Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Stylistic vocabulary based document dating: Frequency-based vs. RoBERTa models

Kuresevic, Ida Zoe LU (2026) STAN40 20261
Department of Statistics
Abstract
The thesis explores historical document dating based solely on vocabulary. That is, we do not consider contextual clues such as years or mentions of historical events in the classification. Both a probability model based on previous research as well as a modern deep learning model using RoBERTa are considered. The thesis demonstrates the difficulty of striking an equilibrium between complexity and precision when working with an unbalanced dataset. Despite its more powerful nature, we
find that RoBERTa does not outperform a simpler frequency-based model in classification of fictional short stories into their publication time intervals.
Please use this url to cite or link to this publication:
author
Kuresevic, Ida Zoe LU
supervisor
organization
course
STAN40 20261
year
type
H1 - Master's Degree (One Year)
subject
language
English
id
9234335
date added to LUP
2026-06-16 10:16:18
date last changed
2026-06-16 10:16:18
@misc{9234335,
  abstract     = {{The thesis explores historical document dating based solely on vocabulary. That is, we do not consider contextual clues such as years or mentions of historical events in the classification. Both a probability model based on previous research as well as a modern deep learning model using RoBERTa are considered. The thesis demonstrates the difficulty of striking an equilibrium between complexity and precision when working with an unbalanced dataset. Despite its more powerful nature, we
find that RoBERTa does not outperform a simpler frequency-based model in classification of fictional short stories into their publication time intervals.}},
  author       = {{Kuresevic, Ida Zoe}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Stylistic vocabulary based document dating: Frequency-based vs. RoBERTa models}},
  year         = {{2026}},
}