The Unified Astronomy Thesaurus: Semantic Metadata for Astronomy and Astrophysics

The Unified Astronomy Thesaurus is the most recent iteration of efforts to develop a keyword classification system that can be used to describe the astronomical concepts and entities found within literature, software, conferences, proposals, data sets, and other astronomical resources. The International Astronomical Union (IAU) Thesaurus,⁶ compiled by Robyn Shobbrook of the Australian Astronomical Observatory and Robert Shobbrook of the University of Sydney in 1992, is often considered the first such attempt (Shobbrook & Shobbrook 1992). The IAU Thesaurus was last updated in 1995 and contained 2,551 terms organized into a loose hierarchy. In 2008, an update to the IAU Thesaurus was created by Rick Hessman and labeled the International Virtual Observatory Alliance, or IVOA, Thesaurus.⁷ Hessman's work added about 300 terms, bringing his proposed IVOA Thesaurus to 2,889 terms.

Around the time of the initial development of the IAU Thesaurus, editors of the major astronomy and astrophysics journals worked together to create an organized set of keywords for classifying and indexing journal articles. This system is called the Astronomical Subject Keywords⁸ (ASK) and was adopted for use in 1992 (Accomazzi et al. 2014). Although the limitations of this classification scheme have become apparent in the intervening years, this set of terms is still currently in use. The Astronomical Subject Keywords were last updated in 2013, and contain about 370 concepts organized by general subject area.

The Physics and Astronomy Classification Scheme⁹ (PACS) is a classification scheme containing roughly 400 astronomical terms, developed by the American Institute of Physics (AIP) in the 1970s to annotate the physics literature using a hierarchical set of codes. PACS was proposed in 1975, updated every two years, and served as the main organizational scheme for the major physics journals until it was decommissioned in 2011 in favor of developing a new system that would support modern indexing and searching technology.

In 2010, major publishers in physics and astronomy, together with other stakeholders from the IVOA, the NASA Astrophysics Data System (ADS), and the Harvard-Smithsonian Center for Astrophysics Library, began to explore new approaches to the various classification schemes in use. Specifically, the Institute of Physics (IoP) and AIP decided to collaborate, funding a project to merge the astronomy portion of their vocabularies with the enhanced IVOA Thesaurus to create a more robust thesaurus for astronomy. The resulting work was named the Unified Astronomy Thesaurus and was donated to the American Astronomical Society (AAS), who hold ownership and copyright. In order to further develop and promote the UAT, the AAS established a Steering Committee, which currently includes members from the AAS, the ADS, the library community, and other astrophysics institutions.

The UAT has been designed as an open, interoperable, and community-supported project that embraces modern standards for thesaurus curation and linked data formats. It uses methods and tools that make it easy for members of the astronomical community to suggest term additions, refinements, revisions, and deletions, allowing open contributions for the maintenance of the thesaurus. The thesaurus is currently distributed as a SKOS¹⁰ (Simple Knowledge Organization System) document, a standard format that uses RDF (Resource Description Framework) for structured controlled vocabularies. This will allow for the UAT to be integrated into a variety of applications, acting as a bridge between disparate systems.

Although there are a range of options for organizing a controlled vocabulary—all the way from a flat list of unconnected concepts to a structured arrangement of precise relationships—the format of a thesaurus was determined to best support the goals of search and findability. According to the ANSI/NISO standard on controlled vocabularies, a thesaurus is a "controlled vocabulary arranged in a known order and structured so that the various relationships among terms are displayed clearly and identified by standardized relationship indicators" (ANSI/NISO Z39.19, 2010). This allows us to define some general relationships between concepts using standard SKOS properties (e.g., "broader," "narrower," "related," "preferred label," and "alternate label") while avoiding overly complex structures.

Linking concepts together in child/parent relationships allows a user to follow a path from broader terms to narrower terms and to discover more specific terms to assist in finding the concepts that best describe their resource. This structure also helps to give additional context to each term, which can help clarify its meaning. More importantly, this hierarchy can inform searching and indexing applications that child terms are subsets of their parent terms, and can be used to suggest related, alternative results to the user.

