ttta: Tools for Temporal Text Analysis

In its current state, the ttta package includes diachronic embeddings, dynamic topic modeling, and document scaling. These tools can be used to track changes in language use, identify emerging topics, and explore how the meaning of words and phrases has evolved over time. Our dynamic topic model approach is based on the model RollingLDA (Rieger et al., 2021), which is a modification of the classic Latent Dirichlet Allocation (Blei et al., 2003), that allows for the estimation of topics over time using a rolling window approach. We additionally implemented the model LDAPrototype (Rieger et al., 2020), serving as a more consistent foundation for RollingLDA than a common LDA. With these models, users can uncover and analyze topics of discussion in temporal data sets and track even rapid changes, which other dynamic topic models struggle with. This ability to track rapid changes in topics is further used in the Topical Changes model put forth by Rieger et al. (2022) and Lange et al. (2022) that identifies change points in the word topic distribution of RollingLDA. Figure 1 visualizes the changes observed by the Topical Changes model in speeches from the German Bundestag (Lange & Jentsch, 2023), which can be analyzed further using leave-one-out word impacts provided by the model or, as Lange et al. (2025) proposed, by asking Large Language Models to interpret the change and relate it to a possible narrative shift.