A study of constructed paths based on big data analysis facing the framework of modern and contemporary literature education

A concept word is a terminological lexicon used primarily as a label for concepts. It consists of a concept name and an alias or synonym that describes the concept at the word level.

The concept name is the identification of the concept, and what should be paid attention to in its naming is accuracy and uniqueness. 1)

Accuracy should be able to exactly express the substantive meaning of the concept, avoid ambiguity, and adopt formal names and scientific names as far as possible. Generally speaking, the subject of the knowledge is a high concept of knowledge, often a very concise few words, such as “modern Chinese language”, “ancient literature”, etc., as far as possible, the subject of the knowledge as the name of the concept.

Attributes are static characteristics of the thing represented by the concept. Attributes are data items that reflect the essential features of the concept. Attributes are represented as sets, with n component representing n sides to characterize the concept. 1A={a1,a2,⋯an}$$A = \left\{ {{a_1},{a_2}, \cdots {a_n}} \right\}$$

The discovery of attributes lies in the care and observation of things. In general, what one cares about may be used as an attribute; what one observes from a perspective may be used as an attribute from that perspective. In an instructional system, attributes can be selected from a perspective that is applicable to instruction. The selection of attributes can be based on the following principles. 1)

Distinguish the feature saliency queue, keep the most significant if the thousands, ignore the others, for example, in “C Language Programming”, “parameter” is a significant attribute of “function”.

A learner can favorite a page of the courseware when using the mobile client for online learning. Therefore, it is possible to determine which pages contain key points or general knowledge points of interest to that learner by the status of the learner’s favorites of the pages, and record the set MR={gp|Limarkedgp}$$MR = \left\{ {{g_p}|{L_i}} \right.marked\left. {{g_p}} \right\}$$ of pages that are favorited by learner L_i, where g_p represents page p.

If there is only one learner, then when his dwell time on consecutive pages exceeds the threshold or collects consecutive pages, then it will be determined that these consecutive pages contain the same knowledge point. If there are only two learners and their collections of consecutive pages are A and B and A ∩ B ≠ ∅ respectively, then the collections can be merged to A ∪ B and considered A ∪ B to contain the same knowledge point together. However, when the number of learners increases indefinitely, pages in the courseware that do not contain the same knowledge point may be merged and considered to contain the same knowledge point. Therefore, when the number of learners increases, the judgment of whether consecutive pages contain the same knowledge point becomes stricter. Therefore, we can maintain a list of consecutive numbers S={s1,s2,…,sP−1}$$S = \left\{ {{s_1},{s_2}, \ldots ,{s_{P - 1}}} \right\}$$ and P as the total number of pages in the courseware, and whenever a learner collects g_p and g_p+1 pages consecutively, or when the residence time of g_p and g_p+1 pages exceeds the threshold, s_p = s_p + 1. Finally, we divide each consecutive number in the list of consecutive numbers s_p by the number of learners m to get the list of consecutive numbers S′={s1m,s2m,…,sP−1m}$$S\prime = \left\{ {\frac{{{s_1}}}{m},\frac{{{s_2}}}{m}, \ldots ,\frac{{{s_{P - 1}}}}{m}} \right\}$$, and when spm≥0.2$$\frac{{{s_p}}}{m} \geq 0.2$$, then it is determined that g_p and g_p+1 pages contain the same knowledge point.

The key points are the important knowledge in the courseware that is present throughout the chapter or even the entire course, and most learners will bookmark the page containing the key points; the general knowledge points are less important than the key points. Most learners do not add it to their favorites, but it is still connected to subsequent knowledge.

