From Words to Networks and Back

Abstract

Digital text has revolutionized how we consume and produce information, and also provides seemingly limitless sources of data from Twitter feeds to online historical archives. Such new data challenge traditional boundaries between quantitative and qualitative research, and exciting horizons have emerged. New analytic approaches are warranted, however, given the typically unstructured, respondent-generated format of such data. In this article, I examine how sociologists have handled text data prior to digitization. Building on recent advancements in computational linguistics and computational social science, I then offer a network-based model and approach for analyzing similarities and locating emergent, general themes in digitized text. I provide a case in point by analyzing the United States’ presidential inaugural addresses. This analysis illustrates how sociologists can take advantage of both the breadth of new digital sources of data and the richness that such qualitative material provides. Indeed, the digitization of texts represents a possible and stimulating sea change in how we tell socio-cultural and historical stories. The greatest potential in these regards rests at the nexus of new computational methods and in-depth, qualitative strategies.

Words ranking and Hirsch index for identifying the core of the hapaxes in political texts

“…• In [40] the analysed data consists of the corpus of United States' presidential Inaugural Addresses from 1789-2009. Also in this case, the dataset is a sensible subset of our corpus.…”

The interconnectedness of the economic content in the speeches of the US Presidents

“…In the context of the US Presidents' communications studies, there are not many works devoted to the analysis of such a wide Presidents' speeches corpus with a combination of text mining and network analysis approaches strictly comparable to our methodological combination. For example, in Light (2014), the author combined text mining and network analysis to analyse the Presidents' Inaugural Addresses, but he has used the Stanford POS-Tagger and a different similarity measure. Hence, we use some excellent studies as reference for validating our approach.…”

Computational genealogy: Continuities and discontinuities in the political rhetoric of US presidents

“…Inaugurals in the same cluster are also part of the same time interval, which is something we also observe as a strong temporal bias in the data. Light (2014) uses part-of-speech tagging (PoS) to develop a sophisticated cluster network analysis of the Inaugurals. He creates a "co-word network for the inaugural addresses" using the "147 most prominent (…) words".…”

“…Again, length normalisation is applied. We employ Pearson, as it has also been used in (Light 2014) to analyse the Inaugurals. For all words in two documents x and y, the Pearson correlation divides the sum of their frequencies in x and y minus their respective mean frequencies m(x) and m(y) by the square roots of the squares of (xm(x)) and (y -m(y)): .…”