Knowledge Mapping Analysis of the Evolution of Internationalisation Development in Higher Education

In the perspective of globalization, the problems and challenges faced by higher education in various countries are somewhat different [1], for China, higher education in the long-term development and change has formed a unique model [2], the trend of internationalization of higher education still brings a certain impact, manifested in the concept of schooling, cultural values and educational resources, educational models and other levels [3-4], which is not conducive to the smooth advancement of its internationalization process, but also It is not conducive to the smooth promotion of its internationalization process, and also hinders the overall improvement of its education quality [5].

The trend of globalization promotes the close contact and cooperation among countries around the world in the fields of economy, science and technology, and culture [6]. As the second largest economy in the world, China is not only affected by domestic development factors, but also driven by the trend of globalization and the challenge of international competition [7-8], and the internationalization of higher education is an inevitable choice for China to adapt to globalization, and also an important way to promote international exchanges and cooperation, and to enhance international competitiveness [9]. China has put forward the innovation-driven development strategy in the new era, emphasizing the key role of innovation in national development, and higher education is the core source of innovation [10], and the internationalization development can attract leading international talents and innovation resources, providing important support for the modernization of Chinese higher education and the enhancement of innovation capacity [11].

After a long period of development and practice, China’s higher education has achieved certain results, and has also formed a unique educational model, which also contains the philosophy of school running and target orientation of Chinese characteristics [12-13], however, in the accelerating process of globalization, the internationalization of higher education has become a necessity [14], due to the internationalization of higher education’s unique system and pursuit of the goals of the educational philosophy has formed a profound impact on the concept of higher education, and has even caused a significant influence on the concept of China’s higher education, and even resulted in the development of the internationalization of higher education. Due to the unique system and goal pursuit of internationalization of higher education, its educational philosophy has formed a profound impact on the concept of higher education in China, and even caused a certain impact and challenge [15-16]. Supported and oriented by the traditional philosophy of school running, China’s higher education is unable to realize the smooth transition of internationalization, nor can it improve its internationalized school running quality [17].

In order to effectively deal with all kinds of challenges, China’s education sector, universities and society should work together [18], analyze the reasons for the emergence of all kinds of challenges in the joint consultation, joint construction and joint creation, combine with the actual development of the internationalization of higher education [19], refer to the practices and experiences of developed countries in the internationalization process of higher education, and change the philosophy of school running, pay attention to the cultural self-awareness, and deepen the advantageous complementation of resources and so on, as the main direction and specific strategies. The main direction and specific strategies are to change the concept of school running, pay attention to cultural consciousness, deepen the complementary advantages of resources, etc., so as to effectively cope with all kinds of challenges in the full practice, and to continuously improve the level of internationalization of higher education [20-22].

The purpose of this paper is to extract the visual analysis dimension of the knowledge map from its nature, and propose a relationship structure for the knowledge graph and its visual analysis method. Based on the entity and relationship, the Transd model and the PTranse model are combined to build the knowledge map model based on improving the relationship path. According to the analysis results, the application of the system is completed, and its application in the visual research of the internationalization of higher education is analyzed, and the development of the internationalization of higher education is analyzed, and the effective method is provided for the research and analysis of the international development of higher education.

2

Knowledge graph model based on improved relational paths

2.1

Visual representation of knowledge graphs

Knowledge graph is a kind of knowledge base that describes and represents entities and their relationships in the objective world by means of graph data structure, which organizes knowledge by simulating the way people understand objective things, in which (entity-relationship-entity) ternary is the basic constituent unit of knowledge graph.

From the point of view of visual layout and visual coding methods, the current main forms of knowledge graph expression are node link diagram, adjacency matrix, space filling, and heat map.

Node link graph is the most commonly used expression form of knowledge graph, which usually adopts the form of G = (V, E) binary group, containing vertex set V and edge set E . In graphical expression, it usually utilizes shapes such as circles or dots to represent the nodes, and expresses the relationship between the nodes by the connecting lines between the nodes.

