| Sales forecast is the starting point for enterprises to make production and operation plans,and is the basis for supply chain to carry out procurement,manufacturing,logistics and distribution activities.Over the years,trends such as the globalization of supply chains,the explosion of product variety,the shortening of product life cycles and increasingly competitive markets have made forecasting more complex,yet its role more critical.At the same time,with the rapid development of information technology and ecommerce.massive data on commodities are rapidly generated.Using Internet big data to accurately forecast product sales is a new issue widely concerned by business and academia.Traditional sales forecasting methods usually use the historical sales data of products and the attributes of products themselves,and cannot well forecast short-term and non-stable sales volume.In view of this,this paper uses online reviews.macroeconomic indicators and search engine data to propose two data-driven product sales forecasting methods.These two methods are helpful to provide new ideas for the research of product sales forecast theory,to assist enterprises to make appropriate production and operation plans and correct operation decisions,and to improve the profits and market competitiveness of enterprises.In conclusion,this paper studies the method of product sales forecasting driven by online reviews,macroeconomic indicators and search engine data,and mainly completes the following three aspects:(1)Proposed a product sales forecasting method based on online reviews and macro data.First,a macroeconomic indicator selection algorithm is developed to select indicators that can reflect the long-term trend of sales volume and ensure that there is no multicollinearity among these indicators.Then,combining with the relevant data of online reviews,using prospect theory and dictionary-based sentiment analysis algorithm.the sentiment index which can reflect that negative reviews have greater influence on consumers’ purchase decisions than positive reviews is calculated.Finally,according to the selected macroeconomic indicators and the obtained sentiment index,a logarithmic autoregressive model for product sales forecasting is established,and Adam optimizer is used to estimate the model parameters to realize the forecast.(2)Proposed A product sales forecasting method based on online review and search engine data.First,the sentiment analysis method proposed in(1)is extended to separately calculate the sentiment scores of online reviews concerning various attributes of products by using different sentiment word dictionaries,and the sentiment indexes of each period are obtained by combining the prospect theory and the relevant information of online reviews.Subsequently,relevant Baidu search queries with different lag orders are screened using the time difference correlation method.Lastly,principal component analysis(PCA)is used to reduce the dimension of the processed data,and an improved fruit fly optimization algorithm(DSFOA)is constructed to modify the BP neural network(BPNN).The PCA-DSFOA-BPNN model is constructed.Through data training and fitting,the optimal parameters are obtained to realize the forecasting of product sales.(3)Taking the sales forecast of 3 brands automobiles,such as Audi A6L,in the next 3 quarters and the sales forecast of 14 brands automobiles,such as Lavida.in the next 10 months as examples,the application research of the above two data-driven sales forecasting methods is given,and the validity,accuracy and robustness of the proposed two forecasting methods are verified.Experimental results show that the proposed dataderiven forecasting methods are superior to the existing methods,and can effectively improve the forecast accuracy and have good robustness.It is hoped that the ideas and methods of sales forecasting driven by Internet data proposed in this paper can provide some reference and help for the sales forecasting of enterprises. |
PDF Full Text Request