A cross-region teacher collaboration pairing recommendation method and system based on causal relationship discovery

Abstract

The application discloses a cross-regional teacher cooperative pairing recommendation method and system based on causal relationship discovery, and belongs to the technical field of education informatization and intelligent decision support. The method comprises the following steps: collecting five-dimensional ability scores and basic attribute information of teachers, and constructing a digital portrait vector; based on historical cooperative behavior logs and ability change data, a hybrid causal discovery method combining Granger causality test and convergent cross mapping is adopted to construct a causal knowledge graph between cooperative behavior and ability improvement; for a candidate pairing combination, based on the causal knowledge graph and the digital portraits of both parties, the expected ability gain is calculated in a directional manner, and a two-way empowerment index is obtained by weighted summation; the two-way empowerment indexes of all candidate combinations are sorted to generate an optimal pairing recommendation scheme. The application first introduces a causal discovery method in the field of education cooperative pairing recommendation, so that the recommendation result has causal explanatory power, the dimensional ability gain of the candidate combination can be quantitatively predicted before the implementation of pairing, and the method is suitable for small sample scenes with limited participating teachers, and provides data-driven intelligent decision support for accurate pairing of teachers in cross-regional education cooperation.

Description

Technical Field

[0001] This invention belongs to the field of educational informatization and intelligent decision support technology, specifically involving a cross-regional teacher collaborative pairing recommendation method and system based on causal relationship discovery. It is particularly suitable for cross-regional teacher co-construction scenarios such as educational assistance and East-West educational cooperation, and is used to realize data-driven optimal teacher pairing scheme generation and two-way capability gain prediction. Background Technology

[0002] Cross-regional educational collaboration, particularly teacher exchanges under the paired assistance model, is a crucial national strategy for promoting educational equity and narrowing regional educational disparities. Taking educational assistance as an example, key teachers from eastern regions are sent to universities in western regions to form partnerships with local teachers, jointly undertaking tasks such as curriculum development, teaching reform, and talent cultivation. The core effectiveness of this model lies in whether the assisting teachers and local teachers can effectively complement each other, achieving a paradigm shift from one-way "blood transfusion" to two-way "blood production."

[0003] However, current teacher pairing practices suffer from significant technological gaps. Pairing schemes heavily rely on the subjective experience of administrators—assignments are typically based on coarse-grained rules such as similar subjects or professional titles, lacking a detailed characterization of individual teachers' abilities and failing to predict the skill enhancement effects of specific pairing combinations. This leads to three common problems: 1) Pairing matching lacks quantitative basis. Current practices can only answer "who can pair with whom," not "who pairing with whom yields the best results." Administrators cannot compare the expected benefits of different pairing schemes across the entire combination space; 2) Pairing effects cannot be predicted. Even if a pairing scheme is determined, it is impossible to predict in advance which ability dimensions the supporting and local teachers will improve in, and by how much. This results in a lack of targeted design basis for post-pairing collaborative activities; 3) There is a lack of causal mechanism explanation. What is the causal relationship between post-pairing collaborative behavior and skill enhancement? Which behaviors are the key factors truly driving skill enhancement? Current methods can only provide statistical correlation conclusions, failing to distinguish between causal relationships and spurious associations, leading to a lack of scientific guidance in the design of collaborative activities.

[0004] At the technical level, existing recommender system methods are difficult to directly transfer to the aforementioned scenarios. Classic recommender techniques such as collaborative filtering and matrix factorization rely on large amounts of user rating data, and their recommendation logic is based on similarity rather than complementarity, resulting in preference matching rather than ability gain prediction. In the field of education, some studies have attempted teacher matching methods based on ability models, but these ability models are mostly static descriptions, and the matching rules are manually defined by experts, lacking the ability to automatically learn causal relationships from historical data and dynamically optimize recommendation strategies. More importantly, cross-regional educational collaboration scenarios inherently possess data sparsity—a single aid project typically involves only a few dozen teachers, making deep learning methods that rely on large-scale training difficult to apply.