The process of generating the node link graph is as follows:

1)

Graph modeling, structuring the original graph data.

2)

Graph layout, initialize the node positions, and continuously adjust the node positions according to the layout algorithm to achieve the ideal graph layout.

3)

User perception, through a series of specific scenario tasks to detect whether the graph layout reaches the user’s expected effect, so as to evaluate the effectiveness of the layout algorithm.

From the above steps, it can be seen that the key problem of node-link graphs is how to design a more reasonable layout algorithm to present a more comfortable layout form in a specific scenario. Existing layout algorithms for node-link diagrams can be categorised into force-oriented layout algorithms, stress models, constrained graph layout algorithms, multilayer iterative algorithms, and dynamic graphs.

The force-oriented layout is the most popular layout method for node-link diagrams, and the center coordinate embedding algorithm is the earliest graph drawing method for force-oriented layout.The FR model is proposed, which adds the criterion of “even vertex distribution”, and treats the vertices in a node-link diagram as “atoms, particles or celestial bodies that exert gravitational and repulsive forces on each other”. The term “atoms, particles or celestial bodies” can be expressed by the formula: (1) $f_{a} (d) = \frac{d^{2}}{k}, f_{r} (d) = - \frac{k^{2}}{d}, k = C \sqrt{\frac{a r e a}{n u m b e r o f v e r t i c e s}}$

Where f_a represents the gravitational function between them, f_r is the repulsive function between them, and the optimal distance between two vertices can be obtained according to the distance between two vertices d, k, where AREA represents the whole layout area, and C is a constant determined by experiment.

The visualization charts reflecting the measurement relationship can be divided into the traditional visualization charts expressing low-dimensional data such as bar charts and pie charts and the special charts expressing multidimensional data such as polar coordinate charts and parallel coordinate system charts.

Whether traditional visualization charts or multidimensional data charts, their data structure can be unified with the following structure: (2) $D a t a = (\dim e n s i o n, v a l u e)$

Where Data represents each piece of data, and its specific composition is shown on the right side of the equation, dimension is a dimension variable, which can also be called a position variable, indicating which dimension and which position the data is in, it is different with the different dimensions of the data, the definition of the structure will be different, for a few dimensions of the data, dimension corresponds to a few dimensions of the coordinates. value is the measure variable, which indicates the measure value of the data in this position, so that the measure value can be visualized and analyzed in different dimensions.

For the data visualization expression of network relationship, there is no obvious structural feature within the data itself, and it is not possible to count the traditional measurement relationship from the surface level, so it brings some challenges to the layout of the visualization expression.

According to the network relationship between nodes, its data structure can be expressed as: (3) $\begin{array}{l} D a t a = (N o d e, E d g e), N o d e = (n o d e) n, \\ E d g e = (n o d e s o u r c e, n o \det \arg e t, v a l u e) m \end{array}$

Data represents a dataset with network relationships, similar to the “graph” data structure, the dataset consists of a collection of nodes and edges, the node collection has n node nodes, which may contain various attributes of the node (name, value, color, etc.), the edge collection has m edges, each containing a start node nodesource and an end node nodetarget, as well as the corresponding value or weight value of the edge. The Node collection has A1 node node, which may contain various attributes (name, value, color, etc.) of the node, and the Edge collection has 2 edges, each containing a start node nodesource and an end node nodetarget, as well as the corresponding value or weight value of the edge.

2.2

Knowledge Graph Relational Structure Mining

In order to further explore the relational structure existing within the knowledge graph and combine it with data visualization for a better understanding of the knowledge graph, this paper first starts from the relational structure of the data itself. By definition, the two core elements of a knowledge graph are entities and relationships, so extracting the regular composition and structure between entities and relationships is necessary for analyzing and understanding knowledge graphs.

