Pathways to Improve the Effectiveness of Multidimensional Data Analysis in Decision Support for Instructional Management

The data described by the relational data model is described through two dimensions: “type” and “value”: “type” represents the basic description structure of the data, and its design follows the basic paradigm of the relational data model; The Value represents the specific traces of data generated during the access to the system. Together, “type” and “value” construct a relational data structure, and the relationship between the data can be described as tables that meet the appropriate relationship pattern, and the tables are associated with each other through “foreign keys” defined by attributes. Relational structures are good for representing descriptive and defining data, and the model can be used for teachers, learners, and other participants and the relationships that exist between them.

Resource Description Framework (RDF) is a structure that represents data in graphs and is suitable for representing decisions and rules of online platform Web business data. At the initiation of the decision-making claim of intelligent teaching management, the platform should construct the business process in time and autonomously, and construct the RDF corresponding to the decisions and rules in due time. Here, take the decision-making feedback of the learning process as an example, the whole process involves three types of participants, namely, the learner, the teaching management personnel and the teaching decision-maker, and the decision-making generates a process that will be associated with the first layer, the second layer, and the third layer, and the process contains a number of attributes of the relevant use cases and the guidance and participation links between resources, then the process can be described as a complete RDF, usually expressed in RDF process is relatively simple.

Knowledge Attribute Graph is a data model for representing autonomous relationships, such as social network relationships. Compared with RDF, it can embed built-in services for the attributes of data and the attributes of relationships, and has a richer way of describing the intrinsic mutual constraints and influencing relationships of data. KAG can describe the activity sequences of use cases and the interaction and collaboration relationships, and with the help of concepts such as labels and association attributes, it can give the graph vertices and richer information about the roles. The objects involved in an activity are its roles, the meaning of the relationship is expressed with the help of labels, and the cause and effect of the activity involvement is pinpointed with the help of association attributes.

Traditional university teaching management mainly completes the management of a series of mathematical activities from the professional teaching plan to the semester course plan, from the scheduling of classes to the examination arrangements, from the generation of student grades to the statistical analysis of course grades. Therefore, it mainly includes the following aspects of the module shown in Figure 1.

Figure 1.

Teaching management function module

The traditional academic affairs management system consists of a university-level academic affairs management subsystem and a faculty-level academic affairs management subsystem. The university-level academic affairs management subsystem faces the academic affairs department of the school, focusing on the management and maintenance of academic affairs data, and data analysis and statistics from an overall perspective. Faculty-level teaching management sub-system for faculty teaching staff, focusing on faculty teaching data entry, query, printing, data analysis and data statistics from a local perspective.

Poor data integration and data inconsistency between traditional university teaching management systems are particularly serious. The original database system is difficult to adapt to the different requirements of data granularity of various types of objects. For example, students’ course learning is gradual, and there is a certain correlation between the courses and the sequential relationship. In the study of a more advanced courses must be studied before which courses, if the first course did not learn well, will inevitably affect the learning of subsequent courses and other issues. Higher vocational colleges and universities should synchronise their professional settings and curriculum systems with the needs of society, which requires schools to revise their teaching plans in a timely manner and replace courses that are no longer suitable for the market. The traditional teaching management system is difficult to correlate the above demands and analyse the reasonableness of the curriculum.

Higher education teaching management decision support system is able to use the traditional university teaching management system using the database on the basis of the use of data warehousing technology, OLAF technology and data mining technology to analyse the following aspects of teaching: performance decision analysis, faculty decision analysis, decision analysis of teaching quality assessment, decision analysis of the selection of the course arrangements, decision analysis of teaching workload, the professional course settings decision analysis, enrolment decision analysis, student employment analysis and so on. The data processing within the current teaching management system can be roughly divided into two categories: operational processing and analytical processing. Operational processing, also called transaction processing, refers to the daily operation of the database on-line, usually the query and modification of a record or a group of records, mainly for the specific application of the teaching implementation of the service, people are concerned about the response time, data security and integrity. Analytical processing, on the other hand, is used for decision analysis by administrators, often accessing large amounts of historical data and it requires high speed data processing - instantaneous processing. The conflicting data needs lead to different requirements for data management and storage. The architecture of this solution applied in the education industry is shown in Figure 2.

Figure 2.

Architecture of university teaching management decision support system

ETL in the data warehouse is the most important data collection tools, which includes data extraction, transformation, loading and cleaning, they jointly constructed the data warehouse, the more mainstream ETL tools on the market include Powercenter and Datastage and so on. Teaching management data warehouse for colleges and universities, it is difficult to establish a function of the content of the most complete ETL tool system, because the data mobility of colleges and universities, so the establishment of a small-scale and time nodes of the ETL tool, the establishment of the data market, should be the key to the development of teaching management decision support system in colleges and universities.

