A Trust Evaluation Model Towards Internet-based Lending Platform Based On BP Neural Network And Semi-supervised Learning

The growth of Internet-based lending platform in Internet-based finance continues advancing triumphantly. However, the number of platforms who involve in financial troubles increases at the same time. The trust of a platform itself becomes the biggest hinder which prevent it from developing. The present researches of Internet-based lending focus on users but there are no empirical studies focusing on trust towards the platform itself. This paper aims to build an effective trust evaluation system towards Internet-based lending platform.According to the high cost of fetching trust evaluation index for a platform, a trust evaluation index system based on the public network data was proposed in this paper, including basic data, dynamic data and visit data. Based on the situation the labeled data is hard to fetch, the paper raises a semi-supervised learning method employs unlabeled data to improve regression estimates. The inputs of the model are 31 indexes data of 169 Internet-based lending platforms which was fetched automatically, and the outputs of it are 23 platforms’ labeled data comes from Chinese Academy of Social Sciences Institute of Finance. Two BP-regressors with different parameters were initialized, each labels the unlabeled data for the other one, where the unlabeled data has the highest confidence will be chosen, at last, two regressors are averaged as the final prediction till the iteration ends.The experiment results show that the co-training style semi-supervised learning regression model ultilizes the unlabeled data, which helps improving the regression estimates compared to other regression models. This kind of regression model can be used to estimate the Internet-based lending platform trust.