Innovation ability dynamic evaluation system and method based on multi-dimensional data fusion analysis

Abstract

The invention relates to the technical field of intelligent education, and discloses an innovation ability dynamic evaluation system and method based on multi-dimensional data fusion analysis, and the system comprises a process data collection module, a literacy feature extraction module, a dynamic evaluation engine and a teaching intervention engine. The method corresponds to the system. According to the method, multi-dimensional, dynamic and accurate evaluation of the innovation ability of the students is realized by constructing a complete data processing and evaluation closed loop, and specifically, performance characteristics of the students in innovation practice can be comprehensively captured by fusing and analyzing online and offline multi-modal process data; by means of a multi-mode cross-domain time sequence fusion model, dynamic tracking of the innovation ability development trajectory is achieved; finally, a personalized teaching intervention strategy is generated based on an evaluation result, a complete teaching closed loop of evaluation-diagnosis-intervention is formed, and the accuracy and teaching efficiency of innovative talent training are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent education technology, specifically a dynamic evaluation system and method for innovation ability based on multi-dimensional data fusion analysis. Background Technology

[0002] In the field of higher education, deepening the reform of innovation and entrepreneurship education and cultivating high-quality innovative and entrepreneurial talents has become an important measure to serve the national innovation-driven development strategy. Currently, universities are actively exploring a dual-track approach to cultivating high-quality innovative and entrepreneurial talents. This dual-track model typically refers to a teaching environment that integrates online and offline learning, as well as a training path that combines innovation education with entrepreneurial practice.

[0003] However, in this dual-track innovation and entrepreneurship teaching practice, assessing students' innovative abilities still faces significant challenges. Traditional assessment methods rely heavily on summative evaluations, such as course completion exams, project reports, or competition results, making it difficult to capture students' process-oriented innovative behaviors during blended online and offline learning and project practice. Specifically, existing technologies have the following shortcomings: First, the multimodal data generated during online and offline teaching and practice (such as online discussions, solution iterations, and offline interactive performance) is fragmented, lacking effective means of integrated analysis, and unable to form a unified and comprehensive view of students' innovative abilities; second, assessment methods are mostly static and lagging, unable to dynamically track and quantify the development trajectory of students' innovative thinking and abilities during innovation and entrepreneurship practice; third, assessment results are often disconnected from precise teaching interventions, making it difficult to provide teachers with a basis for personalized teaching adjustments, and also unable to provide students with adaptive learning path guidance, thus limiting the achievement of the goals of integrated professional and innovative education.

[0004] Therefore, there is an urgent need for a more reliable dynamic assessment technology for innovation capabilities. Summary of the Invention

[0005] The purpose of this application is to provide a dynamic evaluation system and method for innovation capability based on multidimensional data fusion analysis, so as to solve the technical problems mentioned in the background.

[0006] To achieve the above objectives, this application discloses the following technical solutions: Firstly, this application discloses a dynamic evaluation system for innovation capabilities based on multi-dimensional data fusion analysis, the system comprising: The process data acquisition module is used to collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; the multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data; The competency feature extraction module, connected to the process data acquisition module, is used to process the multimodal process data and extract and quantify innovative competency features in multiple dimensions; the innovative competency features include problem discovery features, solution iteration features, knowledge integration features, and team leadership features; The dynamic evaluation engine, connected to the literacy feature extraction module, is used to receive the innovative literacy features in multiple dimensions, perform comprehensive calculations through a multimodal cross-domain temporal fusion model, and output a student ability profile with a dynamic innovation ability index and innovative literacy feature scores in each dimension. The teaching intervention engine, connected to the dynamic assessment engine, is used to receive the student's ability profile and generate personalized teaching intervention instructions based on the student's ability profile. The teaching intervention instructions include pushing targeted tutor matching suggestions, learning resource support plans, and innovative training tasks to the teaching management platform.

[0007] Optionally, the process data acquisition module is also used to complete the missing offline data by extracting the semantic features of the online discussion texts from the same period, generating corresponding behavior embedding vectors, and aligning the online data with the offline data in the time series.

[0008] Optionally, the literacy feature extraction module includes: The problem discovery and analysis unit is configured to: extract keywords and perform semantic analysis on the online discussion text and task solutions, and generate problem discovery feature values ​​by calculating the frequency and contextual relevance of learning difficulties, limitations of existing solutions and interdisciplinary knowledge conflicts involved in the text. The scheme iteration analysis unit is configured to: track and analyze the logical structure changes and content additions and deletions between the iterative versions of the task scheme, and generate scheme iteration feature values ​​by calculating the modification magnitude between versions and the number and frequency of new knowledge points introduced.

[0009] Optionally, the literacy feature extraction module further includes: The knowledge integration and analysis unit is configured to: analyze the learning resource application records and task plans, identify and count the types and quantities of interdisciplinary academic resources, experimental resources and data resources called by students, and evaluate the matching degree of the interdisciplinary academic resources, experimental resources and data resources with the application scenarios respectively, and generate knowledge integration feature values; The leadership analysis unit is configured to: perform role recognition on the online discussion text, and perform voice activity detection and body language analysis on the offline classroom interaction and presentation data; and generate team leadership characteristic values ​​by calculating the frequency and effectiveness of students' behaviors in initiating initiatives, coordinating differences, and summarizing statements in the team.

