Knowledge Graph Technology Helps Intelligent Reform of Innovation and Entrepreneurship Guidance Courses for College Students
Data publikacji: 05 lut 2025
Otrzymano: 13 wrz 2024
Przyjęty: 02 sty 2025
DOI: https://doi.org/10.2478/amns-2025-0051
Słowa kluczowe
© 2025 Yeting Chen et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
With the continuous development of information technology, knowledge graph as an emerging knowledge representation and management technology has been widely used in various fields. Knowledge graph is a formal descriptive framework for generic semantic knowledge, representing the knowledge base of entities and their relationships in the objective world in the form of graphs, integrating emerging research methods in different fields such as statistics, information science, semantic analysis, intelligent recognition, etc., and providing an effective technical means for knowledge extraction and visualization of massive information [1–3]. In the field of education, knowledge mapping is considered to have great potential to help teachers better organize and manage teaching content and improve teaching quality [4].
College graduates are both the main force of employment and the force of innovation and entrepreneurship, and devoting high-quality talents to the development of innovation and entrepreneurship in various industries not only creates more employment opportunities but also promotes the development of the industry, which has a high economic and social value [5–6]. China has formulated and implemented a series of important policy documents and deployment arrangements, requiring the cultivation of innovation and entrepreneurship awareness of college students and providing policy guarantees for college students to participate in innovation and entrepreneurship. The opening of innovation and entrepreneurship guidance courses in colleges and universities is an important initiative to serve the national innovation and development strategy, a due intention to continuously deepen the reform of college education, an important way to cultivate students’ innovation and entrepreneurship awareness and practical ability, and can effectively implement entrepreneurship-led employment and further promote the employment of college graduates [7–9]. However, the traditional teaching methods have reached a bottleneck in stimulating students’ interest in learning as well as their understanding of the process of innovation and entrepreneurship, and it is not easy to further enhance it. How to use modern information technology to improve the teaching quality of innovation and entrepreneurship guidance courses is one of the urgent problems to be solved [10–12]. In the “innovation and entrepreneurship guidance” course, the construction of knowledge mapping can comprehensively sort out the course knowledge points, provide a visual display of the knowledge structure, and then realize the organic organization, indepth excavation and comprehensive use of knowledge. This can not only help students better understand the knowledge system and improve their learning effect, but also assist teachers to carry out curriculum construction, textbook construction and teaching reform so as to improve their teaching ability [13–15]. Knowledge mapping techniques With the rapid development of information technology, knowledge mapping, as a new data representation and processing technology, plays an important role in the field of education. Literature [16] designed and developed a technology-enhanced intelligent learning environment by combining knowledge graphs and learning paths and applied it in the field of education with the aim of identifying students ’ weaknesses or knowledge gaps in order to individually help them achieve their goals. Literature [17] points out the dominance of knowledge mapping technological innovations in the field of education and provides a detailed overview of knowledge mapping functionality, knowledge extraction techniques, knowledge base characteristics, resource requirements, evaluation criteria and limitations to fill the gap in knowledge mapping research in the field of education. Literature [18] proposes an educational resource recommendation algorithm based on the knowledge graph of educational resources in grid technology, in which the hybrid similarity fusion technology utilized can quickly match the core resources, solving the problems of data sparsity and cold start that exist in the traditional recommendation algorithms, effectively improving the relevance and efficiency of educational research, and providing scientific and intelligent data analysis methods for subject teaching and research. Literature [19] emphasizes the important link between research education and employability, systematically reviews the application of knowledge mapping in education and employability, and shows through specific examples how knowledge mapping can be used to solve the skill mismatch between education and the job market to help determine the needs of the job market and establish better assessment methods. Intelligent Reform of Innovation and Entrepreneurship Guidance Courses for College Students In recent years, the state has introduced many measures and favorable policies to encourage college students’ innovation and entrepreneurship, and innovation and entrepreneurship have become one of the important life choices for college students. Literature [20] designed an intelligent tutoring system for college students’ innovation and entrepreneurship education based on the computer online teaching method of the Internet and verified the effectiveness and feasibility of the system through teaching practice, which helps to identify the importance and cognition of entrepreneurship