The reason for distinguishing between key and general knowledge points in this paper is that it is difficult and unnecessary to find connections between key knowledge and other knowledge points, so only connections between general knowledge points will be explored. Because the focus is usually related to most of the content of the entire course or the current chapter, it is difficult to identify all of these connections. And because it is the focus, being added to the collection by the learner, it can be found easily, so it is not necessary to find out the association between the real focus of the course and the other one or two pages to prompt the learner. Let the collection of focused knowledge be F, then there is: 2F={ip|NpA≥0.65}$$F = \left\{ {{i_p}\left| {\frac{{{N_p}}}{A}} \right. \geq 0.65} \right\}$$

Where N_p is the number of learners who have bookmarked i_p knowledge points and A is the number of learners who have bookmarked enough knowledge points, i.e., there are learners who are not very fond of bookmarking pages when they are doing m-learning, so the number of this group of learners will not be added in the denominator. By counting the recorded data, it can be assumed that learners who add less than 2 favorites per 100 pages are learners who are not active in favoriting pages, i.e.: 3Np=card({Li|ip∈Li})$${N_p} = card\left( {\left\{ {{L_i}|{i_p} \in {L_i}} \right\}} \right)$$ 4A=card({Li|card(Mi)≤P⋅0.02})$$A = card\left( {\left\{ {{L_i}|card\left( {{M_i}} \right) \leq P \cdot 0.02} \right\}} \right)$$

Knowledge point is a localized collection of relatively complete knowledge of logical significance, which is the basic unit for transmitting teaching information in the process of teaching and learning, and it includes theories, principles, concepts, definitions, examples and conclusions.

In Internet technology, information may be presented in the form of graphics, images, documents, audio, video, etc., but each form of presentation has certain limitations. Therefore, most of the online course knowledge points combine these presentation forms to achieve optimal teaching and learning results.

In the learning process, it is difficult to achieve the ideal learning effect by learning a certain knowledge point alone without combining the relationship between that knowledge point and other knowledge points. Therefore, the relationship between knowledge points is a research hotspot in both traditional and online education. Learners can form an in-depth understanding of the systematic structure of knowledge points by mastering the relationship between knowledge points. The relationship between knowledge points is generally classified into the following types. 1)

Parent-child relationship: A composite knowledge point is formed by combining several knowledge points, and the relationship between the composite knowledge point and these knowledge points is a parent-child relationship. And the father can participate in the composition of other composite knowledge points, which forms a hierarchical structure between the knowledge points, a course in the knowledge points usually have this hierarchical structure.

If there is a pattern between two or more variables, then the pattern between them can try to be explained by association. Association rules are often used to mine the intrinsic correlation between different items in an event, to extract the hidden knowledge structure, and will not change the original distribution of the sample, to avoid the “dimensionality catastrophe” caused by too many labeled items. Association rule mining usually refers to extracting the common probability of occurrence between items or entities, in order to extract potential behavioral patterns of users. Therefore, association rules are commonly employed to analyze the shopping or browsing behaviors of users. For example, in the famous story of “Beer and Diapers”, there is a 75% probability that a customer who buys diapers at the supermarket will buy beer at the same time. The results of this association rule extraction prompted the supermarket to place diapers and beer in close proximity to each other, which proved to be a significant increase in the sales of both. The effectiveness of association rules in uncovering implicit connections between events is self-evident.

From the principle of association rule, suppose D is the whole set of transactions of geographic information domain literature database, I={I1,I2,…IM}$$I = \left\{ {{I_1},{I_2}, \ldots {I_M}} \right\}$$ is the set of all items of M, and T={T1,T2,…,Tk}$$T = \left\{ {{T_1},{T_2}, \ldots ,{T_k}} \right\}$$ denotes the set of transactions, in which each transaction T_i is a non-empty item, i.e., the transaction T is the set of different items such that T ⊆ I. Then the association rule of the whole set of transactions of geographic information domain literature database can be expressed as i_s ⇒ i_t, in which i_s is the antecedent term of the association rule and i_t is the rule is the latter term of the association rule. For item set i_s and item set i_t in i_s ⊂ I, i_t ⊂ I, i_s ∩ i_t = ∅ under the condition of i_s ⇒ i_t, that is to say, it means that when item set i_s appears in a certain piece of literature, it will lead to item set i_t also appears in the literature with a certain probability. It is heavily used by scholars in text mining, internet, biomedical and other fields.

