Integrated knowledge graph and reinforcement learning: a graph modelling approach for international trade behaviour analysis and economic trend prediction

In this paper, a unified framework of knowledge graphs (KG) and reinforcement learning (RL) is suggested to understand the behaviour of international trade and forecast economic trends.The framework builds a dynamic knowledge graph of key entities in international trade and their historical interactions.The unstructured trade reports and economic bulletins are analysed by means of natural language processing and combined with the structured data of the World Bank and the World Trade Organization.The graph neural network encoder pulls out relational representations of the knowledge graph and inputs the information into a deep reinforcement learning agent to maximise prediction of trade policies and produce strategic advice.The reward component takes into consideration such economic indicators as the GDP influence, change in trade balance, and geopolitical risk.Simulation on historical trade conditions show better prediction using simulation as opposed to the econometric model used in the past.