[0010] Optionally, the multimodal cross-domain temporal fusion model includes: The data preprocessing unit is configured to: align and standardize the innovation literacy features across multiple dimensions according to the time series, and output a standardized time-series literacy feature matrix; The temporal feature extraction unit, connected to the data preprocessing unit, is configured to: extract the evolution pattern of the temporal literacy feature matrix within the course cycle through a gated recurrent unit network, and output a high-level temporal feature vector; The cross-domain information fusion unit, connected to the temporal feature extraction unit, is configured to: calculate the association weight between text semantic features and offline interactive behavior features using a cross-domain attention mechanism on the high-level temporal feature vector, and perform cross-domain information fusion on the high-level temporal feature vector based on the association weight, and output the fused cross-domain feature representation. The course stage adaptation unit is configured to generate a course stage adaptation factor based on the current course task type. The capability index calculation unit, which connects the cross-domain information fusion unit and the course stage adaptation unit, is configured to: dynamically weight the cross-domain feature representation based on the course stage adaptation factor to calculate the dynamic innovation capability index and the innovation literacy feature scores of each dimension.

[0011] Optionally, the dynamic evaluation engine is also used for: For the iterative feature values ​​corresponding to the iterative feature of the proposed solution, a preset learning frustration scenario classification model is used to identify the types of frustration encountered by students, such as solution rejection, experiment failure, or data not meeting expectations. Based on the differences in the types of frustration, the weights of iterative enthusiasm and improvement depth are calculated to generate a differentiated resilience index. The dynamic evaluation engine uses the resilience index as a correction factor to correct the dynamic innovation capability index.

[0012] Optionally, the instructional intervention engine includes: The ability deficiency diagnosis unit is used to analyze the innovation literacy characteristic scores of each dimension in the student's ability profile. By comparing the scores with preset ability thresholds or group average scores, it identifies dimensions with scores below the threshold or ranking at the bottom, and defines the identified dimensions as ability deficiency characteristics. The mentor intelligent matching unit, connected to the capability deficiency diagnosis unit, is configured to: calculate the similarity between the capability deficiency features and the professional field tags of mentors in the mentor database, and generate and push a mentor matching recommendation list; The resource precision push unit, connected to the capability deficiency diagnosis unit, is configured to: match relevant academic literature, classic experimental cases or technical tutorials from the knowledge base according to the capability deficiency characteristics, and generate and push learning resource packages. The practical training task recommendation unit, connected to the capability deficiency diagnosis unit, is configured to: match practical tasks for targeted training of the corresponding capability deficiency from a preset practical training task library based on the capability deficiency characteristics, and generate and push a practical training task recommendation list.

[0013] Optionally, the instructional intervention engine also includes: The early warning and intervention unit is configured to continuously monitor the dynamic innovation ability index and send an early warning message to the teacher when the dynamic innovation ability index shows a continuous decline for a preset period or falls below a preset score threshold.

[0014] Optionally, the system also includes a development trajectory display module, which is connected to the dynamic evaluation engine to retrieve the student's historical dynamic innovation ability index and generate and display the student's innovation ability development trajectory in the form of a line graph.

[0015] Secondly, this application discloses a method for applying to the above-mentioned dynamic evaluation system of innovation capability based on multidimensional data fusion analysis, the method comprising: Process data collection steps: Collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; the multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data; The steps for extracting innovative competency features are as follows: The multimodal process data is processed to extract and quantify innovative competency features across multiple dimensions; these innovative competency features include problem discovery features, solution iteration features, knowledge integration features, and team leadership features. Dynamic assessment steps: Receive the innovation literacy characteristics from multiple dimensions, perform comprehensive calculations through a multimodal cross-domain temporal fusion model, and output a student ability profile with a dynamic innovation ability index and innovation literacy characteristic scores for each dimension; Teaching intervention steps: Receive the student's ability profile and generate personalized teaching intervention instructions based on the student's ability profile. The teaching intervention instructions include pushing targeted tutor matching suggestions, learning resource support plans and innovative training tasks to the teaching management platform.

[0016] Beneficial Effects: The innovative ability dynamic evaluation system and method based on multidimensional data fusion analysis proposed in this application achieves multidimensional, dynamic, and accurate evaluation of students' innovative abilities by constructing a complete data processing and evaluation closed loop. Specifically, by fusing and analyzing online and offline multimodal process data, it can comprehensively capture the performance characteristics of students in innovative practices; by using a multimodal cross-domain temporal fusion model, it can dynamically track the development trajectory of innovative abilities; and finally, based on the evaluation results, it generates personalized teaching intervention strategies, forming a complete teaching closed loop of evaluation-diagnosis-intervention, thereby improving the accuracy of innovative talent cultivation and teaching efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A structural block diagram of the dynamic evaluation system for innovation capability based on multidimensional data fusion analysis provided in this application embodiment; Figure 2 A flowchart illustrating the dynamic evaluation method for innovation capability based on multidimensional data fusion analysis provided in this application embodiment. Detailed Implementation

[0019] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0020] Driven by the innovation-driven development strategy, the reform of innovation and entrepreneurship education in higher education has become a core approach to cultivating high-quality talent. The current dual-track model for cultivating innovation and entrepreneurship talent explored by universities encompasses two core elements: first, the dual-track integration of the teaching environment, namely, the deep integration of digital learning spaces built on online collaborative platforms (such as project management tools and online discussion communities) with physical scenarios such as offline laboratories, classroom interactions, and project training; second, the dual-track synergy of the training path, namely, conveying interdisciplinary knowledge and innovative thinking methods through theoretical innovation education, while simultaneously enhancing students' practical operation and problem-solving abilities through concrete entrepreneurial practices (such as academic competitions and real project incubation).

