Incremental Targeted Mining in Sequences

High utility sequential pattern mining (HUSPM) is a critical research topic in data analytics (e.g., smart-city technologies), which takes into consideration three pivotal factors of data: timestamp, internal quantization, and external utility. Recently, a query-enabled HUSPM approach has been proposed, which aims to discover patterns based on a query sequence. However, this approach only works on static data and does not solve the tasks well under dynamic data. When the data is updated, it needs to restart the mining process, which leads to a lot of duplicate calculations and resource consumption. In the paper, to address the mining task of increasing sequence data over time, we develop an Incremental Targeted HUSPM algorithm called ITUS. A tighter upper bound called tight extension sequence utility (TESU) is proposed to determine key candidates, which can avoid the generation of unpromising patterns. By using TESU, a target candidate pattern tree (TCP-tree) is utilized to record the sequence information, and several efficient strategies are implemented to incrementally update the tree. Finally, we extensively evaluate our proposed algorithm on both real-world and synthetic datasets. The experimental results clearly demonstrate that not only does the novel algorithm guarantee the accuracy of the results after multiple database updates, but it also achieves higher efficiency than the baseline approach.