Semantic Artifact Reuse

• Semantic Artifact Reuse is defined as the principled retrieval, adaptation, and integration of self-describing artifacts using formal semantic models.
• It employs layered architectures, feature trees, and graph matching algorithms to ensure efficient, context-aware, and trustable reuse across heterogeneous domains.
• Empirical studies show significant gains in recommendation precision and interoperability, validating its impact in software engineering and data sharing.

Semantic artifact reuse denotes the principled retrieval, adaptation, and integration of information or knowledge artifacts in a manner that preserves and leverages their meaning, structure, and domain semantics, rather than simply their syntactic or physical representation. This paradigm is central to contemporary practices in software engineering, scientific data sharing, ontology management, verification systems, and knowledge curation, where artifacts—ranging from code components, ontologies, metadata schemas, proofs, fragments of documents, and sensemaking summaries—are increasingly designed to be semantically self-describing, modular, and interoperable. The objective is to facilitate efficient, trustable, and context-aware utilization of established intellectual assets across heterogeneous environments and evolving workflows.

1. Formal Models and Definitions

Semantic artifact reuse is grounded in explicit, formal representation of artifacts’ structure and semantics. In source code repositories, reusable artifacts are often characterized by functional descriptions and operational relationships, and organized via feature hierarchies that support requirements-driven selection (Jin et al., 4 Jun 2025). In scientific domains, artifacts such as data products, calibration files, and publications are encapsulated within semantic web formats and annotated with persistent URIs and formal relationships through models like OAI-ORE (0906.2549).

Advanced formal models—such as attributed graphs, as used in learning-infused verification frameworks (Beg et al., 2 Feb 2026), or the General Fragment Model (GFM) (Fiorini et al., 2019)—enable fine-grained description and anchoring of fragments within heterogeneous information objects. In GFM, an indexer $f_o$ maps parameterized tuple tokens to specific fragments $s$ of an artifact $o$, and anchors serve as precise referents for semantic annotation and cross-modal linkage.

Semantic web ontologies formalize reusable specifications by mapping domain concepts and operational properties to OWL classes, object properties, and axioms (e.g., $$\exists \text{hasOperation}.\top \sqsubseteq \text{Artifact}$$, $\text{Usage} \sqsubseteq \text{A%%%%4%%%%A-Core:Artifact}$) (Daoud, 2020).

2. Architectures and Representation Schemes

Architectural frameworks for semantic artifact reuse are typically layered to support both ingestion and discovery. In OntoPortal-Astro (Cecconi et al., 17 Apr 2025), semantic artifact catalogues are built with ingest layers for harvesting OWL, SKOS, and XSD artifacts, RDF triple stores for persistence and query, and FAIR-compliant metadata schemas (MOD) for provenance, versioning, and licensing. Mapping registries align terms across ontologies using both automated lexical similarity and manual curation.

Tree-based abstractions such as multi-level feature trees (Jin et al., 4 Jun 2025) and TreeRec’s hierarchical semantic trees (Jin et al., 23 Nov 2025) provide scalable, navigable indices, organizing artifacts by semantically coherent features. In formal verification, artifact graphs encode entities, logical formulas, and refinement relations, supporting both symbolic matching and semantic enrichment via LLM-driven embeddings (Beg et al., 2 Feb 2026).

Fragment models permit composition and anchoring of subparts (characters, pixels, intervals, cells) across modalities, while reuse frameworks like Strata (Liu et al., 2021) capture provenance signals and behavioral metadata from the author’s interaction history.

3. Reuse Methodologies and Algorithms

Semantic artifact reuse employs a range of methodologies from graph matching to LLM-driven abstraction:

• Feature tree construction: Algorithms cluster artifacts using text or embedding similarity, then summarize shared features at each abstraction level via LLM prompts, recursively constructing semantic hierarchies. These trees support navigation and recommendation by matching requirements to feature summaries and traversing towards leaf artifact candidates (Jin et al., 4 Jun 2025, Jin et al., 23 Nov 2025).
• Graph-based semantic matching: Artifacts are parsed into typed attributed graphs, nodes are enriched with semantic vector embeddings, and alignment is performed through hybrid structural and semantic metrics ($$d(G_1, G_2) = \alpha d_{\text{struc}} + \beta d_{\text{embed}}$$). Symbolic constraints and refinement obligations enforce soundness of reuse (Beg et al., 2 Feb 2026).
• Metric learning: Integrated retrieval systems learn latent projections ($U$, $V$) aligning code features and text features into shared semantic spaces, with loss terms enforcing content-based regularization and graph smoothness (Wu et al., 2014).
• Ontology modularization and import: Ontology engineering best practices decouple logical interfaces from protocol bindings, employ modularization, and maximize domain vocabulary reuse via formal imports and alignment axioms (Daoud, 2020, Cecconi et al., 17 Apr 2025).
• Fragment anchoring and composition: The GFM framework enables systematic instantiation of anchors and indexers for fragment specification, supporting cross-modal semantic linkage (Fiorini et al., 2019).

