基于并行Adaboost-BP网络的大规模在线学习行为评价

EVALUATION OF LARGE-SCALE ONLINE LEARNING BEHAVIOR BASED ON PARALLEL ADABOOST-BP NETWORK

针对传统的在线学习行为评价方法在处理大规模数据集时面临的问题,提出一种基于并行AdaboostBP神经网络的在线学习行为评价方法。将BP神经网络作为弱预测器,由Adaboost算法组合15个BP神经网络的输出,构建了强预测器;充分利用了Hadoop平台下Map Reduce并行编程模型,提出了大规模在线学习行为的自动评价模型,设计了并行Adaboost-BP神经网络算法的Map和Reduce任务。多组实验表明,提出的算法准确率高、运行耗时少,取得了良好的加速比,效率大于0.5,适合大规模在线学习行为的自动评价。

Aiming at the problems that traditional online learning behavior evaluation methods face when dealing with large-scale data sets, an online learning behavior evaluation method based on parallel Adaboost-BP neural network is proposed. The BP neural network was used as the weak predictor, and 15 BP neural networks were combined by the Adaboost algorithm to construct the strong predictor. The MapReduce parallel programming model of Hadoop platform was fully utilized. An automatic evaluation model of large-scale online learning behavior was proposed. The Map and Reduce tasks of parallel Adaboost-BP neural network algorithm were designed. The experimental results show that the proposed algorithm has high accuracy rate, low running time and good speedup ratio. The efficiency is more than 0.5, which is suitable for the automatic evaluation of large-scale online learning behavior.