Citation recommendation can quickly find supplementary references from a large number of academic papers, making citations more comprehensive and accurate. Current methods often use the reference sequence at the end of a paper to recommend citations, neglecting the actual order and repeated citations within the body text of the paper, both of which are referred to as the “full-text citation sequence.” This leads to the loss of context information around citations. Moreover, current research has not yet fully explored the effects of different full-text citation sequences on citation recommendation. Therefore, this paper proposes a framework for citation recommendation based on semantic embedding of full-text citation sequences. First, we construct multiple types of citation sequences using different strategies to citation order and repeated citations extracted from the full text of the paper. Second, we apply different deep learning methods to generate semantic embedding for each citation using its rich context information. Finally, we compute the semantic similarity between citation embeddings for citation recommendation and evaluate the performance through link prediction. The results in the field of artificial intelligence show that each method achieves the best performance when using citation sequence with repetition (CSR) for citation recommendations. Compared to citation sequences extracted from reference (CSER) and citation sequences without repetition (CSWR), the AUC increased by 15.3
Patent infringement analysis (PIA) is very important to ensure the compliance of product design. However, prevailing methods predominantly rely on post-hoc structural comparisons, which not only contradict the function-oriented paradigm of conceptual design but also prove inadequate for large-scale infringement detection across extensive functional knowledge repositories. To address these limitations, this study proposes an intelligent patent infringement analysis framework based on the patent-function-principle-structure (PFPS) network. First, patent knowledge is extracted and annotated with infringement risk labels according to claims to construct the PFPS network. Second, a morphological matrix linking functional requirements to functional knowledge nodes is established; guided by the all-elements rule, hash functions are employed for preliminary screening of node combinations, followed by a hybrid similarity metric to identify potential equivalent infringement risks. Third, criteria derived from the reverse doctrine of equivalents are integrated to regulate the function-principle-structure mapping process, thereby enabling systematic patent design-around. Finally, the feasibility of the proposed method is validated through a case study on an irregular workpiece transfer device. Results demonstrate that the method effectively makes up for the shortcomings of infringement analysis and design-around within patent knowledge-driven conceptual design.
Scientific academies play long-standing roles in scholarly recognition, science advice, public communication, and knowledge dissemination, yet comparable evidence about these organizations remains fragmented across open registries, institutional websites, and bibliographic authority systems. This paper introduces an open catalogue of global scientific academies that integrates three linked evidence layers: organizational identity, public-facing web presence, and academy-linked scholarly communication infrastructure, operationalized through journal records. The identity layer represents academies as persistent organizational entities and connects them to open registries and knowledge bases through stable identifiers and validated cross-system mappings. The web layer uses navigation menus, sitemap-derived URL hierarchies, Formal Concept Analysis, and graph pruning to construct a shared taxonomy of academy website functions, supporting comparative analysis of web-structural patterns, imitation, and differentiation. The bibliographic layer links academies to Zeitschriftendatenbank records through curated corporate-body authority mappings and uses journal title histories to examine the right-censored lifecycles of observable academy-linked journals. The analysis shows that registry identifiers provide uneven but informative evidence for organizational identity anchoring, and that academy websites differ in the functional structures through which governance, communication, cooperation, and knowledge resources are made visible. Academy-linked journal records further document historically layered journal title trajectories, with uneven observability and persistence across institutional contexts. By treating organizational records, website structures, and journal title histories as distinct but linkable evidence, the catalogue provides a reusable data foundation for comparative research on scientific academies across heterogeneous open research information systems.
While existing scientometric research has examined how individual network position and team composition affect citation-based performance, less is known about how the structural configuration of socio-semantic networks relates to collective publication output. We propose that publication output is associated with configurations that foster mutual visibility between research topics (memes) and author communities. Drawing on mutualistic network theory from ecology, we hypothesise that lower bipartite modularity (H1) and higher nestedness (H2) are associated with higher publication output. We analyse 6,406 publications across three ecosystem approaches to health (EcoHealth, One Health, Planetary Health) over 18 approach × period observations (2004–2023), linking author communities to 20 research topics identified by Latent Dirichlet Allocation. Our analyses show an empirical asymmetry: bipartite modularity is negatively and robustly associated with publication output (Spearman ρ = −0.79, p < 0.001; robust across four community detection algorithms and under leave-one-out by approach), whereas weighted nestedness shows a positive but non-significant association (ρ = +0.39, p = 0.11), in a direction consistent with H2 but without statistical support at the cell level. The modularity result supports the H1 component of the framework: low modularity — author communities touching multiple topics — fosters mutual visibility between authors and themes. Given the observational design, we report these as field-level patterns, not causal mechanisms. The non-significant association for H2, combined with marked algorithmic variability, invites reconsidering how nestedness is operationalised in socio-semantic contexts. By integrating social and semantic dimensions within a mutualistic framework, this approach complements existing scientometric tools.