For the two basic elements of entities and relationships, each element has multiple forms of composition. Among them, the entity elements are composed of one-dimensional entities, two-dimensional entities, three-dimensional entities, and high-definition entities, and in this paper, the collection of multiple entities describing the same type of objective things is called a dimension.

One-dimensional entity indicates that only one entity category exists within the knowledge graph, and this structure of the knowledge graph itself will be shown as a series of scattered points, the whole structure is clear at a glance, without much visualization meaning, and the structure is rarely seen in practical applications, so this situation will not be repeated in the discussion of one-dimensional entities and entities of other dimensions.

In two-dimensional entities, there is only one possible relationship structure, R_x : E_a ~ E_b = {r_x1,r_x2,...,r_x|R|} , which is the correspondence between one-dimensional entity E_a = {e_a1,e_a2,...,e_a|E|} and another-dimensional entity E_b = {e_b1,e_b2,...,e_b|E|}.

The relationship structure of three-dimensional entities is diverse, and its relationship structure mainly includes hierarchical relationship, shared relationship, cyclic relationship, and disassembly and downgrading.

1)

Hierarchical relationships

Specify E_a as the top entity type, E_b as the middle entity type, E_c as the bottom entity type, and there may be relations R_x : E_a ~ E_b = {r_x1,r_x2,...,r_x|R|} and R_y : E_b ~ E_c = {r_y1,r_y2,...,r_y|R|} between the three of them, then specify R_x as the relation between the top entity Ea and the middle entity E_b , and R_y as the relation between the middle entity E_b and the bottom entity E_c . Structures such as these where there is a hierarchical or flow relationship pattern from one side (layer) entity to another (layer) entity are referred to as hierarchical relationship structures of three-dimensional entities.

2)

Shared Relationships

Provide that E_a is a public entity type, E_b and E_c are edge entity types, and there may be relations R_x : E_a ~ E_b = {r_x1,r_x2,...,r_x|R|} and R_y : E_a ~ E_c = {r_y1,r_y2,...,r_y|R|} between the three of them, then provide that R_x and R_y are relations between public entity E_a and edge entities E_b and E_c respectively. Structures such as this where two different types of entities have relationships with the same type of entity are referred to as shared relationship structures for three-dimensional entities.

3)

Cyclic Relationships

Similarly, there are three-dimensional entities E_a , E_b , and E_c , and there may be relationships E_c as the bottom entity type, and there may be relations R_x : E_a ~ E_b = {r_x1,r_x2,...,r_x|R|} , R_y : E_b ~ E_c = {r_y1,r_y2,...,r_y|R|} , and R_z : E_c ~ E_a = {r_y1,r_y2,...,r_y|R|} between them. The structure of closed-loop relationships between three-dimensional entities such as this is called the toroidal relationship structure of three-dimensional entities.

In addition, the idea of disassembling and downgrading can be utilized to interpret the knowledge graph corresponding to the above three-dimensional entity relationship structure. For example, the hierarchical relationships and sharing relationships of 3D entities can be downgraded or disassembled into the corresponding relationships of 2D entities, and the optimal choice can be made by combining the knowledge graphs of different domains.

2.3

Knowledge representation learning based on improved relational paths

The traditional method of knowledge representation in data processing is the solo thermal representation. In the sole heat representation, the number of words represents the length of the word vector, and each word is represented by a vector, and the relevance is judged by calculating the size of the distance between words. In this representation, as the number of words increases, the computational complexity will be higher and higher, so it is not applicable to large-scale data processing.

In large-scale knowledge mapping, this paper embeds entities and relations into a low-dimensional dense vector space through knowledge representation learning, in which the closer the distance between two entities represents a higher degree of similarity, as a way to improve the semantic relevance of entities, relations and their interactions. The proposed knowledge representation learning can effectively improve the performance of tasks such as knowledge fusion and knowledge reasoning. Translation model is a representative approach in knowledge representation learning, and its basic idea is to regard the relationship between each pair of entities as a translation operation from the head entity to the tail entity, and measure the similarity by calculating their distances.

