Multimodal intelligent education scene dynamic sensing method and system based on field domain theory

Abstract

The invention relates to the field of wisdom education, in particular to a multi-modal wisdom education scene dynamic sensing method and system based on a field domain theory. The method comprises the following steps: acquiring a production and education cooperation multi-modal behavior data set, performing cooperation role positioning and relationship network construction according to the production and education cooperation multi-modal behavior data set, analyzing structural power, knowledge flow and emotion connection dimensions, and generating a school-enterprise relationship health degree information set; according to the school-enterprise relationship health degree information set, carrying out multi-dimensional coupled collaborative situation awareness and risk mode identification, and generating a collaborative barrier risk mode set; and according to the collaborative barrier risk mode set, executing root cause traceability and adaptive strategy recommendation operation, and generating an intelligent education dynamic perception state report. In the education scene state perception process, early recognition and diagnosis of potential collaborative barriers and risk modes can be realized, and superficialization and formalization of collaborative processes are further solved.

Description

Technical Field

[0001] This application relates to the field of smart education, and in particular to a method and system for dynamic perception of multimodal smart education scenarios based on field theory. Background Technology

[0002] In the field of smart education and industry-education integration, building an efficient and healthy school-enterprise collaborative education model has become a core link in promoting the construction of a modern vocational education system. The ability to dynamically perceive and precisely control the collaborative process is directly related to the quality of talent training, the efficiency of knowledge innovation, and the effectiveness of resource allocation. It is a key cornerstone for building a modern industrial college and an innovative social ecosystem.

[0003] However, existing collaborative status assessment methods lack a mechanism for deep integration and dynamic attribution of key dimensions based on field theory when facing complex collaborative fields with multiple subjects, multiple modalities, and dynamic evolution. This not only makes it difficult to systematically deconstruct and diagnose potential collaborative barriers and risk patterns, but also easily leads to the superficiality and formalization of the collaborative process, ultimately resulting in innovation stagnation and internal resource consumption. Summary of the Invention

[0004] This application provides a method and system for dynamic perception of multimodal smart education scenarios based on field theory to solve the above-mentioned technical problems.

[0005] Firstly, this application provides a method for dynamic perception of multimodal smart education scenarios based on field theory. The method includes: acquiring a multimodal behavior dataset of industry-education collaboration; based on the multimodal behavior dataset of industry-education collaboration, performing collaborative role positioning and relationship network construction, analyzing structural power, knowledge flow, and emotional connection dimensions, and generating a health information set of school-enterprise relationships; based on the health information set of school-enterprise relationships, performing multi-dimensional coupled collaborative situational awareness and risk pattern recognition, and generating a set of collaborative barrier risk patterns; based on the set of collaborative barrier risk patterns, performing root cause tracing and adaptive strategy recommendation operations, and generating a dynamic perception status report of smart education.

[0006] The aforementioned technical solutions enable a comprehensive and dynamic perception of the industry-education collaboration ecosystem, overcoming the limitations of traditional methods' one-sided and static assessments. Secondly, through multi-dimensional coupling analysis and early risk warning, potential collaboration barriers can be proactively identified, significantly reducing the risk of collaboration failure and resource waste. Furthermore, based on root cause analysis and precise strategy recommendations, a shift from "treating the symptoms" to "addressing the root cause" has been achieved, effectively improving collaboration efficiency and quality. Finally, smart education management is transformed from passive response to proactive and adaptive scientific decision-making, greatly promoting the sustainability and healthy development of industry-education integration.

[0007] Optionally, the acquisition of the industry-education collaboration multimodal behavior dataset, based on which collaborative role positioning and relationship network construction are performed, and the dimensions of structural power, knowledge flow, and emotional connection are analyzed to generate a health information set of school-enterprise relationships, includes: based on a preset industry-education collaboration project task list, analyzing the role definitions and collaborative contractual relationships of multiple stakeholders, including enterprises, universities, and students; based on the role definitions and collaborative contractual relationships, deploying multi-source heterogeneous data sensors at key nodes in the physical and digital spaces of project collaboration to form a panoramic behavior perception network for acquiring the industry-education collaboration multimodal behavior dataset; the industry-education collaboration multimodal behavior dataset This includes voice stream communication data, text stream interaction data, and video stream dynamic data; based on the voice stream communication data and the text stream interaction data, key features of discourse dominance and decision convergence are extracted to generate a power situation information set; based on the text stream interaction data and the video stream dynamic data, a dynamic network of knowledge supply and demand response is constructed to generate a knowledge flow information set; based on the video stream dynamic data and the voice stream communication data, the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence is calculated to generate an emotional connection information set; and based on the power situation information set, the knowledge flow information set, and the emotional connection information set, a health information set of the school-enterprise relationship is constructed.

