Digital marketing is the avenue of electronic communication that is used by marketers to dorse goods and services in the marketplace. Digital marketing includes a wide range of marketing tactics to promote products, services, and brands. It is crucial for businesses to modify their marketing methods in order to help businesses looking to improve their digital marketing efforts in order to customize their marketing plans.
Companies must comprehend the variables influencing consumers' preferences and consumer behavior change toward digital channels. It is very important to identify the most influential factors, that drive consumers' preference for product characteristics and digital marketing channels. With the rapid advancement of e-commerce, knowing the consumer's propensity to purchase different products has become a significant challenge for a product manager who determines marketing strategies for a marketplace platform. Consumer preference analysis explains what aspects of a product affect and how they affect a consumer’s purchasing decision [1].
Consumer preference reconstruction from sales data can assist the product manager to know consumer preferences in the context of e-commerce. Recently, customer preferences are usually measured by the distribution of customer willingness to pay (WTP) [2]. We will assume that customer preferences follow some cumulative distribution function, thus, the purchasing probability of each customer is P(x)=1-F(x). When the WTP distribution is initially unknown, suitable methods of data analysis are used for its estimation. Machine Learning methods are widely used to forecast customer demand. The corresponding models mainly include customer arrival rate and distribution of customer willingness to pay (WTP).
In this paper, we propose a method for the reconstruction of customer preference functions based on the kernel method of machine learning [3,4].
Each product item is characterized by a feature vector zk=(xk1,xk2, ...,xkm, pk), that includes individual indicators measured on the appropriate scales and the price of the corresponding product item. Let us perform a preliminary data clustering by dividing the set of each product’s units for sale into a set of clusters zk ∈ Ωi (z ̄i), k=¯(1,M), i=¯(1,n) with centers z ̄i, that combine products with similar characteristics. We will use a training dataset that includes the number of sales of a certain product on a fixed sales horizon for each cluster si, i=¯(1,n). The conditional expectation of the number of sales is defined as
where is the customer arrival rate to the marketplace.
To build an additive model, we perform a logarithmic transformation log(si )=logv T+logP (z ̄i).
We choose the model in the quasilinear form logv T=w_0, logP (z)=φT (z)w, where, in accordance with the kernel machine learning methodology, the coordinate functions φ(z) are selected as φT (z)φ(z*)=κ(z,z*)=σ-2 exp{-‖z-z* ‖}.
Then the measurement model will be presented in the form
where εn- vector characterizing measurement errors, and 1=[1 1... 1]T.
Estimates of the model parameters can be found as a solution to the regularized constrained optimization problem
Using the Lagrange function with indefinite multipliers λ
we get the solution (4) from kernel matrix Kn=[κ(z ̄i,z ̄j)]i,j=1n,n
Finally, estimates of the buyers' preference function and the intensity of their arrival are obtained in the form
The proposed methods can be used in digital marketing systems and pricing strategies to improve sales efficiency.
References
1.Mengzhuo Guo, Xiuwu Liao, Jiapeng Liu, Qingpeng Zhang. Consumer preference analysis: A data-driven multiple criteria approach integrating online information, Omega, vol. 96, 2020, 102074.
2.Yang Yang, Wan-Ling Chu, Cheng-Hung Wu. Learning customer preferences and dynamic pricing for perishable products, Computers & Industrial Engineering, vol. 171, 2022, 108440.
3.L. Lyubchyk, A. Galuza, G. Grinberg. Semi-supervised Learning to Rank with Nonlinear Preference Model, In: Shahbazova, S., Sugeno, M., Kacprzyk, J. (Eds), Recent Developments in Fuzzy Logic and Fuzzy Sets. Studies in Fuzziness and Soft Computing, vol. 391, 2020, Springer, Cham.
4.L. Lyubchyk, G. Grinberg. Pairwise kernel-based preference learning for multiple criteria decision making, 2017 IEEE First Ukraine Conference on Electrical and Computer Engineering (UKRCON), 2017, pp. 818-821.
|
Голосування 
Ця робота ліцензується відповідно до Creative Commons Attribution 4.0 International License
Знайшли помилку? Виділіть помилковий текст мишкою і натисніть Ctrl + Enter