Transparency to Visibility (T2V): Network Visualization in Humanities Research

FAIN: HAA-261070-18

University of Texas, Austin (Austin, TX 78712-0100)
Samuel Scott Graham (Project Director: January 2018 to June 2021)

The development of a set of tools to automatically extract and visualize relationships in large textual corpora, with a focus on making “hidden” relationships more visible.

Humanities researchers have long studied how power and influence circulate through cultural systems. Advances in network visualization tools support this work, allowing scholars to create graphical representations of complex systems. However, extracting and preparing relational data for visualization can present significant technological challenges when working with the kinds of textual artifacts commonly studied by humanists. This project will develop and test an innovative approach for efficiently curating and visualizing relationships in ways that align with humanities research. Using sample texts from medical research, a digital and medical humanities team will develop, test, and enhance a new toolkit for automatically extracting and visualizing relationships in large textual corpora. The project team will create both a graphical user interface for the toolkit and an open-source code repository to support use by digital humanities scholars.

Media Coverage

Reprint fees linked to conflicts (Media Coverage)
Author(s): Jeffrey Brainard
Publication: Science
Date: 8/7/2020
Abstract: News in brief discussion of Plos-One article.
URL: https://science.sciencemag.org/content/369/6504/604

Associated Products

Transparency to visibility (T2V): Digital resistance and the medical-industrial complex. (Conference Paper/Presentation)
Title: Transparency to visibility (T2V): Digital resistance and the medical-industrial complex.
Author: S. Scott Graham
Author: Dave Clark
Author: Zoltan Majdki
Abstract: Bioethicists and humanistic researchers alike have long been concerned over the effects of unchecked industry money on biomedical cultures. Corporate-funded clinical trials, free lunches, free travel, and industry honoraria for scientists have been shown to adversely affect the integrity of biomedical research. Despite the broad recognition of these hazards, efforts to address them have largely been limited to disclosure requirements and related training. Current standards require clinical researchers to report certain conflicts of interest alongside published scholarship. Unfortunately, a growing body of evidence indicates that disclosure statements often result in unintended and pernicious effects. Ultimately, a significant shift in our understanding of conflicts of interest is required, and this is where the insights of digital humanities (DH) can be most beneficial. However, in order for this kind of theoretically-informed relational network modeling to be of use to humanities research, scholars need effective and efficient tools to transform the dense prose of disclosure statements, financial reports, or database outputs into useable visualizations. This is precisely what the Transparency to Visualization (T2V) tool is designed to do. Our toolkit combines natural language processing, machine-learning enhanced named-entity recognition, and regular expressions to create a script that "reads" biomedical disclosure statements for economic relationships and prepares extracted data for subsequent visualization. The proposed presentation will explain and demonstrate the toolkit while reflecting on how these kinds of resistant DH projects can contribute positively to issues of social concern.
Date: 9/27/2019
Conference Name: Digital Frontiers

Network Modeling in R (Public Lecture or Presentation)
Title: Network Modeling in R
Abstract: Network modeling in R workshop. Focuses on data cleaning and preparation and network visualization using iGraph and VisNetwork.
Author: S. Scott Graham
Date: 9/20/2019
Location: University of Texas at Austin

Transparency to visibility (T2V) Git Repository (Web Resource)
Title: Transparency to visibility (T2V) Git Repository
Author: S. Scott Graham
Author: Zoltan Majdik
Author: Dave Clark
Abstract: Git repository for the T2V project.
Year: 2019
Primary URL: https://gitlab.com/grahamss/transparency2visibility
Primary URL Description: Git repository for the T2V project.

Transparency to visibility: Bioethics and computational rhetoric (Conference Paper/Presentation)
Title: Transparency to visibility: Bioethics and computational rhetoric
Author: S. Scott Graham
Abstract: Rhetoricians of medicine have long been concerned over the effects of industry money on biomedical practices. Corporate-funded clinical trials, free lunches, free travel, and industry honoraria for scientists have been shown to adversely affect the integrity of biomedical research. Despite the broad recognition of these hazards, efforts to address them have largely been limited to disclosure requirements. Current standards require clinical researchers to report certain conflicts of interest alongside published scholarship. Unfortunately, a growing body of evidence indicates that disclosure statements have been shown to cause audiences to actually extend more trust to those holding conflicts of interest, as disclosure provides an opportunity to display both honesty and expertise. disclosure requirements embody some of the most concerning aspects of neoliberal oversight in that they focus attention on individual behavior rather than cultural conditions. Ultimately, a significant shift in our understanding of conflicts of interest is required, and this is where the insights of computational rhetorics can be beneficial. Accordingly, this paper presents findings derived from a computational analysis of 34 million biomedical research articles. Specifically, this paper explores the development a new integrated machine-learning, named-entity recognition toolkit designed to extracting financial relationships data from natural language and visualize those relationships using interactive network diagrams. The ultimate aim here is to allow rhetoricians of medicine and allied researchers to visualize the systemic aggregation of biomedical funding for any given health area of interest. This research has the potential to effectively support future rhetorics of medicine scholarship on medicine and industry funding.
Date: 11/17/2019
Conference Name: National Communication Association

Conflict of Interest: Article XML (Database/Archive/Digital Edition)
Title: Conflict of Interest: Article XML
Author: S. Scott Graham
Author: Zoltan Majdik
Author: David Clark
Abstract: Included: XML file and dictionary for conflict of interest data. (2020-01-20)
Year: 2020
Primary URL: https://doi.org/10.18738/T8/VSWAJY
Primary URL Description: DOI link to Texas Data Repository
Secondary URL: https://dataverse.tdl.org/dataset.xhtml?persistentId=doi:10.18738/T8/VSWAJY
Secondary URL Description: Direct link to Texas Data Repository
Access Model: CC0 - "Public Domain Dedication"

