Research on Optimizing the Curriculum System of College Students’ Innovation and Entrepreneurship Education under the Perspective of Internet Plus

First, standardize the construction of the curriculum. Under the background of “Internet +”, the construction of dual innovation education curriculum in colleges and universities should be centered on the following modules:

First, create basic and general courses to help students understand the framework of innovation and entrepreneurship, mobilize their interest in innovation, and strengthen their entrepreneurial thinking.

Second, create specialized courses. Institutions of higher education should combine the characteristics of faculties and majors to set up dual-creation education courses that meet the teaching requirements of their majors, so that students can optimize their knowledge structure while laying a good foundation for professional learning, and lead them to participate in entrepreneurial practices related to their majors.

Thirdly, create practical platforms, such as creating a cultural atmosphere of dual-creation through school clubs, cultivating students’ practical entrepreneurial ability through laboratories, and strengthening students’ participation awareness through dual-creation projects. Second, set up individualized courses. As all students have their own personality characteristics and growth needs, the construction and practice of the dual-creation education curriculum system should take into account the development direction of different types of students and provide differentiated guidance. Finally, diversified curriculum implementation. Colleges and universities should integrate professional education and dual-creation education with their own orientation and teaching philosophy, so that the first classroom and the second classroom can be integrated, and students’ dual-creation ability can be cultivated in multiple forms and through multiple channels. At the same time, colleges and universities should also combine students’ learning needs and growth characteristics with the needs of regional economic development and modern social construction, so that the dual-creation education program can be implemented in reality [15].

First, offline platform. The so-called offline is to provide a place for students to personally participate in the experience and feeling with the help of the physical platform existing in real life, which is of positive significance to the cultivation of students’ dual-creation practice skills. The main purpose of creating an offline platform is to provide a place for entrepreneurship education, practice and service for the student body, and through cooperation with government organizations and enterprises, to enhance entrepreneurial interaction ability and establish a correct sense of employment, in order to quickly integrate into social practice.

Second, online platform. Based on the background of “Internet +”, the advantage of information technology provides a solid guarantee for the creation of a dual-creation education curriculum system in Chinese institutions of higher education. Using the Internet as a medium, colleges and universities are able to integrate all kinds of high-quality resources to provide college students with timely learning content and diversified guidance services, and to promote a better mastery of dual-creation knowledge among students.

Under the perspective of “Internet +”, when colleges and universities reform the education system of dual-creation courses, they should take the reform of the teaching mode as the starting point to ensure that the teaching mode presents the characteristics of wholeness, freedom, openness and epoch. First of all, the three-dimensional teaching mode to improve students’ comprehensive literacy. Entering the Internet era, modern students have more ways to obtain information and fast access to information, coupled with the open-mindedness of students, so if colleges and universities want to improve students’ professional competence and comprehensive literacy through dual-creation education, they can flexibly apply three-dimensional teaching method. Further, colleges and universities should actively build a new education model, based on the dimensions of school, grade and students, to carry out dual-creative activities. At the same time, in the implementation of dual-creative education and practical activities, it is also necessary to vary from person to person, combined with the cognitive ability of students, learning level, growth characteristics and professional nature of dual-creative education, in order to help students quickly improve their dual-creative ability. Secondly, ubiquitous learning mode. At this stage, people’s access to information is changing day by day, and when universities carry out innovation education for students, they should implement “ubiquitous learning” activities based on the Internet, so as to strengthen students’ comprehensive ability and improve the teaching effect of the courses. In addition, the information carrier in the era of “Internet +” presents diversified characteristics, and means such as flipped classroom, microclass and catechism have gradually become indispensable teaching methods in institutions of higher education. The implementation of dual-creation teaching through diversified information carriers can efficiently transfer information, and based on the “Internet +” education model, students can also form new concepts and new thinking to keep pace with the development of information technology.

Recommendation algorithms are crucial tools for personalizing students’ learning. On the one hand, recommending suitable learning resources to students helps them save time and improve learning efficiency. On the other hand, suitable recommendations can bring students a good learning experience, improve their satisfaction with the platform, and maintain user stickiness. Applying recommendation technology to the personalized learning recommendation scenario can help solve the difficulty of selecting resources from the massive and rich variety of innovation and entrepreneurship courses.

