Recommender systems are useful tools that help people to filter and explore massive information. While most recommender systems focus on providing recommendations for individuals, people's minds are easily altered and dominated by crowds, especially in a socialized environment. In addition to fulfill personalized intentions, more considerate recommendations, which maximize satisfactions of both individuals and common interests within crowds, are expected in various daily-life scenarios: e.g., scenic spots recommendation to help trip planning making for a group of friends, and movie/TV program recommendation for family members. In this paper, we aim at advancing the group recommendation and propose a novel approach which predicts user preferences with the consideration of "group consensus". We combine observations from real-world group discussions with the model learning and conduct several experiments on a real-world dataset. The results show that the proposed approach benefits both individual and group recommendation and surpasses the state-of-the-art approach in terms of individual preference prediction.