[0008] Optionally, the step of extracting key features of discourse dominance and decision convergence based on the voice stream communication data and the text stream interaction data to generate a power situation information set includes: constructing a topic consensus network based on topic initiation and consensus convergence to analyze the initiator's approval rating; identifying new topics initiated by each subject based on the voice stream communication data and tracking the supporting voice feedback triggered by this topic in subsequent dialogues; quantifying the supporting voice feedback by calculating the aggregation degree of acoustic features of other subjects expressing agreement, elaborating in depth, or providing corroboration in subsequent turns, thereby generating a topic resonance index; simultaneously, based on the text stream interaction data, parsing the core text proposition proposed by the topic initiator and constructing a semantic diffusion tree of this proposition in subsequent text streams; quantifying the degree to which the core proposition is collectively absorbed and developed by analyzing the number, depth, and semantic fidelity of branch nodes in the semantic diffusion tree, generating a proposition inheritance index; and comprehensively calculating the dynamic influence weight of an individual in the collaborative field by combining the individual's topic resonance and proposition inheritance, thereby aggregating the weights of all subjects to generate the power situation information set.

[0009] Optionally, the step of constructing a dynamic network of knowledge supply and demand response based on the text stream interaction data and the video stream dynamic data to generate a knowledge flow information set includes: constructing a knowledge interaction network based on direct interactive feedback to accurately characterize the effectiveness of knowledge flow; identifying knowledge supply events and corresponding demand response events based on the text stream interaction data, mapping the subject to the supply node and response node of the knowledge interaction network; quantifying the heuristic value of the supplied knowledge by analyzing the proportion and depth of three types of interactive behaviors—questioning, supplementing, and correcting—in the demand response events; simultaneously, decoding the micro-behavioral feedback of the subject when receiving knowledge supply from the video stream dynamic data, constructing an instantaneous cognitive absorption index by identifying the temporal patterns of nodding, gazing, recording, and confused expressions, to characterize the penetration and absorption state of knowledge at the receiving end; and cross-modal correlation between the heuristic value and the instantaneous cognitive absorption index to assign a circulation efficiency coefficient to each knowledge flow, thereby generating the knowledge flow information set that can identify knowledge hubs and value blind spots.

[0010] Optionally, the step of calculating the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence based on the video stream dynamic data and the voice stream communication data to generate an emotional connection information set includes: extracting physiological microstates based on facial action units and body posture vectors from the video stream dynamic data, and calculating the instantaneous consistency of the microstate transition probability between different subjects to quantify unconscious, underlying behavioral synchronicity; simultaneously, fusing the acoustic features of the voice stream communication data with the facial expressions of the video stream dynamic data to identify the co-occurrence and transmission events of cross-modal emotional valence in dialogue rounds, and constructing an emotional resonance gradient describing the bidirectional flow of emotion between paired subjects; driving the emotional resonance gradient to propagate and superimpose in the collaborative relationship network based on the underlying connections formed by the behavioral synchronicity, and by simulating this ripple effect, emerging a group emotional tone that characterizes the level of cohesion of the group or even the entire team; and constructing the emotional connection information set according to the behavioral synchronicity, the emotional resonance gradient, and the group emotional tone.

[0011] Optionally, the step of performing multi-dimensional coupled collaborative situational awareness and risk pattern identification based on the health information set of the school-enterprise relationship to generate a collaborative barrier risk pattern set includes: performing cross-dimensional coupled analysis on the power situation information set, the knowledge flow information set, and the emotional connection information set; identifying stable emerging collaborative behavior configurations with specific characteristics to deconstruct multiple typical collaborative patterns; and judging them as benign collaboration patterns or malignant risk patterns based on the configurational characteristics and collaborative effectiveness of the typical collaborative patterns, and integrating the identified malignant risk patterns to generate the collaborative barrier risk pattern set.