[0021] However, the current assessment system for students' innovative abilities is ill-suited to this dual-track, dual-innovation teaching scenario. Traditional assessments are primarily summative, relying solely on static results such as course exam scores, final project reports, or competition rankings, which fail to reflect students' dynamic innovative performance throughout the learning and practical process. Specifically, existing assessment methods have three major limitations: First, online discussion texts, solution iteration records, and resource application data are independent of offline multimodal data such as classroom interaction performance, group presentation effects, and practical operation processes. This lack of effective integration and analysis leads to a fragmented perspective on assessing students' innovative abilities, making it difficult to form a comprehensive understanding. Second, assessments often occur at the end of the teaching phase, constituting post-hoc evaluation. This makes it impossible to track the evolution of students' innovative thinking and the growth trajectory of their innovative abilities in different course stages and practical tasks in real time, resulting in a significant lag in assessment results. Third, assessment results serve only as conclusive judgments of students' abilities and are not linked to adjustments in teaching. Teachers cannot accurately identify students' weaknesses and implement personalized guidance based on assessment results, and students cannot obtain learning path suggestions tailored to their individual circumstances. Ultimately, this significantly diminishes the integration of innovation education and entrepreneurial practice, hindering the achievement of the goal of cultivating innovative and entrepreneurial talents.

[0022] Therefore, this embodiment provides an innovation capability assessment technology that can adapt to dual-line scenarios, integrate multi-source data, dynamically track capability development, and support precise intervention.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application. Secondly, in this document, the term "comprising" is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements, but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0024] This embodiment provides, in its first aspect, a dynamic evaluation system for innovation capability based on multi-dimensional data fusion analysis, such as... Figure 1 As shown, it includes a process data acquisition module, a competency feature extraction module, a dynamic assessment engine, and a teaching intervention engine, as detailed below.

[0025] The process data acquisition module is used to collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; the multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data.

[0026] In practical implementation, online collaboration platforms can utilize commonly used learning management systems in universities (such as Chaoxing Learning Platform and Rain Classroom), project collaboration tools (such as Trello and Lark Docs), or professional innovation and entrepreneurship platforms, capturing relevant data in real time via API interfaces. Learning resource application records are extracted from university library digital resource platforms (such as CNKI and Web of Science university mirror sites) through user behavior log interfaces. Offline data is collected using hardware such as classroom interaction recorders and group discussion recording devices, including speaking duration, speaking volume, body language, and peer evaluation scores.

[0027] The competency feature extraction module, connected to the process data acquisition module, is used to process multimodal process data and extract and quantify innovative competency features across multiple dimensions. These innovative competency features include problem discovery features, solution iteration features, knowledge integration features, and team leadership features.

[0028] In practice, the unstructured data collected by the process data acquisition module is converted into standardized feature scores of 0-100 using the core quantification formula described later, and feature scores for each dimension are output.

[0029] The dynamic assessment engine, connected to the literacy feature extraction module, receives innovation literacy features from multiple dimensions. It performs comprehensive calculations through a multimodal cross-domain temporal fusion model, and outputs a student competency profile with a dynamic innovation ability index and scores for innovation literacy features in each dimension.

[0030] In practice, the capability profile is presented in the form of a structured report, which includes an analysis of the overall level and the strengths and weaknesses of each dimension.

[0031] The teaching intervention engine, connected to the dynamic assessment engine, receives student competency profiles and generates personalized teaching intervention instructions based on these profiles. These instructions include pushing targeted tutor matching suggestions, learning resource support plans, and innovative practical training tasks to the teaching management platform. Based on the above, the innovation capability dynamic assessment system based on multi-dimensional data fusion analysis in this embodiment constructs a complete closed loop from data collection to assessment and intervention through the collaborative operation of the process data acquisition module, the competency feature extraction module, the dynamic assessment engine, and the teaching intervention engine. This effectively solves the fragmentation and static nature of traditional assessments. Specifically, the process data acquisition module breaks down the barriers between online and offline data, providing a rich foundation for comprehensive assessment; the competency feature extraction module transforms abstract innovation capabilities into quantifiable features through innovation quantification formulas, achieving process characterization; the dynamic assessment engine accurately captures the temporal changes in capability development; and the teaching intervention engine deeply integrates assessment results with teaching practice, improving the accuracy and efficiency of innovation and entrepreneurship talent cultivation.

[0032] As an optional implementation of this embodiment, the process data acquisition module is also used to complete the missing offline data by extracting the semantic features of the online discussion texts from the same period, generating corresponding behavior embedding vectors, and aligning the online data with the offline data in the time series.

[0033] In practice, semantic feature extraction uses a BERT pre-trained model (selecting the BERT-base-chinese version, with a pre-training corpus containing over 1 million discussion texts from university innovation and entrepreneurship courses), generating 256-dimensional behavior embedding vectors. This vector can accurately map students' cognitive activity and participation during the same period.

[0034] Secondly, this embodiment uses the following data completion formula to complete the missing offline data. Calculation: in, It is the Sigmoid activation function. This is a scaled dot product attention mechanism used to compute the embedding vector. With online data vectors The correlation weights are used to highlight the impact of highly relevant online data on the completion results. This represents the theoretical maximum value for offline data (examples: maximum classroom interaction frequency of 20 times / week, maximum presentation score of 100 points, maximum group collaboration speaking time of 30 minutes / task).

[0035] Furthermore, time series alignment uses course task nodes (such as submission of the initial draft, mid-term report, final defense, etc.) as anchor points, and follows... The weekly time window links online data with supplementary offline data to ensure temporal and spatial synchronization.

