Abstract: With the problem that traditional collaborative filtering recommendation algorithm does not recommend high quality on sparse data sets, a hybrid recommendation algorithm that based on autoencoder and clustering is proposed. First, the user's project rating data and user demographic data as the input data from the autoencoder to extract user characteristics. Then use the extracted user characteristics to cluster the users to obtain the user categories, so that the search range of the neighbor decreases. Then search for a recommendation algorithm for the same type of users by calculating mean absolute error (MAE), and get a hybrid recommendation model. Finally calculate the target user category and use the hybrid recommendation model to get the recommended results. Experimental results show that this algorithm can effectively improve the quality of recommendation.