Topic Models Conditioned on Arbitrary Features with Dirichlet-multinomial Regression
David Mimno, Andrew McCallum · arXiv (Cornell University) · 2012
Although fully generative models have been successfully used to model the contents of text documents, they are often awkward to apply to combinations of text data and document metadata. In this paper we propose a Dirichlet-multinomial regression (DMR) topic model that includes a log-linear prior on document-topic distributions that is a function of observed features of the document, such as author, publication venue, references, and dates. We show that by selecting appropriate features, DMR topic models can meet or exceed the performance of several previously published topic models designed for specific data.
Cited by 312 works. doi:10.48550/arxiv.1206.3278 Open-access PDF
Authors
References
- Latent dirichlet allocation
- On the limited memory BFGS method for large scale optimization
- Modeling annotated data
- The author-topic model for authors and documents
- Supervised Topic Models
- Topic modeling
- Topics over time
- Pachinko allocation
- Topic and role discovery in social networks
- Mixed-membership models of scientific publications