This paper introduces Locally Consistent Concept Factorization (LCCF) for document clustering, which improves clustering performance by leveraging local geometric structures of data. By employing a graph model and a graph Laplacian for regularization, LCCF extracts document concepts that correspond to locally connected components in a manifold, leading to enhanced clustering accuracy. Experimental results demonstrate that LCCF outperforms traditional methods such as Non-negative Matrix Factorization (NMF) and Concept Factorization (CF) in terms of clustering quality.