Research And Implementation Of P2P Internet Lending Risk Identification Based On Reinforcemnt Leanring

P2P internet lending has emerged in the financial industry during the last few decades.Although it raises the social benefits,P2P lending risk never fails to explode at an astonishing speed.Especially,as a result of poor credit risk identification,malignant events occur frequently.This phenomenon not only causes economic losses to the lenders and borrowers,but also affects the healthy development of the P2P internet lending and the stability of the financial industry.Therefore,it is crucial to implement the risk identification of internet lending institution.In order to solve the problem of P2P internet lending platform can obtain small data scale and the data amount of categories are unbalanced,this thesis introduces the reinforcement learning model.Through the research and analysis of various reinforcement learning algorithms,this thesis selects a reinforcement learning model that can solve risk identification of P2P internet lending institutions as the baseline model.which regards the task of risk identification as text classification problem.However,in the process of action selection,the existed reinforcement learning model and reward function do not take the global information into account.Therefore,this thesis puts forward three improved reinforcement learning models using the information gain and attention mechanism to optimize the reinforcement learning reward function and select action processes.The effectiveness of the improved models are verified by experiments.Experimental results show that the performance of the three reinforcement learning models proposed in this thesis are greatly improved compared with the baseline model.Finally,according to the software engineering process,this thesis implements an experimental system of P2P internet lending risk identification.