E-learning Based Data Mining System for Chinese Teaching Information

China's talent policy has a tangible implementation link in higher education. China has actively pushed for educational reform and the use of technology in the classroom in recent years [13]. It will increase the likelihood that higher education will grow [26]. According to the survey, most colleges and universities only use a single application for their network platforms, and most are now used as major colleges and universities' teaching or network library platforms. The survey also revealed that most platforms are not being used as intended for teaching [11]. It is impossible to fully replace manual teaching with a network teaching platform, given the current state of technology and the requirements of the educational tasks. Nonetheless, the network teaching platform can be used as a tool for manual instruction for teaching optional courses.

The swift advancement of contemporary information technology, encompassing big data, artificial intelligence, and mobile Internet, has become a crucial catalyst for transformative educational changes. Big data technology offers tailored and intelligent learning, artificial intelligence technology stimulates the development of new Chinese teaching concepts and learning methodologies, and the widespread use of mobile Internet brings the Chinese learning environment closer to students. Chinese instruction must adapt its traditional methods and reverse the teacher-student dynamic to shift the teacher from a leader to a guide and make the students the primary factor influencing the effectiveness of the instruction. This is due to the influence of contemporary teaching technologies such as the flipped classroom, large-scale online open classroom, and microgrid classroom. Due to multimedia classrooms and the Internet, college and university teaching environments have evolved from closed, insular spaces to intelligent, networked, and digitalized open spaces. To change and investigate the direction of Chinese teaching from a fresh angle, Chinese teachers should, therefore, take into account the usage of contemporary information technology to improve and reform traditional Chinese teaching.

The symbiotic link created by numerous species between instructional elements and resources is known as teaching information data. The principles of symbiosis, evolution, and optimal resource use are at their foundation. Following the implementation of this system in Chinese instruction, its instructional materials will serve as the cornerstone for constructing a symbiotic connection in which mutual influence and direction are essential for preserving the teaching data. Numerous elements, including students, teachers, multimedia, resources, classrooms, schools, and society, are viewed as interrelated in Chinese teaching information data. The ever-evolving culture and technology make it impossible for traditional Chinese teaching approaches to keep up, and a new ecosystem of Chinese instruction needs to be created.

To raise the caliber and standard of Chinese instruction, it is clear from an analysis of this literature that researchers have mostly concentrated on the experiences of teachers, collaborative practice, teaching methodology, Kinect-based somatic learning system, ecological teaching model, innovative teaching model, assistive teaching system, etc.

To create a teaching norm that is marked by direction, motivation, and active participation, Chinese teachers should adhere to the law of learning, adopt suitable and efficient teaching techniques, completely take into account each student's unique learning preferences and individual variances by the course topic, and embody the teaching concept. Enhancing teaching strategies should also incorporate the most recent findings from applied linguistics research, updating concepts often and employing instructional strategies catering to individual student's requirements. Students are overwhelmed by today's digital technologies' abundance of learning options. These pupils put in a lot of time, but their learning is subpar. Thus, a pressing issue in Chinese teaching is balancing the interactions between students, teachers, the learning environment, and learning methodologies. For this reason, big data technology and artificial intelligence techniques are included in Chinese education, and data mining techniques are used based on an analysis of the features and developmental stage of Chinese teaching. It is envisaged that by using association rule algorithms, it will be possible to identify the critical elements influencing teaching quality, offer a strong foundation for learning arrangement and teaching administration, and aid in improving Chinese instruction.

Finding information and hidden knowledge within a large volume of noise, ambiguous, random, incomplete, and unknown data that may be helpful for data fusion, analysis, and decision support is known as data mining. Knowledge discovery is a familiar concept, and the database industry uses the term. Relational databases are the primary objects of data mining. These data are organized. Generally speaking, data mining is the act of modeling and identifying connections among vast amounts of data that may be utilized with a range of analytical and analytical tools to generate predictions and conclusions. An essential area of study in data mining research is the support of large-scale data analysis procedures and methods and choosing or creating an appropriate one.

Among the activities carried out in data analysis include time series models, clustering, classification, outlier identification, forecasting, and correlation analysis. A strong structural design is necessary for these functions to be realized. College Chinese test data has been subjected to data mining. The prototype data warehouse system served as the foundation for the development of the system. See in Figure 1.

Figure 1.

Data mining foundations.

Data integration and cleansing are just two of the many data transformation techniques needed for the data preprocessing stage. The typical technique for cleansing data is a mean approach.

Figure 2.

College Chinese test data mining system flowchart

The w_i representation of the data point weights is more significant than the averaging technique.

After data cleaning is finished, data normalization is necessary. Max-min normalization is frequently utilized [12]: (2) v′=v−minAmaxA+minA(new_maxA+new_minA)−new_minA $$v' = {{v - {{\min }_A}} \over {{{\max }_A} + {{\min }_A}}}(new\_{\max _A} + new\_{\min _A}) - {\rm{new}}\_{\min _A}$$ The completion of data preprocessing is seen as completing data sampling.

