A Dynamic Pricing Algorithm Of Niche Goods Based On Multi-Armed Bandit Model

The Long tail phenomenon refers to such a fact that with the rapid development of Internet technology,in especial automatic matching mechanism and recommendation algorithm,the search cost of goods is reduced,the non-hot products get more exposure,and the sales volume is increased.Moreover,due to the variety of non-hot products is more abundant,the aggregate sales volume can match or even be higher than popular products,forming a long tail on the sale-product diagram.This constitutes the basis of the application environment for real-time dynamic pricing of niche products as a part of the long tail,and the automatic dynamic pricing of niche products has become an important part of revenue management of enterprises,especially e-commerce platforms.Under the big data application scenario of e-commerce platform,the pricing of niche products is facing more and more complex environmental conditions.On the one hand,the probability distribution of sales data caused by niche product characteristics is complex and time-varying,on the other hand,the data is high-dimensional and sparse.Therefore,the conventional dynamic pricing method is difficult to deal with.In recent years,the Multi-armed Bandit(MAB)model of reinforcement learning theory has achieved remarkable results in the field of dynamic pricing.Therefore,this paper studies the dynamic pricing of niche products based on MAB model.Firstly,this paper analyzes the O2 O model of niche products in online e-commerce market,and constructs a real-time dynamic pricing model for niche products based on the basic MAB model and the characteristics of niche products.Then,a dynamic pricing algorithm for niche products is designed from the two perspectives of improving the mainstream solution Upper Confidence Bound(UCB)based on MAB and bayesian demand learning.Finally,the simulation environment of dynamic pricing of niche products is constructed,and the experiment and result analysis are carried out with real data.The experimental results show that the improved algorithm based on UCB and the algorithm based on Bayesian framework perform well in the simulation environment,and they are also robust to real data distribution changes and random risks.Therefore,these two angles are feasible.And the combination of the two algorithms will also bring better results under some conditions,which has good application value for the dynamic pricing of niche products in the big data environment.