Intelligent planning of e-commerce transactions

Software agents in e-commerce scenarios mostly show static and reactive behavior. However, only the customization of agent behavior at runtime exploits the entire flexibility of an agent approach. Situated behavior requires on the one hand goal directed reasoning in order to pursue individual transaction plans and on the other hand learning mechanisms to adapt the planning process to the agent's changing environment. Efficient transaction plans are generated with planning algorithms like STRIPS and GRAPHPLAN. These planning algorithms are based on heuristics, which have to be adjusted cyclically to the current environmental situation. As the relationship between the influencing factors and the goal affecting attributes in e-commerce scenarios are typically unknown, in our approach a neural network is used to estimate these interdependencies.