education. Literature [21] highlights the growing demand for innovative talents in society, refines the innovation and entrepreneurship education system in terms of “cultivation mode, curriculum reform, teacher education, resource platform and quality evaluation”, and builds a curriculum system that integrates professional education and innovation education in both directions, so as to cultivate more innovative and entrepreneurial talents for the society. Literature [22] emphasizes the key role of deepening the reform of innovation and entrepreneurship education. It indicates that the reform of innovation and entrepreneurship education, which starts from the innovative educational concept, innovative curriculum system and teaching methods, and innovative practice mode, is an inevitable choice to promote the cultivation of talents in colleges and universities and an effective way to improve the quality of education and teaching. The study provides innovative specialists, entrepreneurial top-notch talents, and high-level comprehensive talent cultivation. The study provides a reference value for the cultivation of innovative specialists, entrepreneurial talents and high-level comprehensive talents. Ref. [23] briefly analyzes the problems existing in the current curriculum structure and employment guidance mode of innovation and entrepreneurship education and the positive impact of integrated development, puts forward a new concept of integrated development based on employment and driven by innovation and entrepreneurship, and verifies through experiments that the application effect of this integration strategy is good, which is helpful to alleviate the current situation of “employment difficulty”. The resource personalised recommendation method based on the knowledge graph designed in this paper includes three steps. First, extract the knowledge of innovation and entrepreneurship resources on the Internet and establish a semantic matching matrix based on the knowledge graph. Second, calculate the ternary vector and feature matrix to get the corresponding semantic matching matrix. Third, construct the interest model and set the personalised recommendation parameters to achieve personalised recommendation of innovation and entrepreneurship resources in universities. The study uses the recall evaluation index to validate the recommendation results of this paper’s method and uses the optimised CLASSE scale to evaluate the intelligent reform of innovation and entrepreneurship guidance courses, judging the effectiveness of the method based on the example validation results.
The recommendation algorithm is one of the core technologies for recommending learning resources. This paper introduces the recommendation algorithm based on traditional methods, the recommendation algorithm based on sequences and the recommendation algorithm based on a knowledge graph, and finally chooses the recommendation algorithm based on the knowledge graph to construct the personalised recommendation model of learning resources for college students’ innovation and entrepreneurship guidance courses.
Currently, with the rapid development of the Internet, recommender systems have gradually become an indispensable part of Internet services. The recommender system mines the behavioural characteristics from the user’s historical behavioural records and selects the resource that best meets the user’s interest from a large number of candidate resources for recommendation, and the recommendation results include two forms of top-N recommendation and streaming recommendation. Recommender systems have accumulated a large amount of experience since their introduction, and research on recommender systems can be divided into collaborative filtering-based recommendation algorithms and deep learning-based recommendation algorithms according to whether or not they incorporate deep learning [24].
Sequential recommender systems aim to displayably model users’ sequential behaviours, thereby enhancing the effectiveness of recommender systems. Compared to traditional user- or item-based collaborative filtering, traditional recommender systems often model the interaction between users and items in a static way, which results in limited capture of user preferences and affects the performance of the recommendation system. Sequence-based recommender systems, on the other hand, model the interaction between the user and the item as a dynamic sequence to capture the user’s current preferences as well as long-term preferences, which can better characterise the user’s preferences [25]. Sequence-based recommendation algorithms include standard sequence recommendation algorithms, sequence recommendation algorithms based on long and short-term preferences, and sequence recommendation algorithms based on multi-interest representation.
The basic idea of recommender systems is to analyse the user’s historical data in order to recommend items that the user may like, and the core of this is by modelling user preferences and item characteristics. So, without considering some attribute factors of the user, the essence of the recommender system is to judge the similarity between the collection of historical items and the current new items. Therefore, the recommendation performance of the model can be improved by fusing structural information by combining knowledge graphs in the recommendation system. The traditional recommendation algorithm has the problems of data sparsity and cold start, while the knowledge graph has the performance of being able to represent the higher-order relationships between entities, and the introduction of the knowledge graph as auxiliary information into the recommendation system helps to improve the performance of the recommendation system [26].