[0005] Therefore, there is an urgent need for a pair recommendation method specifically designed for cross-regional teacher collaboration scenarios: under limited data conditions, it should be able to automatically discover the causal relationship between collaborative behavior and ability improvement from historical collaborative experience, encode this causal relationship into computable knowledge, and then quantitatively predict the future ability gain of any candidate pair combination, ultimately outputting the optimal pair recommendation scheme supported by causal explanation. Summary of the Invention

[0006] This invention aims to address the following technical problems existing in current teacher pairing recommendation methods for cross-regional educational collaboration: (1) Lack of pair recommendation models based on causal relationships Existing methods either rely on manual assignment based on administrative experience or on similarity matching based on statistical correlation, which cannot automatically discover the causal relationship of "collaborative behavior → capability improvement" from historical collaborative data, resulting in a lack of causal explanatory power in the recommendation results.

[0007] (2) The bidirectional gain effect of pairing cannot be quantified and predicted. Existing methods cannot quantitatively predict the expected capability gains of candidate pairing combinations across multiple dimensions before using pairing schemes, leaving managers without a basis for decision-making based on a comprehensive comparison of the entire portfolio.

[0008] (3) Insufficient reliability of recommendations in small sample scenarios Cross-regional educational collaboration projects typically involve a limited number of teachers (dozens). Recommendation methods based on collaborative filtering or deep learning rely on large-scale user behavior data, which are poorly applicable in such small-sample scenarios.

[0009] To address the aforementioned issues, this invention proposes a cross-regional teacher collaboration pairing recommendation method based on causal relationship discovery. The core idea is to model teacher collaboration pairing as a causal relationship-driven problem of ability gain prediction and combination optimization. A causal knowledge graph between collaborative behavior and ability improvement is constructed from historical collaboration data. The bidirectional empowerment effect of candidate pairing combinations is quantitatively predicted based on the causal effect coefficient. Finally, the optimal pairing recommendation scheme with the overall combination ranking is output.

[0010] The method includes the following four core steps:

[0011] Collect five-dimensional competency scores and basic attribute information for each teacher to construct a structured digital profile vector of teachers.

[0012] The five dimensions of competence include: digital technology integration ability, cross-cultural teaching adaptability, curriculum innovation and development ability, collaborative leadership, and professional reflective development ability. The scores for each competence are based on a validated five-dimensional competence scale (Cronbach's α > 0.83), and are collected periodically at the beginning and end of each semester through a combination of teacher self-assessment, peer assessment, and evaluation by teaching administrators.

[0013] The basic attribute information includes: teaching experience, subject direction code, and aid / local identity identifier.

[0014] The digital profile vector representation of each teacher is as follows: in, to Assess five dimensions of ability. For years of teaching experience, Code the subject area. For identification purposes.

[0015] Step S2: Construction of a causal knowledge graph for collaborative behavior Based on collaborative behavior logs and teacher competency change data from historical pairing projects, a hybrid causal discovery method combining Granger causality test and convergent cross-mapping is used to construct a causal knowledge graph between collaborative behavior and competency improvement.

[0016] Collect process data within the historical pairing period (usually a full semester), including: Collaborative behavior time series: Record the frequency of collaborative behaviors of each pair of teachers in each time window (e.g., weekly). The behavior types include five categories: joint lesson preparation, mutual lesson observation, joint student guidance, online communication, and co-construction of courses, denoted as the set of behavioral variables. .

[0017] Competency change data: Record the five-dimensional competency scores of each teacher at the beginning and end of the pairing cycle, and calculate the change in each dimension of competency. Let be denoted as the set of variables of ability change. .

[0018] S2.2 Candidate Causal Pair Screening For each pair of behavioral variables With ability change variables A binary vector autoregressive model was established, and Granger causality tests were performed. Specifically, if behavioral variables were introduced... Historical values ​​can significantly reduce the impact of changes in ability. If the prediction error (i.e., the coefficients of the lag terms are jointly significantly non-zero) is found, then it will be marked as a candidate causal pair. The F-statistic of the Granger causality test is used to initially screen candidate causal pairs with time-series predictive power.

