Research On The Method Of Crowdfunding Project Recommendation And Investment Amount Prediction Based On Neural Network

In recent years,with the continuous development of Internet technology,more and more new network applications are emerging,and online crowdfunding is an innovative financial model based on the Internet.With the help of information dissemination advantages of the Internet,the crowdfunding platform is expected to help fund-raisers obtain the financial support needed by the projects from all over the world.Faced with a large number of projects on the crowdfunding platform,it is often difficult for potential investors to find the projects they want to invest.Personalized recommendation system is one of the effective ways to solve this problem.The recommendation system is a tool that can automatically extract effective information from complex data sources,and it has been widely used in many fields.Some classical recommendation algorithms,such as collaborative filtering,matrix decomposition,etc.,have been proved to be very effective in e-commerce recommendation,film recommendation and other fields.In recent years,with the rise of deep learning technology,some recommendation algorithms based on deep learning are put forward.Thanks to advanced model training technology and large-scale data sets,deep learning recommendation models has the ability of features extraction and complex problems solving that traditional recommendation methods can’t match.Therefore,combining deep learning technology with recommendation system to serve for crowdfunding platforms is a research direction.Based on the analysis of the inherent characteristics of crowdfunding platform,the crowdfunding recommendation algorithm based on deep learning technology is studied.The main tasks are as follows.First of all,aiming at the matching between investors and crowdfunding projects,this thesis finds that compared with the traditional recommendation data set,crowdfunding data set has the characteristics of high data quality and high information density.Therefore,a new matching structure of neural network is designed in this thesis.In this framework,we add the results of hidden factor model,a traditional recommended feature extraction technology,to the shallow position in the overall model LFFI-H,which effectively avoids the information loss caused by the bottom layer-by-layer transmission of deep neural network(DNN).In addition,in this framework,investors’ preferences for crowdfunding project categories have also undergone fine-grained hierarchical processing.A large number of experiments are carried out on Kiva public data sets,which verify the performance and effectiveness of the architecture.Then,this thesis attempts to explore investors’ willingness to invest in potential investment projects from another difficult problem "crowdfunding pricing" on the platform of crowdfunding.In order to explore investors’ multi-dimensional preference for crowdfunding projects,this thesis introduces the structure of capsule neural network,combines it with the classical neural network,and integrates matrix decomposition technology,and puts forward a new framework for predicting investors’ investment intention.In order to adapt the capsule network to the task of this thesis effectively,we modified it and combined it with other parts of the architecture.With the help of the capsule network,the user’s preference is no longer a single embedded vector,but a multi-dimensional vector group,which effectively alleviates the problem that the feature dimension is relatively single when the neural network is embedded.In the verification of Indiegogo open source data set,the proposed model shows good performance.