A feasible way to build personalised learning resources for innovation and entrepreneurship guidance courses is knowledge mapping. The construction aspect of knowledge mapping shall be formed by comprehensively applying professional knowledge in the field of education and artificial intelligence technology, combined with educational big data resources. In terms of function, it can provide learners with systematic knowledge management, analysis and service methods, can select and arrange relevant knowledge points and learning resources according to factors such as learners’ interests, abilities and learning goals, can accurately understand learners’ characteristics, and can recommend personalised innovation and entrepreneurship guidance course learning resources using knowledge mapping in order to implement precise teaching. In terms of construction standards, it is necessary to establish standards that unify professional field standards and teaching standards, solve the problem of integration of educational data and professional field knowledge, and form high-quality teaching resources [27]. It must also meet the needs of students’ personalised learning, the needs of teachers’ precise teaching, and the needs of curriculum teaching resources association.
The application of knowledge mapping to personalised learning first requires a comprehensive understanding of the individual learning. A large amount of individual learning data must be collected and processed, including students’ learning behaviours, learning outcomes, interests and hobbies, and the processing and analysis of these data is the basis for personalized recommendation and learning path planning.
Personalised learning requires the construction of a personalised knowledge resource platform with the knowledge graph as the hub. The data is unified and kept, and the big data platform carries out resource scheduling. Structured data is constructed by automatically collecting subject knowledge and individual knowledge to form a mapping type, and unstructured data is constructed by adopting an extraction type for videos, information and literature. The platform can integrate and annotate the knowledge points and conceptual relationships in the learning field and construct and maintain an accurate, comprehensive and complete knowledge network so that the same knowledge point has different levels of difficulty and differentiation for different students, reflecting the personalised learning characteristics of different students. On this basis, based on the user learning big data, for the entity and conceptual information of the knowledge used in the user learning process, through the analysis of the semantic information system, the corresponding knowledge map search to find the corresponding answer, at the same time, according to the user’s feedback and the learning situation, constantly adjust and optimize the algorithms and models, to improve the precision and hitting rate of the innovative entrepreneurship guidance course resources recommendation, and to achieve a better learning effect. The following is a summary of the results of the course.
Based on the knowledge of innovation and entrepreneurship resources, a semantic matching matrix based on knowledge graph is established. Set the triad of knowledge graph, i.e., entity, attribute, and relationship [28]. Let the entity of the knowledge graph triad be
Where:
Let the number of layers in the hidden layer of the knowledge graph be
Calculated by Eq. (3),
Where: the feature matrix
Based on the semantic matching matrix, the innovation and entrepreneurship resource interest model for college students is constructed. The model designed in this paper is the KG-NCF model, which takes the feature vectors of the semantic matching matrix as the objective vectors. The KG-NCF model consists of three modules, which are the recommendation module, the cross-connection module, and the knowledge expression module, and the framework of the KG-NCF model is shown in Figure 1. Among them, the cross-linking module is responsible for extracting the corresponding knowledge of innovation and entrepreneurship resources from the database and delivering it to the knowledge expression module.
KG-NCF Model frame
Students will be interested in their own innovation and entrepreneurship direction of the keywords into the recommendation module, the cross-connection module to achieve the process of personalised recommendation. In this paper, the weighted fuzzy calculation is used to derive the recommendation level
Where:
Where:
The comparison experimental objects used in this paper are the personalised recommendation method based on the traditional method and the sequence-based personalised recommendation method. Based on the convolutional neural network set up convolutional layer, the innovation and entrepreneurship resource information is trained, and the feature parameter
Where:
The personalised recommendation of innovative entrepreneurial resources is achieved by capturing the main preferences of users in the current sequence and eliminating unexpected behaviours. The experiment uses the recall rate to evaluate the performance of the personalised recommendation method, the higher the recall rate, the better the performance of the recommendation method. The recall rate is calculated by the formula:
Where:
The learning resource dataset of college students’ innovation and entrepreneurship guidance course contains 3652 data, including 1563 picture data, 963 text data, and 1126 video data. The data source comes from the user public data of the Bilibili, Zhihu, and Douban platforms.
