Research On Personalized Recommendation Algorithm In Crowdfunding Platform Based On Machine Learning

With the rapid development of internet technology,the crowdfunding platform has become a new way of network financing.The additional value generated by data doesn't synchronous growth with the increasing of the crowdfunding platform and the scale of crowdfunding data.Therefore,the so called “information overload” appears.Personalized recommendation system can solve this problem by mining user preferences from large amounts of data.It has achieved some success in the fields of e-commerce,social media,advertising system,search engines,etc.But in the fields of crowdfunding,there is no sites provide users with professional service of personalized recommendation.According to the analysis of the platform and the research of several commonly used individual recommendation algorithms,the input,output and recommendation algorithm of the system are selected and designed.Using the algorithm of machine learning domain to build personalized recommendation algorithm based on collaborative filtering,and put forward an effective improvement scheme for the problem of it.On the one hand,a collaborative filtering algorithm which is based on the latent factor model is designed for solving the sparseness problem of data.The model optimization problem is solved by using the statistical learning method.Learning the feature of user rating data to get the predicted score which is filled into the original score matrix.The predictive score is obtained based on the collaborative filtering algorithm with the relatively dense scoring matrix as the data source.On the other hand,combined with the user rating and item attribute features of the crowdfunding platform,the collaborative filtering algorithm is improved to solve the cold start problem which is caused by the single data source.The rating data and project attribute data of crowdfunding platform is obtained from the network communication technology.After the algorithm is implemented,the feasibility of the algorithm is verified first,then the parameters in the feature learning model are adjusted.Finally,the average absolute error and accuracy of the improvement before and after are compared.After the experimental study,this paper designs the personalized recommendation algorithm of the crowdfunding platform in the thesis can provide the accurate personalized recommendation service for the users,and can quickly complete the recommendation.It can both provide users with convenience and make the platform develop better.The algorithm of this thesis improves the sparseness of data and cold start problem to a certain extent.Compared to the traditional recommendation algorithms,forecast accuracy has increased significantly.The program can achieve better effect of real-time recommendation based on amendment of user preferences.Moreover,due to the training data are from the real crowdfunding platform,the improved solution mentioned here based on machine learning algorithm is more practical.