Topic modeling is an effective tool for analyzing the thematic content of large collections of text. However, traditional probabilistic topic modeling is limited to a small number of topics (typically no more than hundreds). We introduce fine-grained topic models, which have large numbers of nuanced and specific topics. We demonstrate that fine-grained topic models enable use cases not currently possible with current topic modeling techniques, including an automatic cross-referencing task in which short passages of text are linked to other topically related passages. We do so by leveraging anchor methods, a recent class of topic model based on non-negative matrix factorization in which each topic is anchored by a single word. We explore extensions of the anchor algorithm, including tandem anchors, which relaxes the restriction that anchors be formed of single words. By doing so, we are able to produce anchor-based topic models with thousands of fine-grained topics. We also develop metrics for evaluating token level topic assignments and use those metrics to improve the accuracy of fine-grained topic models.

Topic models are useful tools for exploring large data sets of textual content by exposing a generative process from which the text was produced. Anchor-based topic models utilize the anchor word assumption to define a set of algorithms with provable guarantees which recover the underlying topics with a run time practically independent of corpus size. A number of extensions to the initial anchor word-based algorithms, and enhancements made to tangential models, have been proposed which improve the intrinsic characteristics of the model making them more interpretable by humans. This thesis evaluates improvements to human interpretability due to: low-dimensional word embeddings in combination with a regularized objective function, automatic topic merging using tandem anchors, and utilizing word embeddings to synthetically increase corpus density. Results show that tandem anchors are viable vehicles for automatic topic merging, and that using word embeddings significantly improves the original anchor method across all measured metrics. Combining low-dimensional embeddings and a regularized objective results in computational downsides with small or no improvements to the metrics measured.