[0019] For the candidate causal pairs selected in step S2.2, the convergent cross-mapping method is further used to verify their nonlinear causal relationships. The convergent cross-mapping method is based on dynamic systems theory and verifies the relationship between behavioral variables. Can time series be used to analyze changes in capabilities? The data is reconstructed in the time series (i.e., the predictive power of the cross-mapping converges as the sequence length increases), and the existence of a nonlinear causal driving relationship between the two is determined. Convergence testing is then performed. Candidate causal pairs whose values ​​are greater than a preset threshold are confirmed as true causal pairs.

[0020] S2.4 Calculation of Causal Effect Coefficient For confirmed true causal pairs Calculate its causal effect coefficient The strength of the impact of changes in the frequency of quantitative behaviors on changes in ability: in, Let be the covariance between the behavioral variable and the ability change variable. The variance of the behavioral variable. This is the convergence correlation coefficient for the convergence cross-mapping test, used to perform nonlinear correction on the linear regression coefficients.

[0021] All true causal pairs and their causal effect coefficients The organization is a directed weighted bipartite graph structure, resulting in a causal knowledge graph. , where the node set For collaborative behavior types, node sets For the type of capability change, a directed edge set Indicates behavior Causal driving ability The improvement in driving strength is .

[0022] Step S3: Calculation of the bidirectional empowerment index of candidate pair combinations For any candidate pairing of a teacher A (providing assistance) and a local teacher B, based on a causal knowledge graph... Digital profiles of both parties , Calculate the bidirectional empowerment index of this combination.

[0023] For each capability dimension Calculate the complementarity between teacher A and teacher B in this dimension:

[0024] That is, in terms of capability dimension If teacher A's ability score is higher than teacher B's, then the complementarity is the difference between the two; otherwise, the complementarity is 0 (indicating that A cannot provide effective support to B in this ability). Conversely, the complementarity of B to A is calculated in the same way.

[0025] S3.2 Determination of Collaborative Feasibility Coefficient For each causal path Based on the subject areas of Teacher A and Teacher B , With identity markers , Determine the feasibility coefficient of collaboration When two teachers share the same subject area, the feasibility coefficient for joint lesson planning, mutual lesson observation, and curriculum co-construction is set to 1.0; when their subject areas differ, the coefficient is set to 0.3-0.7 based on subject relevance. The feasibility coefficient for online communication is set to 1.0 by default.

[0026] Define the Bidirectional Empowerment Index (BEI) as the weighted sum of expected capability gains across all causal paths in a causal knowledge graph: This index comprehensively quantifies: after pairing, through which causal paths, in which capability dimensions, and to what extent will both parties achieve bilateral improvement? In the calculation process, the expected gain of B is calculated separately with A as the output and B as the receiver; then, the expected gain of A is calculated with B as the output and A as the receiver; the weighted sum of the two-way gains yields the final BEI value.

[0027] Step S4: Optimal Pairing Solution Generation For all feasible groups of teachers to provide assistance Grouping with local teachers The Cartesian product combination is then processed pairwise, and step S3 is performed to obtain the bidirectional empowerment index of all candidate pair combinations.

[0028] Sort the recommended pairing sequences in descending order of BEI value. Each recommendation record includes: the identifier of the teacher providing assistance, the identifier of the local teacher, the BEI value of the two-way empowerment index, and the expected gain details for each capability dimension (extracted from the intermediate calculation process in step S3.3).

[0029] Meanwhile, the scheme supports one-to-many constraint processing: when a teacher supporting the construction needs to be paired with multiple local teachers, after outputting the optimal pairing in the first round, the remaining pairing slots of the teacher supporting the construction are used as constraints to participate in the next round of selection, generating a globally optimal pairing scheme that satisfies the specified constraints. Beneficial effects

[0030] Compared with the prior art, the present invention has the following beneficial effects: 1. Upgrading from relevance-based recommendation to causal recommendation. For the first time in the field of educational collaborative pairing recommendation, a causal discovery method combining Granger causality test and convergent cross-mapping is introduced. This method automatically extracts causal knowledge of "collaborative behavior → ability improvement" from historical data, giving the recommendation results causal explanatory power and enabling them to clearly answer the question "why this teacher pairing is recommended".

