|Autors: Budakova, D. V., Petrova-Dimitrova V. S., Dakovski L. G.|
Title: Virtual Agents, Learning How to Reach a Goal by Making Appropriate Compromises
Keywords: Intelligent system, Reinforcement learning, Control the way
Abstract: This paper proposes a modification to the model-free Reinforcement learning algorithm Q-learning. It is implemented to train smart gift shopping-cart learning agents (SGSCLA). The aim of the modification is to empower the learning agent to reaching a goal by making appropriate compromises only. That is the way in which the measure models and emotional models, represented as new agent memory matrixes are introduced. This models show how the user perceives and evaluates the environment. The Shopping Center is represented by a multigraph in which the nodes represent three groups of shops. The edges illustrate the connections between the shops; the primary (major) and the secondary (minor) paths between them and the emotions, evoked in the customer under consideration by a visit to a particular shop. The user can be see the route suggested by the virtual agent and the made compromises to the goal. The emotion types chosen for the purpose of the experiment are boredom, joy and worry.
Full text of the publication
Вид: публикация в национален форум