Relationships among Commercial Biases and Author Conflicts of Interest in Biomedical Publishing (Article)
Title: Relationships among Commercial Biases and Author Conflicts of Interest in Biomedical Publishing
Author: S. Scott Graham
Author: Zoltan Majdik
Author: Dave Clark
Author: Molly Kessler
Author: Tristin Brynn Hooker
Abstract: Recently, concerns have been raised over the potential impacts of commercial biases on editorial practices in biomedical publishing. Specifically, it has been suggested that commercial biases may make editors more open to publishing articles with author conflicts of interest (aCOI). Using a data set of 128,781 articles published in 159 journals, we evaluated the relationships among commercial publishing biases and reported author conflicts of interest. The 159 journals were grouped according to commercial biases (reprint services, advertising revenue, and ownership by a large commercial publishing firm). 30.6% (39,440) of articles were published in journals showing no evidence of commercial publishing biases. 33.9% (43,630) were published in journals accepting advertising and reprint fees; 31.7% (40,887) in journals owned by large publishing firms; 1.2% (1,589) in journals accepting reprint fees only; and 2.5 % (3,235) in journals accepting only advertising fees. Journals with commercial biases were more likely to publish articles with aCOI (9.2% (92/1000) vs. 6.4% (64/1000), p = 0.024). In the multivariate analysis, only a journal's acceptance of reprint fees served as a significant predictor (OR = 2.81 at 95% CI, 1.5 to 8.6). Shared control estimation was used to evaluate the relationships between commercial publishing biases and aCOI frequency in total and by type. BCa-corrected mean difference effect sizes ranged from -1.0 to 6.1, and confirm findings indicating that accepting reprint fees may constitute the most significant commercial bias. The findings indicate that concerns over the influence of industry advertising in medical journals may be overstated, and that accepting fees for reprints may constitute the largest risk of bias for editorial decision-making.
Year: 2020
Primary URL: https://www.medrxiv.org/content/10.1101/2020.01.24.20018705v2
Access Model: CC-BY-NC-ND 4.0
Format: Other
Publisher: medRxiv

Relationships among commercial practices and author conflicts of interest in biomedical publishing (Article)
Title: Relationships among commercial practices and author conflicts of interest in biomedical publishing
Author: S. Scott Graham
Author: Zoltan P. Majdik
Author: Dave Clark
Author: Molly M. Kessler
Author: Tristin Brynn Hooker
Abstract: Recently, concerns have been raised over the potential impacts of commercial relationships on editorial practices in biomedical publishing. Specifically, it has been suggested that certain commercial relationships may make editors more open to publishing articles with author conflicts of interest (aCOI). Using a data set of 128,781 articles published in 159 journals, we evaluated the relationships among commercial publishing practices and reported author conflicts of interest. The 159 journals were grouped according to commercial biases (reprint services, advertising revenue, and ownership by a large commercial publishing firm). 30.6% (39,440) of articles were published in journals showing no evidence of evaluated commercial publishing relationships. 33.9% (43,630) were published in journals accepting advertising and reprint fees; 31.7% (40,887) in journals owned by large publishing firms; 1.2% (1,589) in journals accepting reprint fees only; and 2.5% (3,235) in journals accepting only advertising fees. Journals with commercial relationships were more likely to publish articles with aCOI (9.2% (92/1000) vs. 6.4% (64/1000), p = 0.024). In the multivariate analysis, only a journal’s acceptance of reprint fees served as a significant predictor (OR = 2.81 at 95% CI, 1.5 to 8.6). Shared control estimation was used to evaluate the relationships between commercial publishing practices and aCOI frequency in total and by type. BCa-corrected mean difference effect sizes ranged from -1.0 to 6.1, and confirm findings indicating that accepting reprint fees may constitute the most significant commercial bias. The findings indicate that concerns over the influence of industry advertising in medical journals may be overstated, and that accepting fees for reprints may constitute the largest risk of bias for editorial decision-making.
Year: 2020
Primary URL: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0236166
Access Model: Open access
Format: Journal
Periodical Title: Plos One
Publisher: Plos One

Methods for Extracting Relational Data from Unstructured Texts Prior to Network Visualization in Humanities Research (Article)
Title: Methods for Extracting Relational Data from Unstructured Texts Prior to Network Visualization in Humanities Research
Author: S. Scott Graham
Author: Zoltan P. Majdik
Author: Dave Clark
Abstract: Network modelling methodologies in the digital humanities have been be used to advance inquiry in a variety of areas—most commonly those having to do with correspondence, citation, and social media networks. While new technologies have made generating high-quality and even dynamic network visualizations relatively easy, key challenges remain for humanities researchers. Many common objects of humanistic inquiry, such as aesthetic, historiographic, and biographical texts are often not easily transformed into the kinds of data structures necessary for network visualization. The Transparency to Visibility (T2V) Project was initiated to develop new methods and toolkits that can support humanistic researchers who need to extract relationship data from unstructured texts to support network visualization. The T2V team used bioethics accountability statements to pilot and evaluate different methods for extracting relationship data. The resulting machine-learning-enhanced natural language processing (NLP) and metadata-assisted approaches offer promising potential pathways for contemporary digital humanities and future toolkit development.
Year: 2020
Primary URL: https://openhumanitiesdata.metajnl.com/article/10.5334/johd.21/
Access Model: open access
Format: Journal
Periodical Title: Journal of Open Humanities Data
Publisher: Journal of Open Humanities Data

---