Interdisciplinary Knowledge Integration Approach and Computational Modeling in English Curriculum Civics and Politics

With the rapid development of natural language analysis methods, analyzing and mining text data through topic modeling, in-depth to the semantic level of the text can be derived from the technical hotspots and development trends in a certain field, which can provide ideas for industrial development and technology research and development.

Topic extraction and mining is a branch of text-oriented topic extraction research field, it is also a typical and effective application of text mining technology and news topic evolution analysis and topic tracking technology of a carrier, the era of information outbreak, the emergence of topic modeling, so that people from the long text to obtain information has become easy. Not only that, the topic model is an efficient and simple text data mining method that can extract semantic information from news text.

The core of the LDA topic model is to join the classical ideas of the Bayesian school, which adds prior knowledge in the model inference, and also introduces the concept of covariate distribution, the prior distribution and the likelihood function in the LDA topic model are covariate, which also makes the posterior probability able to have mathematical consistency with the prior probability, and this kind of mathematical consistency provides help for the derivation of the model and its solution, and in the model solution, the The model can use the result of the last obtained a posteriori distribution probability as the a priori probability distribution for the next model training, and repeat the iterative process until the model converges.

Four distributions are included in the LDA subject model, and the specific distribution forms are given below: 1)

Binomial distribution: (1)P(K=k)=( n k)pk(1−p)n−k

From the above equations, it can be seen that although these distributions are completely different in expression, it can be seen that there is a conjugate relationship between these distributions, i.e., binomial distribution is conjugate to Beta distribution and polynomial distribution is conjugate to Dirichlet distribution, and there is a mathematical consistency in these distributions.

The main algorithmic idea of the LDA topic model is to carry out continuous Bayesian inference and iteration on the M documents in the whole document data set through the predetermined number of topics K, so that the output of the model is the probability distribution of the topic words of the M documents in the text set, and the probability distribution of the feature words that can represent the topic.

The steps of LDA algorithm are as follows: 1)

For dataset M:

Select K subject-word distribution ϕk→~Dir(β→) $\overrightarrow {{\phi _k}} {\sim}Dir\left( {\vec \beta } \right)$ , where k ∈ {1, 2, …, K}, Dir(β→) denote Dirichlet distributions with parameter β→ .

The complete solution of the LDA subject model is a very complex process, and the existing methods usually use methods such as Gibbs sampling and Bayesian derivation to find an approximate parametric solution. In this study, the Gibbs random sampling algorithm is mainly used in the parameter solution of the LDA subject model. The sampling method is used to simulate the sampling by combining the problem with a probabilistic model using a computer, and then the sampled random samples are solved, and the result of the solution can be used as an approximate solution of the parameters of the LDA model.

The core principle of Gibbs sampling is that, in the derivation of the LDA topic model, the Markov chain-based Monte Carlo method (MCMC) will first fix the other words in the document, so that the topics to which the other words belong will not be changed, and then construct a Markov chain, through which all the pre-subtracted words in the whole document are connected, and then these words are matched with the document Then, we connect all the pre-sorted words in the whole document through this Markov chain, and then correspond these words to each topic of the document one by one, and get the topic of the word through constant sampling, and wait until the topic probability of the model converges, then we can determine the topic to which the word belongs, and finally repeat the algorithm operation until all the words have corresponding topics.

Gibbs sampling is to solve the conditional probability p(Z|W) through the joint probability, and then use the conditional probability to sample the implied variables of each word, so that w_i = w_m,n, the corresponding topic is z_i then the Gibbs sampling algorithm for the LDA model is shown in Equation (5): (5)p(zi=k|Z−i,W,α→′,β→) = p(W,Z)p(W,Z−i) = p(W,Z)p(W−i,Z−i)p(wi) ∝ Δ(β→+nk→)Δ(β→+nk,−i→)⋅Δ(α→+nm→)Δ(α→+nm,−i→) ∝ nk1−i(t)+β∑t=1V(nk,−i(t)+β)⋅nm,−i(k)+α∑m=1M(nm,−i(k)+α)

By first finding the word-topic probability distribution and then finding the text-topic probability distribution. After arriving at the topic to which each word belongs by calculation, an approximation can be made to estimate the topic-word probability distribution parameters and the text-topic probability distribution parameters.

