Dynamic Pricing of Information Products Based on Reinforcement Learning: A Yield-Management Approach

Abstract

Pricing of information services gains an increasing importance in an IT environment, which is characterized by more and more decentralized computing resources (e.g. P-2-P computing). Even if pricing theory represents a kernel domain of economic research the pricing problem related to automated information production processes could not be handled satisfactory. This stems from the combination of high fixed costs with negligible variable costs. Especially in airline industries this problem is addressed by heuristics in the so called “ Yield Management” (YM) domain. The paper presented here, shows the transferability of these methods to the information production and services domain. Pricing a bundle of complementary resources can not be solved by the simple addition of value functions. Therefore we introduce Machine Learning (ML) techniques to master complexity. Artificial Neural Networks (ANN) are used for the joint representation of the multidimensional value functions and Genetic Algorithms (GA) should help train them in a first effort. While this does not lead to outstanding results, we try Reinforcement Learning (RL) in a second approach. This ML method provides encouraging results for efficient adaptive pricing of resource attribution related to the multidimensional YM problem.KeywordsReinforcement LearnDynamic PriceStochastic Dynamic ProgramPolicy IterationPrice TheoryThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.