Predicting user churn and taking personalized measures to retain users is a set of common and effective practices for online game operators. However, different from the traditional user churn relevant researches that can involve demographic, economic, and behavioral data, most online games can only obtain logs of user behavior and have no access to users' latent feelings. There are mainly two challenges in this work: 1. The latent feelings, which cannot be directly observed in this work, need to be estimated and verified; 2. User churn needs to be predicted with only behavioral data. In this work, a Recurrent Neural Network(RNN) called LaFee (Latent Feeling) is proposed, which can get the users' latent feelings while predicting user churn. Besides, we proposed a method named BMM-UCP (Behavior-based Modeling Method for User Churn Prediction) to help models predict user churn with only behavioral data. The latent feelings are names as satisfaction and aspiration in this work. We designed experiments on a real dataset and the results show that our methods outperform baselines and are more suitable for long-term sequential learning. The latent feelings learned are fully discussed and proven meaningful.