Map of science with topic modeling: Comparison of unsupervised learning and human‐assigned subject classification

Abstract

The delineation of coordinates is fundamental for the cartography of science, and accurate and credible classification of scientific knowledge presents a persistent challenge in this regard. We present a map of Finnish science based on unsupervised-learning classification, and discuss the advantages and disadvantages of this approach vis-à-vis those generated by human reasoning. We conclude that from theoretical and practical perspectives there exist several challenges for human reasoningbased classification frameworks of scientific knowledge, as they typically try to fit new-to-the-world knowledge into historical models of scientific knowledge, and cannot easily be deployed for new large-scale data sets. Automated classification schemes, in contrast, generate classification models only from the available text corpus, thereby identifying credibly novel bodies of knowledge. They also lend themselves to versatile large-scale data analysis, and enable a range of Big Data possibilities. However, we also argue that it is neither possible nor fruitful to declare one or another method a superior approach in terms of realism to classify scientific knowledge, and we believe that the merits of each approach are dependent on the practical objectives of analysis.

Comparing grounded theory and topic modeling: Extreme divergence or unlikely convergence?

“…For example, several topics identified different types of triggers for returning to Facebook, but no single topic aligned with the higher‐level theme of triggers. In applying LDA (Blei et al, ) to Finnish scientific papers, Suominen and Toivanen () similarly found that, in most cases, the topic model included finer distinctions than existing classification schemes. For instance, the Organization for Economic Cooperation and Development classification for medical research includes 10 subfields, but documents in that classification came from over 26 different topics (Suominen & Toivanen, , p. 7).…”

“…For instance, topic modeling has been used to uncover the temporal evolution of topics in documents (Song et al, ), to compare between the popularity of scientific themes in open and closed access publications (Hu et al, ), and to detect communities of taggers in social recommendation systems (Li et al, ). Some preliminary work has compared machine classification with existing categorization schemes (Suominen & Toivanen, ).…”

Scientific evolutionary pathways: Identifying and visualizing relationships for scientific topics

“…However, compared to citations, the relationships derived from semantic structures are not as direct and clear as citation links. In recent years, the latent Dirichlet allocation approach (Blei, Ng, & Jordan, 2003) has played an active role in science maps (Suominen & Toivanen, 2015;Yau, Porter, Newman, & Suominen, 2014), as well as ontology and semantic webs (Wang, Lu, & Zhang, 2007). In recent years, the latent Dirichlet allocation approach (Blei, Ng, & Jordan, 2003) has played an active role in science maps (Suominen & Toivanen, 2015;Yau, Porter, Newman, & Suominen, 2014), as well as ontology and semantic webs (Wang, Lu, & Zhang, 2007).…”

Exploring archives with probabilistic models: Topic modelling for the valorisation of digitised archives of the European Commission

“…With that in mind, we resorted to matching the most prominent tokens with EuroVoc 4 terms, allowing us to base our work on a solid, well-documented foundation on the one hand, and to harness the power of a hierarchical, multilingual thesaurus on the other. Despite the fact that LDA provides a distribution of topics for each document, we resorted, in this case study, to hard-partitioning (Suominen and Toivanen, 2015): as this is a proof-of-concept approach to a semi-automatic classification of historical archives, soft-partitioning -associating several topics to a document -would have proven too time-consuming for too low a gain. Indeed, it can be argued that if topic models do detect the main topic of a document, as it is the case in our study, subsequent less-important topics can be assumed to be correct.…”

“…This approach, along with other text-mining routines, has gained momentum for document classification, as pointed out by Suominen and Toivanen (2015) in the specific field of bibliometrics or by Newman et al (2010) in a library context. Similarily, Roe et al (2016) uses LDA in order to draw a map of all human knowledge -as seen by d'Alembert and Diderot -contained in the French Encyclopédie.…”