[0012] Optionally, the process for identifying the benign collaboration model includes: the benign collaboration model must satisfy the following configurational characteristics: power dynamics level: the dynamic influence weights exhibit a decentralized spindle-shaped distribution, and high-influence subjects possess topic resonance and proposition inheritance levels higher than the team average, indicating that influence stems from recognition rather than position; knowledge flow level: the circulation efficiency coefficient of knowledge flow is generally high, cross-subject re-creation links frequently occur in the inspirational value, and the inspirational value is strongly positively correlated with the instantaneous cognitive absorption index; emotional connection level: the behavioral synchronicity and the emotional resonance gradient form a highly connected positive emotional loop in the collaborative relationship network, and the group emotional tone is stably in the positive range.

[0013] Optionally, the process for identifying the malignant risk pattern includes: identifying at least two malignant risk patterns: a unipolar dominant risk pattern: at the power dynamic level, the dynamic influence weight of a single entity is abnormally high, while the topic resonance and proposition inheritance of other entities remain consistently low, forming a decision-making black hole; at the knowledge flow level, the circulation efficiency coefficient issued by this dominant entity is greater than a preset efficiency threshold, and the inspirational value and circulation efficiency coefficient of the reverse knowledge flow that forms feedback are close to zero; at the emotional connection level, behavioral synchronicity is manifested in the unidirectional catering of most entities to this single entity, while the... The emotional resonance gradient and the overall emotional tone of the group exhibit a suppressed state; the floating collaboration risk mode: at the power dynamic level, the topic resonance is higher than the team's historical average, but the proposition inheritance is close to zero, indicating that the discussion is active but consensus cannot be solidified; at the knowledge flow level, the instantaneous cognitive absorption index of knowledge flow remains at a median level due to social etiquette, but the associated inspirational value remains at a low level, revealing that knowledge interaction remains at a superficial level; at the emotional connection level, the behavioral synchronicity is greater than the preset density threshold, but the emotional resonance gradient is weak, and the overall emotional tone of the group is characterized by detached politeness and a lack of deep resonance.

[0014] Optionally, the step of performing root cause tracing and adaptive strategy recommendation operations based on the collaborative barrier risk pattern set to generate a smart education dynamic perception status report includes: based on the collaborative barrier risk pattern set, tracing back its abnormal data configuration in the power situation information set, knowledge flow information set, and emotional connection information set to identify the core dimensions and key subject nodes that lead to the collaborative barrier; generating a set of adaptive strategy intervention vectors that are precisely matched with the root cause dimensions, wherein the intervention vectors include, but are not limited to: a dynamic discourse power rebalancing mechanism designed to break unipolar dominance and a deep knowledge inquiry protocol constructed to eliminate floating collaboration; and generating the smart education dynamic perception status report containing diagnostic conclusions, evolution predictions, and dynamic intervention schemes based on the root cause tracing path and the adaptive strategy intervention vectors.

[0015] Secondly, this application provides a multimodal smart education scenario dynamic perception system based on field theory. The system includes: a health information module, used to acquire a multimodal behavior dataset of industry-education collaboration, and based on the multimodal behavior dataset of industry-education collaboration, to perform collaborative role positioning and relationship network construction, analyze the dimensions of structural power, knowledge flow and emotional connection, and generate a set of health information on school-enterprise relationships; a risk labeling module, used to perform multi-dimensional coupled collaborative situational awareness and risk pattern recognition based on the set of health information on school-enterprise relationships, and generate a set of collaborative barrier risk patterns; and a report generation module, used to perform root cause tracing and adaptive strategy recommendation operations based on the set of collaborative barrier risk patterns, and generate a smart education dynamic perception status report. Attached Figure Description

[0016] 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application; Figure 2 A flowchart illustrating a dynamic perception method for multimodal smart education scenarios based on field theory, provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a multimodal smart education scene dynamic perception system based on field theory, provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0020] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0021] Existing collaborative status assessment methods lack a mechanism for deep integration and dynamic attribution of key dimensions based on field theory when facing complex collaborative fields involving multiple subjects, multiple modalities, and dynamic evolution. This not only makes it difficult to systematically deconstruct and diagnose potential collaborative barriers and risk patterns, but also easily leads to the superficiality and formalization of the collaborative process, ultimately resulting in innovation stagnation and internal resource consumption.

