Next Generation SciVal Topics available as of May 21st 2024

The current set of Topics was introduced back in 2016. Since that time new Topics have been created as needed and new papers have been assigned to the Topic structure using state-of-the-art methods. However, after 7 years, and in light of recent methodological advancements, we are now able to produce more precise and cohesive publication groupings within Topics and so we felt it appropriate to release a new updated set of Topics.

The methodology concept does not change.

We still take the entire direct citation network (about 1.7 billion links) between 59+ million Scopus-indexed publications from 1996 onwards as well as an additional 24+ million non-indexed documents that are cited at least twice, and indexed and non-indexed publications from before 1996 – and break that network into roughly 94,000 Topics and 1,500 Topic Clusters.

A Topic is created where the direct citation linkages within the Topic are strong and the direct citation linkages outside the Topic are weak. Only the Scopus-indexed publications from 1996 onwards are included in Topics available to analyze in SciVal.

Over time however, we have learned how to better separate the signal from noise to form tighter (more strongly connected) sets of publications in the Topics. New Topics result in a noticeably higher proportion of publications being strongly linked to a Topic (71.8% on average vs 63.6% currently).

• More cohesive sets of publications in a Topic

  With our ability to better separate signal from noise when clustering publications into Topics, in the new Topics set a significantly higher proportion of publications are strongly linked to a Topic, or more central to it: 71.8% on average vs 63.6% for the previous generation Topics, providing greater relevance and precise insights.  

  There are five centrality categories defining how strongly a publication is linked to its Topic: 
  1 - Definitive; 2 - Very good; 3- Defensible; 4- Weak; 5 - Other

  We consider a publication to be strongly linked to the Topic to which it is assigned if it falls within either of the first two categories (1-Definitive or 2-Very good). In general, a publication is deemed central if at least 30% (with a minimum of 5 links) or at least 20 (with a minimum of 10%) of its citation links (either references or citations from later publications) are to publications in the same Topic.  

  It is unrealistic to expect all publications to be strongly linked to a Topic. Some publications have few references, while others refer to a broad swath of prior research and/or are cited by a broad set of publications from many Topics. Only a small fraction of publications (6% on average) are weakly linked (4-Weak or 5-Other) to their Topic, with most of these having very few references and/or citations. 

• More granularity and precision for previously very large Topics

  In the old version of Topics, some Topics had grown very large over time. These are important Topics, they grew very large because they are very active. With the new generation of SciVal Topics, we have been able to increase the granularity of the Topic space in these key areas by effectively dividing these very large Topics into multiple, logically coherent clusters of publications.  

  Example:
  Old Topic #4338: Object Detection; Deep Learning; IOU was focused on deep learning and convolutional neural networks (CNN). In 2016, when the previous generation Topics were launched, this Topic only contained about 1,600 publications. By the end of 2023, this Topic had grown to encompass more than 150,000 publications, with more than 45,000 being added in 2023 alone.  

  This exponential growth can be attributed to the absence of other (competing) Topics utilizing the tools created in this Topic for different applications. Thus, as publications from 2017-2023 were added to Topics, the majority of the deep learning / CNN papers were allocated to this Topic, regardless of whether they were about the development of new methods or new applications.  

  With the new set of Topics, the old Topic #4338 had been divided into dozens of new Topics. The largest, new generation Topic #0, contains over 16,000 publications focused on DL/ML/CNN methods. Thirteen other Topics contain over 1,000 publications each from the current Topic #4338. Most of these Topics are also method-focused, with subjects like generative adversarial networks, filter pruning, CNN accelerators, explainable AI, semantic segmentation, etc. However, most of the publications have moved into smaller application-focused Topics where deep learning is not the primary focus but is used to solve other problems, such as for example image analysis for medical diagnostics, autonomous vehicles navigation, climate and weather prediction models, waste classification, pest detection and more.

  While offering more granularity for large Topics, we are also reducing the long tail of small Topics reassigning the publications to Topics with at least 50 publications and displaying stronger connections than previously. 

• Generative AI powered summaries for Topics and Topic Clusters

  Beta: generative AI (GenAI) and LLM technology powered summaries for Topics and Topic Clusters help you quickly understand the research fields and questions being addressed within the Topic. You can access the summaries by clicking the “About this Topic/Topic Cluster” link in the Topic entity header, or ‘i’ icon in the Topics’ listings. We also welcome your feedback.

  Learn more about Topic and Topic Cluster summaries

• Improved names for Topics and Topic Clusters

  For the new generation Topics naming we are using a new OmniScience thesaurus - an in-house developed single unified thesaurus spanning all major disciplines to create a list of standardized concepts, better covering the essence of each cluster of publications making up the Topics.

  Learn more about how the names of Topics and Topic Clusters are generated

For the majority of Topics there is no direct one-on-one mapping from the old to the new Topics. Large or small portions of old Topics’ publications form a large or a small portions of new Topics publications.

However, when comparing the new set of Topics with the old set, there is a strong overlap, especially for the cores of the Topics – groups of publications strongly connected.

With the launch of the new generation Topics and Topic Clusters, the old Topics and Topic Clusters are no longer available. For about 6 months functionality is available to help you transition. The transition functionality will be decommissioned with the first SciVal release of 2025, on January 28th.

Learn more about the transition functionality