This study uses the academic literature and patent document corpus system to process structured text data across knowledge domains. Its purpose is to systematically display the research evolution and technical transformation trajectory of the exosome-hydrogel system in the field of bone regeneration. The data covers from 2016 to 2025, integrating scientific research achievements (in academic aspects) and industrial technology layout (in patent aspects), and realizing cross-source semantic linking between “research” and “application”.

Academic papers were retrieved from major international journal databases based on titles, abstracts, or keywords containing the core terms “exosome/extracellular vesicle,” “hydrogel,” and “bone regeneration/bone repair.” Patent data were collected from open intellectual property databases using identical filtering criteria. After importation, all documents were uniformly formatted to ensure consistency across key fields, including unique identifier, title, abstract, and year of publication or disclosure.

To ensure the consistency and comparability of time, convert the year field to an integer and unify to the same time series. Academic data mainly shows the research focus and mechanical verification, while patent data reflects the scope of the claims of applied technology and the formalization of language. he two corpora jointly provide a panoramic “discovery-verification-translation” knowledge framework (Porter and Cunningham, 2004).

Before topic modeling and before entity analysis, the corpus needs to undergo systematic preprocessing and quality control. Concatenate the titles and abstracts of the documents so that the context can be more coherent. Delete texts with a length of less than 30 characters, which can reduce semantic noise and screen out ambiguous samples. Replace missing values or null values with empty strings to prevent the model from interrupting.

To ensure consistency of vocabulary at the language level the text is lowercased and punctuation is removed stopwords are filtered using standard english lists and domain specific word lists with scientific terms such as “gelatin methacryloyl (GelMA)”, “MSC”, “osteogenesis” retained the standardization process follows traditional natural language processing pipelines such as tokenization stopword management and text normalization (Loper and Bird, 2002).

In order to avoid deviation in time analysis, the annual statistical data will adopt the full year (up to 2025) to prevent false decline due to incomplete data. Then the text is constructed into a standard and quality-controlled corpus for semantic models and entity analysis.

To find the differences in semantic structures between academic texts and patent texts, this paper uses latent Dirichlet allocation (LDA) for topic modeling. The text is vectorized using a bag-of-words model that includes co-occurrences of unigrams and bigrams to enhance the recognition ability of compound terms. To avoid the dilution of topics caused by terms being too frequent or too rare, an upper threshold (≤90% document frequency) and a lower threshold (≥2 occurrences) are set to focus on semantically useful terms. The number of topics was determined based on coherence score and corpus size to ensure semantic stability.

The LDA model produced a document–topic distribution matrix assigning a set of topic weights to each record. The number of topics was adapted to corpus size: 10 topics for >200 documents, eight topics for 80–200 documents, and 5 topics for <80 documents. Each topic included the top 12 weighted keywords for interpretability. Model training used symmetric Dirichlet priors (α = 0.1, β = 0.01) to enhance sparsity and interpretability, consistent with classical applications of LDA in scientific text mining.

To realize the visual semantic pattern, the document-topic matrix is projected into a two-dimensional space by using the method of truncated singular value decomposition (SVD). The first and second principal components (dimension 1, dimension 2) are used as the main and secondary semantic axes, in the same way as latent semantic analysis (Deerwester et al., 1990).

After semantic modeling, entity and relationship extraction form a structured knowledge graph in this field. Using NLP technology to identify/standardize entities, and processing hydrogel pads, exosome-derived cells, and experimental endpoints/biomarkers. Standardize synonyms and names to simplify and remove lexical fragments, for example, merge “Gelatin methacrylate” and “GelMA”, and merge “microcomputed tomography” and “micro-CT.” This is in line with the already constructed biomedical NLP framework (Neumann et al., 2019).

Integrate the extracted entities into annual series to grasp long-term trends. Among hydrogels, GelMA, collagen, chitosan, hyaluronic acid (HA), and polyethylene glycol are the main materials. MSC-derived exosomes still rank first among source cells, and bone marrow–derived mesenchymal stem cell (BMSC), adipose-derived stem cell (ADSC), and immune cells are gradually increasing. In terms of end-point indicators, osteogenic markers (ALP, RUNX2, OCN) and angiogenesis markers (VEGF, CD31) are the most common.

Extracting relations uses a mixed method, combining dependency parsing and lexical-syntactic pattern matching. Generating verb-centered relation triples, with document types (paper and patent) and years, to assist downstream hypothesis identification and evidence quantification. Dependency parsing and relation extraction follow the Stanford natural language processing toolchain, such as word segmentation, part-of-speech tagging, syntactic dependency parsing (Manning et al., 2014). This transforms narrative statements into computable knowledge structures and descriptive evidence into structured data. Typical dependency-based relation extraction patterns were used to identify triplets such as “hydrogel–promotes–osteogenesis” and “exosome–enhances–angiogenesis,” enabling conversion of descriptive statements into computable knowledge structures.

Knowledge extraction after semantic modeling, and analyzing the macro-evolution of papers and patents.

