[Paper Review] A Computational Model For Individual Scholars' Writing Style Dynamics

Teddy Lazebnik, Ariel Rosenfeld|arXiv (Cornell University)|May 1, 2023

Complex Network Analysis Techniques4 citations

TL;DR

This study develops a computational model using temporal co-authorship graphs and NLP to analyze how individual scholars' writing styles evolve over time, revealing that writing style stabilizes around the 13th publication and that early-career scholars are most influenced by co-authors, especially same-gender or same-field collaborators. The model identifies key dynamics in academic writing style development through large-scale analysis of 13.7 million computer science publications.

ABSTRACT

A manuscript's writing style is central in determining its readership, influence, and impact. Past research has shown that, in many cases, scholars present a unique writing style that is manifested in their manuscripts. In this work, we report a comprehensive investigation into how scholars' writing styles evolve throughout their careers focusing on their academic relations with their advisors and peers. Our results show that scholars' writing styles tend to stabilize early on in their careers -- roughly their 13th publication. Around the same time, scholars' departures from their advisors' writing styles seem to converge as well. Last, collaborations involving fewer scholars, scholars from the same gender, or from the same field of study seem to bring about greater change in their co-authors' writing styles with younger scholars being especially influenceable.

Motivation & Objective

• To investigate how individual scholars' writing styles evolve throughout their academic careers.
• To examine the extent to which advisees diverge from their advisors' writing styles over time.
• To analyze the impact of collaborative writing on style change, particularly in relation to gender, field of study, and number of co-authors.
• To develop a computational framework that models writing style dynamics using time-dependent social graphs and deep learning.
• To inform academic training programs by identifying critical periods and factors influencing writing style development.

Proposed method

• The study uses a time-dependent mathematical graph to model co-authorship dynamics across scholars' publication histories.
• It leverages the DBLP database and integrates text retrieval from CrossRef and author profiles from SciProfiles for 13.7 million computer science publications.
• Natural language processing techniques are applied to extract and represent writing style features from manuscripts.
• A deep learning-based style embedding model captures the evolution of individual writing styles over time.
• The model computes style distance between authors and their advisors, and measures style change due to collaboration using statistical analysis.
• The framework tracks style convergence and divergence using publication count as a proxy for career stage.

Experimental results

Research questions

• RQ1 How does a scholar’s writing style significantly change over time?
• RQ2 How do research students (i.e., advisees) part from their advisors’ writing styles?
• RQ3 How is a scholar’s writing style affected by collaborations, particularly in relation to gender, field of study, and number of co-authors?

Key findings

• Scholars’ writing styles stabilize around their 13th published manuscript, indicating a key milestone in style development.
• The divergence from advisors’ writing styles follows a sigmoid-like pattern and converges around the 14th publication, aligning with apprenticeship dynamics.
• Collaborations with fewer co-authors lead to greater writing style change, especially when co-authors are from the same field or gender.
• Younger scholars, defined by lower publication count, are significantly more influenceable in their writing style compared to more experienced authors.
• Same-gender collaborations result in more pronounced style changes than mixed-gender collaborations, which show symmetric but less impactful effects.
• Co-authorship with scholars from the same field has a stronger influence on writing style change than cross-field collaborations, contrary to expectations.