According to the teaching decision-making system business needs and the reality of the existing teaching information system data, decision-making system days ago there are many decision-making analysis theme, here mainly introduces the three analysis theme and the design of the decision-making theme data analysis model: 1)

School registration information data analysis model

School registration information data analysis model consists of regional dimension table, teaching unit dimension table, national dimension table, professional dimension table and school registration information fact table constitutes a star structure.

Data collection tools are used in data warehouses with the characteristics of applicability, extensibility and flexibility, and data sampling and conversion can be achieved with hand-written code. So it is necessary for college teaching to design ETL tools with the help of data warehouse, for example, the following is designed as an ETL tool module for managing academic information, and its function procedure should be: 1)

Procedure SET_SDblink()//apply to set the meta-database link.

1)

ID3 algorithm

The advantages of the ID3 algorithm are: the theory of the algorithm is clear, the method is simple, and the learning ability is strong. The disadvantage is that it is only effective for smaller data sets and is more sensitive to noise, and when the training data set is increased, the decision tree may change accordingly.

The C4.5 algorithm is a further invention by Quinlan that improves on the ID3 algorithm by introducing the concept of information gain rate, which is able to deal with consecutive attributes as well as the problem of attribute value vacancies. Some of the problems related to C4.5 are described below:

Concept 1 defines n messages that occupy the same probability, where the probability of a single message is P = 1/n, then I(P)=−∑nPiLog2(Pi) is called the amount of information transmitted by the message.

Concept 2: When there is an event y_j, followed by the probability of the occurrence of time x_i in the case of y_j is the conditional probability, this result is p(x_i ∣ y_j), we use the negative of the logarithm of the conditional probability to calculate the amount of conditional self-information, that is, the entropy of the decision tree algorithm, which represents a kind of uncertainty. As shown in equation (1): (1)I(xi|yj)=−∑i,j=1nP(xi|yj)Log2P(xi|yj)

Concept 3: Conditional entropy, which is the uncertainty of a random variable under a condition, is found according to equation (1). Given the outcome value X, the conditional entropy under condition y_j is shown in equation (2): (2)H (X,yj)=∑p(xi | yj)*I(xi | yj)

Concept 4: Use X to represent the outcome value and A to represent one of the decision attributes. Based on the above three concepts the information gain can be calculated as in equation (3): (3)Gain(X,A)=H(X)−H(X,A)

In the improved C4.5 algorithm, let: there are m positive example attribute and n negative example attributes in the category attributes. Then it is informative: (4)I(m,n)=−mm+nlog,mm+n−nm+nlog2nm+n

If A is used as a property of the test and A has p different values, then: (5)E(A) = ∑i=1pmi+nim+n*(−mimi+nilogmimi+ni−nimi+nilog2nimi+ni) = 1(n+m)ln2∑i=1p(−yilnmini+mi−nilnnini+mi)

For simplicity of calculation, the 1(n+m)ln2 to be counted at each step is omitted as a constant.

Then: (6)E(A)=∑i=1p(−milnmini+mi−nilnnini+mi)

Using the formula in higher numbers, if x is small, then ln(1 + x) = x, which can be obtained as follows: (7)lnmini+mi=ln(1−nini+mi)≈−nini+mi (8)lnnini+mi=ln(1−mini+mi)=−mini+mi

Simplifying Eq. (8) with Eqs. (6) and (7) gives the following: (9)E(A)=∑i=1p[−mi*(−nini+mi)−ni*(−mini+mi)]=2∑i=1pnimini+mi

By omitting the constant, the: (10)E(A)=∑i=1pnimini+mi

As: (11)SplitInfo(A,D)=−mimi+nilogmimi+ni−nimi+nilog2nimi+ni≈∑i=1pmini+mi

The formula to derive the information gain rate is as follows: (12)GainRatio(A,D)=SplitInfo(A,D)−E(A)SplitInfo(A,D)

The distribution of evaluation scores, divided into five intervals and statistical distribution of evaluation scores: 18% of the courses with scores of 90 and above, 65% of the courses with scores of 85-90, 15% of the courses with scores of 80-85, 1% of the courses with scores of 75-80, and 1% of the courses with scores of 75 or below. The following courses accounted for 1%, indicating that the overall rating of the participating students for the courses offered is high, and the ratings show a normal distribution trend.

Statistical analysis of the evaluation scores of different course natures, using SPSS software to analyse the mean and standard deviation of the ratings of courses of different natures in the two semesters, the statistical results are shown in Table 1.