[0022] Based on this, this application provides a method and system for dynamic perception of multimodal smart education scenarios based on field theory. First, multimodal behavioral data generated during industry-education collaboration are collected and integrated to construct a dynamic relationship network among participating roles. Then, based on field theory, this network is quantitatively modeled and analyzed from three core dimensions: structural power, knowledge flow, and emotional connection, thereby generating a comprehensive information set reflecting the health of school-enterprise relationships. On this basis, through coupled correlation analysis of multi-dimensional indicators and machine learning pattern recognition, the evolution of the collaborative situation is dynamically perceived, and potential collaborative barrier risk patterns are accurately identified. Finally, for the identified risks, in-depth root cause analysis is conducted to locate structural or relational contradictions in the field, and intervention measures are adaptively matched and recommended from a strategy library. Ultimately, the entire process data and insights are integrated to generate a dynamic perception status report for smart education, which is then output to school-enterprise managers to provide decision support, forming a closed loop of "perception-diagnosis-intervention". This solution achieves a comprehensive and dynamic perception of the industry-education collaboration ecosystem, overcoming the limitations of traditional methods' one-sided and static assessments. Secondly, through multi-dimensional coupling analysis and early risk warning, it can proactively identify potential collaboration barriers, significantly reducing the risk of collaboration failure and resource waste. Furthermore, based on root cause tracing and precise strategy recommendations, it achieves a shift from "treating the symptoms" to "addressing the root cause," effectively improving collaboration efficiency and quality. Finally, it elevates smart education management from passive response to proactive and adaptive scientific decision-making, greatly promoting the sustainable and healthy development of industry-education integration.

[0023] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In the process of state perception in educational scenarios, the method provided in this application can achieve early identification and diagnosis of potential collaborative barriers and risk patterns, and further address the superficiality and formalization of collaborative processes.

[0024] Specifically, the method of this application is applied to any server that communicates with multi-source heterogeneous data sensors. First, it collects and integrates multimodal behavioral data generated during industry-education collaboration to construct a dynamic relationship network among participating roles. Then, based on field theory, it quantitatively models and analyzes this network from three core dimensions: structural power, knowledge flow, and emotional connection, thereby generating a comprehensive information set reflecting the health of the school-enterprise relationship. On this basis, through coupled correlation analysis of multi-dimensional indicators and machine learning pattern recognition, it dynamically perceives the evolution of the collaborative situation and accurately identifies potential collaborative barrier risk patterns. Finally, for the identified risks, it conducts in-depth root cause analysis, locating structural or relational contradictions in the field, and adaptively matches and recommends intervention measures from a strategy library. Ultimately, it integrates all process data and insights to generate a smart education dynamic perception status report, which is then output to school-enterprise managers to provide decision support, forming a closed loop of "perception-diagnosis-intervention."

[0025] For specific implementation details, please refer to the following examples.

[0026] Figure 2 This is a flowchart illustrating a dynamic perception method for multimodal smart education scenarios based on field theory, provided in one embodiment of this application. The method of this embodiment can be applied to servers in the above scenarios. Figure 2 As shown, the method includes: S201. Obtain a multimodal behavior dataset of industry-education collaboration. Based on the multimodal behavior dataset of industry-education collaboration, perform collaborative role positioning and relationship network construction, analyze the dimensions of structural power, knowledge flow and emotional connection, and generate a set of information on the health of school-enterprise relationships.

[0027] The industry-education collaboration multimodal behavior dataset can be a collection of multimodal behavior data involved in the industry-education collaboration process. Industry-education collaboration is a cooperative and interactive process between schools and enterprises based on education, training, research and development, etc. It includes text stream interaction data (such as course records, meeting minutes, project documents), voice stream communication data (such as interview recordings, classroom lectures), and video stream dynamic data (such as surveillance videos, online meeting videos). The data comes from multi-source heterogeneous data sensors.

[0028] Collaborative role positioning and relationship network construction can be achieved by using data analysis techniques to identify participating roles in industry-education collaboration (such as students, teachers, corporate mentors, and managers) and construct a network structure of social and cooperative relationships among them. The structural power dimension can characterize the power distribution among parties in industry-education collaboration based on factors such as organizational status and resource control. The knowledge flow dimension can characterize the process of knowledge (such as technical skills and theoretical achievements) being transferred, shared, and innovated among roles in industry-education collaboration. The emotional connection dimension can characterize the strength of emotional bonds formed based on emotional interaction (such as trust, satisfaction, and willingness to cooperate) in industry-education collaboration. The information set on the health of school-enterprise relations can be a comprehensive set of information reflecting the health status of school-enterprise cooperation, including a power dynamics information set, a knowledge flow information set, and an emotional connection information set.