[0031] 2. Achieve multidimensional quantitative prediction of pairing effectiveness. Based on the causal effect coefficient and capability complementarity, construct the Two-Way Empowerment Index (BEI), which can quantitatively predict the capability gains of candidate combinations in each dimension before pairing is implemented, providing managers with data basis for comparing and selecting the best combination.

[0032] 3. Reliable operation in small sample scenarios. The data input for this method consists of competency scale scores and behavior logs collected once per semester, and does not require large-scale user behavior data. The causal discovery method can operate effectively with a sample size of several dozen, making it suitable for the practical conditions of limited teacher participation in cross-regional educational collaboration projects.

[0033] 4. Forming a data closed loop to support continuous optimization. After each pairing cycle, the newly generated collaborative behavior data and capability change data can be fed back into the causal knowledge graph to update the causal effect coefficients, enabling the recommendation model to continuously evolve as the project progresses and gradually improve recommendation accuracy. Attached Figure Description

[0034] Figure 1 This is an overall flowchart of the cross-regional teacher collaborative pairing recommendation method based on causal relationship discovery as described in this invention.

[0035] Figure 2 This is a flowchart of the sub-steps for constructing a causal knowledge graph of collaborative behavior in this invention.

[0036] Figure 3 This is an example diagram of the causal knowledge graph structure for collaborative behavior and capability enhancement constructed in this invention. Detailed Implementation

[0037] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0038] Example This example uses an educational assistance project between an eastern university and a western university as an application scenario. The project involves 12 teachers from the eastern university providing assistance (referred to as...). ) and 24 local teachers from the west (referred to as Both parties will jointly undertake the tasks of curriculum development and teaching reform in majors such as Computer Science and Technology, Data Science and Big Data Technology, according to their paired relationship.

[0039] Step S1: Building a Digital Profile of Teachers This step involves collecting two types of data to construct a digital profile of the teacher.

[0040] (1) Five-dimensional ability assessment Before the pairing period begins (at the beginning of the semester), all participating teachers are assessed using a validated five-dimensional competence scale. The specific definitions of the five dimensions of competence are: Digital and intelligent technology integration capabilities The ability to skillfully use digital teaching tools and platforms and integrate technologies such as artificial intelligence and big data into the entire teaching process. Cross-cultural teaching adaptability The ability to understand and respect the learning characteristics of students from different cultural backgrounds, and to adjust teaching content and methods to adapt to regional differences. Curriculum innovation and development capabilities The ability to independently or collaboratively develop new courses, restructure teaching content, and integrate cutting-edge knowledge and industry needs into the curriculum system Collaborative Leadership The ability to build and lead teaching teams, integrate cross-regional educational resources, and effectively handle collaborative conflicts. Professional Reflection and Development Capabilities The ability to objectively assess one's own teaching effectiveness, proactively learn new knowledge and skills, and systematically summarize practical experience. The scoring uses a 5-point Likert scale, and the final score is a weighted composite of three factors: teacher self-assessment (weight 0.3), peer assessment (weight 0.4), and teaching administrator evaluation (weight 0.3).

[0041] (2) Collection of basic attribute information Export teachers' teaching experience, subject area, and identity identifier from the personnel management system. The subject area coding rules are: Computer Science and Technology = 1, Data Science and Big Data Technology = 2, Artificial Intelligence = 3, Automation = 4, Software Engineering = 5. Identity identifiers: Teachers supported by aid projects = 1, Local teachers = 0.

[0042] (3) Vectorization of digital profiles The above data will be integrated into a digital profile vector. (This will be used to support teachers.) For example: The score indicates that the individual has 4.5 points for digital technology integration ability, 3.2 points for cross-cultural teaching adaptability, 4.3 points for curriculum innovation and development ability, 4.1 points for collaborative leadership, and 3.7 points for professional reflective development ability. The individual has 10 years of teaching experience, specializes in computer science and technology, and is a teacher supporting the construction project.

[0043] With local teachers For example: The score indicates that the individual has 3.1 points for digital technology integration ability, 4.6 points for cross-cultural teaching adaptability, 3.3 points for curriculum innovation and development ability, 3.5 points for collaborative leadership, and 3.9 points for professional reflective development ability. The individual has 8 years of teaching experience, specializes in computer science and technology, and is a local teacher.

