基于数据挖掘的电商促销活动效应与销量预测研究

发布时间：2018-01-23 11:05

本文关键词： 电商促销活动活动销量预测活动效应支持向量回归机关联规则　出处：《东华大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着互联网及电子商务的快速发展,商家彼此之间的竞争越来越激烈,电商平台和商家都会采用各种各样的销售运营手段来抢占市场。为了促进销售,平台会提供和举办的各类型促销活动供商家进行参与,同时随着信息化的飞速发展,商家的数据积累也已达到一定的规模。在这样的时代背景下,本文致力于通过历史数据,基于数据挖掘和数据分析的角度对电商促销活动效应和活动销量进行分析和预测,通过数据挖掘和分析,从而提高企业管理经营决策的科学化程度和智能化程度,使本文研究工作在具有一定理论意义的同时又具有重要的现实意义。首先,本文对电商促销活动短期效应进行分析,主要分析商家较为关心的销售和客流两方面,并且对销售和客流两方面在活动期间和活动前后期的效应展开分析。基于某天猫旗舰店的历史数据,通过T检验和回归等统计方法分析得出,在活动销售效应方面,活动期间商品的销量会显著上升,同时随着活动时间的进行,其刺激销售的能力在逐渐减弱;而活动前后期商品的销量会有一定的回落。在客流效应方面,活动期间客流量会显著上升,而活动前后期客流量不会有明显的变化,销量的降低主要是因为这段时间转换率较低。其次,结合上述分析,针对企业是否要参加某次活动的管理经营决策问题,基于整体利润的角度进行决策建模分析,需综合考虑参加活动的商品和未参加活动的商品在活动期间的收益和活动前后期的损失,以及平台收取的佣金费用。通过对各决策变量的分析发现,商品的日常期间和活动前后期的销量变化通常会相对平稳,且较容易确定,可以采用移动平均法进行计算。但商品在活动期间的销售变化相对较大,对此本文提出采用支持向量机为基准模型,结合粒子群参数优化和灰色综合关联因素分析的综合预测模型对参加活动的商品进行销售预测;同时采用基于兴趣度约减的关联规则分析未参加活动商品与参加活动商品之间的交互影响。再次,对本文基于历史数据分析中的数据准备工作进行阐述,包括数据源分析、数据仓库设计以及数据ETL实现。然后结合本文模型方法进行实例分析,得出综合预测模型相对于单一支持向量机预测模型在预测精度上有一定的提升,结合兴趣度可以更好的发现有效的关联规则,同时考虑活动前后期利润损失等的活动决策模型较为全面和客观,有助于企业更好地进行管理决策。最后,对本文的研究内容进行了总结,针对本文研究中的不足进行分析和展望。
[Abstract]:In recent years, with the rapid development of the Internet and e-commerce, the competition between businesses is becoming more and more fierce, e-commerce platforms and businesses will use a variety of sales operations to seize the market in order to promote sales. The platform will provide and organize various types of promotional activities for merchants to participate in, and with the rapid development of information technology, the business data accumulation has reached a certain scale. Through historical data, based on data mining and data analysis, this paper analyzes and predicts the effect of e-commerce promotion activities and activity sales, and through data mining and analysis. In order to improve the scientific and intelligent degree of enterprise management and management decision-making, the research work in this paper has a certain theoretical significance but also has important practical significance. This paper analyzes the short-term effect of e-commerce promotion activities, mainly analyzes the two aspects of sales and passenger flow that are more concerned by merchants. Based on the historical data of a Tmall flagship store, this paper analyzes the effects of sales and passenger flow during the activities and before and after the event, through T-test and regression statistical methods. In the aspect of activity sales effect, the sales volume of goods will increase significantly during the activity period, and the ability to stimulate sales will weaken gradually with the time of activity. In the aspect of passenger flow effect, the passenger flow will increase significantly during the activity, but there will be no obvious change in the passenger flow before and after the activity. The decrease in sales volume is mainly due to the low conversion rate of this period of time. Secondly, combined with the above analysis, whether the enterprise should participate in a certain activity management decision. Based on the analysis of overall profit, it is necessary to consider the profit of the participating commodity and the loss of the former and the later period of the activity. Through the analysis of each decision variable, it is found that the change of sales volume during the daily period and before and after the activity is usually relatively stable and easy to determine. The moving average method can be used to calculate, but the sales of goods during the activity is relatively large, so the support vector machine (SVM) is used as the benchmark model in this paper. Combined with particle swarm optimization and grey comprehensive correlation factor analysis, the comprehensive forecasting model was used to predict the sales of the participating commodities. At the same time, the association rules based on interest reduction are used to analyze the interaction between non-participating and participating commodities. Thirdly, the data preparation in this paper based on historical data analysis is described. Including data source analysis, data warehouse design and data ETL implementation. Compared with the single support vector machine prediction model, the comprehensive prediction model has a certain improvement in the prediction accuracy, combined with interest can better find the effective association rules. At the same time, considering the loss of profit before and after the activities, the activity decision-making model is more comprehensive and objective, which is helpful for enterprises to make better management decisions. Finally, the research content of this paper is summarized. In view of the deficiency of this paper, the analysis and prospect are carried out.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F724.6;F274

【参考文献】