After the data has been prepared, the computation for data mining is codified. The two primary techniques are the k-means algorithm for cluster analysis and the decision tree algorithm for classification.

Each sample in a data sample has its expected information computed as follows: (3) I( s1, s2, s3,…, sm)=∑i=1m+1pilog2(pi) \[I(~{{s}_{1}},~{{s}_{2}},~{{s}_{3}},\ldots ,~{{s}_{m}})=\underset{i=1}{\overset{m+1}{\mathop \sum }}\,{{\log }_{2}}({{p}_{i}})\] P_i denotes the likelihood that a given sample is a member of C_i. A is the test attribute, and each sample has m attributes. After splitting A into subsets, the entropy is computed as follows: (4) E(A)=∑j=1vs1j+...+amjsI(s1j,…,smj) \[E(A)=\underset{j=1}{\overset{v}{\mathop \sum }}\,\frac{{{s}_{1j}}+...+{{a}_{mj}}}{s}I({{s}_{1j}},\ldots ,{{s}_{mj}})\] Where v denotes how many attribute values there are in A. The purity increases with the subset's entropy decreasing. Given a subset, the following exist: (5) I( s1j, s2j, s3j,…, smj)=−∑i=1m+1pijlog2(pij) $${\rm{I}}({{\rm{s}}_{1{\rm{j}}}},{{\rm{s}}_{2{\rm{j}}}},{{\rm{s}}_{3{\rm{j}}}}, \ldots ,{{\rm{s}}_{{\rm{mj}}}}) = - \mathop \sum \limits_{{\rm{i = 1}}}^{{\rm{m}} + 1} {{\rm{p}}_{{\rm{ij}}}}{\log _2}({{\rm{p}}_{{\rm{ij}}}})$$ The branch receives coded data, which consists of: (6) Gain(S,A)=I(S1,S2,…,Sm)−E(A) \[Gain(S,A)=I({{S}_{1}},{{S}_{2}},\ldots ,{{S}_{m}})-E(A)\] Where benefit (A) is the anticipated entropy compression brought about by being aware of the value of attribute A.

The primary method for clustering is the k-means algorithm. When n objects are classified into k clusters using k as a parameter, the degree of similarity between the clusters increases and decreases. The cluster's mean value—regarded as its center of gravity—is used to compute the similarity between the objects in the cluster. Typically applied to similarity, the function is defined as follows: (7) E=∑i=1K∑p∈Ci|p−mi|2 \[E=\underset{i=1}{\overset{K}{\mathop \sum }}\,\underset{p\in {{C}_{i}}}{\mathop \sum }\,|p-{{m}_{i}}{{|}^{2}}\] One such way data mining is used in Chinese instruction is to create a diagnostic internship system for university students using SPSS data processing software. The primary technique combines the test paper algorithm and knowledge point correlation analysis.

First, the data are categorized. We must examine the association rules between various knowledge points in light of different sample sizes if we are varying sizes. The project team, therefore, divided the percentage of each knowledge point into four groups before mining based on the sample size of each knowledge point. Table 1 displays the outcomes of the stratification.

After extracting the association rules, various maximum numbers of front-end items are also selected for testing to increase the results' correctness and dependability. The outcomes are displayed in Figure 3.

As the maximum number of anterior elements is achieved, Figure 3 illustrates how the number of excavation correlation rules rises and then tends to fall. At that point, the maximum number of frontend terms increases while the number of association rules stays constant. The experiment demonstrated that while an excessive number of precondition terms affects excavation efficiency and confuses exam paper recommendations, it does not affect the number of relational rules to be excavated. While maintaining mining efficiency, the test approach enhances the final association rule mining outcomes.

Figure 3.

Experimental folding of the maximum number of antecedent words setting in a line graph.

Among the techniques used in data analysis include databases, neural networks, statistical techniques, and machine learning. Genetic algorithms and inductive analysis techniques (decision trees, rule generalization, etc.) are examples of machine learning techniques. Examples of statistical techniques include multiple regression, autoregression, and discriminant analysis techniques (Bayesian, Fisher, and non-parametric). Feedforward neural networks (BP algorithms) and self-organizing neural networks are examples of neural network techniques. Multivariate data analysis and OLAP techniques are examples of database techniques.

The boosting approach is its foundation; in each iteration, new decision trees are constructed to decrease the residual gradient, progressively enhancing the system's capacity for generalization. Decision trees using gradient boosting as their foundation may recognize distinguishing. Consequently, contextual attributes that impact the user's preferences can be identified to understand the user's needs better and provide more individualized information recommendations. Figure 4 displays the conceptual diagram of the decision tree data mining.

Figure 4.

Decision tree data mining conceptual diagram.

With contextual attributes, every context occurrence is discretized, converted into an input attribute, and fed into a decision tree with gradient enhancement. The meaning of learning support services has to change in the big data and artificial intelligence era. Personalized Chinese language instruction, course administration, and learning assessment services will replace conventional, fixed, and uniform learning support services. Using a greedy top-down technique, each decision tree in the gradient tree selection method isolates the features that best represent the classification effect at each node.

