Entrepreneurial project recommendation by implicit user feature extraction and DNN-MF

In order to solve the problems that it is difficult to capture the user’s interest and the traditional matrix factorization model has limited expressive ability, this study proposes an implicit feature enhanced Deep Neural Network (DNN) matrix factorization model (IFE-DNN-MF for short). Firstly, the model extracts hidden features from users’ multi-source data, including basic attribute features, behavior sequence features and statistical features, and processes different types of data through fully connected networks and gated circulation units to mine users’ deep interest expressions. Then the extracted hidden features are integrated into the matrix decomposition framework enhanced by DNN, and the interaction between users and projects is modeled from multiple angles, and finally accurate entrepreneurial project recommendation is realized. Compared with the best-performing baseline model, the Root Mean Square Error (RMSE) of this model is reduced by 6.8%, and the rate of recall@10 is increased by 16.9%. The ablation experiment verified the effectiveness of each module. After removing the user’s hidden feature extraction module, the rate of recall@10 decreased by 14.5%, and after removing the hidden feature enhancement path, it decreased by 12.5%, which proved that the hidden feature and its enhancement path played a key role in improving the recommendation performance. The study provides an effective solution for the recommendation of entrepreneurial projects, which is of positive significance for promoting the intelligent development of entrepreneurial service platform.