[0036] Based on the above, the innovation capability dynamic evaluation system based on multi-dimensional data fusion analysis in this embodiment can ensure the integrity and coherence of the data by completing missing offline data. Furthermore, the data completion formula incorporates a scaling dot product attention mechanism, which can adaptively mine the inherent correlation between online data and missing offline data, avoiding deviations caused by simple interpolation and ensuring the rationality and accuracy of the completed data. Time series alignment further guarantees the integrity and coherence of the dataset, providing reliable data support for subsequent evaluation and analysis, and improving the comprehensiveness of the innovation capability evaluation.

[0037] As an optional implementation of this embodiment, the competency feature extraction module includes a problem discovery and analysis unit, a solution iteration analysis unit, a knowledge integration and analysis unit, and a leadership analysis unit.

[0038] The problem discovery and analysis unit is configured to: extract keywords and perform semantic analysis on online discussion texts and task solutions; and generate problem discovery feature values ​​by calculating the frequency and contextual relevance of learning difficulties, limitations of existing solutions, and interdisciplinary knowledge conflicts involved in the text.

[0039] In practice, keyword extraction uses the TF-IDF algorithm, with a word frequency statistics window of 500 characters and an IDF smoothing coefficient. The keyword selection threshold is 0.3. Semantic analysis uses the LDA topic model, with 5 topics, 1000 iterations, α=0.1, β=0.01, covering core topics related to problem discovery.

[0040] Secondly, this embodiment calculates the eigenvalues ​​for problem discovery using the following formula. : in, These correspond to three types of problems: learning difficulties, limitations of the solution, and knowledge conflicts. This refers to the percentage of keyword frequency for the corresponding category (e.g., if a student's discussion text contains 12 occurrences of the keyword "difficulty" and a total of 30 occurrences of the keyword, then...). ), The context relevance score is calculated by taking the cosine similarity between the core keyword and the context, with a value ranging from [0,1]. For example, if the keyword "interdisciplinary conflict" has a similarity of 0.85 with the context, then... ), The information entropy calculation function (which quantifies the dispersion of the frequency distribution of the three types of questions to reflect the diversity of the types of questions students discover) avoids the one-sidedness of a single dimension.

[0041] The solution iteration analysis unit is configured to: track and analyze the changes in logical structure and the addition and deletion of content between the iterative versions of the task solution, and generate solution iteration feature values ​​by calculating the modification range between versions and the number and frequency of new knowledge points introduced.

[0042] In practical implementation, the iterative eigenvalues ​​of the scheme are calculated using the following formula. : in, This represents the total number of iterations of the solution (example: if a student submits 5 versions of the solution, then...). ), To improve the edit distance (because the replacement of the core content of the solution logic has a greater impact on innovation than simple insertion / deletion, it can be designed with a weight of 1.5 for replacement operations, 1.2 for insertion operations, and 1.0 for deletion operations), The number of knowledge points in version i (determined by matching with a pre-set academic dictionary containing 1000+ interdisciplinary core knowledge points, reviewed by 3 associate professors or above; for example, version V1 contains 8 knowledge points, and version V2 contains 12 knowledge points). , ), Let $\frac{i}{i+1}$ be the iteration time interval between version $i$ and version $i+1$ (in days, e.g., if V1 is submitted on day 7 and V2 is submitted on day 10, then $\frac{i}{i+1}$ is the iteration time interval between version $i$ and version $i+1$. The exponent is the rate of knowledge point growth. This formula accurately reflects the students' enthusiasm and innovation in iterating their solutions.

[0043] The knowledge integration and analysis unit is configured to: analyze learning resource application records and task plans; identify and count the types and quantities of interdisciplinary academic resources, experimental resources and data resources called by students; evaluate the matching degree between interdisciplinary academic resources, experimental resources and data resources and application scenarios; and generate knowledge integration feature values.

[0044] In practical implementation, the knowledge integration feature value is calculated using the following formula. : in, The number of interdisciplinary resource types accessed by students (e.g., engineering papers, science datasets, management case studies). ), The maximum number of subject types involved in the course (the maximum number of subject types is determined based on the course syllabus, such as interdisciplinary courses). Typically 3-5, single-subject courses For example, the course "Artificial Intelligence + Entrepreneurship Management" ), The matching degree for a single resource is calculated using the Analytic Hierarchy Process (AHP). In the analysis, relevance is weighted at 0.5, applicability at 0.3, and innovation at 0.2. The score is determined by five experts in innovation and entrepreneurship education based on the Delphi method. For example, if a resource scores 0.9 for relevance, 0.8 for applicability, and 0.7 for innovation, then... ), The total number of resources (e.g., if students use 10 resources in total, then...) This formula reflects the cross-domain nature and adaptability of resources through a composite calculation of the proportion of subject types and the average matching degree.

[0045] The leadership analysis unit is configured to: identify roles in online discussion texts and detect voice activity and body language in offline classroom interaction and presentation data; and generate team leadership characteristic values ​​by calculating the frequency and effectiveness of students' actions in initiating initiatives, coordinating disagreements, and summarizing presentations within the team.

[0046] In practical implementation, role recognition adopts a rule-based named entity recognition method. For example, statements containing "suggestion," "proposal," or "I think" are identified as initiating initiatives; statements containing "is it possible?" or "what do you all think?" are identified as resolving disagreements. Voice activity detection uses the VAD algorithm, and body language analysis uses the OpenPose model. For example, it focuses on detecting 18 key points in the head, hands, and torso to determine body language activity. For example, a gesture frequency of ≥5 times / minute is considered active.