In this study, keywords with a frequency of more than 50 and a high degree of centrality are derived, and the high-frequency words in the learning resource dataset of college students’ innovation and entrepreneurship guidance courses are shown in Table 1. The frequency of “entrepreneurship education” reached 253, with a high degree of centrality (0.17), indicating that “entrepreneurship education” is a hot spot in the research of college students’ innovation and entrepreneurship guidance courses. “Talent training (0.17)”, “innovation (0.14)”, “professional education (0.12)”, “curriculum system (0.12)” and “entrepreneurship (0.09)” also have a large degree of centrality, which is also the focus of the research on innovation and entrepreneurship guidance courses for college students.
The course learning resource data set high-frequency words
| Serial number | Key words | Frequency | Center degree |
|---|---|---|---|
| 1 | Entrepreneurship education | 253 | 0.17 |
| 2 | Talent culture | 200 | 0.17 |
| 3 | Innovate | 165 | 0.14 |
| 4 | Professional education | 160 | 0.12 |
| 5 | Curriculum | 139 | 0.12 |
| 6 | Start a business | 137 | 0.09 |
| 7 | Game | 125 | 0.08 |
| 8 | Education mode | 121 | 0.08 |
| 9 | Education system | 112 | 0.07 |
| 10 | Local university | 107 | 0.07 |
| 11 | Innovation education | 89 | 0.07 |
| 12 | Melding | 89 | 0.06 |
| 13 | Path | 85 | 0.06 |
| 14 | Internet+ | 82 | 0.05 |
| 15 | Education reform | 77 | 0.05 |
| 16 | Ideological and political education | 73 | 0.04 |
| 17 | Question | 69 | 0.03 |
| 18 | Practice | 69 | 0.03 |
| 19 | Higher education | 67 | 0.02 |
| 20 | Private colleges | 66 | 0.01 |
| 21 | Teaching reform | 63 | 0.01 |
| 22 | Practical teaching | 59 | 0.05 |
| 23 | Status quo | 55 | 0.06 |
Data pre-processing. After completing the data collection, it will be subjected to simple preprocessing, including standardizing the text format, adjusting the image size and quality, and ensuring the uniformity of the video file format.
Feature extraction. By analyzing the file names, keywords are extracted as input for the recommendation algorithm. These keywords can reflect the content of the file, such as “entrepreneurship education”, “talent development” and other terms, in order to more accurately infer the characteristics and uses of the file.
Combining the curriculum knowledge map. We perform various graph database operations to read important information such as labels, names, and semantic relationships of entities in the knowledge graph of college students’ innovation and entrepreneurship guidance courses, which provides inputs for constructing the innovation and entrepreneurship resource interest model of college students based on the semantic matching matrix.
Simulate recommendation. Simulate the process of user recommendation of learning resources by comparing the recommendation algorithm based on collaborative filtering (Method 1), recommendation algorithm based on deep learning (Method 2), standard sequence recommendation algorithm (Method 3), sequence recommendation algorithm based on long and short-term preferences (Method 4), sequence recommendation algorithm based on multiinterest representation (Method 5), and recommendation results of the course knowledge graph-based recommendation algorithm (KG-NCF) designed in this paper to evaluate them in comparison. Verify the applicability and effectiveness of the KG-NCF model in recommending learning resources.
Innovative entrepreneurial resource information is imported into six personalised recommendation algorithms. The recall assessment index is used to analyse the five comparison methods as well as the method of this paper. Through the comprehensive consideration of the assessment indexes, the performance of each algorithm in recommending learning resources can be objectively assessed, and the assessment results are shown in Figure 2. The recommendation algorithm based on the course knowledge graph designed in this paper consistently has the highest recall rate with 9 different numbers of resources input. When the number of input resources is 250, the recall can reach 89%.