There are two vector spaces in the TransR model, i.e., the entity vector space and the relation vector space, where entities are mapped from the entity space to the relation space by defining a mapping matrix M_r , which distinguishes the type characteristics of the entities and relations by making the head and tail entities with relation r close together in the relation space and the non-relationship entities farther apart in the relation space. The scoring function in TransR is defined as: (4) $f_{r} (h, t) = {‖ h M_{r} + r - t M_{r} ‖}_{L_{1} / L_{2}}$

Two mapping matrices are set up in the TransR model to model the head entity and the tail entity respectively, and then the modeled entities are mapped to the relation space respectively as a way to solve the problem of relational semantic diversity. In TransD, the mapping vector of ternary (h,r,t) is represented by (h_p,r_p,t_p ), and the mapping matrix is constructed by the mapping vector as: (5) $M_{r h} = r_{p} h_{p}^{T} + I^{m \times n}$ (6) $M_{r t} = r_{p} t_{p}^{T} + I^{m \times n}$

Therefore, the TransD scoring function is defined as: (7) $f_{r} (h, t) = {‖ M_{r h} h + r - M_{r t} t ‖}_{L_{1} / L_{2}}$

In translation-based model representation learning, TransD model enhances the semanticity of entities and relations by mapping them to different semantic spaces with a mapping matrix for each pair of entities and relations, taking into account the diversity of both entities and relations, and reduces the complexity of the model by replacing matrix operations with vector operations in the TransD model. Among a series of translation-based models, the TransD model has obvious advantages and is suitable for large-scale knowledge graph representation learning.

Taking into full consideration the advantages of TransD model in modeling complex relationships such as 1-N, N-1, N-N, etc. and the advantages of PTransE in modeling relational paths, this chapter proposes to combine the TransD model with the PTransE model to constitute the dynamic mapping and relational paths based representation learning. For direct relationships between two entities, the TransD model is used for modeling.

In PTransDW model, the knowledge graph is denoted as G = (E,R,S), E = {e₁,e₂,...,e_|E|} is the set of entities consisting of |E| entities, R = {r₁,r₂,...,η_|R|} is the set of relations consisting of |R| relations, S ∈ E × R × E denotes the set of triples in the knowledge graph, with (h,r,t) denoting the triple consisting of the head entity h , the tail entity t , and the relation r , and with (h,r,t) denoting the vectorial representation of the triple (h,r,t).

The entities are mapped into the relation space by mapping matrix in the PTransDW model, and translation operations are performed in the relation space for multi-step relation paths. Inspired by the defined form of the scoring function of the TransE model, the scoring function of the representable learning model is defined as: (8) $G (h, r, t) = f_{r} (h, t) + E (h, p, t)$

where f(h,t) is the scoring function for modeling direct relationships using the TransD model and E(h,p,t) is the scoring function for modeling relational paths for multi-step relational paths.

Relationships are concerned with different attributes of head entities and head entities, and it is necessary to define two dynamic mapping matrices M_n ∈ R^n×n , M_n ∈ R^m×n that can represent the characteristics of entities and relationships, and map entities into the relationship space through the mapping matrices. The dynamic mapping matrices M_rh , M_r are defined as: (9) $M_{r t} = r_{p} h_{p}^{T} + I^{m \times n}$ (10) $M_{r t} = r_{p} t_{p}^{T} + I^{m \times n}$

where h_p , t_p ∈ Rⁿ, and r_p ∈ R^m represent the projection vectors of (h,r,t) in the entity vector space, respectively, and I^I×n represents the unit matrix of dimension m×n . The head and tail entities of the entity space can then be defined as the mapping vectors in the relation space through the mapping matrices: (11) $h_{⊥} = M_{r t h} h$ (12) $t_{⊥} = M_{r t} t$

