Topic models towards high performance data mining and analysis

While unimaginable amounts of data are continuously stored recording our transactions, conversations, connections, movements, behavior, personality, emotions, and opinions, data has been termed “the new oil”. The process of “refine-ment” and knowledge extraction from data is the core of data mining. Advances in automated algorithms and models for extracing knowledge about human behavior will ultimately measure the value of data. This work discusses the use of probabilistic latent topic models, particularly Latent Dirichlet Allocation (LDA) [2], for data mining and explores its application on various sorts of large-scale data, focusing on the advantages and disadvantages of their use. While topic models have been shown to provide a promising new tool for data mining, one current open issue is with respect to developing methods for implementing them in high performance computing platforms.