Performance assessment results
In this study, the “Course-Level Student Engagement Survey Questionnaire” (CLASSE) was introduced into the evaluation of the intelligent reform of innovative entrepreneurship guidance courses and was localized. The questionnaire is designed with four dimensions: “active cooperative learning”, “teacher-student interaction”, “course challenge”, and “process evaluation”, and a total of 20 questions. A four-point Likert scale was used, with 1 to 4 representing the respondents’ opinion of the option, “1=very little, 2=some, 3=more, 4=very much”. The optimized CLASSE scale was used to effectively evaluate the students’ behavior and teachers’ behavior in the innovation and entrepreneurship guidance course under the personalized resource recommendation model based on a knowledge graph to judge the effectiveness of this intelligent reform.
This study focuses on investigating the degree of student’s commitment to the innovation and entrepreneurship guidance course under the personalised resource recommendation model based on a knowledge graph as a basis for evaluating the effectiveness of the course, which is targeted at students of all majors in all four grades in three research universities (denoted as A, B, and C, respectively) that offer the innovation and entrepreneurship guidance course. The basic statistical characteristics of the sample are shown in Table 2. Of the 2,000 students surveyed, 47.3 per cent were male, and 52.7 per cent were female. The proportion of students in the first year of university was the highest, accounting for 44.6 per cent, while students in arts and history, science and technology, and arts accounted for 15.6 per cent, 71.2 per cent, 13.2 per cent, respectively, which is in line with the distribution of the number of disciplines in the sample schools.
Sample basic statistics
| Index | Classification | Sample size | Proportion/% |
|---|---|---|---|
| Gender | Female | 1054 | 52.7 |
| Male | 946 | 47.3 | |
| Grade | First grade | 892 | 44.6 |
| Second grade | 304 | 15.2 | |
| Third grade | 436 | 21.8 | |
| Fourth grade | 368 | 18.4 | |
| Disciplines | Literary history | 312 | 15.6 |
| Technologists | 1424 | 71.2 | |
| Art class | 264 | 13.2 | |
| School | A | 672 | 33.6 |
| B | 684 | 34.2 | |
| C | 644 | 32.2 |
The descriptions of the 20 question items designed in the questionnaire are given below.
A total of 7 question items were designed under the active collaborative learning dimension, which is as follows:
Asking questions in the innovation and entrepreneurship guidance course. Participating in classroom discussions in an innovation and entrepreneurship mentoring course. Making classroom presentations in the innovation and entrepreneurship mentoring course. Carrying out projects with other students in the innovation and entrepreneurship mentoring course. Mentor other students in an innovation and entrepreneurship mentoring programme. Prepare class assignments with fellow students outside of the Innovation and Entrepreneurship Mentoring course. Communicating with classmates, friends, etc., outside of the innovation and entrepreneurship mentoring course about ideas generated during the course.
A total of four questions were designed under the teacher-student interaction dimension, which are:
Discussing grades or assignments with the instructor. Discuss career development planning with the teacher. Discuss ideas generated in readings or courses with the instructor outside of the Innovation and Entrepreneurship Orientation course. Receive written or verbal feedback from the instructor on student performance.
There are seven questions designed under the course challenge dimension, which are:
Worked harder than expected to meet the teacher’s standards or expectations. Completed assigned pre-reading assignments prior to class. Recognised facts, ideas or methods from the course being taught. Analysed a particular case, idea or theory in depth. Synthesise different perspectives, information or experiences and integrate them into new and more complex explanations and relationships. Evaluate the value of information, arguments or methods. Apply theories or concepts to solve practical problems and generalise to new situations. Organisation of teaching and learning activities in a variety of forms during the teaching and learning process.
There are 2 questions designed under the process evaluation dimension, respectively:
Emphasis on students’ classroom interaction assessment. Require students to submit process learning outcomes other than the closing assignment.