In the existing literature dataset, from the disciplinary characteristics of journals, it is possible to understand the disciplines to which each piece of literature belongs intuitively, and from the classification number of the literature, it is possible to understand roughly which literature belongs to the fusion of two disciplines, such as the literature in the Journal of Intelligence, which has the classification number of “TP3”. However, since the classification number is given by the author or editor, it is highly subjective, so it is defective to use this method to identify the fusion literature of two disciplines. In order to overcome the defects of this method with human subjective will, it is necessary to find a fusion disciplinary literature identification method based on the essential characteristics of data.

Definition 1: Assume that the literature set A contains a total of two disciplines, which are denoted as Discipline I and Discipline II. Based on the disciplinary characteristics of the journals, A is divided into two subsets, A₁ and A₂, and C is said to be the fusion literature of the two disciplines if ∃φ, such that A1≫A2→φ(U1,U2)C , where φ is a fuzzy division of the fuzzy subordinate degree function U of the dingbatin fuzzy division.

Definition 2: Assuming that K₁ is the set of all keywords of A₁, K₂ is the set of all keywords of A₂, K₃ = K₁ ≫ K₂, K_{1_3} = K₁\K₃, and K_{2_3} = K₂\K₃, we call K₁ a subject I feature word, K₂ a subject II feature word, and K₃ a subject fusion feature word. Clearly, K_{1_3} ≫ K₃ = Φ and K_{2_3} ≫ K₃ = Φ.

Definition 3: The evaluation function G will be φ divided into evaluation criteria, G defined as follows: (6)G=F1+F2=2P1R1(P1+R1)+2P2R2(P2+R2)

Among them, P_i, R_i, and F_i are the detection rate, accuracy rate, and F indicators of the i(i = 1, 2) categories. G is the sum of the F values of the two categories, and the principle of φ division is to ensure that a certain number of them are classified into the fusion category, but to ensure that the G value is higher.

Based on the above definitions, the discovery algorithm for fusion discipline literature is designed:

STEP1: Identify the category number of all literature based on the disciplinary characteristics of the journal.

STEP2: Extract the keywords of the two categories of literature and construct the discipline-specific keyword sets K_{1_3}, K_{2_3} and K₃.

STEP3: Calculate the sum of the frequencies of the keywords in the first feature keyword set in the title, Chinese keywords and Chinese abstract of each literature respectively to divide the frequency.

STEP4: Construct VSM matrix, with literature as rows and three features as columns to construct the matrix, and the matrix element is frequency (the contribution of literature to the features).

STEP5: Specify the data, since each element represents the frequency of its occurrence in each column and the three feature spaces have different dimensions, specification of the data is done to eliminate the bias of the magnitude.

STEP6: The specified VSM matrix is clustered using FCM algorithm.

STEP7: For the clustered results, the affiliation function is adjusted and the results are output when the evaluation function is optimal.

Disciplinary integration is an important form of contemporary scientific research, and understanding and measuring the degree of disciplinary integration in interdisciplinary research is a difficult and hot topic in contemporary bibliometrics. The current understanding of disciplinary fusion research only stays on the surface, i.e., there is a great lack of overall understanding and essential research on the development of interdisciplinary research. To study and measure interdisciplinarity from the perspective of bibliometrics and to construct an interdisciplinary measurement index system based on multidisciplinarity, specialization, disciplinary fusion and author cooperation, the proposed disciplinary fusion indexes are based on citations and do not take into account the amount of disciplinary publications and disciplinary co-words contributing to disciplinary fusion, which has some limitations, respectively, from the amount of disciplinary fusion literature and disciplinary fusion co-words to measure the integration relationship between disciplines.