In order to distinguish different mapping regions under different relationship types, it is necessary to introduce a relationship weight w_r . Referring to the definition of relationship weights in the TransM model, the relationship weight w_r is defined as: (13) $w_{r} = \frac{1}{\log (h_{r} p t_{r} + t_{r} p h_{r})}$

where h_rpt_r represents the average number of head entities corresponding to each tail entity and t_rph_r represents the average number of tail entities corresponding to each head entity. Based on the mapping vectors of entities in the relationship space and the defined weight matrix, the scoring function f_r(h,t) in direct relationship modeling can be defined as: (14) $f_{r} (h, t) = - {‖ w_{r} h_{⊥} + r - w_{r} t_{⊥} ‖}_{L 1 / L 2}$

On the other hand, multistep relationship modeling between entities is considered, and a semantic combination approach needs to be chosen to combine all relationships on a relationship path. Suppose that the i st relationship path can be represented as p_i = (r_i1,r_i2,...,r_ii)(i ∈ M) and the combination of relationships for the i rd path is represented as: (15) $p_{i} = r_{i 1} \circ r_{i 2} \circ \dots \circ r_{i l}$

The PTransDW model in this paper uses the sum operation to realize the semantic combination, and the relation combination operation of the i st relation path is shown in Equation (16): (16) $p_{i} = r_{i 1} + r_{i 2} + \dots + r_{i l}$

In the PTransDW model, the scoring function for modeling multi-step relationships is defined as: (17) $E (h, p, t) = {‖ h + p - t ‖}_{t_{1} / t_{2}}$

And the calculation of the direct relationship has minimized ||h + r − t|| thus leading to the conclusion of r ≈ t – h . Hence it can be introduced: (18) $E (h, p, t) = {‖ p - (t - h) ‖}_{L_{1} / L_{2}} = {‖ p - r ‖}_{L_{1} / L_{2}}$

When the direct relationship is consistent with the multi-step relational path, the smaller the loss value of the scoring function and thus the better the performance. According to the direct relationship scoring function and multi-step relationship path scoring function can be defined the scoring function of the whole PTransDW model is defined below as follows: (19) $G (h, r, t) = f_{r} (h, t) + \frac{1}{Z} \sum_{p \in P (h, t)} R (p | h, t) {‖ p - r ‖}_{L_{1} L_{2}}$

Where $Z = \sum_{p \in P (h, t)} R (p | h, t)$ denotes the normalization factor and R(p|h,t) is used to measure the reliability of path p. In this paper, Path Constraint Based Resource Allocation Algorithm (PCRA) is used to measure the magnitude of correlation between two entities.The schematic diagram of PCRA resource allocation is shown in Fig. 1.

The basic idea of the path constraint based resource allocation algorithm is that there are several paths between the head entity and the tail entity, for each relational path p , there is only one entity in the confidence calculation of the head entity, i.e., the tail entity of the path of the head entity h , i.e., itself, and hence the confidence is set to R_p(h) = 1 . For the tail entity, the number of messages passed from the head entity h to the tail entity t is denoted by R_p(t), i.e., the confidence level of the path. When entities h and t are given, the reliability of the path is denoted by R_p(t), i.e., R(p|h,t)=R_p(t) . Assuming that there is a relational path $W_{0} \overset{n}{\to} W_{1} \overset{r_{2}}{\to} \dots \overset{n}{\to} W_{l}$ , h = W₀, t ∈ W_l from the head entity h to the tail entity t , where there is an entity e , and e ∈ W_i, the formula for calculating the confidence level is as follows: (20) $R_{p} (e) = \sum_{n \in W_{i - 1} (\cdot, e)} \frac{1}{| W_{i} (n, \cdot) |} R_{p} (n)$

where W_{_i−1}(·,e) is the direct antecedent of entity e along relation r_i and W_i(n,·) is its direct successor, n∈W_i–1.

