Context-aware recommender systems are specifically designed to exploit contextual information during the recommendation process to achieve higher efficiency than traditional non-contextual recommender systems. However, one of the major challenges in building and deploying recommender systems is data sparsity, which directly affects the quality of recommendations. Although using contexts provides more useful information in the recommendation process, it also increases the data sparsity and computational complexity. In this paper, we propose a new context-aware collaborative filtering method that comprehensively integrates contextual information, minimizes data sparsity’s impact, and maintains low computational complexity. This is resolved by combining an improved contextual user-splitting technique and a method of graph-based spreading activation. Practical experiments on several datasets show that the proposed method improves prediction accuracy significantly and is feasible in terms of implementation time compared to baseline context-aware collaborative filtering methods.