In this paper, we propose an Embedding and Mapping Approach with Domain-Independent Latent (EMCDR-DiL) for cross-domain recommender. The proposed EMCDR-DiL framework is improved compared to EMCDR by using matrix triple factorization to capture domain-related knowledge, thus preventing the negative transfer effect caused by domain knowledge transfer. Extensive experiments on recommendation of Douban to Movielens show that EMCDR-DiL improves over EMCDR on MAE, RMSE, and other performance metrics.