3

Knowledge Graph Visualization System for Internationalization of Education

3.1

Knowledge graph system design process

The design process of the higher education internationalization knowledge map is shown in Figure 2, which is based on text analysis and knowledge map construction. The text analysis explored the characteristics of higher education internationalization literature texts in terms of the form, structure and function of policies, which provided reference for domain specialization in the process of higher education internationalization knowledge mapping construction. The construction of the knowledge map mines the deeper meanings of the texts, enhances the connotation of the textual information expression, and provides support for the depth of the application of the knowledge map of internationalization of higher education. The process of constructing and designing higher education internationalization knowledge map in this study mainly involves four main research tasks: domain knowledge collection and output, knowledge unit determination, knowledge storage design and knowledge visualization and function design. Data acquisition and pre-processing Domain knowledge collection and output Knowledge extraction and optimization Knowledge unit identification Knowledge Storage Knowledge store design Visual Interaction and System Functions Knowledge visualization and functionality design

3.2

Process of developing knowledge mapping system for internationalization of education

The development process of Higher Education Internationalized Knowledge Graph is consistent with the design process, and also contains four major steps: data collection and preprocessing, knowledge extraction and optimization, knowledge storage, and knowledge visualization interaction and system function implementation, and the development process of Higher Education Internationalized Knowledge Graph system is shown in Figure 3. The knowledge extraction and optimization phase involves extracting and optimizing knowledge units from the data set. Among them, there is a difference between the extraction methods of data knowledge units without semantics and data knowledge units with semantics, and the method of data knowledge units with semantics is more complicated, but it belongs to the core of the knowledge extraction phase. After extraction and optimization, the entity, relationship, and attribute data that make up the policy knowledge graph will be output.

The knowledge storage phase involves storing the output of the knowledge extraction and optimization phases. The relevant attributes are stored in a relational database, and the entities and relationships are stored in a graph database as graph data.

The Knowledge Visualization Interactive System Functional Implementation phase involves using database language to filter, retrieve, and manipulate data to achieve the desired system functionality. Once the functions are implemented, they are integrated into the visualization system along with algorithms and data to complete the entire internationalized knowledge mapping system for higher education.

3.3

Framework for Implementing Internationalized Knowledge Mapping System for Higher Education

The implementation framework of the higher education internationalization knowledge mapping system is shown in Figure 4. The whole higher education internationalization knowledge mapping system is based on the support of data collection and preprocessing, knowledge extraction and optimization, and knowledge storage. The data processed in the first stage is used for the calculation and extraction in the second stage, which forms the entities, relations and attributes of the policy knowledge map, and they are then stored according to certain rules. The whole system interacts in the form of WEB, the back-end is implemented in Python, the application framework uses Flask, the front-end uses uikit and the graph visualization tool ECharts, which is mainly used for graph visualization. The connection between Python and the graph database uses the Py2neo library, which allows the graph database to be used in Python. The Python connection to the graph database uses the Py2neo library, so that the data can be interacted with in Python using the graph database’s native Cypher statements, and the results of the interaction can be synchronized to the front-end visualization of the graph to achieve the functionality of the whole system.

With the use of the knowledge map, we can fully grasp the main characteristics of the international development of higher education in different periods, and provide the basis for the analysis of its subsequent evolution.

4

Research and Analysis of Internationalization Development of Higher Education Based on Knowledge Mapping

4.1

Testing of knowledge mapping systems for internationalization of higher education

The system functions of the Higher Education Internationalization Knowledge Graph are the embodiment of the system’s capabilities, which determine the way and effect of the system’s role in research applications. In the Higher Education Internationalization Knowledge Mapping System, in addition to the basic functions of searching, time range filtering, policy text viewing and policy basic information viewing, some of the policies on the internationalization and development of higher education are also classified by characteristics and themes. According to the system functions, it can be divided into A filtering retrieval, B mapping visualization effect, C keyword retrieval, D keyword node, E theme retrieval, F development policy classification and G time classification. A questionnaire survey was carried out on 50 experts after using this system for one month, which was conducted using a scoring system with a score out of 5, the higher the score the better the evaluation. The validity rate of this questionnaire was 100%, and the results of the questionnaire were analyzed in both directions: system functionality and satisfaction.