[0029] Specifically, field theory emphasizes how the positions, capital, and interactions of various parties in social relationships within a specific field (such as the educational field) affect the overall dynamic equilibrium. In the context of smart education, industry-education collaboration is a key link, aiming to improve educational quality and practical skills through deep cooperation between schools and enterprises. However, existing industry-education collaboration management often relies on static assessments or single data sources, making it difficult to comprehensively capture the complex dynamic changes in the collaboration process, leading to low collaboration efficiency, resource waste, and frequent conflicts. This step collects multimodal data to construct a network of collaborative roles and, based on field theory, conducts quantitative analysis from three dimensions: structural power, knowledge flow, and emotional connection, generating a health information set of school-enterprise relationships. Through root cause analysis and adaptive strategy recommendations, it can specifically address barriers to industry-education collaboration, improve collaboration efficiency and quality, and provide actionable insights through a smart education dynamic perception status report, enhancing the scientific nature and agility of education management and promoting the development of industry-education integration towards intelligence and adaptability.

[0030] S202. Based on the health information set of school-enterprise relationship, conduct multi-dimensional coupled collaborative situational awareness and risk pattern identification to generate a collaborative barrier risk pattern set.

[0031] Collaborative situational awareness and risk pattern recognition can be achieved through comprehensive perception and pattern recognition based on multiple dimensions (such as structural power, knowledge flow, and emotional connection) of the health information set of school-enterprise relationships, in order to dynamically capture changes in collaborative situational awareness and identify risk patterns. The collaborative barrier risk pattern set can be a collection used to identify the types of obstacles or risks existing in the process of industry-education collaboration. Collaborative barriers specifically include communication barriers, uneven resource allocation, conflicting goals, and cultural differences. This set includes unipolar dominant risk patterns and floating collaboration risk patterns.

[0032] Specifically, in the process of industry-education collaboration, collaborative barriers often stem from the interaction of multiple dimensions. For example, an imbalance in power structure may inhibit knowledge flow, and weak emotional connections may exacerbate communication barriers. However, existing methods typically employ single-dimensional analysis or post-event evaluation, making it difficult to perceive the dynamic changes in the collaborative situation in real time or to identify complex risks in advance, leading to delays or failures in collaborative projects. Furthermore, internal conflicts and risks are not caused by a single factor but are the result of the "coupling" effect of capital games and relational tensions across different dimensions. For instance, the root cause of impeded knowledge flow may lie in the monopoly of structural power, exacerbated by weak emotional connections. This step performs coupled correlation analysis on multi-dimensional indicators within the health information set, using pattern recognition technology to dynamically perceive the collaborative situation and identify and pattern potential collaborative barrier risks. Through multi-dimensional coupled analysis, the industry-education collaboration situation can be perceived in real time, potential barrier risks can be identified in advance, the probability of collaborative failure can be reduced, the risk response capability and adaptability of the smart education system can be improved, and an intuitive risk view can be provided to education decision-makers.

[0033] S203. Based on the collaborative barrier risk pattern set, perform root cause tracing and adaptive strategy recommendation operations to generate a smart education dynamic perception status report.

[0034] Adaptive strategy recommendation can be a process of tracing the root causes of risks through data analysis and recommending adaptive adjustment strategies based on those causes. A smart education dynamic perception status report can be a report that comprehensively reflects the dynamic perception status of industry-education collaboration, including diagnostic conclusions, evolution predictions, and dynamic intervention plans.

[0035] Specifically, after identifying collaborative barriers and risks, merely focusing on patterns without delving into the root causes makes it difficult to formulate effective interventions. Furthermore, the complex and ever-changing environment of industry-education collaboration requires adaptive strategies for dynamic adjustment. However, existing methods lack systematic root cause analysis and strategy recommendation mechanisms, often resulting in insufficient solution generalization or poor implementation. This step traces the causal relationships of risk patterns to pinpoint the root causes, then adaptively matches and recommends intervention strategies from a strategy library, and integrates all processes and results to generate a dynamic perception status report. Through root cause tracing and adaptive strategy recommendation, industry-education collaboration barriers can be addressed in a targeted manner, improving collaboration efficiency and quality. The intelligent education dynamic perception status report provides actionable insights, enhances the scientific rigor and agility of education management, and promotes the intelligent and adaptive development of industry-education integration.