The discipline integration factor is proposed based on the number of disciplinary publications, which is used to describe the degree of integration between two disciplines. Discipline integration factor is based on keywords. Since the research direction of disciplines is based on keywords, the relationship between keywords can be used to measure the degree of integration between two disciplines in terms of research direction.

Definition 4: Discipline Integration Factor: assuming that the set of keywords in Discipline I literature is k₁, the set of keywords in Discipline II literature is k₂, and the set of keywords common to the two disciplines is k₃, the integration of the two disciplines is defined as: (7)a=k3k1+k2−k3

Discipline fusion knowledge visualization is an aspect of knowledge visualization, focusing on multiple disciplines fusion direction i.e. disciplinary commonality research, the main indexes of visualization include: disciplinary fusion based on the proportion of common words, disciplinary fusion view based on clustering perspective and so on.

Keywords are the core and focus of each piece of literature and are usually highly generalized. In the visualization analysis, the recurring keywords and theme words can be approximated as the research hotspots in the field. Using the “Keyword” node in CiteSpace, the time slice is 1 year, and the frequency and centrality of keywords are analyzed by pruning the slice network, and the relationship between frequency and centrality is shown in Table 1. The node size usually indicates the frequency size and degree centrality of node keywords, this paper selects the degree centrality to draw the co-occurrence mapping, and the node whose degree centrality is greater than 6 is marked by the text focus. The degree centrality of a node refers to the number of all the shortest paths passing through the node in the mapping network, which is a measure of the size of the node’s connectivity role in the overall network, so the higher the centrality of a node, the more it appears on the shortest paths in the overall network, and the greater its influence and importance. Through the co-occurrence map, it can be seen that in addition to the limited keywords of the two research fields of “English” and “curriculum ideology and politics”, the node size of “teaching design” is the largest, while the nodes of keywords such as “cultural confidence”, “moral cultivation” and “university” are slightly smaller, and through the distribution of colors in the nodes, it can be seen that the two keywords of “English” and “curriculum ideology and politics” span the entire research period, while other smaller keyword nodes usually only exist for one to two years. In addition to degree centrality, CiteSpace can calculate the intermediary centrality between nodes, and when the intermediary centrality of a node is greater than 0.1, it can usually be regarded as a key node in this field. Only the two keywords “English” and “curriculum ideology and politics” have an intermediary centrality greater than 0.1, but the intermediary centrality of the two keywords “curriculum reform” and “cultural self-confidence” are 0.09 and 0.07, respectively, which are closest to 0.1 in the list, so they can be approximated as sub-critical nodes. Therefore, in the field of ideology and politics in English courses, the main research direction is the education of moral cultivation centered on curriculum reform and design, as well as cultural self-confidence.

The co-occurrence map is shown in Figure 1, the clustering map of the clustering analysis keywords can indicate different research hotspots in the field, and the clustering analysis map can be obtained according to the clustering calculation in CiteSpace, and the clustering analysis can be obtained after screening out the four categories with more sub-elements, as shown in Figure 1. Through the cluster analysis chart, it can be seen that the keywords in this field can be divided into four categories: “English”, “ideological and political education”, “cultural self-confidence” and “ideological and political elements”. Since “English” and “curriculum ideology and politics” are the keywords that define the research field, it can be seen that the keyword “English” appears in the category of ideological and political education, while the keyword “curriculum ideology and politics” appears in the English category. In addition, it can be seen that the four categories are not independent of each other, but have certain connections and intersections, such as the intersection of “ideological and political elements” and “ideological and political education”, and the intersection of “English” and “cultural self-confidence”. The diagram mainly shows the connections within the categories, but there are also cross-category connections. Although the field can be divided into four main research categories, there are also rich exchanges and associations between them.

Figure 1.

Key word clustering pattern