4.1.1

Survey and Analysis of Knowledge Graph System Functions

By analyzing the questionnaire survey of this paper for the seven functions of the higher education internationalization knowledge graph system, including A filtering retrieval, B graph visualization effect, C keyword retrieval, D keyword node, E topic retrieval, F development policy classification and G time classification, the scoring results of each function of the higher education internationalization knowledge graph system are shown in Figure 5. The scoring range is indicated by the box-shaped body, while the green data points represent the scoring situation, and the blue line segment represents the average score. It can be observed that the average scores of the seven functions of A Filtering Retrieval, B Graph Visualization Effect, C Keyword Retrieval, D Keyword Nodes, E Topic Retrieval, F Development Policy Classification, and G Temporal Classification are 4.166, 4.149, 4.054, 4.024, 4.029, 4.036, and 4.035, respectively, which are higher than 4 points. Particularly, the system achieves better results in and visualization, indicating that each function of the higher education internationalization knowledge mapping system in this paper performs well and can effectively filter and retrieve relevant information and visualize it to meet the research needs.

4.1.2

Knowledge Graph System Satisfaction Survey Analysis

Further to analyze the satisfaction of the higher education internationalization knowledge mapping system of this paper, according to the satisfaction scoring results of 50 experts on the knowledge mapping of this paper, the scores of the system in five aspects, such as convenience, ease of use, completeness, effectiveness and interactivity, are counted, and the satisfaction scores of the knowledge mapping system are shown in Figure 6. The average scores of the constructed higher education internationalization knowledge mapping system in the five aspects of convenience, ease of use, completeness, effectiveness and interactivity all come to more than 4, among which, the average scores of the system’s convenience and ease of use are relatively high. The experts are highly satisfied with the use of the higher education internationalization knowledge mapping system developed in this paper. The system is intuitive and easy to operate, which has received excellent ratings.

4.2

Research on the development of internationalization of education based on knowledge mapping system

4.2.1

Analysis of the evolution of the development of internationalization of higher education

In this section, the literature related to internationalization of higher education from 2011 to 2023 is retrieved and visualized using the knowledge mapping system of this paper, and the statistics of the literature on internationalization of higher education based on the knowledge mapping system is shown in Figure 7. It can be visualized that from 2011 to 2023, the overall number of research articles on internationalization of higher education is on an upward trend. It increases from 330 articles in 2011 to 8201 articles in 2023. In terms of the annual volume of publications, there was a small decline in the volume of Chinese and foreign publications in 2014 and 2019, followed by a steady growth until it peaked in 2023, with 534 Chinese publications and 526 foreign publications.

According to the trend of internationalization of higher education in terms of time and depth, it can be divided into three stages.

The first stage is 2011~2014, the slow development stage. At this stage, the research on internationalization of higher education was initially noticed, and some scholars carried out research on it, but fewer core papers were published, and the volume of foreign publications in China was lower than 190.

The second stage is 2015~2018, the awakening stage. In this stage, the research on internationalization of higher education has received higher attention, the research topics have increased, and the number of published results has been significantly improved.

The third stage is 2019~2023, the recovery stage. The research on internationalization of higher education has been growing rapidly after a small decline in attention in 2019, and the annual publication volume in China and abroad exceeds 380 articles and continues to grow. The research on internationalization of higher education has been effectively promoted, and the research on it has been fundamentally and rapidly improved in both quantity and quality.