The UAT has been licensed under a Creative Commons Attribution-ShareAlike license (CC-BY-SA). This license was chosen by considering both the nature of the Thesaurus as a community-supported free cultural work and the intended process for its use and development. The CC-BY-SA license grants anyone the freedom to copy, use, or improve the work while requiring that derivative works be made available under the same license. This license reflects the collaborative process that has led us to the creation and steady improvement of the UAT and safeguards its future openness. Incidentally, having a license for the UAT was recognized as a necessity early on: the lack of licensing terms for the progenitors of the UAT has been a cause of concern for several publishers interested in reusing these works in their production systems.

Early on, it was decided that the UAT should contain all the terms necessary to completely describe the primary topics found in the mainstream refereed literature of astronomy and astrophysics. As the discipline expands over time, so, too, will the thesaurus; for instance, additional terms describing exoplanets and gravitational waves will surely need to be incorporated into the Thesaurus within the next few years. In addition to deciding what should be included in the UAT, the boundaries of the Thesaurus have been intentionally delineated to define what must be excluded. Without a clearly articulated definition of scope, the task of maintaining a vocabulary can quickly get out of control as additional terms are incorporated. Concepts that are only tangentially connected to astronomy, or methods and techniques that are not specific to astronomy, are better described in external vocabularies, which can be referenced by, rather than incorporated into, the UAT. Similarly, existing taxonomies that detail astronomical objects, instruments, and facilities are best kept outside of the UAT and managed by the existing groups that are currently curating them.

The curatorial decisions limiting which concepts are represented in the UAT and which are kept out do not necessarily mean that descriptiveness needs to be compromised for the sake of expedience: other thesauri can provide the concepts needed to properly describe a multidisciplinary work that could not be fully described using an astronomy-focused thesaurus. One of the features of the Semantic Web (Berners-Lee et al. 2001) is the ability to interlink distributed concept schemes, and SKOS provides ways to formally map, reuse, or even delegate entire branches of a thesaurus to external knowledge systems.

Concept mapping provides a way to establish semantic similarity between a term in the UAT and an external, potentially broader, knowledge base. Using concept mapping allows a curator for a project such as WikiData (Vrandečić & Krötzsch 2014) to link concepts in Wikipedia to corresponding concepts in the UAT, stating, for instance, that the UAT concept for "G stars" is equivalent to the Wikipedia category "G-type star." As more resources are linked to WikiData, such curation allows for cross-vocabulary mapping, which will increase interoperability across knowledge systems and disciplines. This is the approach behind the Linked Data effort (Heath & Bizer 2011), which attempts to publish and interlink structured data on the web. While this is, in general, a good thing, one should be aware of the fact that any such mapping is subject to interpretation and approximation, which means that the result of these efforts may present new challenges to those attempting to leverage them.

Reusing a concept allows a system to "borrow" it from an external source via a simple referencing mechanism. A publisher could use this mechanism to say that the top-level UAT concepts represent the astronomy branch of a higher-level physics vocabulary, which could also include high-energy physics concepts to fully describe an interdisciplinary cosmology paper. Finally, the SKOS standard allows one to extend a thesaurus by delegating parts of it to outside knowledge systems rather than incorporating their contents. One useful application of this capability for the UAT could be the delegation of object types to the SIMBAD ontology (Derriere et al. 2010). This delegation works both ways, and allows a publisher such as IoP to simply adopt the UAT itself as the astronomy branch of its physics thesaurus. It should be noted that the choice of relying on external knowledge systems involves making certain compromises, the most notable of which is a delegation of authority, and therefore control, over part of the domain being covered.

The concepts found in the Astronomical Subject Keywords (ASK) were approved and adopted by the editors of major astronomy journals such as the Astronomical Journal, Astronomy & Astrophysics, Monthly Notices of the Royal Astronomical Society, the Publications of the Astronomical Society of the Pacific, and the Astrophysical Journal (including Letters and Supplements). The Keywords were originally introduced in 1973 by the Astrophysical Journal under the name of "Subject Headings" (AAS 1972) and were later adopted by the other journals. Revisions and updates to the Astronomical Subject Keywords were made by general agreement and consensus among the editors of the above journals every few years based on emailed suggestions. The last of these updates was in 2013, after which a general consensus emerged that these keywords were not as useful as originally envisioned.