[0047] Secondly, in this embodiment, the team leadership characteristic value is calculated using the following formula. : in, These correspond to three types of leadership behaviors: initiating initiatives, coordinating differences, and summarizing and presenting findings. The frequency of the corresponding behavior (e.g., if a student initiates an initiative 4 times, and the maximum frequency within the same team is 5 times, then...) , ), This represents the highest frequency of this behavior among all students in the same team. The effectiveness score is calculated by weighting the teacher's score (60%) and the peer review score (40%). For example, if the teacher's score is 90 and the peer review score is 85, then... This formula takes into account both frequency percentage and behavior quality, objectively reflecting the team's leadership ability.

[0048] Based on the above, the innovation ability dynamic evaluation system based on multi-dimensional data fusion analysis in this embodiment, with its problem discovery analysis unit and solution iteration analysis unit, achieves quantitative extraction of the core dimensions of students' innovative literacy, solving the problem of the difficulty in concretizing innovative behavior in traditional assessments. Furthermore, the problem discovery feature value formula is implemented through an information entropy function, taking into account both the intensity and diversity of problem identification types, avoiding the one-sidedness of a single dimension; the solution iteration feature value formula integrates improved edit distance and knowledge point growth rate, accurately characterizing the quality of iteration and the incremental innovation.

[0049] Secondly, the knowledge integration analysis unit and leadership analysis unit improve the innovation literacy characteristic system, covering two key capabilities: cross-domain knowledge application and teamwork. Furthermore, the knowledge integration characteristic value formula, through a non-linear composite of subject type proportion and AHP matching degree, reflects the depth and adaptability of cross-domain integration better than traditional counting methods. The leadership characteristic value formula takes into account both behavior frequency proportion and effectiveness, avoiding the bias of judging success solely by frequency.

[0050] As an optional implementation of this embodiment, the multimodal cross-domain temporal fusion model includes a data preprocessing unit, a temporal feature extraction unit, a cross-domain information fusion unit, a course stage adaptation unit, and a capability index calculation unit, as detailed below.

[0051] The data preprocessing unit is configured to align and standardize the innovation literacy features across multiple dimensions according to the time series, and output a standardized time-series literacy feature matrix. In specific implementation, the standardized features are normalized using Z-score.

[0052] The temporal feature extraction unit, connected to the data preprocessing unit, is configured to extract the evolution pattern of the temporal literacy feature matrix within the course cycle through a gated recurrent unit network (GRU), outputting a high-level temporal feature vector. In specific implementation, the GRU contains two hidden layers, has a dimension of 128, undergoes 100 training iterations, has a learning rate of 0.001, uses Adam as the optimizer, has a dropout rate of 0.2, and is determined based on 5-fold cross-validation.

[0053] The cross-domain information fusion unit, connected to the temporal feature extraction unit, is configured to: calculate the association weights between textual semantic features and offline interaction behavior features using a cross-domain attention mechanism on the high-level temporal feature vector; and perform cross-domain information fusion on the high-level temporal feature vector based on these association weights, outputting the fused cross-domain feature representation. In specific implementation, the cross-domain feature representation is calculated using the following formula. : in, It is a text semantic feature sub-vector (the first 64 dimensions of the high-level temporal feature vector). This is the offline behavioral feature sub-vector (the last 64 dimensions of the high-level temporal feature vector). The average weight for cross-domain attention is obtained by taking the row mean after calculating the association matrix between text and behavioral features through a scaled dot product attention mechanism. The value ranges from [0,1]. For example, if the association degree between a student's text and behavioral features is 0.7, then... This formula enables deep fusion of multimodal data, breaking down data barriers.

[0054] The course stage adaptation unit is configured to generate course stage adaptation factors based on the current course task type. In practice, based on the teaching objectives of each stage, the experimental and practical stage focuses on the practical application of innovation; therefore, the adaptation factors... This could be, for example, the theoretical learning stage. Project design phase Experimental and practical stage Comprehensive display phase .

[0055] The capability index calculation unit, connecting the cross-domain information fusion unit and the curriculum stage adaptation unit, is configured to: dynamically weight the cross-domain feature representation based on the curriculum stage adaptation factor to calculate the dynamic innovation capability index and the innovation literacy feature scores for each dimension. In specific implementation, the dynamic innovation capability index is calculated using the following formula. : in, The output range is 0-100 points. , For the pre-trained stage-adapted weight matrix (e.g., all dimensions are 128×128, trained using historical data from 500+ courses, covering engineering, humanities, and business, with complete data from at least 30 students per course, and the training loss function is MSE), this formula accurately outputs the standardized ability index through stage-based dynamic weighting and Sigmoid nonlinear transformation.

[0056] Based on the above, the innovation capability dynamic evaluation system based on multidimensional data fusion analysis in this embodiment achieves dynamic tracking and comprehensive calculation of innovation capability through a multimodal cross-domain temporal fusion model, breaking through the limitations of traditional static evaluation. Furthermore, the cross-domain feature representation and the calculation formula for the dynamic innovation capability index enhance the model's adaptability and accuracy. Specifically: the cross-domain information fusion formula adaptively integrates text and behavioral features through cross-domain attention average weights, solving the core problem of fragmented multimodal data; the dynamic innovation capability index formula is based on a pre-training stage-adapted weight matrix, adapting to the teaching characteristics of different course stages, and achieves standardized score output through Sigmoid mapping.

[0057] Based on the aforementioned multimodal cross-domain temporal fusion model, as a further optional implementation method in this embodiment, the dynamic evaluation engine is also used for: For the iterative feature values ​​corresponding to the iterative feature characteristics of the solution, the preset learning frustration scenario classification model is used to identify the types of frustrations encountered by students, such as solution rejection, experiment failure, or data not meeting expectations. The weights of iterative enthusiasm and improvement depth are calculated according to the differences in frustration types to generate differentiated resilience indices. The dynamic evaluation engine uses the resilience index as a correction factor to correct the dynamic innovation capability index.

