arXiv subjects

Paul Hofmarcher

Publications and source records attributed to Paul Hofmarcher.

5 recordsLinked to original sources

arXiv2025

Seeded Poisson Factorization: leveraging domain knowledge to fit topic models

Topic models are widely used for discovering latent thematic structures in large text corpora, yet traditional unsupervised methods often struggle to align with pre-defined conceptual domains. This paper introduces seeded Poisson Factorization (SPF), a novel approach that extends the Poisson Factorization (PF) framework by incorporating domain knowledge through seed words. SPF enables a structured topic discovery by modifying the prior distribution of topic-specific term intensities, assigning higher initial rates to pre-defined seed words. The model is estimated using variational inference with stochastic gradient optimization, ensuring scalability to large datasets. We present in detail the results of applying SPF to an Amazon customer feedback dataset, leveraging pre-defined product categories as guiding structures. SPF achieves superior performance compared to alternative guided probabilistic topic models in terms of computational efficiency and classification performance. Robustness checks highlight SPF's ability to adaptively balance domain knowledge and data-driven topic discovery, even in case of imperfect seed word selection. Further applications of SPF to four additional benchmark datasets, where the corpus varies in size and the number of topics differs, demonstrate its general superior classification performance compared to the unseeded PF model.

stat.ME↗

arXiv2025

Evolving Voices Based on Temporal Poisson Factorisation

The world is evolving and so is the vocabulary used to discuss topics in speech. Analysing political speech data from more than 30 years requires the use of flexible topic models to uncover the latent topics and their change in prevalence over time as well as the change in the vocabulary of the topics. We propose the temporal Poisson factorisation (TPF) model as an extension to the Poisson factorisation model to model sparse count data matrices obtained based on the bag-of-words assumption from text documents with time stamps. We discuss and empirically compare different model specifications for the time-varying latent variables consisting either of a flexible auto-regressive structure of order one or a random walk. Estimation is based on variational inference where we consider a combination of coordinate ascent updates with automatic differentiation using batching of documents. Suitable variational families are proposed to ease inference. We compare results obtained using independent univariate variational distributions for the time-varying latent variables to those obtained with a multivariate variant. We discuss in detail the results of the TPF model when analysing speeches from 18 sessions in the U.S. Senate (1981-2016).

stat.ME↗

arXiv2025

Revisiting Group Differences in High-Dimensional Choices: Method and Application to Congressional Speech

Gentzkow, Shapiro and Taddy, Econometrica Vol 87, No 4, 2019 (henceforth GST) use a supervised text-based regression model to assess changes in partisanship in U.S. congressional speech over time. Their estimates imply that partisanship is far greater in recent years than in the past, and that it increased sharply in the early 1990s. The paper at hand provides a replication in the wide sense of GST by complementing their analysis in three ways. First, we propose an alternative unsupervised language model, which combines ideas of topic models and ideal point models, to analyze the change in partisanship over time. We apply this model to the Senate speech data used in GST ranging from 1981-2017. Using our model we replicate their results on the specific evolution of partisanship. Second, our model provides additional insights such as the data-driven estimation of evolvement of topical contents over time. Third, we identify key phrases of partisanship on topic level.

econ.GN↗

arXiv2024

A Structural Text-Based Scaling Model for Analyzing Political Discourse

Scaling political actors based on their individual characteristics and behavior helps profiling and grouping them as well as understanding changes in the political landscape. In this paper we introduce the Structural Text-Based Scaling (STBS) model to infer ideological positions of speakers for latent topics from text data. We expand the usual Poisson factorization specification for topic modeling of text data and use flexible shrinkage priors to induce sparsity and enhance interpretability. We also incorporate speaker-specific covariates to assess their association with ideological positions. Applying STBS to U.S. Senate speeches from Congress session 114, we identify immigration and gun violence as the most polarizing topics between the two major parties in Congress. Additionally, we find that, in discussions about abortion, the gender of the speaker significantly influences their position, with female speakers focusing more on women's health. We also see that a speaker's region of origin influences their ideological position more than their religious affiliation.

stat.ME↗

arXiv2013

Model trees with topic model preprocessing: An approach for data journalism illustrated with the WikiLeaks Afghanistan war logs

The WikiLeaks Afghanistan war logs contain nearly $77,000$ reports of incidents in the US-led Afghanistan war, covering the period from January 2004 to December 2009. The recent growth of data on complex social systems and the potential to derive stories from them has shifted the focus of journalistic and scientific attention increasingly toward data-driven journalism and computational social science. In this paper we advocate the usage of modern statistical methods for problems of data journalism and beyond, which may help journalistic and scientific work and lead to additional insight. Using the WikiLeaks Afghanistan war logs for illustration, we present an approach that builds intelligible statistical models for interpretable segments in the data, in this case to explore the fatality rates associated with different circumstances in the Afghanistan war. Our approach combines preprocessing by Latent Dirichlet Allocation (LDA) with model trees. LDA is used to process the natural language information contained in each report summary by estimating latent topics and assigning each report to one of them. Together with other variables these topic assignments serve as splitting variables for finding segments in the data to which local statistical models for the reported number of fatalities are fitted. Segmentation and fitting is carried out with recursive partitioning of negative binomial distributions. We identify segments with different fatality rates that correspond to a small number of topics and other variables as well as their interactions. Furthermore, we carve out the similarities between segments and connect them to stories that have been covered in the media. This gives an unprecedented description of the war in Afghanistan and serves as an example of how data journalism, computational social science and other areas with interest in database data can benefit from modern statistical techniques.

stat.AP↗