Exploratory factor analysis and validation factor analysis were conducted using SPSS24, and the questionnaire reliability statistics are shown in Figure 3. The mean of the scale with items deleted (MSDI), the variance of the scale with items deleted (VSID), the corrected item-total correlation (CITI), and the Cronbach’s alpha with items deleted were examined separately, and the Cronbach’s alpha coefficient exceeded 0.7. The results showed that the validity and reliability of the revamped measurement instrument performed well.
A program survey of innovative entrepreneurship courses
The questionnaire reliability test is shown in Table 3. By analysing the four dimensions of active cooperative learning, teacher-student interaction, course challenge degree, and process evaluation, the remaining three dimensions were all greater than 0.9, except for the clombach Alpha of process evaluation, which was 0.894, indicating that the scale reliability was high and the question item design was of high consistency.
Survey questionnaire reliability test
| Dimension | Cronbach’s Alpha | Term number |
|---|---|---|
| Active cooperative learning | 0.916 | 7 |
| Student interaction | 0.932 | 4 |
| Course challenge | 0.928 | 7 |
| Process evaluation | 0.984 | 2 |
| Overall performance | 0.976 | 20 |
A comparison of the discriminant validity of the Innovative Entrepreneurship Orientation Course Survey is shown in Figure 4, where the correlation coefficients are less than the square root of the AVE, except for Active Cooperative Learning and Teacher-Student Interaction (0.882), which is a good discriminant validity of the scale.
Questionnaire validity test
The results of the innovative entrepreneurship course survey are shown in Table 4, and the overall mean score of the questionnaire is 3.37 (standard score of 2), which indicates that the overall teaching effectiveness of the innovative entrepreneurship course under the personalised resource recommendation model based on knowledge graph is much higher than the standard level.
Results of innovative entrepreneurship course survey
| Dimension | Item | Min | Max | M | SD | Mean average | Dimensional mean |
|---|---|---|---|---|---|---|---|
| Active cooperative learning | T1 | 1 | 4 | 3.57 | 0.775 | 3.39 | 2 |
| T2 | 1 | 4 | 3.23 | 0.786 | |||
| T3 | 1 | 4 | 3.56 | 0.779 | |||
| T4 | 1 | 4 | 3.17 | 0.801 | |||
| T5 | 1 | 4 | 3.46 | 0.805 | |||
| T6 | 1 | 4 | 3.63 | 0.773 | |||
| T7 | 1 | 4 | 3.13 | 0.784 | |||
| Student interaction | T8 | 1 | 4 | 3.29 | 0.807 | 3.23 | 2 |
| T9 | 1 | 4 | 3.12 | 0.784 | |||
| T10 | 1 | 4 | 3.18 | 0.806 | |||
| T11 | 1 | 4 | 3.34 | 0.785 | |||
| Process evaluation | T12 | 1 | 4 | 3.4 | 0.807 | 3.39 | 2 |
| T13 | 1 | 4 | 3.2 | 0.781 | |||
| T14 | 1 | 4 | 3.44 | 0.8 | |||
| T15 | 1 | 4 | 3.49 | 0.795 | |||
| T16 | 1 | 4 | 3.12 | 0.784 | |||
| T17 | 1 | 4 | 3.66 | 0.8 | |||
| T18 | 1 | 4 | 3.44 | 0.766 | |||
| Overall performance | T19 | 1 | 4 | 3.76 | 0.81 | 3.47 | 2 |
| T20 | 1 | 4 | 3.18 | 0.806 |
This paper designs the KG-NCF model, which utilizes knowledge graph technology, to provide personalized learning resource recommendations for college students’ innovation and entrepreneurship guidance courses. The recall rates of personalized recommendation methods based on traditional methods and sequence-based personalized recommendation methods are compared with this paper’s method. In the case of inputting 9 different resource numbers, the recall rate of this paper’s method is always the highest, and the average recall rate reaches 0.738, which can effectively improve the recommendation results. After students applied this method to the innovative entrepreneurship guidance course, the evaluation mean value reached 3.37 points, which is 1.37 points higher than the standard score. The KG-NCF model has high applicability and effectiveness in recommending learning resources.