[0044] After this step is completed, a complete digital profile library of 36 teachers is formed, which serves as the input basis for subsequent steps.

[0045] like Figure 2 As shown, this step constructs a causal knowledge graph from historical paired data, and includes five sub-steps.

[0046] Collect process data from all 12 paired teachers during the previous pairing cycle (Fall semester of the 2024-2025 academic year, a total of 18 weeks): Collaborative behavior time series: The frequency of each pair of teachers in five types of collaborative behavior was recorded in weekly time windows. Table 1 provides an example of the weekly collaborative behavior records for a pair of paired teachers.

[0047] Table 1. Examples of time series of collaborative behaviors (paired teacher pairs: - ) Week 1 2 0 0 3 0 Week 2 1 0 1 2 0 Week 3 2 1 0 4 1 ... ... ... ... ... ... Week 18 1 1 1 3 0 Competency change data: The difference in five-dimensional competency scores for each teacher at the beginning and end of the pairing period. Table 2 provides examples of pre-test, post-test, and change values.

[0048] Table 2 Example of Ability Change Data (Teachers: ) Digital technology integration 3.1 3.7 +0.6 Cross-cultural teaching adaptation 4.6 4.7 +0.1 Curriculum Innovation Development 3.3 3.8 +0.5 Collaborative Leadership 3.5 3.7 +0.2 Professional reflection and development 3.9 4.1 +0.2

[0049] Data from all 12 paired teachers were compiled, and using all 36 teachers as the analysis sample, behavioral variables for each pair were analyzed. With ability change variables Perform Granger causality testing.

[0050] The specific operation is as follows: For behavioral variables... and ability change variables Establish the following two autoregressive models: Constraint Model:

[0051] Unconstrained model:

[0052] in, To determine the maximum lag order, this embodiment takes... Calculate the F-statistic for the sum of squared residuals of the two models: If the F-statistic is at the preset significance level ( If we reject the null hypothesis, then we will... Mark them as candidate causal pairs.

[0053] After testing, 12 candidate causal pairs that met the significance criteria were selected from 5×5=25 possible behavior-ability combinations.

[0054] S2.3 Nonlinear Causality Verification For each of the 12 candidate causal pairs selected in step S2.2, convergent cross-mapping verification is performed.

[0055] Candidate causal pairs (joint lesson preparation) Changes in curriculum innovation and development capabilities Taking (e.g., CCM verification) as an example, the specific steps are as follows: Take behavioral variables time series and ability change variables time series .

[0056] Build Shadow manifold : Employs time-delayed embedding, embedding dimension Time delay .

[0057] for Each observation in ,exist Find it Use the nearest neighbor points and their corresponding values. The values ​​are locally weighted and averaged to obtain the cross-mapping estimate. .

[0058] Calculate cross-mapping prediction skills Actual value Compared with the estimated value The Pearson correlation coefficient between them. Take the maximum sequence length. ,observe Does the value follow It increases and converges.

[0059] like And if it shows a clear convergence trend, then it is confirmed. This is a true cause-and-effect relationship.

[0060] According to CCM verification, 8 out of 12 candidate causal pairs were confirmed as true causal pairs, and 4 were removed.

[0061] For the eight confirmed true causal pairs, calculate the causal effect coefficient. (Joint lesson preparation) Curriculum innovation and development capabilities For example: behavioral variables variance behavioral variables With changes in ability covariance CCM convergence correlation coefficient Causal effect coefficient Similarly, the effect coefficients of the remaining causal pairs were calculated, and the results are summarized in Table 3.

[0062] S2.5 Construction of Causal Knowledge Graph The eight true causal pairs and their causal effect coefficients are organized into a directed weighted bipartite graph. ,like Figure 3 As shown in Table 3, a complete causal knowledge graph is presented.

