基于深度神经网络的高维数据序列化推荐算法

High Dimensional Data Serialization Recommendation Algorithm Based on Deep Neural Network

随着互联网、超大数据时代的到来,如何在海量数据中快速找到有用信息,关系着数据推荐系统的好坏。为了解决传统数据推荐算法存在的数据维数低、推荐准确率不高、用户匹配性不好等问题,提出一种基于DNN深度神经网络、多层注意力机制结合的数据推荐方法。方法的核心思想在于利用卷积神经网络对前后数据信息进行多层维度提取;利用引入的注意力机制,利用多通道注意力机制获取的低维特征和深度网络获取的非线性特征进行求和,实现数据特征的融合,完成用户喜好的外围数据挖掘,从而提高数据挖掘深度和推荐性能。最后采用Amazon公开数据进行仿真验证,通过与多种模型和算法的实验数据对比,结果表明上述算法不仅能够提升数据推荐性能,同时还能够有效发掘数据的潜在信息。

With the advent of the Internet and ultra-large data era,how to quickly find useful information in massive data is related to the quality of the data recommendation system.In order to solve the problems of low data dimension,low recommendation accuracy and poor user matching in traditional data recommendation algorithms,a data recommendation method based on DNN deep neural network and multi-layer attention mechanism is proposed.The core idea of this method is to use convolution neural network to extract the multi-layer dimensions of the front and rear data information.Using the introduced attention mechanism,the low dimensional features obtained by the multichannel attention mechanism and the nonlinear features obtained by the deep network are summed to realize the fusion of data features and complete the peripheral data mining of user preferences,so as to improve the depth of data mining and recommendation performance.Finally,the open data of Amazon is used for simulation verification.Through the comparison with the experimental data of various models and algorithms,the results show that the algorithm can not only improve the data recommendation performance,but also effectively explore the potential information of the data.