At present, the more common personalized recommendation model is based on the recommendation model based on association rules, the personalized recommendation model based on association rules requires a large amount of data to be analyzed, and the generation of association rules is more difficult, the accuracy is also lower, and it can not really achieve personalized recommendation [16]. Therefore, it is necessary to construct a recommendation system based on the collaborative filtering personalized recommendation model to achieve the purpose of personalized recommendations of innovative entrepreneurship course resources for students.

According to the degree of interest of student users in innovation and entrepreneurship course resources to give the course resources a different score value, the score value is expressed as an integer number 0 to 5. The higher the score, the greater the interest of the student user in this course resource, and the greater the expectation of the student user for similar courses in this course. The lower the score, the less interest the student user has in this course, and the lower the student user’s expectation. When the score is 0, it means that the student user has not yet scored this course resource. The student-course scoring matrix is shown in equation (1): (1)R(m,n)=[ r11 r12 ... r1n r21 r22 ... r2n ... ... ... ... rn1 rn2 ... rnn]$$R(m,n) = \left[ {\begin{array}{c} {{r_{11}}}&{{r_{12}}}&{...}&{{r_{1n}}} \\ {{r_{21}}}&{{r_{22}}}&{...}&{{r_{2n}}} \\ {...}&{...}&{...}&{...} \\ {{r_{n1}}}&{{r_{n2}}}&{...}&{{r_{nn}}} \end{array}} \right]$$

The student-course scoring matrix of Equation (1) in which m denotes the number of users n denotes the number of courses, and r_mn denotes the scoring score of the mth student user for the nth course. To facilitate data collection and calculation, the score attributes in the matrix are expressed in binary variables (0, 1), where 0 in the binary variable indicates that the student user dislikes the course and 1 indicates that the student user likes the course. Based on the student-course score matrix, it is possible to identify the set of neighbors of the target student user. The similarity between the student users and the neighbor set is key to the accuracy of the recommended courses, so the similarity between the users needs to be calculated. Considering the user’s scoring score for the course resource as a space vector, there is cosine similarity as shown in Equation (2): (2)sim(i,j)=cos(i,j)=i¯⋅j¯‖i¯‖⋅‖j¯‖$$sim(i,j) = \cos (i,j) = \frac{{\bar i \cdot \bar j}}{{\left\| {\bar i} \right\| \cdot \left\| {\bar j} \right\|}}$$

In Equation (2), sim(i, j) denotes the similarity between user i and user j, and i¯,j¯$$\bar i,\bar j$$ denotes the scoring score of a course by user i and user j, respectively. If Pearson’s correlation coefficient is used to calculate the similarity between users, assuming that user i and user j have scored the same collection of course resources, the similarity between them is shown in Equation (3): (3)sim(i,j)=∑s=Rs+i(Ri,s−R¯i)(Rj,s−R¯j)∑s=Rs+i(Ri,s−R¯i)2∑s=Rs+i(Rj,s−R¯j)$$sim\left( {i,j} \right) = \frac{{\sum\limits_{s = {R_{s + i}}} {\left( {{R_{i,s}} - {{\bar R}_i}} \right)} \left( {{R_{j,s}} - {{\bar R}_j}} \right)}}{{\sqrt {\sum\limits_{s = {R_{s + i}}} {{{\left( {{R_{i,s}} - {{\bar R}_i}} \right)}^2}} } \sqrt {\sum\limits_{s = {R_{s + i}}} {\left( {{R_{j,s}} - {{\bar R}_j}} \right)} } }}$$

In Equation (3), R¯i$${\bar R_i}$$, and R¯j$${\bar R_j}$$ denote the average value of the scores of user i and user j on the course resources, respectively. By using the above two formulas, the similarity of course preferences between users can be calculated, and based on the similarity, the set of users with close similarity can be found, and relevant course resources can be recommended to the members of the user set. The collaborative filtering algorithm predicts the user’s preference for innovation and entrepreneurship course resources based on the scoring of the user’s neighbors on the innovation and entrepreneurship course resources, as shown in Equation (4): (4)Pi,n=R¯i+∑j=NNisim(i,j)*(Rin−R¯j)∑j=NNi(|sim(i,j)|)°$${P_{i,n}}\: = \:{\bar R_i}\: + \:\frac{{\sum\limits_{j = N{N_i}} s im{{\left( {i,j} \right)}^*}\left( {{R_{in}}\: - {{\bar R}_j}} \right)}}{{\sum\limits_{j = N{N_i}} {\left( {\left| {sim\left( {i,j} \right)} \right|} \right)} }}_\circ $$

In Equation (4), NN_i denotes the set of neighbors of user i, and P_Ln denotes the predicted scoring score of user i for innovative entrepreneurship course resource u.

