I am getting negative cosine similarity value between two documents in Latent Semantic analysis. How should it be treated?
After applying LSA the cosine similarity should yield values between -1 and 1. Why exactly is this a problem for you?
See this answer for a good explanation regarding your question.
Cosine similarities could be negative for sure. If you're trying to interpret this (not that you think it's problematic to have negative values), then I think it means these two documents are talking about opposite things.