[0036] The method provided in this embodiment first collects and integrates multimodal behavioral data generated during the industry-education collaboration process to construct a dynamic relationship network among participating roles. Then, based on field theory, the network is quantitatively modeled and analyzed from three core dimensions: structural power, knowledge flow, and emotional connection, thereby generating a comprehensive information set reflecting the health of the school-enterprise relationship. On this basis, through the coupling and correlation analysis of multi-dimensional indicators and machine learning pattern recognition, the evolution of the collaborative situation is dynamically perceived, and potential collaborative barrier risk patterns are accurately identified. Finally, for the identified risks, in-depth root cause tracing is conducted to locate structural or relational contradictions in the field, and intervention measures are adaptively matched and recommended from the strategy library. Ultimately, the data and insights from the entire process are integrated to generate a smart education dynamic perception status report, which is then output to school-enterprise managers to provide decision support and form a closed loop of "perception-diagnosis-intervention". This solution achieves a comprehensive and dynamic perception of the industry-education collaboration ecosystem, overcoming the limitations of traditional methods' one-sided and static assessments. Secondly, through multi-dimensional coupling analysis and early risk warning, it can proactively identify potential collaboration barriers, significantly reducing the risk of collaboration failure and resource waste. Furthermore, based on root cause tracing and precise strategy recommendations, it achieves a shift from "treating the symptoms" to "addressing the root cause," effectively improving collaboration efficiency and quality. Finally, it elevates smart education management from passive response to proactive and adaptive scientific decision-making, greatly promoting the sustainable and healthy development of industry-education integration.

[0037] In some embodiments, based on a pre-defined task list for industry-education collaboration projects, the roles and collaborative contractual relationships of multiple stakeholders—enterprises, universities, and students—are analyzed. According to these roles and contracts, multi-source heterogeneous data sensors are deployed at key physical and digital nodes in the collaborative project space to form a panoramic behavior perception network, used to acquire a multimodal behavior dataset for industry-education collaboration. This dataset includes voice stream communication data, text stream interaction data, and video stream dynamic data. Based on the voice stream communication data and text stream interaction data, key features of discourse dominance and decision convergence are extracted to generate a power dynamics information set. Based on the text stream interaction data and video stream dynamic data, a dynamic network of knowledge supply and demand response is constructed to generate a knowledge flow information set. Based on the video stream dynamic data and voice stream communication data, the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence is calculated to generate an emotional connection information set. Based on the power dynamics information set, knowledge flow information set, and emotional connection information set, a health information set of the university-enterprise relationship is constructed.

[0038] A task list for industry-education collaboration projects can be a predefined document containing project objectives, task allocation, timelines, and expected outcomes, used to guide collaborative activities among enterprises, universities, and students. Multi-source heterogeneous data sensors can be physical or virtual devices used to collect different types of behavioral data, including audio acquisition devices (such as microphones), text acquisition devices (such as keyloggers or logging systems), and video acquisition devices (such as cameras or screen recording tools). A panoramic behavior perception network can be a distributed data acquisition system composed of multi-source heterogeneous data sensors, designed to cover the entire process and all scenarios of industry-education collaboration projects, enabling continuous monitoring of the behavior of multiple stakeholders. Voice stream communication data can include audio information such as meeting recordings and telephone conversations. Text stream interaction data can include text information such as emails, instant messages, and document editing records. Video stream dynamic data can include visual information such as video conference recordings and on-site event recordings. A power dynamics information set can be a collection of information reflecting the distribution of discourse power and decision-making power in industry-education collaboration projects, generated by analyzing key features in voice and text data (such as speaking duration, topic control frequency, and the number of times decision-making instructions appear). The knowledge flow information set can be a collection of information describing the dynamics of knowledge transfer and absorption in industry-education collaboration projects. It is generated by constructing a network of knowledge nodes in text and video data (such as document sharing links, question-and-answer sequences, and skill demonstration events). The emotional connection information set can be a collection of information characterizing the intensity and quality of emotional interaction in industry-education collaboration projects. It is generated by calculating the co-occurrence frequency of nonverbal behaviors (such as body language synchronicity and facial expression consistency) and emotional valence (such as the emotional tendency of voice tone) in video and audio data.