[0063] Table 3. Causal Knowledge Graph of Collaborative Behavior and Capability Enhancement Joint lesson preparation → Curriculum Innovation Development 0.26 5.32** 0.72 Joint lesson preparation → Digital technology integration 0.31 6.18*** 0.81 mutual observation → Cross-cultural teaching adaptation 0.24 4.87* 0.68 mutual observation → Professional reflection and development 0.18 4.21* 0.59 Jointly guide students → Collaborative Leadership 0.19 4.45* 0.63 Online communication → Cross-cultural teaching adaptation 0.15 3.98* 0.55 Course co-construction → Curriculum Innovation Development 0.28 5.76*** 0.78 Course co-construction → Digital technology integration 0.22 4.64* 0.66 Note: *p<0.05, **p<0.01, ***p<0.001 The following causal knowledge can be gleaned from the causal knowledge graph: For example, the behavior of "joint lesson preparation" causally drives the "ability to integrate digital and intelligent technologies" (…). ) and "curriculum innovation and development capability" The improvement of "mutual observation" causally drives the improvement of "cross-cultural teaching adaptability". ).

[0064] Teachers who receive assistance with local teachers Taking this candidate pairing as an example, we calculate its bidirectional empowerment index.

[0065] S3.1 Calculation of Complementarity of Capabilities Five-dimensional capability scores are extracted from the digital profile vectors of both parties to calculate complementarity. Capabilities are then integrated using digital intelligence technologies. For example: of , of ,because complementarity .

[0066] Conversely, adaptability in cross-cultural teaching For example: of , of ,at this time Stronger in this dimension In the For the output side, When calculating complementarity for the receiver, .

[0067] and The subject area codes are all 1 (Computer Science and Technology), indicating they share the same subject area. Therefore, joint lesson planning is necessary. mutual observation Course co-construction Cooperative feasibility coefficient Jointly guide students Feasibility factor is set at 0.8; online communication The feasibility factor is set to 1.0.

[0068] by → direction( Empower Taking as an example, traversing the causal knowledge graph Ability score higher than The causal edge corresponding to the capability dimension: Retrieval from the causal knowledge graph in Table 3: Joint lesson preparation → Digital and intelligent technology integration , Complementarity 1.4, Feasibility 1.0, Contribution = 0.31 × 1.4 × 1.0 = 0.434; Joint lesson planning →Curriculum Innovation Development , Complementarity 1.0, Feasibility 1.0, Contribution = 0.26 × 1.0 × 1.0 = 0.260; Co-construction of Courses → Digital and intelligent technology integration , Complementarity 1.4, Feasibility 1.0, Contribution = 0.22 × 1.4 × 1.0 = 0.308; Co-construction of Courses →Curriculum Innovation Development , Complementarity 1.0, Feasibility 1.0, Contribution = 0.28 × 1.0 × 1.0 = 0.280.

[0069] → Subtotal of directional empowerment = 0.434 + 0.260 + 0.308 + 0.280 = 1.282.

[0070] by → direction( Empower ), traversal Ability score higher than The causal edges corresponding to the dimensions: mutual observation →Cross-cultural teaching adaptation , Complementarity 1.4, Feasibility 1.0, Contribution = 0.24 × 1.4 × 1.0 = 0.336; Mutual observation of classes →Professional Reflection on Development , Complementarity 0.2, Feasibility 1.0, Contribution = 0.18 × 0.2 × 1.0 = 0.036; Online communication →Cross-cultural teaching adaptation , Complementarity 1.4, Feasibility 1.0, Contribution = 0.15 × 1.4 × 1.0 = 0.210.

[0071] → Subtotal of directional empowerment = 0.336 + 0.036 + 0.210 = 0.582.

[0072] Two-way empowerment index BEI ( , = 1.282 + 0.582 = 1.864.

[0073] Table 4 summarizes the complete calculation process.

[0074] Table 4. Calculation details of the bidirectional empowerment index for candidate pair combinations ( - ) → 0.31 → 1.4 1.0 0.434 → 0.26 → 1.0 1.0 0.260 → 0.22 → 1.4 1.0 0.308 → 0.28 → 1.0 1.0 0.280 → 0.24 → 1.4 1.0 0.336 → 0.18 → 0.2 1.0 0.036 → 0.15 → 1.4 1.0 0.210 Two-way Empowerment Index (BEI) 1.864

[0075] For all 288 pairs of teachers (12 from the aid program and 24 from the local program), step S3 was performed one by one to calculate the BEI value for each pair.