The collaborative filtering recommendation model can achieve innovative and entrepreneurial personalized course recommendations for student users, but it suffers from cold start problems and data sparsity problems. The cold-start problem is divided into the problem of new student users and new course resources. The new student user problem is that a newly registered student has not yet evaluated and scored the course resources, and there is no corresponding historical browsing record, so the collaborative filtering recommendation model can not predict the interested course resources of the student user, and can not recommend the course resources that the student may be interested in for the student [17].

The data sparsity problem can easily reduce the recommendation quality and recommendation effect of collaborative filtering recommendation model. The collaborative filtering recommendation model is based more on students’ scores on course resources to determine their level of interest in course resources, and then suggests the appropriate course resources for them. When the number of students’ evaluations and scores on course resources is small, the recommendation accuracy of the collaborative filtering recommendation model cannot be guaranteed, and as the number of student users and the number of course resources continue to rise, the problem of data sparsity will continue to expand, and the student-course scoring matrix will become even sparser [18]. Therefore, it is also necessary to optimize the collaborative filtering recommendation model in order to better personalize the recommended course resources for student users.

K-mean clustering is a commonly used segmentation clustering method, which is based on the principle that K random objects in a data set are used as the clustering centers, and other data objects in the data set are automatically grouped into a class with the nearest clustering center based on their distance from these K data objects. These classes are then iterated so that the data objects move through the classes and the average is calculated based on the update of the data in the class and the data objects are reassigned so that the classes are improved until the maximum number of iterations is reached or no more new clusters are generated.

The disadvantage of the K-means clustering algorithm is that it is too dependent on the initial clustering centers and is prone to fall into a local optimum, so it is optimized using genetic algorithms (GA) to enable the GA-K-means algorithm to converge to the best clusters [19]. The steps to optimize the K-means algorithm by genetic algorithm are, firstly, the attributes of the student users are represented by chromosomal binary strings, and a random initial population is generated according to the genetic algorithm, which is used to search for the global optimum. Secondly, the fitness function is used to determine whether the clustering result of the K-means algorithm is the global optimum. Finally, the genetic algorithm uses genetic operations like crossover and mutation to update the initial clustering seeds continuously, and repeats the fitness function judgment and genetic operations until the conditions are satisfied. The fitness function is shown in equation (5): (5)F=∑a=1k∑p=cs|p−ma|2$$F = \sum\limits_{a = 1}^k {\sum\limits_{p = {c_s}} {{{\left| {p - {m_a}} \right|}^2}} }$$

In Equation (5), p is a point in the n-dimensional space that represents the user, m_a represents the clustering center generated by the K-means algorithm, k represents the number of clustering centers generated by the K-means algorithm, and c_a represents the number of optimal clustering centers. The workflow of collaborative filtering personalized recommendation model optimized using genetic K-means algorithm is shown in Fig. 1. The collaborative filtering personalized recommendation model optimized with the genetic K-means algorithm can find a suitable set of neighbors for the target student users based on the initial attributes of the students when they register, such as age, grade, major, gender, etc. Even if the target students have just registered and have not graded any course resources, the collaborative filtering personalized recommendation model can recommend the innovation and entrepreneurship course resources for the target student users, and can better serve the students, and can provide better service for the students. The personalized recommendation model of collaborative filtering can recommend innovation and entrepreneurship course resources to target students even if they have just registered and not scored any course resources.

Figure 1.

Improved collaborative filtering personalized recommendation model

In this paper, the advantages and disadvantages of recommender system related algorithms are evaluated by two metrics: mean absolute error (MAE) and accuracy, and the profile coefficient is used as a criterion for evaluating the clustering effect. 1)

For a user u and item i, r_wi in the test set representing the actual ratings of user u for item i, and r^wi$${\hat r_{wi}}$$ representing the predicted ratings computed by the recommendation algorithm model, MAE the prediction error is computed using an absolute value, which is defined as: (6)MAE=Σu,i∈T|rui−rui||T|$$MAE = \frac{{{\Sigma _{u,\:i \in T}}|{r_{ui}} - {r_{ui}}|}}{{|T|}}$$

Where T′(u) is the number of all objects in the recommended result set. 3)

Contour coefficient is a kind of index for evaluating the good or bad effect of clustering, which can be understood as an index describing the clarity of the contour of each category after clustering. It consists of two factors - cohesion and separation. The degree of cohesion can be understood as reflecting the degree of closeness of a sample point to the elements within the class. Separation can be understood as a reflection of how closely a sample point is associated with elements outside the class.

