基于阿里巴巴电商数据的商品需求预测研究

Commodity Demand Forecast Research Based on Alibaba E-commerce Dataset

在电商供应链的自动化管理过程中,高质量的商品短期需求预测是供应链管理的基础和核心功能。基于电商大数据对商品需求进行准确预测,可以帮助企业降低成本,提升用户体验,对整个电商行业的效率提升都会起到重要作用。由于受用户喜好、价格调整、季节变化等各种动态与间歇性因素影响,使得该项任务具有一定的挑战性。首先创新性的使用商品历史销量走势图作为商品特征,通过深度学习的迁移学习获取图像特征并对商品进行聚类分析,目的是针对相似的商品进行预测;接着将短时段内(天为单位)的商品时序动态特征转为二维矩阵数据,通过卷积神经网络(CNN)自动提取特征,在与商品的固有特征融合后,连接到较长时序(周为单位)的长短期记忆神经网络(LSTM),并通过混合密度网络(MDN)进行预测输出。最后使用阿里巴巴菜鸟网络下的真实电商数据验证了所提算法更加的精确和有效,该算法符合大多数以综合性商品销售为主的电商平台,具有较好的可扩展性和普遍的应用价值。

In the process of e-commerce supply chain automation management, high-quality commodity short-term demand forecasting is the basis and core function of supply chain management. Accurate prediction of commodity demand based on e-commerce big data can help enterprises reduce costs,improve user experience, and play an important role in improving the efficiency of the entire e-commerce industry. Due to the user preferences, price adjustment, seasonal changes and other dynamic and intermittent factors, this task has a certain degree of challenge. First of all, we adopted the historical sales trend chart of commodities as the characteristics of commodities innovatively, acquired the image characteristics through in-depth learning transfer learning and cluster analysis of commodities, with the purpose of predicting similar commodities;then we transformed the dynamic characteristics of commodity time series in a short period of time(days as a unit) into two-dimensional matrix data, and automatically extracted the characteristics through convolution neural network(CNN), which was used to extract the characteristics of commodities in the same trade after the product’s inherent characteristics were fused. It was connected to the long and short term memory neural network(LSTM) with long time sequence(week as unit), and the prediction output was made through the mixed density network(MDN). Finally, the real e-commerce data of Alibaba rookie network was used to verify that the algorithm was more accurate and effective.