[0058] In practice, the learning frustration scenario classification model adopts a CNN-LSTM hybrid model. The CNN layer has a convolution kernel size of 3×3 and a number of 64, and a pooling window of 2×2. The LSTM layer has a hidden layer dimension of 64 and a dropout rate of 0.3. The training set accounts for 80%, the validation set 10%, and the test set 10%. The training optimizer is RMSProp.

[0059] Secondly, in this embodiment, the toughness index is calculated using the following formula. : in, The value range is [0,1]; The learning frustration type (identified by a CNN-LSTM hybrid model); The differentiated weights for iterative motivation were determined using the Delphi method, a method developed by experts in innovation and entrepreneurship education. To improve the depth-based differential weights (satisfying) (This was determined by the Delphi method, an expert in innovation and entrepreneurship education;) To incentivize iteration, this quantifies the level of initiative in modifying the solution per unit of time; for example, if modifications are made ≥3 times per week, then... If it is version 1.0 and modified twice a week, then... If the value is 0.7 and it is modified once a week, then... If the value is 0.3 and there are 0 modifications per week, then... =0; To improve the depth of the analysis, this is used to quantify the effectiveness of the modifications in addressing the issues. For example, if the core knowledge points are modified at a rate of ≥50%, then... If the core knowledge points are modified in version 1.0, then... If the score is 0.7 and the percentage of changes to core knowledge points is less than 30%, then... It is 0.3; The time delay (in days) for the first iteration after a setback is automatically recorded and statistically analyzed by the scheme iteration analysis unit (e.g., if a revised version is submitted on the 3rd day after the setback). ); The average iteration delay (in days) after setbacks for students in the same course is based on historical data statistics (example: if the average delay for a course is 4 days, then...). ).

[0060] Furthermore, in this embodiment, the modified dynamic innovation capability index is calculated using the following formula. : in, This is the revised dynamic innovation capability index. It is a dynamic innovation capability index.

[0061] Based on the above, the innovation capability dynamic evaluation system based on multi-dimensional data fusion analysis in this embodiment corrects the dynamic innovation capability index through the resilience index, supplementing the dimension of resilience that is ignored in traditional evaluation, making the capability evaluation more comprehensive. Furthermore, by identifying the type of setback based on the CNN-LSTM hybrid model and correcting the dynamic innovation capability index, the system achieves precision and differentiation in resilience evaluation.

[0062] As an optional implementation of this embodiment, the teaching intervention engine includes a competency deficiency diagnosis unit, an intelligent tutor matching unit, a precise resource push unit, and a practical training task recommendation unit, as detailed below.

[0063] The competency gap diagnosis unit is used to analyze the scores of innovation literacy characteristics in various dimensions of students' competency profiles. By comparing the scores with preset competency thresholds or group average scores, it identifies dimensions with scores below the threshold or ranking at the bottom, and defines the identified dimensions as competency gap characteristics.

[0064] In practice, the ability threshold is based on conventional standards of passing in educational evaluation, combined with historical data statistics from over 1000 students, ensuring that approximately 80% of students' initial scores are higher than this threshold. For example, the passing score for each dimension is 60 points, with a maximum score of 100 points. The group average score can be the arithmetic mean of the corresponding dimensions for students in the same grade and course, calculated and updated every two weeks to ensure timeliness. An example of an ability weakness is: a student's knowledge integration characteristic score is 55 points, lower than 60 points, thus their knowledge integration dimension is considered a weakness.

[0065] The mentor intelligent matching unit, connected to the capability deficiency diagnosis unit, is configured to: calculate the similarity between capability deficiency features and the professional field tags of mentors in the mentor database, and generate and push a mentor matching recommendation list.

[0066] In practice, cosine similarity calculation can be any existing technology. When determining a matching mentor, if the similarity threshold is set to 0.6, mentors with cosine similarity results higher than this threshold are considered suitable. An example mentor matching recommendation list could be sorted by similarity from high to low, pushing 3-5 mentors. Examples of mentor tags in the database could be: "Problem Discovery Ability Guidance - Mechanical Engineering", "Interdisciplinary Knowledge Integration - Artificial Intelligence + Management", and "Team Collaboration Guidance - Marketing".

[0067] The resource precision push unit, which connects to the capability deficiency diagnosis unit, is configured to: match relevant academic literature, classic experimental cases, or technical tutorials from the knowledge base based on the characteristics of capability deficiencies, and generate and push learning resource packages.

[0068] In practice, the learning resource package will provide 5-8 core resources for each weakness. The types of resources can be as follows: academic literature includes SCI / core journal papers from the past 5 years; case studies include award-winning projects from the National College Students Innovation and Entrepreneurship Competition; and technical tutorials include high-quality MOOC course videos.

[0069] The training task recommendation unit, connected to the capability deficiency diagnosis unit, is configured to: match targeted training tasks for the corresponding capability deficiency from a preset training task library based on the characteristics of the capability deficiency, and generate and push a training task recommendation list.

[0070] In practice, the training task recommendation list will push 2-3 suitable tasks for each weakness. The gradient can be as follows: basic task "single discipline resource integration exercise" → intermediate task "interdisciplinary project design" → advanced task "incubation of cross-disciplinary innovation projects".

[0071] Based on the above, the innovation ability dynamic assessment system based on multi-dimensional data fusion analysis in this embodiment generates personalized teaching intervention instructions through the teaching intervention engine, realizing the linkage between assessment and intervention. It solves the problem of the disconnect between traditional assessment and teaching, providing teachers with a clear direction for teaching adjustments and students with adaptive learning path support. It effectively promotes the filling of students' ability gaps and the comprehensive improvement of their innovative literacy, significantly enhancing the teaching effect of innovation and entrepreneurship education.