4. Use Cases and Empirical Impact

Semantic artifact reuse is demonstrated across multiple domains:

• Scientific Data Interoperability: OntoPortal-Astro facilitates interdisciplinary astronomy research by cataloguing and aligning ontologies, vocabularies, and metadata schemas, allowing concepts like “exoplanet” to be mapped and reused in annotation and workflow pipelines (Cecconi et al., 17 Apr 2025).
• Multi-Agent Systems: Semantic transformation of device descriptions enables agents in IoT environments to reuse functional artifact interfaces, with plans remaining invariant under device replacement (Daoud, 2020).
• Software Artifact Recommendation: Hierarchical semantic trees (FTBuilder, TreeRec) cut artifact selection time and improve LLM-powered recommendations by up to 235% precision and 26% efficiency over official baselines (Jin et al., 4 Jun 2025, Jin et al., 23 Nov 2025).
• Formal Verification: Artifact graph matching and adaptation enable reuse of invariants and contract proofs, with soundness enforced by semantic alignment and refinement checks (Beg et al., 2 Feb 2026).
• Fragmented Document Linkage: GFM’s formal language supports semantic cross-linking and querying of data fragments within geological, multimedia, and tabular domains (Fiorini et al., 2019).
• Knowledge Evaluation and Sensemaking: Strata’s signal-driven visualization of trust, context, and thoroughness improves users’ reuse accuracy, rationale quality, and speed (+32.5% faster decisions, +75.6% more valid rationales) (Liu et al., 2021).

5. Semantic Alignment, Mapping, and Soundness

Alignment of semantic artifacts requires both automated and manual mechanisms:

• Lexical and Structural Mapping: OntoPortal-Astro utilizes string similarity metrics ($\text{sim}(l_1, l_2) = 2|LCS|/(|l_1|+|l_2|)$), supports formal SKOS and OWL equivalence axioms, and includes mapping registries for clean separation of curated alignments (Cecconi et al., 17 Apr 2025).
• Embedding-based Similarity: LLM-based embeddings and pooling operate as inputs to cosine similarity evaluations in both tree-based search and graph matching (Jin et al., 23 Nov 2025, Beg et al., 2 Feb 2026).
• Soundness and Proof Obligations: Formal reuse is contingent on meeting refinement and implication obligations; transfer of predicates and transitions must preserve logical semantics ($\phi_{n_2} \models \phi_{n_1}$, $\mathrm{Refine}(G_1, G_2)$) (Beg et al., 2 Feb 2026).
• Modularization Guidelines: Decoupling logical artifact interfaces from implementation details ensures functional equivalence and discoverability across replacement and runtime configuration (Daoud, 2020).

6. Challenges, Best Practices, and Future Directions

Semantic artifact reuse faces technical and community-lifecycle challenges:

• Heterogeneous Ecosystems: In astronomy and open-source environments, fragmented artifact registries, varied formats, and uneven adoption of semantic standards hamper cross-domain reuse (Cecconi et al., 17 Apr 2025, Jin et al., 23 Nov 2025).
• Scalability and Latency: Brute-force LLM scorers suffer prohibitive inference costs (e.g., 600 s per query on large corpora), mitigated by tree-guided pruning that reduces search to $O(\log N)$ (Jin et al., 23 Nov 2025).
• Metadata and Provenance: Robust reuse depends on comprehensive metadata, provenance trail, and versioning in line with FAIR principles (Cecconi et al., 17 Apr 2025).
• Ontology Engineering: Best practices mandate modular ontological design, vocabulary reuse, and publication of linked usage/context graphs to enable agent reasoning and semantic discovery (Daoud, 2020).
• Tooling and Federation: Authoring tools, registry federation, and domain working-group engagement are essential to scale reuse and ensure sustainability (Cecconi et al., 17 Apr 2025, 0906.2549).
• Soundness vs. Coverage: For formal verification, tuning matching hyperparameters and advancing rule-learning for adaptation balances extensibility against guaranteed correctness (Beg et al., 2 Feb 2026).

Ongoing research seeks to expand datasets, formalize adaptation across heterogeneous logics, and integrate AI-assisted semantic mapping (e.g., OPAL, OSCARS projects) (Cecconi et al., 17 Apr 2025). Persistent challenges include automated aggregation, traceability, stability of semantic embeddings, and certification for critical infrastructure.

7. Conclusions and Research Outlook

Semantic artifact reuse is a foundational construct in modern knowledge-centric systems, underpinning the interoperability, efficiency, and quality of multi-agent, software, scientific, and formal verification environments. State-of-the-art methodologies formalize semantic relationships via ontologies, attributed graphs, feature trees, and fragment models, while recommender and retrieval systems leverage hybrid symbolic and learning-based alignment to promote intent-preserving adaptation. Empirical evidence from multiple domains indicates substantial gains in recommendation accuracy, interoperability, knowledge evaluation, and workflow automation. Continued refinement of best practices, federation architectures, and soundness criteria will be necessary to realize semantically robust, FAIR-compliant, and scalable reuse across the computational sciences and engineering.