For any node i, denote by a(i) the average spacing of i from other points of the same class and by b(i) the average spacing of i to points not of the same class, whichever is the lowest. Then the contour coefficient s(i) for i is given by: (8)s(i)=b(i)−a(i)max{a(i),b(i)}$$s(i) = \frac{{b(i) - a(i)}}{{\max \{ a(i),b(i)\} }}$$

The dataset used in this paper is the online open dataset Creative star. In the experiment, 196,352 ratings were selected for 3524 innovative entrepreneurship course videos from 1963 users in the Creative Star dataset. This selection provides a practically meaningful large-scale data environment for performance evaluation of recommendation algorithms, which not only cover a wide range of user groups and diversified course videos, but also have a standardized range of ratings to provide users with a rich options to precisely express their preferences for course videos, thus helping to capture the nuances of user preferences.

By clustering the 1963 users in the Creative Star dataset using the GA-K-means clustering algorithm, parameter optimization experiments are conducted to finalize a value for the number of clusters K. The number of clusters K is calculated as 6 sets of numbers, and a clustering operation is performed for each value of K. The clustering results for each K value are used to calculate the contour coefficient of the sample points using the contour coefficient formula. The contour coefficient values under different number of clusters K are shown in Fig. 2. The GA-K-means clustering algorithm has the highest contour coefficient when the number of clusters is 20, i.e., the algorithm has the highest performance at this time. Therefore, in the next study, experiments will be conducted in cluster number 20.

Figure 2.

The contour coefficient of different clustering

This subsection is intended to experiment with the GA-K-means clustering algorithm proposed in this paper. Using Creative star’s dataset, which is divided into training set (70%) and test set (30%), the number of clusters K is taken as 20, and the number of neighbors is divided into 6 different categories (5-30 with interval of 5), the GA-K-means clustering recommendation algorithm, traditional collaborative filtering algorithm (Method 1), personal feature similarity weighted recommendation algorithm (Method 2), collaborative filtering algorithm based on matrix decomposition (Method 3), and collaborative filtering algorithm based on K-Means clustering (Method 4) were analyzed and compared from the point of view of average absolute error (MAE) and accuracy.

4.2.1

Mean Absolute Error of Recommendation for Different Number of Neighbors

Calculate the average absolute error (MAE) of GA-K-means clustering recommendation algorithm, K-Means clustering based collaborative filtering algorithm, personal feature similarity weighted recommendation algorithm, traditional collaborative filtering algorithm, and matrix decomposition based collaborative filtering algorithm when the number of nearest neighbors is different to examine the effect of different number of nearest neighbors on the recommendation results, and the comparison of experimental results of MAE values of the five algorithms is shown in Figure 3. . With the increase of the number of neighbors, the MAE value of each algorithm has a tendency to rise and then level off, but the GA-K-means clustering recommendation algorithm always maintains a relatively low level relative to the other four algorithms, with the MAE value ranging from 0.652 to 0.688. This shows the stability and superiority of this algorithm in dealing with the number of neighbors of different sizes, and also indicates that the score predicted by the improved algorithm in this paper is closer to the actual score.

Figure 3.

Compared with the MAE value experiment of five algorithms

4.2.2

Accuracy of Recommendation with Different Number of Neighbors

The accuracy rates of five algorithms’ recommendations are calculated when the number of nearest neighbors varies to examine the impact of different proportions of the number of nearest neighbors on the recommendation results, and a comparison of the accuracy experimental results of the five algorithms is shown in Figure 4. The results show that the GA-K-means clustering recommendation algorithm in this paper demonstrates a higher accuracy rate compared with the other four algorithms, and when the number of nearest neighbors is increased to 30, the accuracy rate under this paper’s method is 90%. This means that the algorithm is able to capture the user’s points of interest more accurately and provide more relevant recommendations.

Figure 4.

The accuracy of the five algorithms was compared

The purpose of this subsection is to investigate the educational impact of the online educational course recommendation system based on collaborative filtering technology applied to the innovative entrepreneurship education curriculum system. A longitudinal design is adopted to measure the entrepreneurial attitudes and entrepreneurial self-efficacy of Z-school juniors at the beginning and the end of the semester after the optimization of the innovation and entrepreneurship education curriculum system by means of “pre-test-post-test” comparisons.