Through Table 1, it is found that: firstly, the standard deviation of most of the courses is small, between 1-3 points, with little difference, and the types of courses with large standard deviation include the compulsory public foundation courses, the elective courses of subject foundation, and the elective courses of practical sessions, indicating that there is obvious variability in students’ ratings for these three types of courses; secondly, the types of courses with the bottom two rankings in the mean value are the compulsory public foundation courses (with the rating The largest variance), general education elective courses, in which: the public basic compulsory courses include English, high mathematics, probability, linear algebra and other basic courses, such courses have an important role in supporting the students’ professional course learning and further study, the students’ expectation level is higher, at the same time, the difficulty of passing such courses is relatively large, there will be a certain degree of failing the course phenomenon, so the lower ratings indicate that the students’ expectations on the At the same time, these courses are relatively difficult to pass, and there is a certain degree of failure, so the lower ratings indicate that students’ expectations have a greater impact on the ratings, and also indicate that the ratings of these courses are more polarised, and that the teaching method of tailoring the teaching to the students’ needs needs needs to be explored.

Through the students’ evaluation data of teachers, after each comment is processed, the corresponding emotional tendency is obtained, and then added into each pre-processed comment Here, the value of the minimum support degree and the minimum confidence degree is set to 0.01, and about 896 association rules are obtained According to the actual situation, after careful screening, the association rules shown in Table 2 are obtained. From the table, it can be seen that the vast majority of the emotional tendency obtained from the evaluation of teaching is positive (confidence level close to 1), indicating that the vast majority of teachers in this school are of high level, strong teaching ability, appropriate teaching methods, and liked by the majority of students. It can also be seen that the teaching styles and methods that are liked by the students are: lecturing with humour, vividness, seriousness, and have a patient and careful style. However, there are also individual correlation rules that show that there are some styles of lecturing that are not liked by the students. Therefore, it is recommended that teachers, in addition to enriching their own teaching ability and basic knowledge, should find ways to improve the atmosphere of learning and discussion in each lesson, and combine their own teaching methods and methods with their own years of teaching, and call on the students’ motivation to learn, which is a win-win situation for both teachers and students.

To verify the algorithmic gap between the C4.5 algorithm before and after improvement, the confusion matrix of the C4.5 algorithm before and after improvement is shown in Table 3-Table 4. Comparing the two tables it is found that the number of correct classifications of C4.5 algorithm increases by 5 after improvement. Four evaluation metrics, accuracy, precision, recall and F1-Measure value, are calculated based on the confusion matrix to assess the performance of the C4.5 algorithm before and after improvement. The performance comparison of the post C4.5 algorithm before and after improvement is shown in Fig. 3. From the figure, it can be seen that the improved C4.5 algorithm improves the accuracy rate by 1.32% relative to the original C4.5 algorithm, the precision rate by 0.91% relative to the original C4.5 algorithm, the recall rate by 3.07%, and the F1-Measure by 1.81% relative to the original C4.5 algorithm. Summarising the results of the experimental evaluation, the performance of the final stage grade prediction model constructed by the improved C4.5 algorithm is better than that of the final stage grade prediction model constructed by the original C4.5 algorithm.

Figure 3.

The C4.5 assessment index was compared before and after the improvement

The following training of the learning data before the deadline of the students’ weekend achievement test generates the decision tree for the third stage as shown in Fig. 4.

Figure 4.

Decision tree generated in the fifth stage

The learning data before the students’ end-of-month test deadline was trained to generate the decision tree for the fifth stage, and the decision tree generated in the fifth stage is shown in Fig. 5. Since the tree constructed in the fifth stage is relatively large, only the left branch is shown.

Figure 5.

Decision tree generated in the third phase

For the generated stage 3 decision tree and stage 5 decision tree, one path from the root node to the leaf node at a time is a classification rule, and all the classification rules are integrated together as the stage performance prediction model.

When evaluating a multiclassification model, instead of evaluating the model with precision, recall, F1-Measure, and accuracy for a specific class, the stage-achievement prediction model is evaluated by the accuracy of the model as a whole. The accuracy rate is calculated by the formula: (18)Accuracy=correctClassification (testSet)total(testSet)

Where total(testSet) denotes the total number of test sets and correctClassification(testSet) denotes the total number of correct classifications in the test set.

The accuracy of each stage performance prediction model can be calculated according to Equation (18), and the stage performance prediction model accuracy is shown in Figure 6.

Figure 6.

The accuracy of stage grade prediction model

The figure shows that with the increase in the number of learning behavioural features at each stage, the stage-based performance prediction model accuracy rate is roughly on the rise, but most of the overall accuracy rates are around 52%. There are two main reasons for the low prediction accuracy of the stage-based grade prediction model: (1) Because the stage-based grade prediction model predicts the final grade interval rather than the final grade pass or fail, and predicts the final grade interval in more detail, the stage-based grade prediction model has a lower accuracy rate. (2) The accuracy is relatively low due to incomplete characteristics of the stage-based learning process.