The annual number of paper publications and the number of patent disclosures form approximately parallel time-series curves. The form and synchronization of the curves reflect the growth rate of research and application, and can evaluate the time-coupling relationship between scientific exploration and technological deployment.

The frequency of key entities in the annual period is tracked by entity-level analysis, and trend lines are drawn for the top 12 high-frequency elements within the categories of matrices, cellular sources, and endpoint categories. These curves show the changes of research focuses over time.

Build a matrix to grasp the pattern of experimental configurations. Rows are hydrogel pads, columns are functional endpoints, and cells are co-occurrence frequencies. Heatmaps visualize high-density clusters as experimental paradigms, and sparse areas are areas of experiments or emerging research, which is the same as the co-word analysis classification of scientific knowledge graphs (Van Eck and Waltman, 2010).

Finding innovative directions with strong academic evidence but low industry adoption requires cross-field comparison at three levels. Sort the relationships not appearing in patents in academic literature according to the citation frequency to obtain the attribute of high evidence and unpatented. Typical examples are “hydrogel-encapsulated exosomes,” “GelMA-enhanced osteogenesis,” “exosome-enhanced angiogenesis.” The differences between academic knowledge and patent claims have long been studied in scientometrics and patentology.

Decompose the results, and decompose into mature themes and emerging themes. For each relationship, calculate and compare the total number of evidences and the number of evidences in the past 3 years. The relationship where both dimensions are at a high level is a sustained hot spot (stable trajectory), and the relationship where the total number of evidences is at a medium level but the recent growth is relatively fast is an emerging frontier. Measure the priority of research focus through two-dimensional sorting, and divide into established consensus areas and new innovative frontiers.

This framework transforms traditional qualitative research into quantitative analysis of knowledge prioritization, and also identifies innovation nodes of “science and mathematics fields but insufficient institutional support.” It provides data-driven guidance for future research directions, experimental verification patent strategies. Finally, this study constructs a traceable workflow, including corpus modeling, relationship identification and evidence ranking, and realizes the closed-loop connection between academic inquiry and technological transformation (Jaffe, 2002). This work is entirely data-driven, integrating information from academic publications and patent records.

To demonstrate the research achievements and technological transformation dynamics of exosome-hydrogel systems in the past decade, an annual aggregation analysis of the literature and patents published from 2016 to 2025 is carried out (Figure 2).

The result shows a “bimodal pattern,” with stable academic growth coexisting with concentrated patent expansion, reflecting a phased shift from theoretical accumulation to application transformation. In terms of academia, since 2016, the number of papers has steadily increased, growing from early exploratory research to more than 40 papers in 2025, showing a continuous upward trend. This continuous growth indicates that the understanding of the exosome-hydrogel system in regenerative medicine in the field of scientific communication is continuously deepening, marking the shift from conceptual exploration to mechanism elucidation and experimental verification.

Early research focused on the isolation, purification, and physical and chemical analysis of exosomes. After 2019, research shifted to carrying out work related to vector construction and functional verification. For example, exosomes from mesenchymal stem cells were applied in hydrogels for bone regeneration. After 2022, matters related to exosomes in interdisciplinary areas such as immunomodulation-tissue repair began to emerge, which indicates that the research scope has expanded in the direction of multi-mechanism integration.

In 2021, there was a significant increase in industrial patent output, which increased by approximately two times in 3 years and remained at a high level from 2023 to 2025. The rapid growth reflects the maturity of technology and the increase in industrial investment. The time of some patents and academic achievements is consistent, which indicates that there is a mode of co-evolutionary research - patent between research institutions and enterprises. This synchronism is relatively rare in the field of biomaterial research, which shows that the exosome - hydrogel system is feasible and its market potential is widely recognized.

The time trajectories of papers and patents are positively correlated. Continuous academic output lays the foundation for technology transfer, and active patent applications promote the in-depth development of research focuses. Essentially, a two-way coupling between academia and industry has been formed: academia takes the lead in mechanical discoveries and method innovations, and industry is in a leading position in technical standards and intellectual property protection. This cyclic process “scientific exploration → engineering verification → industrial application” is the driving force behind the research progress of exosome hydrogels.

Using LDA topic model and truncated SVD dimensionality reduction visualization to compare academic resources and patent layouts in the merged corpus (Figure 3). This method can intuitively explain the differences in narrative patterns, semantic concentration, and technical diversity between the two fields.

After dimensionality reduction, it is observed that academic samples are relatively closely clustered within the semantic space, while patents are in a dispersed state. This difference reflects that there are differences in narrative logic: scholars, based on mechanism, attach importance to causal chains and experimental verification; patents use non-standard claim terms to carry out extensive technical coverage. Academic texts have high semantic density and focus on content, while patents have a relatively wide semantic range and strong universality.

Further topic statistics indicate systematic divergence in research emphasis (Table 1). On the academic side, Topic 2 and Topic 3 dominate, with mean topic weights of 0.3870 and 0.4935, respectively—significantly higher than 0.0720 and 0.0804 on the patent side. Keywords for these topics include “MSC,” “hydrogel,” “osteogenesis,” “bone regeneration,” “scaffold,” and “exosome,” forming the core narrative of academic research. This structure indicates that academic discourse is semantically convergent, focusing on the mechanistic core of exosome–hydrogel systems in bone regeneration, reflecting a mature research consensus.