[0076] Sort the pairs in descending order of BEI value to generate the optimal pairing recommendation scheme. Considering practical management, a constraint is set that each supporting teacher can be paired with a maximum of two local teachers. A greedy selection strategy is adopted: in each round, the pair with the highest current BEI value that does not violate the constraint is selected, until all the supporting teacher pairing slots are used up. The final output of the optimal pairing recommendation scheme is shown in Table 5.

[0077] Table 5. Optimal Pairing Recommendations (Partial) 1 1.864 Digitalization +0.31 / Cross-cultural interaction +0.24 2 1.726 Curriculum Innovation +0.28 / Collaboration +0.19 3 #imgpt244# 1.593 Digitalization +0.22 / Professional Reflection +0.16 4 #imgpt245# #imgpt246# 1.481 Curriculum Innovation +0.25 / Cross-cultural Interpretation +0.18 5 #imgpt247# #imgpt248# 1.405 Collaboration +0.20 / Digitalization +0.19 ... ... ... ... ... At the end of the pairing cycle for the new semester, the collaborative behavior logs and post-test data generated in this cycle will be incorporated as new historical data into the causal discovery process in step S2, updating the causal edge structure and effect coefficients in the causal knowledge graph. As the number of pairing rounds increases, the causal knowledge graph gradually becomes more refined, and the recommendation accuracy continues to improve.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cross-regional teacher collaborative pairing recommendation method based on causal relationship discovery, characterized in that, Includes the following steps: Step S1: Collect the five-dimensional ability scores and basic attribute information of each teacher to construct a structured digital profile vector of teachers; the five-dimensional abilities include digital technology integration ability, cross-cultural teaching adaptability, curriculum innovation and development ability, collaborative leadership ability, and professional reflection and development ability; the basic attribute information includes teaching experience, subject direction code, and identity identifier. Step S2: Based on the collaborative behavior logs and teacher ability change data in historical pairing projects, a hybrid causal discovery method combining Granger causality test and convergent cross mapping is used to construct a causal knowledge graph between collaborative behavior and ability improvement. Step S3: For any candidate pairing of a teacher supporting the construction project and a local teacher, based on the causal knowledge graph and the digital profile vectors of both parties, calculate the expected capability gain of the two empowerment directions with the teacher supporting the construction project as the output and the local teacher as the receiver, and with the local teacher as the output and the teacher supporting the construction project as the receiver, and obtain the two-way empowerment index by weighted summation. Step S4: Sort the bidirectional empowerment indices of all candidate pair combinations to generate the optimal pair recommendation scheme.

2. The method according to claim 1, characterized in that, Step S2 includes the following sub-steps: S2.1, Collect the collaborative behavior logs and teacher competency change data within the historical pairing period; wherein, the collaborative behavior logs record the frequency of collaborative behaviors of each pair of teachers within each time window, and the behavior types include joint lesson preparation, mutual lesson observation, joint student guidance, online communication, and course co-construction; the teacher competency change data is the difference in the five-dimensional competency scores of each teacher at the beginning and end of the pairing period; S2.2, Perform Granger causality test on each pair of behavioral variables and ability change variables, and mark the combinations that meet the preset significance level as candidate causal pairs; S2.3, Perform convergent cross-mapping verification on the candidate causal pairs, and confirm the candidate causal pairs with convergent correlation coefficients greater than a preset threshold and showing a convergent trend as true causal pairs; S2.4 For the true causal pair, calculate the causal effect coefficient, which is the ratio of the covariance of the behavioral variable and the ability change variable to the variance of the behavioral variable, multiplied by the convergent correlation coefficient of the convergent cross mapping. S2.5, the true causal pairs and their causal effect coefficients are organized into a directed weighted bipartite graph to obtain the causal knowledge graph, wherein the node set includes collaborative behavior types and capability change types, the directed edges represent collaborative behavior causally driving capability improvement, and the weight of the edge is the causal effect coefficient.