[0039] Specifically, in industry-education collaboration projects, the health of the relationship between schools and enterprises directly affects the success or failure of the project and the sustainability of cooperation. However, traditional evaluation methods often rely on subjective reports or a single data source, which cannot comprehensively capture the dynamic interactions of multiple stakeholders in dimensions such as structural power, knowledge flow, and emotional connection. This leads to evaluation bias or delayed response. For example, enterprises may dominate decision-making due to their resource advantages, but if the distribution of discourse power is not quantified, the participation of universities and students may be underestimated, affecting the fairness of collaboration. If knowledge flow is evaluated only through document transmission, ignoring real-time responses and absorption during interactions, it may lead to low efficiency in knowledge sharing. Emotional connection, as an implicit dimension, is difficult to identify team cohesion and potential conflicts without analysis of multimodal data (such as nonverbal behavior in videos). To address the aforementioned issues, this step involves pre-setting a task list for industry-education collaboration projects, systematically analyzing the role definitions and collaborative contractual relationships of each entity, and providing a contextual basis for behavioral data collection. Based on the analysis results, multi-source heterogeneous data sensors are deployed at key nodes in physical and digital spaces (such as decision-making meeting venues and online collaboration channels) to form a panoramic behavioral perception network, ensuring the comprehensiveness and real-time nature of data acquisition. Based on the acquired multimodal behavioral dataset of industry-education collaboration, discourse dominance features (such as dominant speaking frequency and appearance of decision-making keywords) and decision convergence features (such as the speed of agenda resolution) are extracted from speech and text data, respectively, to generate a power dynamics information set to reveal the power structure. The study aims to achieve a balance between knowledge supply and demand response. It constructs a dynamic network of knowledge supply and demand response from text and video data, generating a knowledge flow information set by analyzing document flow paths, problem response times, and skill matching to assess knowledge sharing efficiency. It also calculates nonverbal behavior synchronicity (e.g., the degree of gesture imitation, frequency of eye contact) and emotional valence co-occurrence frequency (e.g., the synergy of positive emotional expression) from video and audio data to generate an emotional connection information set to measure emotional cohesion. Finally, based on the power dynamics information set, knowledge flow information set, and emotional connection information set, it constructs a health information set of university-enterprise relationships through weighted fusion or rule-based reasoning, providing data support for subsequent risk identification and strategy recommendations.

[0040] The method provided in this embodiment quantifies structural power through power situation analysis, promoting collaborative fairness and decision-making transparency; assesses knowledge flow through knowledge transfer networks, optimizing knowledge sharing and innovation capabilities; and strengthens relationship tightness through emotional connection measurement, enhancing team cohesion and collaborative sustainability. This comprehensively improves the accuracy and practicality of assessing the health of school-enterprise relationships, providing a reliable data foundation for identifying collaborative barriers and recommending adaptive strategies, effectively preventing project failures and relationship breakdowns, and simultaneously promoting the deep integration of industry and education.

[0041] In some embodiments, a topic consensus network based on topic initiation and consensus convergence is constructed to analyze the initiator's approval rating: Based on voice stream communication data, new topics initiated by each subject are identified, and the supporting voice feedback triggered by this topic in subsequent dialogues is tracked; the supporting voice feedback is quantified by calculating the aggregation degree of acoustic features of other subjects expressing agreement, elaborating, or providing corroboration in subsequent turns, thereby generating a topic resonance index; simultaneously, based on text stream interaction data, the core text proposition proposed by the topic initiator is parsed, and a semantic diffusion tree of this proposition in subsequent text streams is constructed; by analyzing the number, depth, and semantic fidelity of branch nodes in the semantic diffusion tree, the degree to which the core proposition is collectively absorbed and developed is quantified, generating a proposition inheritance index; combining the individual's topic resonance and proposition inheritance, its dynamic influence weight in the collaborative field is calculated, and then the weights of all subjects are aggregated to generate a power situation information set.

[0042] A topic consensus network can be a network model used to analyze how topics are initiated and converged during collaboration. It quantifies the level of acceptance and consensus by identifying topic initiators and supporters. Supportive voice feedback refers to the positive responses of other subjects to the topic initiator during voice communication, including expressing agreement, elaborating, or providing supporting evidence. The topic resonance index quantifies the degree of support a topic initiator receives in collaboration, calculated based on the frequency and intensity of supportive voice feedback. A core text proposition refers to a key viewpoint or claim raised by the topic initiator in text interaction, typically guiding the direction of discussion. A semantic diffusion tree is a tree-like structure used to describe the propagation and evolution of core text propositions in subsequent text interactions, quantifying the degree of proposition inheritance by analyzing the number, depth, and semantic fidelity of branch nodes. The proposition inheritance index quantifies the degree to which core text propositions are collectively absorbed and developed, calculated based on the structural features of the semantic diffusion tree.