[0072] Based on the aforementioned teaching intervention engine, as a further optional implementation of this embodiment, the teaching intervention engine also includes: The early warning and intervention unit is configured to continuously monitor the dynamic innovation capability index and send an early warning message to the teacher's end when the dynamic innovation capability index shows a continuous decline for a preset period or falls below a preset score threshold.

[0073] In practice, the continuous preset cycle can be set as 3 course cycles. Based on the course teaching progress, which is typically 16-18 weeks, 3 weeks constitutes a small teaching unit, facilitating timely intervention and preventing the accumulation of problems. The score threshold can be set at 50 points (out of 100). A score below this indicates a minimum of 10 points below the aforementioned passing mark, signifying a severe deficiency in ability requiring urgent intervention.

[0074] Secondly, the warning information includes the student's basic information, an index trend chart, characteristics of ability deficiencies, and preliminary intervention suggestions (example: "Student XXX, knowledge integration ability has declined for 3 consecutive weeks, it is recommended to match with an interdisciplinary mentor and push the 'interdisciplinary resource integration' special resource package"). It is pushed through pop-up notifications on the teaching management platform and email reminders to the linked email address (the email subject is marked "Innovation Ability Warning - Student XXX - Course XXX").

[0075] Based on the above, the innovation ability dynamic evaluation system based on multi-dimensional data fusion analysis in this embodiment realizes real-time monitoring and proactive intervention of students' innovation ability development through the early warning intervention unit. This avoids the continuous expansion of ability shortcomings, breaks the limitations of traditional post-assessment remediation, realizes the transformation from pre-warning to in-process intervention, effectively curbs the further decline of students' abilities, and provides strong support for ensuring the quality of innovation and entrepreneurship education.

[0076] As an optional implementation of this embodiment, the system also includes a development trajectory display module, which is connected to the dynamic evaluation engine to retrieve the student's historical dynamic innovation ability index and generate and display the student's innovation ability development trajectory in the form of a line graph.

[0077] In practical implementation, the trajectory graph parameters can be designed as follows: the horizontal axis time granularity can be selected as week / month (default is week), the vertical axis score precision is 1 point, and it supports hovering the mouse to view the score at a specific time point and the contribution ratio of each dimension (example: hovering over week 8, the index is displayed as 75 points, of which problem discovery accounts for 25%, solution iteration accounts for 30%, knowledge integration accounts for 20%, and team leadership accounts for 25%); the trajectory graph is overlaid with the average index curve of the same group (same grade and same course) (displayed as a dashed line), and it supports viewing each dimension of innovation literacy separately; access permission design: students can only view their own trajectory, teachers can view the trajectory of students in their classes, and administrators can view the school-wide data summary (no personal privacy information).

[0078] Based on the above, the innovation ability dynamic evaluation system based on multi-dimensional data fusion analysis in this embodiment presents the development process of students' innovation ability in a visual way through the development trajectory display module, making the changes in ability intuitive and perceptible. This not only helps students intuitively understand their own progress and shortcomings and adjust their learning strategies, but also provides teachers with intuitive feedback on teaching effectiveness, making it easier to accurately locate teaching problems, promoting effective communication between teachers and students, and driving the virtuous cycle of assessment-diagnosis-intervention-improvement.

[0079] This embodiment provides a method for a dynamic evaluation system of innovation capability based on multidimensional data fusion analysis in a second aspect, which is applied to the system described above for a dynamic evaluation system of innovation capability based on multidimensional data fusion analysis, such as... Figure 2 As shown, the method includes: Process data collection steps: Collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data; The steps for extracting innovative competency features are as follows: Multimodal process data is processed to extract and quantify innovative competency features across multiple dimensions; these features include problem identification, solution iteration, knowledge integration, and team leadership characteristics. Dynamic assessment steps: Receive innovation literacy characteristics from multiple dimensions, perform comprehensive calculations through a multimodal cross-domain temporal fusion model, and output a student ability profile with a dynamic innovation ability index and scores for innovation literacy characteristics in each dimension; Teaching intervention steps: Receive student ability profiles and generate personalized teaching intervention instructions based on these profiles. These instructions include pushing targeted tutor matching suggestions, learning resource support plans, and innovative training tasks to the teaching management platform.

[0080] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0081] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A dynamic evaluation system for innovation capability based on multidimensional data fusion analysis, characterized in that, The system includes: The process data acquisition module is used to collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; the multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data; The competency feature extraction module, connected to the process data acquisition module, is used to process the multimodal process data and extract and quantify innovative competency features in multiple dimensions; the innovative competency features include problem discovery features, solution iteration features, knowledge integration features, and team leadership features; The dynamic evaluation engine, connected to the literacy feature extraction module, is used to receive the innovative literacy features in multiple dimensions, perform comprehensive calculations through a multimodal cross-domain temporal fusion model, and output a student ability profile with a dynamic innovation ability index and innovative literacy feature scores in each dimension. The teaching intervention engine, connected to the dynamic assessment engine, is used to receive the student's ability profile and generate personalized teaching intervention instructions based on the student's ability profile. The teaching intervention instructions include pushing targeted tutor matching suggestions, learning resource support plans, and innovative training tasks to the teaching management platform.

2. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 1, characterized in that, The process data acquisition module is also used to complete the missing offline data by extracting the semantic features of the online discussion texts from the same period, generating corresponding behavior embedding vectors, and aligning the online data with the offline data in the time series.

3. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 1, characterized in that, The literacy feature extraction module includes: The problem discovery and analysis unit is configured to: extract keywords and perform semantic analysis on the online discussion text and task solutions, and generate problem discovery feature values ​​by calculating the frequency and contextual relevance of learning difficulties, limitations of existing solutions and interdisciplinary knowledge conflicts involved in the text. The scheme iteration analysis unit is configured to: track and analyze the logical structure changes and content additions and deletions between the iterative versions of the task scheme, and generate scheme iteration feature values ​​by calculating the modification magnitude between versions and the number and frequency of new knowledge points introduced.

4. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 3, characterized in that, The literacy feature extraction module also includes: The knowledge integration and analysis unit is configured to: analyze the learning resource application records and task plans, identify and count the types and quantities of interdisciplinary academic resources, experimental resources and data resources called by students, and evaluate the matching degree of the interdisciplinary academic resources, experimental resources and data resources with the application scenarios respectively, and generate knowledge integration feature values; The leadership analysis unit is configured to: perform role recognition on the online discussion text, and perform voice activity detection and body language analysis on the offline classroom interaction and presentation data; and generate team leadership characteristic values ​​by calculating the frequency and effectiveness of students' behaviors in initiating initiatives, coordinating differences, and summarizing statements in the team.

5. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 1, characterized in that, The multimodal cross-domain temporal fusion model includes: The data preprocessing unit is configured to: align and standardize the innovation literacy features across multiple dimensions according to the time series, and output a standardized time-series literacy feature matrix; The temporal feature extraction unit, connected to the data preprocessing unit, is configured to: extract the evolution pattern of the temporal literacy feature matrix within the course cycle through a gated recurrent unit network, and output a high-level temporal feature vector; The cross-domain information fusion unit, connected to the temporal feature extraction unit, is configured to: calculate the association weight between text semantic features and offline interactive behavior features using a cross-domain attention mechanism on the high-level temporal feature vector, and perform cross-domain information fusion on the high-level temporal feature vector based on the association weight, and output the fused cross-domain feature representation. The course stage adaptation unit is configured to generate a course stage adaptation factor based on the current course task type. The capability index calculation unit, which connects the cross-domain information fusion unit and the course stage adaptation unit, is configured to: dynamically weight the cross-domain feature representation based on the course stage adaptation factor to calculate the dynamic innovation capability index and the innovation literacy feature scores of each dimension.

6. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 5, characterized in that, The dynamic evaluation engine is also used for: For the iterative feature values ​​corresponding to the iterative feature of the proposed solution, a preset learning frustration scenario classification model is used to identify the types of frustration encountered by students, such as solution rejection, experiment failure, or data not meeting expectations. Based on the differences in the types of frustration, the weights of iterative enthusiasm and improvement depth are calculated to generate a differentiated resilience index. The dynamic evaluation engine uses the resilience index as a correction factor to correct the dynamic innovation capability index.

7. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 1, characterized in that, The instructional intervention engine includes: The ability deficiency diagnosis unit is used to analyze the innovation literacy characteristic scores of each dimension in the student's ability profile. By comparing the scores with preset ability thresholds or group average scores, it identifies dimensions with scores below the threshold or ranking at the bottom, and defines the identified dimensions as ability deficiency characteristics. The mentor intelligent matching unit, connected to the capability deficiency diagnosis unit, is configured to: calculate the similarity between the capability deficiency features and the professional field tags of mentors in the mentor database, and generate and push a mentor matching recommendation list; The resource precision push unit, connected to the capability deficiency diagnosis unit, is configured to: match relevant academic literature, classic experimental cases or technical tutorials from the knowledge base according to the capability deficiency characteristics, and generate and push learning resource packages. The practical training task recommendation unit, connected to the capability deficiency diagnosis unit, is configured to: match practical tasks for targeted training of the corresponding capability deficiency from a preset practical training task library based on the capability deficiency characteristics, and generate and push a practical training task recommendation list.

8. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 7, characterized in that, The instructional intervention engine also includes: The early warning and intervention unit is configured to continuously monitor the dynamic innovation ability index and send an early warning message to the teacher when the dynamic innovation ability index shows a continuous decline for a preset period or falls below a preset score threshold.

9. The dynamic evaluation system for innovation capability based on multidimensional data fusion analysis according to claim 1, characterized in that, The system also includes a development trajectory display module, which is connected to the dynamic evaluation engine. This module retrieves the student's historical dynamic innovation ability index and generates and displays the student's innovation ability development trajectory in the form of a line graph.

10. A method for applying to a dynamic evaluation system for innovation capability based on multidimensional data fusion analysis as described in any one of claims 1-9, characterized in that, The method includes: Process data collection steps: Collect multimodal process data of students in course practice tasks from online collaboration platforms and offline physical spaces; the multimodal process data includes online discussion texts, iterative versions of task plans, learning resource application records, and offline classroom interaction and presentation data; The steps for extracting innovative competency features are as follows: The multimodal process data is processed to extract and quantify innovative competency features across multiple dimensions; these innovative competency features include problem discovery features, solution iteration features, knowledge integration features, and team leadership features. Dynamic assessment steps: Receive the innovation literacy characteristics from multiple dimensions, perform comprehensive calculations through a multimodal cross-domain temporal fusion model, and output a student ability profile with a dynamic innovation ability index and innovation literacy characteristic scores for each dimension; Teaching intervention steps: Receive the student's ability profile and generate personalized teaching intervention instructions based on the student's ability profile. The teaching intervention instructions include pushing targeted tutor matching suggestions, learning resource support plans and innovative training tasks to the teaching management platform.

Images