Research On Entrepreneurship Project Recommendation System Based On Deep Neural Network

In recent years,with the slogan of “Mass entrepreneurship and innovation” and the economic and social development,the number of entrepreneurial projects has increased dramatically.Offline approach is gradually unable to meet the needs of people's rapid access to entrepreneurial project information.Therefore,many online entrepreneurial project information platforms have been established in order to improve the efficiency of entrepreneurial project information acquisition.Typical examples include itjuzi.com,cyzone.cn,and Angel List.However,with the continuous expansion of the scale of information platform for entrepreneurial projects,it is increasingly difficult for users to fully browse all the entrepreneurial projects and quickly find the projects that they are interested in,causing serious information overload problems to users.In reality,the recommender system is commonly used to alleviate the problem of information overload.Recommender system enables users to obtain their required information from a vast amount of information in a short period,reducing the impact of information overload,and thereby improving the user experience of the information platform.Based on the above background,this paper studies the recommendation system for entrepreneurial projects.This paper firstly makes in-depth analysis and research on current mainstream recommendation technologies and applications,and summarizes their applicable fields,advantages and disadvantages.Then,this paper analyzes and summarizes the research status and characteristics of the entrepreneurial project recommendation system,and expounds the feasibility and necessity of applying personalized recommendation in the information platform of the venture project.Based on this,this paper proposes two recommendation algorithms for entrepreneurial projects based on deep neural network:(1)Entrepreneurship project recommendation algorithm based on the deep neural network and matrix factorization.This algorithm starts from the characteristic information of the entrepreneurship project,and combines a convolutional neural network,word embedding,and one-hot coding techniques to construct a deep neural network for extracting the entrepreneurship project's hidden features.Then,this algorithm uses the probability matrix factorization algorithm to extract the user's hidden features from the scoring matrix.Finally generate a recommendation according to the user's latent factors and the entrepreneurship project's latent factors.(2)Entrepreneurship project recommendation algorithm based on a multi-model.This algorithm establishes two deep neural networks to extract user and entrepreneurship project latent factors,respectively,where the project's implicit feature extraction is the same as the algorithm(1).For the user's implicit feature extraction,the user scoring data is modeled using a Restricted Boltzmann Machine(RBM),and other features are modeled using a convolutional neural network,a word vector technique,and a one-hot encoding technique.The algorithm further solves the user's cold start problem and improves the accuracy of the recommendation results.This paper first introduce the above two algorithms' recommendation process,recommendation principle and the generation method of recommendation results.Then,according to the principle of the algorithm,the program is implemented,and the off-line experiment is performed using the data set of the Entrepreneurship project collected from the Internet.The experimental results are compared with PMF and ConvMF.The experimental results show that the two proposed algorithms are superior to other algorithms.In addition,it can better complete the tasks recommended by the entrepreneurial project and effectively solve the cold start problem of users and projects.