基于余切相似度和BP神经网络的相似度快速计算

A Fast Similarity Calculation Method Based on Cotangent Similarity and BP Neural Network

摘要:

相似性度量在大数据相关应用中具有重要的意义，然而传统余弦相似度遍历计算方法的准确性和时效性较差，具有较大局限性，无法为海量高维数据的质量评估提供有效依据。针对上述问题，利用余切三角函数和数据维度差值构造2种余切相似度公式，提高相似度计算的准确性；借助后向传播（BP）神经网络建立一个能够逼近数据集相似度映射关系的网络模型，降低相似度计算的时间复杂度。实验表明，改进的相似度快速计算方法具有良好的准确性和时效性，而且应用在大规模数据集时的性能提升更显著。

Abstract:

Similarity measurement is of great significance in big data related applications. However， the traditional cosine similarity traversal calculation method has a poor accuracy and timeliness， which cannot provide an effective basis for the quality assessment of massive high-dimensional data. To improve the accuracy of similarity calculation， two types of cotangent similarity formulas with cotangent trigonometric function and data dimensional differences was constructed. Besides， a back-propagation（BP） neural network model approximating the similarity mapping relationship of datasets was established to reduce the time complexity. The experimental results demonstrate that the improved fast similarity calculation method has a good accuracy and timeliness. Moreover， it has a more significant performance improvement when applied to large-scale datasets.