Discovering and Visualizing Efficient Patterns in Cost/Utility Sequences

Many algorithms have been proposed to discover interesting patterns in sequences of events or symbols, to support decision-making or understand the data. In sequential pattern mining, patterns are selected based on criteria such as the occurrence frequency, periodicity, or utility (eg. profit). Although this has many applications, it does not consider the effort or resources consumed to apply these patterns. To address this issue, this paper proposes to discover patterns in cost/utility sequences, in which each event/symbol is annotated with a cost, and where a utility value indicates the benefit obtained by performing each sequence. Such sequences are found in many fields such as in e-learning, where learners do various sequences of learning activities having different cost (time), and obtain different utility (grades). To find patterns that provide a good trade-off between cost and benefits, two algorithms are presented named CEPDO and CEPHU. They integrate many optimizations to find patterns efficiently. Moreover, a visualization module is implemented to let users browse patterns by their skyline and visualize their properties. A case study with e-learning data has shown that insightful patterns are found and that the designed algorithms have excellent performance.