The n-gram model has been widely used to capture the local ordering of words, yet its exploding feature space often causes an estimation issue. Local Topic Models introduces a new concept of locality, local contexts, which provide rich word representations that generate locally coherent topics and document representations. Based on its novel features, we employ a nonprobabilistic topic modeling technique, sparse coding, to effectively handle large feature space. Sparse coding efficiently finds topics and representations by applying numerous fast-methods (such as greedy coordinate descent). The model is useful for discovering local topics and the semantic flow of a document and constructing predictive models.
Seungyeon Kim, Joonseok Lee, Guy Lebanon and Haesun Park. Local Context Sparse Coding. Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI). 2015. pdf