Cross-Work Theme Identification in Long Novels via Nonnegative Tensor Factorization

Read full text Summarise with VIRAT
Authors: Yiying Chen, Mei Liu, Yuning Yang
Cited by: 0
References: 30

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Identifying the major themes and recurring motifs of an author’s long novels is a basic task in literary studies. To support this task in a scalable way while retaining within-novel narrative variation, we model an author corpus as a third-order nonnegative tensor indexed by work × narrative segment × vocabulary. For such narrative tensors, we propose a tailored nonnegative tensor factorization model that reduces redundancy among topics via an orthogonality-promoting penalty and promotes smooth topic variation along contiguous narrative segments via an ℓ2 total-variation penalty. We develop a block proximal linearization algorithm for the resulting optimization problem and show that every limit point of the generated sequence satisfies the KKT conditions. Experiments on Toni Morrison’s long novels, including comparisons of the results of the proposed model with those of NMF and LDA, suggest that the cross-work themes extracted by the proposed approach exhibit qualitative patterns broadly consistent with thematic concerns discussed in existing literary scholarship. Additional experiments on the corpora of Ernest Hemingway and Graham Swift provide further validation of the proposed model.

Fields of Study & Subjects

Tensor decomposition and applications Topic Modeling Computational and Text Analysis Methods

Publication details

Journal / source
Mathematics
Volume
14
Issue
7
Pages
1106–1106
Publication date
25 March 2026
Language
EN