Research On Personalized Investment Recommendation For Crowdfunding Platforms

Crowdfunding as a new way of financing is gaining popularity and growing rapidly both in domestic and international.The increase in crowdfunding projects has made it more difficult to navigate through the projects,making it difficult for investors to pick out the ones they really like among the many projects.Most crowdfunding platforms only provide fixed recommendation lists such as popular project recommendations,which makes it difficult to meet the individual needs of investors,resulting in many projects not finding the right investors,which is an important reason for the low success rate of crowdfunding project financing.Therefore,a personalized recommendation system needs to be designed for crowdfunding platforms to tap into investors’ investment preferences so as to achieve accurate and rich recommendations.Although current collaborative filtering algorithms have achieved certain recommendation effects,they lack quantitative analysis of user preferences and still suffer from problems such as sparse data and cold starts.Based on this,two deep collaborative filtering recommendation algorithms for crowdfunding platforms are proposed in the paper.(1)A deep collaborative filtering recommendation algorithm based on investor preference modeling.The algorithm incorporates the interaction information of investors and projects and their respective attribute information,and fully learns the general investor preferences and investor attribute preferences to obtain the final overall investor preferences.On the basis of this,the linear and nonlinear interactions between investorinvestor,project-project and investor-project are fully learned by deep collaborative filtering,and the information of three different modalities,namely,numerical,subtype and textual,is fused after effective processing,so as to improve the recommendation accuracy.(2)A dual attention mechanism-based investor preferences representation learning recommendation algorithm.The algorithm is based on the deep collaborative filtering recommendation algorithm based on investor preference modeling and introduces a dual attention mechanism.The first attention is used to further quantify investor attribute preferences by assigning different weights to the interaction between each attribute of the investor and each attribute of the project.The second attention is used to fuse investor general preferences and investor attribute preferences according to the different weights to obtain the final investor preference representation.The paper details the principles and components of the two algorithms and conducts offline experiments on a real crowdfunding platform Indiegogo dataset.The effects of hyperparameters on the recommendation performance of the two algorithms are compared,and the need for a dual attention mechanism is demonstrated by ablation experiments.Experiments are conducted on datasets with different sparsity to verify the effectiveness of the algorithms proposed in the paper on extremely sparse crowdfunding data.Finally,comparison experiments are conducted with baseline algorithms such as LR,Deep FM and NFM,and the experimental results show that dual attention mechanismbased investor preferences representation learning recommendation algorithm proposed in the paper achieves the best recommendation results on the Indiegogo dataset.Therefore,the algorithm proposed in the paper can improve the recommendation effect and thus increase the success rate of project financing.