In contrast, the patent domain shows a dispersed thematic structure. Topics 4, 6, 7, 8, 9, and 10 are patent-dominant, featuring keywords such as “treatment,” “composition,” “system,” “method,” “device,” and “embodiment”—all characteristic of claim-oriented expressions. This pattern reflects the horizontal expansion strategy of patents, which prioritize breadth of coverage and claim protection over mechanistic specificity. Frequent patent keywords such as “broad therapeutic application,” “system and process design,” and “composition formulation” exemplify this institutionalized generalization.

In conclusion, academic research is in a state of vertical deepening (focusing on refinement), and patents are in a state of horizontal expansion (broadening the concept boundaries). The two are not in an opposing relationship but complementary stages of knowledge transformation: academic research endows mechanisms with depth, and patents transform mechanisms into technically feasible forms that can be popularized and protected.

To clarify the long-term evolution of the exosome-hydrogel system, the annual trajectories of key research entities are analyzed from three dimensions: exosome-derived cells, hydrogel matrices, and evaluation endpoints or biomarkers (Figure 4). They jointly constitute a three-layer framework of biological input, matrix carrier, and functional output.

The hydrogel material appears together with biological endpoints to demonstrate the structural relationship between the material and biological results (Figure 5). The results show an obvious pattern with a higher center and lower periphery, indicating that there are some stable experimental configurations. The high - density regions match the GelMA or collagen containing osteogenic markers (ALP, RUNX2, OCN), forming a standard and repeatable model of exosome-hydrogel co-culture, which becomes the benchmark for related research.

Conversely, emerging explorations exist in the low-density peripheral region. Combinations like GelMA-VEGF, HA-CD31, chitosan-TNF-α/IL-10 have low frequencies but relatively fast development speeds, suggesting the potential to define a new wave of research paradigms. These co-occurrence patterns reveal stable experimental configurations and emerging functional relationships, which collectively provide the empirical foundation for understanding subsequent translational trends.

Building upon the material–outcome structure observed above, we next examined how these scientific findings have been translated into patented applications over time, and also displayed them with a scatter plot (Figure 6). Most data points are above the diagonal (y = x), which indicates a trend that research occurs earlier than patents. For example, classic cases such as “MSC exosome-GelMA composite material” and “exosomes promote bone repair” first appeared in academic literature around 2018–2019, but most of them obtained patents after 2021, which reflects that the expression at the institutional level is obviously lagging behind.

Points below the diagonal show the phenomenon of patent priority, especially for widely applied situations. Early applications often come from industry or academic-industry alliances, which reflects the situation of foresight and risk tolerance, meaning that the exosome-hydrogel technology is regarded as having commercial prospects before obtaining sufficient academic verification.

A small part near the diagonal points has synchronous development, which is a typical industry-university-research cooperation model. In order to realize the parallelism of paper publication and patent application, this model improves the transformation efficiency and also reflects the maturity of the organization.

Overall, a progression from unidirectional lag to multi-modal synchronization is observed, encompassing three coexisting patterns.

Research-led, patent-following (knowledge-driven translation)

These dynamics reflect a hierarchical and interactive innovative ecosystem, seeking balance between scientific depth and transformation speed, thereby enhancing the continuity from research to application in the exosome-hydrogel system.

To find innovative opportunities that have been academically verified but not fully reflected in business, we compare the semantic triples (head entity, relation, tail entity) of the paper corpus and the patent corpus. Only selecting the relations that appear in highly cited academic literature, several “high-evidence unpatented” propositions are found (Figure 7).

There are groups that appear with three main mechanisms.

The functional association of structures, such as “hydrogel encapsulating exosomes” and “hydrogel enabling exosome release,” highlights the mechanisms of spatial distribution and related kinetics.

Propositions verified in multiple studies (appearing more than 30 times on average) have reproducibility and consensus, but most are not protected by patents. The gap between evidence and protection reflects the asymmetry between academic precision and patent generality. From a strategic perspective, the unpatented relationship with high evidence support is the main goal of intellectual property development and transformation application, and it builds a bridge from scientific mechanisms to technical expressions to patent protection.

To distinguish between persistent hotspots and emerging growth points, a time decomposition of high - credibility, Not privatized relationships is carried out (Figure 8). By comparing the total number of evidences and the number of evidences in the recent period (2023–2025), there are two main clusters at present.

“GelMA enhanced osteogenesis” and “hydrogel-encapsulated exosomes” are continuous hotspots. For many years, the evidence level has been relatively high, and they are the anchors of long-term research. This reflects the standardization of the method, and it is also continuously verified as a “consensus scaffold” in various models.

The results show that the research has shifted from material structure repair to biological integrated functional regeneration. The continuous hotspots make the exosome-hydrogel research stable and repeatable, and the new growth points determine the innovation frontiers and patent opportunities. They provide a priority framework for the academic and translational agendas of future exosome-hydrogel research.