# Is it ok to get negative Cosine Similarity using LSA?

I am getting negative cosine similarity value between two documents in Latent Semantic analysis. How should it be treated?