Source-linked AI summary
Extracting Semantics from Cattle Reporting Categories for Data Interoperability and Findability
Kassy Raymond, Andrew Hamilton-Wright, Deborah Stacey
TL;DR
Livestock data are fragmented, while metadata-centered interoperability can require substantial effort and standards that may be unavailable or unsuitable. This paper extracts and normalizes reporting terms to build a bottom-up vocabulary from cattle data; it captures granularity absent from AGROVOC, while the study’s national-source age categories remain limited in scope.
Problem
Livestock datasets are fragmented, and metadata may be absent or require substantial curation against standards that are unavailable or poorly suited to stakeholders.
Source: §1 · p. 2Method
The study extracts and normalizes reporting terms to build a bottom-up vocabulary from livestock data, using cattle as its pilot.
Source: §1 · p. 3, §2.6 · p. 7Results
The vocabulary captures age, sex, and production granularity in cattle reporting terms that is not present in AGROVOC.
Source: §3.3 · p. 17Takeaways & Limitations
The approach offers a faster, less resource-intensive pathway to improving cattle data findability and interoperability before metadata or standards are created, but does not replace contextual metadata needed for dataset reuse.
Source: §3.3 · p. 17Because Ethiopia was the only national source included, additional national reporting systems may reveal age categories not captured in this extensible vocabulary.
Source: §3.1.2 · p. 14
Abstract
from arXiv · showhide
Livestock population data disaggregated by age, sex, and production are important inputs to calculations and models that inform our understanding of global health, yet these data are fragmented across disparate sources. Bridging data siloes to improve the findability of data requires interoperability. Conventional approaches to improving the findability and interoperability of data include indexing standardized metadata. However, creating metadata is time and resource-intensive and is often difficult in domains such as livestock, which lack standards that address the needs of broad user groups. When metadata exist, they typically need to be standardized against a pre-existing vocabulary, ontology, or thesaurus, requiring a technique known as `crosswalking'. To overcome issues in the absence of metadata, the lack of standards, and the resource-intensive solutions that currently exist, this study uses a bottom-up approach. By leveraging real-world reporting categories in datasets, the composition and semantics of terms already present in the data were extracted and analyzed. Using cattle data as a pilot, we find the age, sex, and production modifiers present across cattle terms from five datasets from four data sources capture granularity not present in AGROVOC, the largest agricultural vocabulary in the world. We discuss how the composition and semantics of these terms can be used to improve the interoperability and findability of data without first requiring metadata to be generated or standards to be created. Rather than forcing datasets to conform to an existing vocabulary, this approach uses the semantics embedded in terms already present in datasets, allowing systems to make data more discoverable and interoperable while maintaining culturally and dataset-specific terminology.
1. Introduction
Livestock data are siloed despite their value for health and agricultural analyses, while metadata-centered discovery depends on resources and standards that may be missing or unsuitable. The paper proposes extracting and normalizing reporting terms to build a vocabulary from livestock data, capturing semantics already embedded in reporting systems.
1. Introduction
- Livestock census and survey data support emissions, antimicrobial-use, economic, yield, and biomass analyses, yet siloed repositories limit their FAIR findability and reuse.These challenges matter as the sector seeks to use AI-ready data.Source: §1 · p. 2
- Semantic interoperability is needed to make disaggregated datasets from national and intergovernmental portals collectively findable and retrievable.Conventional approaches rely on metadata, which can be indexed and searched or crosswalked to common standards and KOS concepts.Source: §1 · p. 2
- Metadata may be incomplete, absent, or governed by unavailable, underdeveloped, contested, or poorly suited standards, complicating discovery and reuse.Addressing these gaps through conventional practices requires human curation, financial resources, and domain expertise.Source: §1 · pp. 2–3
- The proposed approach extracts and normalizes dataset terms to describe livestock data without requiring prior metadata creation or conformity to a predefined standard.It uses semantics already embedded in reporting systems to identify differences in granularity, regional terminology, and semantic coverage, and to support later alignment and mapping.Source: §1 · pp. 3–4
- The study investigates whether terms in heterogeneous livestock datasets can form a bottom-up vocabulary that captures real-world classification granularity and variation.Its questions cover cattle terminology and modifiers, and compare the resulting vocabulary’s granularity and coverage with AGROVOC.Source: §1 · p. 3
2. Methods
The methods acquire cattle population data, harmonize its structure, extract reporting and AGROVOC terms, and normalize both sets for comparison and mapping.
