Visualizing the Topical Structure of the Medical Sciences: A Self-Organizing Map Approach by André Skupin, Joseph R. Biberstine, Katy Börner. (Skupin A, Biberstine JR, Börner K (2013) Visualizing the Topical Structure of the Medical Sciences: A Self-Organizing Map Approach. PLoS ONE 8(3): e58779. doi:10.1371/journal.pone.0058779)
We implement a high-resolution visualization of the medical knowledge domain using the self-organizing map (SOM) method, based on a corpus of over two million publications. While self-organizing maps have been used for document visualization for some time, (1) little is known about how to deal with truly large document collections in conjunction with a large number of SOM neurons, (2) post-training geometric and semiotic transformations of the SOM tend to be limited, and (3) no user studies have been conducted with domain experts to validate the utility and readability of the resulting visualizations. Our study makes key contributions to all of these issues.
Documents extracted from Medline and Scopus are analyzed on the basis of indexer-assigned MeSH terms. Initial dimensionality is reduced to include only the top 10% most frequent terms and the resulting document vectors are then used to train a large SOM consisting of over 75,000 neurons. The resulting two-dimensional model of the high-dimensional input space is then transformed into a large-format map by using geographic information system (GIS) techniques and cartographic design principles. This map is then annotated and evaluated by ten experts stemming from the biomedical and other domains.
Study results demonstrate that it is possible to transform a very large document corpus into a map that is visually engaging and conceptually stimulating to subject experts from both inside and outside of the particular knowledge domain. The challenges of dealing with a truly large corpus come to the fore and require embracing parallelization and use of supercomputing resources to solve otherwise intractable computational tasks. Among the envisaged future efforts are the creation of a highly interactive interface and the elaboration of the notion of this map of medicine acting as a base map, onto which other knowledge artifacts could be overlaid.
Impressive work to say the least!
But I was just as impressed by the future avenues for research:
It appears that the use of indexer-chosen keywords, including in the case of a large controlled vocabulary-MeSH terms in this study-raises interesting questions. The rank transition diagram in particular helped to highlight the fact that different vocabulary items play different roles in indexers’ attempts to characterize the content of specific publications. The complex interplay of hierarchical relationships and functional roles of MeSH terms deserves further investigation, which may inform future efforts of how specific terms are handled in computational analysis. For example, models constructed from terms occurring at intermediate levels of the MeSH hierarchy might look and function quite different from the top-level model presented here.
Future user studies will include term differentiation tasks to help us understand whether/how users can differentiate senses of terms on the self-organizing map. When a term appears prominently in multiple places, that indicates multiple senses or contexts for that term. One study might involve subjects being shown two regions within which a particular label term appears and the abstracts of several papers containing that term. Subjects would then be asked to rate each abstract along a continuum between two extremes formed by the two senses/contexts. Studies like that will help us evaluate how understandable the local structure of the map is.
There are other, equally interesting future research questions but those are the two of most interest to me.
I take this research as evidence that managing semantic diversity is going to require human effort, augmented by automated means.
I first saw this in Nat Torkington’s Four short links: 13 March 2013.