As a result, the references for this StudyStudy use each context instance to calculate the context instance's influence: (9) t. { ∑r=1suryrj∑i=1mvixij≥1(j=1,2,3,…,n)vi≤0(i=1,2,3,…, m)ur≤0(r=1,2,3,…, s) \[t.~\{\begin{matrix} \frac{\mathop{\sum }_{r=1}^{s}{{u}_{r}}{{y}_{rj}}}{\mathop{\sum }_{i=1}^{m}{{v}_{i}}{{x}_{ij}}}\ge 1(j=1,2,3,\ldots ,n) \\ {{v}_{i}}\le 0(i=1,2,3,\ldots ,~m) \\ {{u}_{r}}\le 0(r=1,2,3,\ldots ,~s) \\ \end{matrix}\] Based on the user's preference information for choosing information resources, it is assumed that the GBDT algorithm generates. Context instances are then obtained under the context attribute, and the scenario instances are used to determine the extent to which the context instances contributed: (10) Q={Y=x1a1+x2a2+x3a3+x4a4+x5a5,x1+x2+x3+x4+x5=1,0<x<1 $${\rm{Q}} = \{ \matrix{ {{\rm{Y}} = {{\rm{x}}_1}{{\rm{a}}_1} + {{\rm{x}}_2}{{\rm{a}}_2} + {{\rm{x}}_3}{{\rm{a}}_3} + {{\rm{x}}_4}{{\rm{a}}_4} + {{\rm{x}}_5}{{\rm{a}}_5},} \cr {{{\rm{x}}_1} + {{\rm{x}}_2} + {{\rm{x}}_3} + {{\rm{x}}_4} + {{\rm{x}}_5} = 1,} \cr {0 < {\rm{x}} < 1} \cr } $$ The following formula determines the node and node Gini indices, and the mean value of the Gini coefficient remains unchanged: (11) { vij(t−1)=vij(t)+c1r1(t)(pij(t)−xij(t))+c2r2(t)(pgj(t)+xij(t)),mij(t−1)=mij(t)+vij(t−1) \[\{\begin{matrix} {{v}_{ij}}(t-1)={{v}_{ij}}(t)+{{c}_{1}}{{r}_{1}}(t)({{p}_{ij}}(t)-{{x}_{ij}}(t))+{{c}_{2}}{{r}_{2}}(t)({{p}_{gj}}(t)+{{x}_{ij}}(t)), \\ {{m}_{ij}}(t-1)={{m}_{ij}}(t)+{{v}_{ij}}(t-1) \\ \end{matrix}\] Typically, information gain is used to determine which decision tree characteristic is prioritized: (12) Z1=-0.002 Z2-0.0869 Z3+1.6 Z4+0.385 Z5 \[{{Z}_{1}}=-0.002~{{Z}_{2}}-0.0869~{{Z}_{3}}+1.6~{{Z}_{4}}+0.385~{{Z}_{5}}\] Furthermore, the ideal qualities are frequently categorized using the Gini index and the gain ratio, with the latter represented by the symbol (13) maxhj0=∑r=1nuryrj0∑i=1mvrxrj0 \[\max {{h}_{j0}}=\frac{\mathop{\sum }_{r=1}^{n}{{u}_{r}}{{y}_{rj0}}}{\mathop{\sum }_{i=1}^{m}{{v}_{r}}{{x}_{rj0}}}\] Decision trees are frequently trimmed to increase generalization and avoid overfitting: (14) t=1∑i=1mvixijwi=tviμr=tur \[t=\frac{1}{\mathop{\sum }_{i=1}^{m}{{v}_{i}}{{x}_{ij}}}{{w}_{i}}=t{{v}_{i}}{{\mu }_{r}}=t{{u}_{r}}\] Pre-pruning and post-pruning are two types of pruning. Pre-pruning is the process of estimating, using information gain calculations, whether keeping nodes will enhance the model's capacity for generalization. The method of post-pruning involves starting with the lowest leaf nodes and working your way. Examine each node and determine whether to keep it. The degree of affiliation indicates how much a sample belongs to a specific category when it can be classified into multiple categories simultaneously. Fuzzy mathematics can be used to express the ambiguity of objects and connections. Based on this, a fuzzy fault detection model is designed better to handle the intricate relationships between fault sources and signals.

Given the state of technology and the demands of instructional duties, it is not practical to fully replace manual instruction with online teaching platforms. Nonetheless, the network teaching platform can be used as an additional tool for manual instruction or even a platform for teaching optional courses. A sophisticated network teaching platform is becoming increasingly necessary for modern instruction due to the rapid advancement of network technology. Network teaching systems must be created and designed in response to this need and progressively included in instructional activities. Additionally, college students' educational experiences must be elevated. The learning system's structure is built on a three-layer B/S structure selected for its superior functionality and smoother functioning. More and more educators are starting to see the benefits of e-learning due to the advancements in information and network technology, and an increasing number of e-learning systems are being used in classroom instruction.