3. The method according to claim 2, characterized in that, The Granger causality test in step S2.2 includes: establishing a constrained autoregressive model and an unconstrained autoregressive model for the behavioral variable and the ability change variable, respectively, wherein the unconstrained model introduces the lagged term of the behavioral variable; calculating the F-statistic of the sum of squared residuals of the two models; if the F-statistic rejects the null hypothesis at a preset significance level, then the combination of the behavioral variable and the ability change variable is marked as a candidate causal pair.

4. The method according to claim 2, characterized in that, The convergent cross-mapping verification in step S2.3 includes: taking the time series of the behavioral variable and the time series of the capability change variable; constructing a shadow manifold of the behavioral variable time series using time delay embedding; for each observation of the capability change variable, finding the nearest neighbor in the shadow manifold and performing a local weighted average to obtain the cross-mapping estimate; calculating the correlation coefficient between the actual value of the capability change variable and the cross-mapping estimate as the convergent correlation coefficient; and determining whether the convergent correlation coefficient converges as the sequence length increases.

5. The method according to claim 1, characterized in that, Step S3 includes the following sub-steps: S3.1 For each ability dimension, calculate the complementarity of the two teachers' abilities; where, if the first teacher's ability score in that dimension is higher than that of the second teacher, the complementarity is the difference between the two, otherwise the complementarity is zero. S3.2, Based on the subject areas and identities of the two teachers, determine the collaborative feasibility coefficient for each type of collaborative behavior; S3.3 Calculate the bidirectional empowerment index, which is the sum of the products of the causal effect coefficient, the capability complementarity, and the synergistic feasibility coefficient on all causal paths in the causal knowledge graph.

6. The method according to claim 5, characterized in that, The feasibility coefficient for collaboration mentioned in step S3.2 is determined according to the following rules: when the two teachers have the same subject direction, the feasibility coefficient for joint lesson preparation, mutual lesson observation, and course co-construction is 1.0; when the subject directions are different, the value is between 0.3 and 0.7 according to the subject relevance; the feasibility coefficient for online communication is 1.0 by default.

7. The method according to claim 1, characterized in that, The optimal pairing scheme generation in step S4 includes: calculating the bidirectional empowerment index of the Cartesian product combination of all supporting teachers and all local teachers; sorting the bidirectional empowerment indices in descending order and outputting the recommended pairing sequence in sequence; wherein, each recommended record includes the supporting teacher identifier, the local teacher identifier, the bidirectional empowerment index value, and the expected gain details of each capability dimension.

8. The method according to claim 7, characterized in that, Step S4 also includes one-to-many constraint processing: setting the maximum number of pairs for each teacher providing assistance; after outputting the optimal pairing in the first round, using the remaining available pairs of the assigned teachers as constraints to participate in the next round of selection, generating a globally optimal pairing scheme that satisfies the specified constraints.

9. The method according to claim 1, characterized in that, The method also includes a data closed-loop iteration step: after a new pairing cycle ends, the newly generated collaborative behavior logs and teacher ability change data are incorporated into step S2 to update the causal edge structure and causal effect coefficient in the causal knowledge graph.

10. A cross-regional teacher collaborative pairing recommendation system based on causal relationship discovery, characterized in that, include: The teacher digital profile construction module is used to collect the five-dimensional ability scores and basic attribute information of each teacher to construct a structured teacher digital profile vector; the five-dimensional abilities include digital technology integration ability, cross-cultural teaching adaptability, curriculum innovation and development ability, collaborative leadership ability, and professional reflection and development ability. The causal knowledge graph construction module is used to construct a causal knowledge graph between collaborative behavior and ability improvement based on collaborative behavior logs and teacher ability change data in historical pairing projects, using a hybrid causal discovery method that combines Granger causality test and convergent cross mapping. The two-way empowerment index calculation module is used to calculate the expected capability gain in two empowerment directions for any candidate pairing of teachers supporting construction and local teachers, based on the causal knowledge graph and the digital profile vectors of both parties, and then sum them up by weight to obtain the two-way empowerment index. The optimal pairing scheme generation module is used to sort the bidirectional empowerment index of all candidate pairing combinations and generate the optimal pairing recommendation scheme.

Images