一种面向流数据频繁项挖掘的降载策略*

流数据产生速率具有不可预见性，当其速率超过系统处理能力时，部分数据元素不能被实时处理。降载技术是处理此问题的关键技术之一。分析了目前降载技术的不足，提出了一种面向挖掘流数据频繁项集的降载策略。该策略采用了基于元组出现频率的语义删除策略，优先删除出现频率相对较低的元组，从而有效解决了在挖掘流数据中的频繁项所遇到系统超载时所出现的问题，同时采用了根据流数据产生速率自动地控制是否启动降载策略，有效地解决了降载的适应性问题。最后，通过实验和分析，证明了该策略在流数据频繁项挖掘中有效性。

Load-shedding strategy for data stream frequent item mining

With the unpredictability of data stream generation rate, when the rate exceeds system capacity, some of the data elements cannot be real-time processing. Load shedding techniques is one of the key technologies to deal with this issue. The deficiencies of current load shedding techniques are analyzed and a new load-shedding strategy for data stream frequent data item mining is proposed in this paper. This strategy adopts the semantics of tuple deletion based on data item frequency to delete tuples with relatively low frequency as possible, thus it can effectively solve the problems when mining the frequent data item while the system is overloaded. Moreover, starting and stopping load shedding strategy is controlled automatically based on the data stream rate, so it is effectively address the problem of load shedding adaptability. According to our experiments and analysis, the proposed strategy has the effectiveness of mining frequent items in data stream.