2. Methods
- The pipeline acquires livestock data, harmonizes formats, extracts dataset and AGROVOC terms, normalizes them, and maps dataset terms to the KOS.The study’s scripts were planned for public availability upon publication.Source: §2 · p. 4
2.1. Data Acquisition
- Four sources supplied livestock population data: FAOSTAT, ESS, Eurostat, and WOAH, including two FAO datasets.The sources represent international organizations and a national statistics agency.Source: §2.1 · p. 4
2.2. Technical Interoperability
- Acquisition methods were tailored to machine readability: APIs enabled scripted access, while portal downloads, OCR, and scraping handled sources without APIs.ESS data came from PDF reports.Source: §2.2 · p. 4
2.3. Structural Interoperability
- Dataset-specific scripts harmonized rows, columns, and headers so differently formatted data could be processed by an automatic term-extraction pipeline.Source: §2.3 · p. 7
2.4. Term Extraction
- The extraction pipeline retained cattle terms and excluded other livestock terms, which may require different normalization rules.Source: §2.4 · p. 7
2.5. Reference Vocabulary Extraction
- AGROVOC served as the agricultural reference vocabulary; its English cattle concept and child terms were extracted through recursive traversal of the SKOS hierarchy.The term list was manually reviewed for relevance to cattle.Source: §2.5 · p. 7
2.6. Normalization
- Normalization lowercased terms, removed whitespace and punctuation, and used TextBlob to singularize plurals after testing found stemming and lemmatization inconsistent for irregular forms.Age expressions were standardized before stopword removal; age ranges were normalized to labels, months converted to years, negation cues mapped to a NOT prefix, and selected stopwords removed.Source: §2.6 · pp. 8–9
3. Results and Discussion
Real-world cattle reporting categories encode reusable root, age, sex, and production semantics. Their comparison with AGROVOC reveals granularity gaps, while the study argues that extracted terms can support findability before metadata is generated.
3.1. Terminology and Modifiers that Characterize Real-World Cattle Root Terms
- The 63 cattle terms combine root terms with age, sex, and production modifiers; their composition reflects institutional priorities and can serve as descriptive metadata for search.Table 2 presents term components, and Figure 4 shows their hierarchical composition.Source: §3.1 · pp. 10, 12, Table 2 · p. 10, Figure 4 · p. 11
3.1.1. Sex and Production Modifiers.
- Sex may be explicit or implicit in a root term, while production cues vary in reliability: dairy can imply female, but beef alone cannot reliably determine sex.The paper recommends flagging terms for manual intervention when systems cannot reliably extract sex information.Source: §3.1.1 · p. 12
3.1.2. Age Modifiers.
- Four age rules normalize ranges, discrete ages, minimum bounds, and maximum bounds into common labels while preserving reported age information.The examples include converting “10 years and older” to “age over 10 year.”Source: §3.1.2 · pp. 12–13, Table 3 · p. 13
- Normalized age components can support age-based retrieval and mappings to modelling categories, but the vocabulary is extensible rather than exhaustive because only Ethiopia represented national sources.Maintaining provenance can help analysts compare mappings and assess model-output sensitivity to alternative age classifications.Source: §3.1.2 · pp. 13–14
3.2. Comparison to AGROVOC
- AGROVOC maps many specific reporting categories to the broader concept cattle, leaving age, sex, and production distinctions unrepresented and risking information loss.Some root terms and explicitly expressed production terms match directly, but other combinations match only to cattle.Source: §3.2 · pp. 14–16, Table 4 · p. 15
3.3. Conclusion
- Bottom-up extraction can improve cattle-data findability and interoperability before metadata or standards are available, but it does not replace contextual metadata needed for dataset reuse.The conclusion also identifies this approach as a way to address legacy datasets where top-down standardization may lose information.Source: §3.3 · p. 17