In order to solve the problem of information asymmetry of agricultural products on-line, combined with the advantages of deep learning and the characteristics of transaction data and put forward the sales predicting model:imperial crown model(ICM).Firstly,it established factor evaluation index, sales could be divided into four levels.Secondly, it adopted two layers autoencoder network to extract feature, and generated the new feature vector.Then it trained the classifier with labeled sample set and used the classifier to classify the unlabeled training samples.Finally, it fine-tuned the optimal parameters of the whole network parameters by using the backward propagation algorithm got the minimize value of the loss function, and achieved online sales of agricultural products classification predict.By simulation analysis, the classification accuracy of the ICM verifies as high as 88%, ICM accuracy is significantly higher than other shallow classifiers.It proves that ICM has a better ability of incremental self-learning and the ability of the cognitive level.%针对线上农产品销售存在的信息不对称问题,提出一种结合深度学习算法优势和涉农电商销售数据特点的皇冠模型(ICM).首先建立因素评价指标,将销量分为四个类别;其次采用两层自编码网络提取样本特征,并生成新的特征向量;然后利用带标签样本集训练分类器并对无标签训练样本分类;最后利用BP微调整个网络参数得到使损失函数值达到最小的最优参数,实现线上农产品的销量分类预测.经仿真分析,验证了ICM的分类准确率高达88%,明显高于其他未将数据进行特征学习的浅层分类器,证明了ICM具有较好的增量自学习能力和层次认知能力.
展开▼