The association rule contains three metrics: support, confidence and enhancement. The support of the association rule is expressed as the ratio of the number of documents containing both itemset i_s and itemset i_t in the transaction set to the ratio of the entire geographic information domain literature database transaction set |D|$$\left| D \right|$$, i.e., the probability of the simultaneous occurrence of itemset i_s and itemset i_t. Confidence, on the other hand, is expressed as the ratio of the number of documents containing both itemset i_s and itemset i_t to the number of documents containing only itemset i_s, i.e., it reflects the probability that a given document contains itemset i_s and at the same time itemset i_t. In contrast, the degree of enhancement is often considered as an evaluation metric that expresses the ratio of the probability that a document contains itemset i_t while also containing itemset i_s to the probability that itemset i_t occurs overall. 5Support(is⇒it)=P(isit)=count(is∪it)|D|$$Support\left( {{i_s} \Rightarrow {i_t}} \right) = P\left( {{i_s}{i_t}} \right) = \frac{{count\left( {{i_s} \cup {i_t}} \right)}}{{\left| D \right|}}$$ 6Confidence(is⇒it)=P(is|it)=Support(is⇒it)Support(is)$$Confidence\left( {{i_s} \Rightarrow {i_t}} \right) = P\left( {{i_s}|{i_t}} \right) = \frac{{Support\left( {{i_s} \Rightarrow {i_t}} \right)}}{{Support\left( {{i_s}} \right)}}$$ 7Lift(is⇒it)=P(is|it)P(is)=Confidence(is⇒it)Support(it)$$Lift\left( {{i_s} \Rightarrow {i_t}} \right) = \frac{{P\left( {{i_s}|{i_t}} \right)}}{{P\left( {{i_s}} \right)}} = \frac{{Confidence\left( {{i_s} \Rightarrow {i_t}} \right)}}{{Support\left( {{i_t}} \right)}}$$

Mining of association rules i_s ⇒ i_t refers to finding the association rules with support and confidence greater than the minimum support (Minsup)$$\left( {Mi{n_{\sup }}} \right)$$ and minimum confidence (Minconf)$$\left( {Mi{n_{conf}}} \right)$$ given by the user in the entire set of geographic information domain literature database transactions D, at which time, the corresponding itemset {is,it}$$\left\{ {{i_s},{i_t}} \right\}$$ is called the frequent itemset. Usually, when the confidence and support of a rule are higher than the thresholds Min_sup and Min_conf, respectively, the association rule is considered valid and feasible and is called a strong association rule; alternatively, in the case where the lift is greater than 1, then the association rule i_s ⇒ i_t is considered to be a strong association rule that is valid and feasible, and conversely, it is considered to be an invalid strong association rule.

The FP-growth algorithm is realized based on a special prefix tree FP-tree, which needs to be established before executing the FP-growth algorithm and is defined as follows.

Define FP-tree: FP-tree consists of an item prefix subtree and a frequent item header table; 1)

Item prefix subtree, the nodes in the tree include: item name, item count value and node chain. The item name indicates the name of the item represented by the node, the item count value indicates the number of transactions in the path to reach the node, and the node chain connects the next node in the tree with the same item name, and the value is null if the next node does not exist.

In FP-tree, for a certain node i, there exists a path from the root node to reach node i, in which the portion of the path that removes node i is called the prefix sub-path of i, and i is referred to as the suffix of the path. Starting from item i in the frequent item header table, the first node i in the FP-tree can be located by a pointer to the first node, and all prefix subpaths containing node i can be found by the node chain, and these prefix subpaths form the conditional pattern base of node i. And the FP-tree constructed with the conditional pattern base of node i is called the conditional pattern tree of node i.

