The amount of social work publications has vastly increased within the last 30 years - and it continues to
increase (Eckl, Ghanem, &
Löwenstein, 2019; Perron et al.,
2017). It is nearly impossible for researchers to
get a full overview of the literature published for one's own area of interest. The mass of publications but
also fragmentation through different databases, publishers and increasingly specialized journals are
considerable obstacles to efficient research. If scholars cannot get a full overview of the discourses, how
should students or practitioners get one without being familiar with search engines and often complex
strategies for how to identify topic-specific publications. The Social Work Research Map (SWORM) aims to
solve these problems by mapping publications relevant to social work from the last 60 years and key
scientific journals. On one single interactive website, the user is able to identify scholarly literature
from across varied disciplines and sources related to one’s own interest. We also hope to contribute to an
internationalization of social work science. Since many discourses are characterized by an ‘inward gaze’ and
insufficiently acknowledge publications in other countries, SWORM can increase the visibility and
accessibility of research from different geographical areas.
How was SWORM developed?
SWORM represents the results of a quantitative text analysis using Latent Dirichlet Allocation (LDA) by
Blei, Ng and Jordan (2003). This type
of topic modeling is especially suitable for huge data corpora, as
researchers using this statistical procedure do not have to subjectively determine or specify topics;
rather, the probability of certain topics is determined based on the words that occur together in the
documents. LDA is based on a complex statistical model that allows for the investigation of semantic
structures in the text. For an introduction in the background and application of LDA in social work see
Eckl, Ghanem and Spensberger (2021).
The underlying data comprise bibliometric information and abstracts from 23 social work-relevant journals
from 1960 to 2020. We used the top twenty journals assessed by Hodge, Yu and Kim
(2020). We added three
journals that appear within the top ten list based on impact factor scores in the Clarivate Analytics report
from 2019 (when you only consider social work journals listed in the Social Work Research Database; Perron
et al., 2017). In sum, we used 24,472 publications for the analyses.
The LDA identified 40 topics in our corpus, which means that, technically, each document is represented by a
40-dimensional vector. Each entry in such a vector denotes the probability that the corresponding topic
occurs in the document. In order to visualize these 40-dimensional data points on the 2-dimensional plane,
we applied a method called “t-distributed stochastic neighbor embedding” (t-SNE), which, loosely speaking,
aims to reduce the dimensionality of the data while preserving as much of its structure as possible.
The recommendation system provided to authenticated users is largely inspired by the “ArXiv Sanity
Preserver” developed by Karpathy (2019).
Internally, it is based on a machine learning model called support
vector machine (SVM). The intuitive idea is that an SVM is trained to discriminate between the documents a
particular user has added to their personal library, and those that they have not.
It does so by drawing a decision boundary such that added documents reside on one side of the boundary,
while others reside on the other.
The documents that a user has not added and that are closes to this decision boundary are recommended to the
user, since they tend to be more similar to the documents that are already in the library.
You can find the results of this approach under the Map index tab. We hope that many will find it
helpful for their purpose.
Who we are?
We are a research group from three different German universities connected through our interest in the
interconnection between social work and computer science. The aim of our collaboration is to contribute to
the field of social work by using digital technology for research questions from the discipline.
Social worker (B.A., M.A.) and professor at the Nuremberg Institute of
Technology. His research interests mainly comprise professional judgement, evidence-based practice
and offender
rehabilitation.
Computer Scientist (M.Sc.) and currently a PhD candidate at University of Magdeburg. His research
interests include machine learning in general, and its application in anomaly detection in
particular.
