly detection model based on Consume2Vec model. Experiments were conducted on a large-scale card consumption data，verify the performance of two specific Consume2Vec models，and divide the students into different groups from different dimensions for comparative analysis，and find some consumption rules and characteristics of students.

Key words: campus card big data, Transformer, Long Short-Term Memory (LSTM), Local Outliner Factor (LOF) algorithm, consumption anomaly detection