FP-growth algorithm is an association rule mining algorithm proposed by Chia-Wei Tsao et al. The algorithm compresses the frequent itemsets in the transaction database into a frequent pattern tree FP-tree by one scan, while retaining the association information in it, and then divides the FP-tree into a number of conditional libraries corresponding to the length of the frequent itemsets of length L, and finally mines these conditional libraries separately. The algorithm is divided into two main parts, in the first step, the transaction database is scanned and FP-tree is constructed. In the second step, frequent patterns are mined from FP-tree by FP-Growth algorithm. It has been proved that the FP-growth algorithm is well adapted to rules of different lengths, and the FP-growth algorithm only scans the transaction database twice and does not generate a candidate itemset, which is a huge improvement in algorithmic efficiency compared to the Apriori algorithm.

The content of the next book of History of Modern Chinese Literature 1915-2022 (Fourth Edition) by Peking University Press is analyzed, in which the introduction, exercises and chapter guides in the textbook are not counted in the scope of knowledge point extraction. According to the definition and extraction method of knowledge points in the concept definition, the specific content of the textbook is deconstructed, and a total of 208 knowledge points are extracted, each of which has a specific corresponding paragraph in the textbook, and if a certain paragraph belongs to a certain topic, then the corresponding knowledge point belongs to a certain topic, and it is difficult to show the number of knowledge points completely due to the large number of knowledge points, so the number of statistical data of various types of knowledge points and the topics they belong to are as follows Table 1.

According to the rules of confirmation of knowledge point connection, 208 knowledge points in the research content of the next book of “History of Modern Chinese Literature 1915-2022 (Fourth Edition)” of Peking University Press were analyzed, and the neighbor matrix (208 horizontal rows and 208 vertical rows, both indicating 208 knowledge points) was constructed by using Excel software for statistics, and the analysis process was as follows: starting from the first Knowledge points to start, if the first knowledge point and the nth knowledge point related to the first line of the nth column space marked as 1, analyze each knowledge point one by one for statistics. At the same time, note that due to the sequence of knowledge points, if the nth knowledge point and the n+1th knowledge point is related, then only in the nth row of the n+1st column marking, no longer in the n+1st row of the nth column marking, so as to derive a network adjacency matrix that can be constructed without weights. (Note: using this method to record is actually recorded a directed network, but the subsequent analysis using simple undirected network analysis, so the subsequent adjustment in the Pajek software can be).

The above knowledge point extraction and judgment of knowledge point association were carried out by the two researchers based on the conceptual definition respectively, and the results of the extraction and judgment reached 90.5% agreement, and the subsequent network mapping and analysis was carried out after reaching a unification of the different judgments.

1)

Node point degree centrality

The distribution of node point degree centrality in the knowledge network of the statistical textbook is shown in Figure 3 below. Combined with Figure 3, it can be seen that the degree of 208 knowledge points ranges from 1-18, in which the number of knowledge points increases and then decreases as the degree increases, the number of knowledge points with a degree of 3 and 4 is the largest number of knowledge points, the overall majority of the knowledge point degree centered on 2-4 or so, accounting for about 59% of the total number of knowledge points when the degree value reaches 10 and after the number of knowledge points was once reduced to a total of 17, accounting for 8% of the total number of knowledge points or so.

As a whole, through the software calculations, the average point of knowledge points of the center of the degree of 4.83, which means that a knowledge point is associated with about 5 knowledge points on average, from a local point of view, the number of knowledge points with a degree equal to and more than 10 accounted for only 8%, and less than 2% of the knowledge points accounted for 2%. The above data show that most of the knowledge points in the knowledge network of the textbook can be associated with two or more knowledge points, and the association between the knowledge points is close, while the knowledge points with a degree value of more than 10 can be associated with a very large number of knowledge points, which is the skeleton knowledge of the textbook knowledge network, and the percentage of this part of the knowledge is relatively small.

Figure 3.

Figure of number of nodes in point degree center degree

Figure 4.

Distribution of knowledge point proximity to center

Figure 5.

