This paper introduces an alternative to the popular machine learning algorithm known as Latent Dirichlet Allocation, or LDA for short. In this paper we derive the theory behind this alternative algorithm and demonstrate a specific use case for it with sample results. We call this new algorithm "keyword-assisted LDA". It works by taking a set of constraints which are set based on prior knowledge of the underlying topic structure within a corpus and then ensuring that they are maintained. Depending on one’s underlying implementation of LDA, keeping these constraints in order takes a variety of forms. This paper delves into the details for implementations using Gibbs sampling or Expectation-Maximization.