Domain-topic models with chained dimensions: Charting an emergent domain of a major oncology conference.

Alexandre Hannud Abdo,Pascale Bourret,Alberto Cambrosio,Jean‐Philippe Cointet

doi:10.1002/asi.24606

Abstract

This paper presents a contribution to the study of bibliographic corpora through science mapping. From a graph representation of documents and their textual dimension, stochastic block models can provide a simultaneous clustering of documents and words that we call a domain‐topic model. Previous work investigated the resulting topics, or word clusters, while ours focuses on the study of the document clusters we call domains. To enable the description and interactive navigation of domains, we introduce measures and interfaces that consider the structure of the model to relate both types of clusters. We then present a procedure that extends the block model to cluster metadata attributes of documents, which we call a domain‐chained model, noting that our measures and interfaces transpose to metadata clusters. We provide an example application to a corpus relevant to current science, technology and society (STS) research and an interesting case for our approach: the abstracts presented between 1995 and 2017 at the American Society of Clinical Oncology Annual Meeting, the major oncology research conference. Through a sequence of domain‐topic and domain‐chained models, we identify and describe a group of domains that have notably grown through the last decades and which we relate to the establishment of “oncopolicy” as a major concern in oncology.

Highlights

Within the tradition of co-word analysis (Callon et al, 1983), understood as the first attempt at using the content of documents to capture the dynamics of scientific activity, a variety of methods have been adopted to reveal meaningful relationships between words, documents, and other dimensions in a corpus, among which are semantic maps and topic models (Chavalarias & Cointet, 2013; Hecking & Leydesdorff, 2019)
In the present paper we explore the fact that these network models can be employed to simultaneously infer document clusters, and we introduce procedures and tools to systematically interpret these clusters in combination with the topic model
We investigate the abstracts of the annual meetings of the American Society for Clinical Oncology (ASCO) from 1995 to 2017, and show how we can lay out 23 years of the world's largest oncology conference in terms of research domains evolving through a sequence of periods where different domains rise and fall

Summary

Introduction

Within the tradition of co-word analysis (Callon et al, 1983), understood as the first attempt at using the content of documents to capture the dynamics of scientific activity, a variety of methods have been adopted to reveal meaningful relationships between words, documents, and other dimensions in a corpus, among which are semantic maps and topic models (Chavalarias & Cointet, 2013; Hecking & Leydesdorff, 2019). In the present paper we explore the fact that these network models can be employed to simultaneously infer document clusters, and we introduce procedures and tools to systematically interpret these clusters in combination with the topic model. Our goal in this paper is not to quantitatively compare our results to those procedures, but to highlight and explore the qualitative possibilities and tools that follow from the simplicity and flexibility of the present approach

Objectives

Methods

Results

Conclusion