4.2.2

Analysis of Hot Spots for the Development of Internationalization of Higher Education

Through the knowledge mapping system of this paper, the co-occurrence analysis of keywords related to higher education internationalization literature from 2011 to 2023 was carried out, and the 1349 keywords retrieved were ranked according to the frequency of occurrence in the visualization map, and the keywords located in the top 10 were selected for hotspot analysis. The keyword ranking of internationalization of higher education is shown in Table 1, and the keywords with high frequency are higher education, international students, university, internationalization, global, students, education, policy, model and perspective. Among these keywords, there are six with a frequency greater than 50, and the keyword with the highest frequency is higher education, which is used 230 times. This is followed by the keyword international students with a frequency of 185, the keyword university with a frequency of 95, and the keywords globalization and students with a comparable frequency of 70 and 68, respectively.The rest of the keywords with a high frequency of occurrence are education and policy, which all have a frequency of 30 or more. Through the frequency of these keywords, it can be seen that in addition to internationalization of higher education, which is the search keyword of this study, the frequency of the keywords “international students” and “universities” is quite high, which indicates that international students and universities have become an important part of the research on internationalization of higher education. This indicates that international students and universities have become important research objects in the study of internationalization of higher education. As the two main subjects of higher education, students, as active participants, undoubtedly promote cooperation and exchange between schools and become an important driving force for the internationalization of higher education. And universities, as the training bases of higher education, play an unignorable role in the process of internationalization of higher education. In addition, the frequency of policies and models is also high, indicating that policies and models are likely to be research hotspots in the field of higher education internationalization research. Overall, the scope of higher education research is relatively broad and not limited to a single subject. The research on internationalization of higher education is complex and involves all aspects of higher education.

Table 1.

Key words of higher education internationalization

Ranking	Keyword	Frequency
1	Higher education	230
2	International student	185
3	University	100
4	Internationalization	95
5	Globalization	70
6	Student	68
7	Education	45
8	Policy	36
9	Model	28
10	Perspective	25
Total		882

5

Conclusion

A knowledge graph model with better relational paths was made using the TransD model and the PTransE model structure in this study. For internationalizing higher education, a knowledge graph visualization system was also made. The visualization study on the trend of higher education internationalization in lectures is being carried out.

1)

The system’s screening retrieval and mapping visualization effect scores are relatively high at 4.166 and 4.149, respectively, and each function and system satisfaction index is higher than 4 points. The knowledge mapping system in this paper shows that it can effectively find relevant information, screen it, and display it. The system is also useful for the research on the internationalization of higher education.

2)

The research literature on the internationalization of higher education, in terms of time and depth, is divided into three stages, including the slow development stage from 2011 to 2014, the awakening stage from 2015 to 2018, and the recovery stage from 2019 to 2023. Among them, the number of domestic and international publications in the slow development stage is less than 190. The annual publication volume in the recovery stage is more than 380 articles and continues to grow. This indicates that the research on the internationalization of higher education has been effectively promoted from 2011 to 2023 and also verifies the practical application of this paper’s system.

3)

According to the analysis of hot research, higher education is the most frequent keyword with 230 times. It is followed by international students 185 times, universities 95 times, globalization 70 times, and students 68 times. It shows that higher education is the most important hotspot in the research of internationalization and development of higher education, and international students and universities in higher education as the two major carriers of higher education are important research objects. From the perspective of keyword frequency, it is indicated that the scope of higher education internationalization research is relatively wide, which further validates the application value of the knowledge graph visualization system in this paper.

Langue:: Anglais

Périodicité:: 1 fois par an
Sujets de la revue:: Sciences de la vie, Sciences de la vie, autres, Mathématiques, Mathématiques appliquées, Mathématiques générales, Physique, Physique, autres

RSS Feed de la revue

Knowledge Mapping Analysis of the Evolution of Internationalisation Development in Higher Education

Lijie Chen

Guanping Liang

Publié en ligne: 17 mars 2025

Reçu: 25 oct. 2024

Accepté: 03 févr. 2025

DOI: https://doi.org/10.2478/amns-2025-0254

Mots clésVisual representation, Relational structure mining, Improved relational path, Knowledge mapping system, Internationalization of higher education

© 2025 Lijie Chen et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Mots clés
Visual representation, Relational structure mining, Improved relational path, Knowledge mapping system, Internationalization of higher education