A1118
Title: Time-varying weighted latent Dirichlet allocation
Authors: Louisa Kontoghiorghes - Kings College London (United Kingdom) [presenting]
Ana Colubi - University of Giessen (Germany)
George Kapetanios - Kings College London (United Kingdom)
Abstract: The time-varying weighted Latent Dirichlet Allocation (tvwLDA), a probabilistic topic modeling method that can track the evolution of topics in a series of documents, is introduced. This approach combines a topic model method, the Latent Dirichlet Allocation (LDA), with time-varying weights to estimate the parameters with the weighted log-likelihood. The tvwLDA estimates the topics using the rolling window method at each time index, where terms in more recent documents have a greater weight when estimating the term-topic distribution and the topic-document density parameter. This method allows the number of topics to vary for each time index, capturing the evolution of topics even when the themes appear inconsistently or new topics emerge. Additionally, a new metric, the dominance index, is introduced. This metric combines the prevalence metric with the Simpson index to assess topic expansion based on a defined set of keywords. The methodology is applied to assess the evolution of the topics from the COST Action HiTEc in the abstracts of the series of conferences CFE-CMStatistics.
