We previously proposed an unsupervised model using the inclusion-exclusion principle to compute sentence information content. Though it can achieve desirable experimental results in sentence semantic similarity, the computational complexity is more than O(2ⁿ). In this paper, we propose an efficient method to calculate sentence information content, which employs the thinking of the difference set in hierarchical network. Impressively, experimental results show that the computational complexity decreases to O(n). We prove the algorithm in the form of theorems. Performance analysis and experiments are also provided.