Abstract
This chapter proposes a novel method that deploys non-negative matrix factorization to extract topic models from texts. This topic modeling process reveals how terms and DocuScope Language Action Type Analysis (LATs) align, providing robust information on what texts are about and how they are organized rhetorically. Moreover, the non-negative nature of the topics means that each derived topic can be viewed as a sum of topical features, which can greatly ease the interpretive process. To elucidate and benchmark this method, I apply it to a well-known 20 Newsgroups dataset and sample the results.
Publisher
John Benjamins Publishing Company
Reference58 articles.
1. Angelov , D. ( 2020 ). Top2vec: Distributed representations of topics . arXiv:2008.09470 .
2. A practical algorithm for topic modeling with provable guarantees;Arora,2013
3. Computing a Nonnegative Matrix Factorization---Provably
4. Unsupervised sentence representations as word information series: Revisiting TF–IDF