Throughout these revisions, the Astronomical Subject Keywords remained a list of a relatively small number of concepts organized into broad categories. Furthermore, the ASK is a simple text list, utilizing none of the modern standards and formats to support machine readability and usability. Innovations in semantic markup and knowledge representation have brought about standards such as SKOS. Systems that follow these standards can leverage these relationships and use them to build organizational structures that benefit from the explicit links. For example, a system tasked to search for very specific leaf concepts at the bottom of the hierarchy, such as "UAT: Bailey type stars," might suggest that a user broaden their results by searching for the concept's parent, "UAT: RR Lyrae variable stars."

In order to facilitate a transition from the Astronomical Subject Keywords to the UAT and as a way to validate the completeness of the latter, we decided to attempt to map ASK terms to UAT concepts. For each term in the Astronomical Subject Keywords, we searched for similar terms found in the Unified Astronomy Thesaurus. To perform the mapping, we used well-known SKOS Mapping Properties¹³ to describe the relationship between ASK terms and UAT concepts, as defined in Table 1.

The analysis defined 557 relationships between terms in the Astronomical Subject Keywords and concepts in the Unified Astronomy Thesaurus: 187 exact matches, 9 close matches, 265 narrow matches, and 96 related matches. We also found that 62 terms in the ASK had no analogs to concepts in the UAT.

Overall, terms in the Astronomical Subject Keywords tend to describe broader concepts or multiple concepts, whereas the Unified Astronomy Thesaurus uses a single term to describe a single, specific concept. For example, the single term "ASK: meteorites, meteors, meteoroids" describes three separate UAT concepts ("UAT: Meteorites," "UAT: Meteors," and "UAT: Meteoroids"). Similarly, "ASK: accretion, accretion disk" describes both the process of accretion and the generic concept of an accretion disk, and several UAT concepts are a narrow match, such as "UAT: Accretion," "UAT: Galaxy accretion," and "UAT: Stellar accretion disk," among others. Further examples of relationship matches between ASK and UAT can be seen in Table 2.

The high number of close, partial, and narrow matches, along with the 96 related concept links between the ASK and the UAT, implies two things:

• 1.  
  The UAT has a lack of generalized concepts; and
• 2.  
  For the most part, terms in the ASK are included as concepts in the UAT.

Whereas the ASK has only top-level categories and second-level concepts, the UAT has 10 levels of depth. Likewise, the UAT has 1,843 concepts where the ASK has only 302 terms. It is apparent that the UAT delves into more topics, and covers them in greater detail, than does the ASK. The exception to this rule is that the ASK allows for users to include specifically named astronomical objects. For example, if an author wished to use the keyword ASK: comets: individual (..., ...) they would be expected to include the individually named comet in parentheses, i.e., "ASK: comets: individual (ShoemakerLevy 9)." In contrast to ASK, and in accordance with best practices for controlled vocabularies (ANSI/NISO Z39.19, 2010), the UAT does not allow for open-ended, on-demand user input and only includes a few specifically named examples.

The lack of open-ended and generalized concepts in the UAT is a feature rather than a bug. That being said, this comparison highlights 62 terms in the ASK that have no match of any kind with concepts in the UAT. For example, where the ASK includes concepts describing different facets of publishing (such as "ASK: editorials, notices" and "ASK: errata, addenda"), there are no corresponding terms in the UAT. However, they could easily fit into the existing UAT section on "Astronomical research." Similarly, the ASK covers modern topics on mathematical and computational methods for processing data and images much more thoroughly than the UAT, which includes only a few very general concepts on these topics. Whether these concepts warrant inclusion in the UAT or not will depend on their relevance to astronomy literature and the existence of external SKOS vocabularies that the UAT could link to that cover these concepts in greater depth.

