Application of the LDA Model for Obtaining Topics from the WIKICORPUS

Gerardo Martínez Guzmán, María Beatríz Bernábe Loranca, Carmen Cerón Garnica, Jonathan Serrano Pérez, Etelvina Archundia Sierra


A fundamental problem in text analysis of great amount of information is to discover the topics described in the documents. One of the most useful application involves the extraction of topics from documents corpus. Such is the case of Wikicorpus that consists of approximately 250,000 documents totaling in 250 millions of words. In this work, a system based on the Latent Dirichlet Allocation (LDA) model has been developed to carry out the task of automatically selecting the words of the corpus and, based on their frequency in the documents, it would indicate that they may or not belong to certain topic, classifying words without human intervention. Due to the large amount of information of the corpus, a Serial-Parallel Algorithm (SPA) in C/C++ and OpenMP have been used to perform parallel programming, since in parallel stages all threads must share certain variables, so the design architecture was shared memory


Corpus, generative model, Dirichlet distribution, latent topics, parallelization, algorithm, C/C++ programming

Full Text: PDF