Distribution of center degree of knowledge center

The goal of this section is to mine the association between simultaneous wrong answers to test questions, the question bank of about 6000 questions, 150 questions are extracted from each paper, the probability of each question being extracted is about 0.025, the probability of answering the right answer to the wrong test question is obtained according to the teacher’s law of the questions of the course are about 0.5, according to empirical calculation of the average probability of the appearance of each item in the dataset is 0.015. In the case of determining the degree of support, according to the confidence level. The change curve of the number of correlation rules in the interval [0.25,0.65] found that the number of correlation rules began to decrease in the confidence level of 0.4, and then the decrease tends to level off, combined with the specific number of correlation rules in the dataset, the confidence level of 0.45 was selected.

After you have determined a method to use frequent pattern mining and support-confidence levels, you need to implement correlation rule mining on your computer. The FP-growth algorithm is a typical algorithm based on the support-confidence framework, which achieves good results by limiting the generation of candidates to find frequent itemsets, but cannot avoid the “negative correlation defect” of the support-confidence framework, for example, an association rule is obtained by using the FP-growth algorithm, as shown in equation (8). 85610→5782conf:0.6,sup:0.65$$5610 \to 5782conf:0.6,\sup :0.65$$

The implication of this rule is that if the student answers question 5610 correctly, then the probability of his correct answer to question 5782 is 0.6, but in fact, the support degree of question 5782 is 0.65, that is, the probability of answering question 5782 correctly in the total data set is 0.65, which means that the former question has a negative impact on the latter question, and these “negative correlations” need to be screened out in the mining results, so it is necessary to introduce a boost degree to solve the “negative correlation” defect in the support-confidence framework.

Setting the support s=0.015 confidence c=0.45 for the association between wrongly answered test questions, the data of wrongly answered test questions got 978 association rules, which were sorted according to the confidence of the rules, and the top 10 are shown in Table 2.

According to the association rules in Table 2, we found that if students answered the pre-test questions incorrectly, the rate of answering the post-test questions incorrectly also increased relatively, which shows that there is a knowledge association between the pre-test questions and the post-test questions. Further analysis revealed that students were prone to give incorrect answers to the pre-test questions that examined the works of specific literary authors, and correspondingly, to give incorrect answers to the post-test questions on the superordinate concepts such as the “literary genres” to which these specific authors belonged. From this, we can find that students are easy to answer the knowledge points of the first few units of the textbook incorrectly when they are studying History of Modern Chinese Literature 1915-2022. It is speculated that the reason for this may be that students studied these units at the beginning of the semester, and the time interval between them and the final exam is long, which leads to the situation that the relevant knowledge points are forgotten more. Teachers can improve their teaching planning accordingly to address this situation.

The association rule mining between simultaneous right-answer questions has the same method as the association rule mining between simultaneous wrong-answer questions in 4.2.1, based on which, the association between simultaneous right-answer questions is directly analyzed in this section.

The association rules between simultaneous right answer questions are counted, and the content of the questions is viewed to investigate the students’ perception of the difficulty of the questions, and the questions that appear as the top 5 frequencies of the antecedent and the consequent items are obtained as shown in Table 3.

The analysis of the topics in Table 3 reveals that the topics frequently appearing as the antecedent are basically examining the basic knowledge points despite their varying difficulty, and that answering the topics examining the basics indicates that the students’ mastery of the basic knowledge points is firm, and naturally, there is a high probability of answering the other topics correctly; and the questions with high frequency of the latter items are generally simple, which indicates that answering the antecedent problems correctly has a high probability of answering the simpler questions correctly after the latter items as well.

To analyze the knowledge points of specific test questions, the knowledge points in the middle or back unit of the textbook are relatively easy for students, and the reason for this may be that most teachers will focus on the middle or back unit, and the time interval between the study of the knowledge points in the middle or back unit and the final exam is relatively short, and students’ memory and review of the relevant knowledge points are more in place. Teachers have the ability to modify their teaching plans accordingly.