Finally, though the UAT has many concepts about various physical processes, they tend to be specific and collocated with astronomical phenomena. In contrast, the ASK allows these concepts to stand alone so that they can be discussed in general rather than only as they relate to specific objects or events. This difference leads to an important practical distinction between the two systems: terms in the ASK may be ambiguous and their meaning can only be inferred when considered alongside other terms assigned to the same resource. Instead, the UAT uses precise language to define each concept so that its meaning does not depend on its context. As an example, consider the term "ASK: accretion, accretion disks" When considering this term alone, there is no way for a reader to know whether it refers to stellar, galactic, or Bondi accretion processes, which correspond to very different concepts in the UAT. While a human reader may rather quickly find the answer to the question, things are more complex for automated systems which will use this information to classify content and provide recommendations based on them. Since one of the goals of the UAT is to be machine actionable, the greater precision of its semantics is a significant advantage over the ASK.

With the imminent launch of the James Webb Space Telescope (JWST), the Space Telescope Science Institute (STScI) has begun the work of updating and preparing the Astronomer's Proposal Tool, which is used to write, validate, and submit proposals. A specific change under consideration for this work is to update the two keyword lists used to describe proposals and targets within the Proposal Tool. One keyword list, the Target Descriptions,¹⁴ has already been updated with a subset of UAT keywords and will be used to index JWST data going forward. The other list under consideration to be updated is the JWST Scientific Keywords,¹⁵ which is currently based on the keywords used to describe proposals for the Hubble Space Telescope.

The JWST Scientific Keywords (JSK) list currently stands at about 140 concepts arranged into 7 top-level categories and second-level keywords. MAST collaborators recently mapped the JSK terms to the UAT in order to investigate the potential of aligning the two. The analysis completed by STScI focused on the viability of using the UAT with the Astronomer's Proposal Tool. We found that nearly two-thirds of the JSK were well represented in the UAT, either as exact matches (∼31%) or as synonymous matches (∼34%), where the concepts had the same meaning but used a different term or syntax. The remaining JSK terms were split between concepts missing entirely in the UAT (approx. 14%), and concepts requiring further analysis to make a determination (∼21%). See Table 3 for an excerpt from this analysis.

The last group also contains JSK terms that described multiple concepts (e.g., "JSK: Bulges, Spheroids, and Ellipticals"), which could easily match multiple concepts within the UAT if they were split into their component parts (e.g., "UAT: Galaxy bulges," "UAT: Dwarf spheroidal galaxies," and "UAT: Elliptical galaxies"). In some cases, the compound JSK terms still had no matches even when split into separate concepts. Careful review of all the JSK terms with multiple matches in the UAT is recommended before making any further determinations of which terms might need updating in either the JSK or the UAT.

Much of the reason why additional review is required for concepts in the JSK is that the concepts do not have unique identifiers throughout the system, leading to semantic ambiguity. The concept "JSK: Astrometry" is found under the categories "JSK: Stellar Physics" and "JSK: Stellar Populations," but in both cases, the meaning of "JSK: Astrometry" is clearly the same. It could be said that Astrometry as found in the JSK terms simply has two parent concepts (also called a polyhierarchical relationship). However, the concept of "Dust" stands in stark contrast to this: "JSK: Dust" is found under "JSK: Stellar Physics," "JSK: Galaxies and the IGM," and "JSK: Stellar Populations"—three different parent concepts! To compound the issue, the concept of "Dust" clearly has different meanings relative to "Stellar Physics" versus "Galaxies/IGM." In the former case ("JSK: Stellar Physics, Dust"), it is equivalent to "UAT: Interstellar dust," but in the latter case ("JSK: Galaxies and the IGM, Dust"), it is clear that "UAT: Intergalactic dust clouds" is a more appropriate analog. As for the third case ("JSK: Stellar Populations, Dust"), it is unclear whether "Dust" shares a meaning with "UAT: Intergalactic dust clouds" or with "JSK: Interstellar dust."

Just as with the ASK, this analysis uncovered that there are small but significant gaps in UAT content, even though the majority of JSK terms are well-represented. This analysis demonstrated that the UAT lacks concepts regarding observational techniques, modern computational methods for analysis, and niche concepts relating to galaxies, galactic processes, and black holes. At the same time, the comparison showed shortcomings in the current JSK system, where ambiguous terms are used to describe semantically different concepts. While the context in which the terms appear explains the difference in their meaning, it is best practice not to have a vocabulary reuse the same term for different concepts.