[0043] Specifically, in industry-education collaboration scenarios, the interactions among multiple stakeholders, including enterprises, universities, and students, often involve complex power dynamics. These dynamics directly affect collaboration efficiency and educational outcomes. Traditional methods typically rely on subjective observation or simple questionnaires to assess power relationships, but these methods suffer from problems such as lag, strong subjectivity, and limited coverage. For example, in cross-organizational collaboration, the power structure may dynamically change as the topic progresses and consensus is formed, but traditional methods cannot capture these changes in real time, leading to untimely adjustments to collaboration strategies, which in turn can cause decision-making biases, resource waste, or breakdowns in cooperation. Furthermore, industry-education collaboration scenarios involve multimodal data (such as voice, text, and video), but existing technologies often analyze this data in isolation, lacking a comprehensive quantification of power dimensions, making it difficult to fully reflect the health of collaboration. To address the aforementioned issues, this step first involves real-time acquisition of voice and text stream interaction data using multi-source heterogeneous sensors (such as microphones and text log systems) deployed in the industry-education collaboration scenario. The voice stream data undergoes preprocessing, including noise reduction, speech segmentation, and speaker identification. Automatic speech recognition technology is then used to convert the speech into text. Simultaneously, the text stream data is preprocessed, including word segmentation, stop word removal, and entity recognition. Next, a topic consensus network is constructed based on the preprocessed data to identify new topics initiated by each subject (e.g., by detecting initial statements such as "I suggest") and to track supporting voice feedback. This is achieved through acoustic feature analysis (e.g., pitch rising to 0.8 units and speech rate stabilizing at 120 words / minute) and semantic analysis (e.g., sentiment analysis). The polarity score (0.9) quantifies the feedback intensity, and the topic resonance index (e.g., a comprehensive score of 0.75) is calculated. Simultaneously, the core text propositions are parsed, and a semantic diffusion tree is constructed. The proposition inheritance index (e.g., a score of 0.80) is calculated by analyzing the number of branch nodes (e.g., 5 branches), depth (e.g., depth 3), and semantic fidelity (e.g., similarity 0.95). Next, the topic resonance index and proposition inheritance index are combined, and a weighted average method (e.g., weights of 0.5 each) is used to calculate the dynamic influence weight of each subject (e.g., a weight value of 0.78). Finally, the dynamic influence weights of all subjects are aggregated to generate a power dynamics information set, which is stored in a structured data format (e.g., JSON) for easy subsequent analysis.

[0044] By utilizing the method provided in this embodiment and continuously analyzing voice and text interactions, the system can instantly capture changes in topic initiation, support feedback, and proposition inheritance, thereby reflecting the dynamic adjustment of the power structure. This enables collaborative managers to promptly detect signs of power imbalance or conflict and take intervention measures to prevent a decline in collaborative efficiency.

[0045] In some embodiments, a knowledge interaction network based on direct interactive feedback is constructed to accurately characterize the effectiveness of knowledge flow: Based on text stream interaction data, knowledge supply events and corresponding demand response events are identified, and subjects are mapped as supply nodes and response nodes of the knowledge interaction network; by analyzing the proportion and depth of three types of interactive behaviors in demand response events—questioning, supplementation, and correction—the heuristic value of the supplied knowledge is quantified; simultaneously, from video stream dynamic data, the micro-behavioral feedback of subjects when receiving knowledge supply is decoded, and by identifying the temporal patterns of nodding, gazing, recording, and confused expressions, an instantaneous cognitive absorption index is constructed to characterize the penetration and absorption status of knowledge at the receiving end; the heuristic value and the instantaneous cognitive absorption index are cross-modal correlated to assign a circulation efficiency coefficient to each knowledge flow, thereby generating a knowledge flow information set that can identify knowledge hubs and value blind spots.

[0046] A knowledge interaction network can be a network model used to describe the flow of knowledge among subjects, representing knowledge supply and demand response relationships through nodes and edges. Knowledge supply events can refer to the actions of a subject providing knowledge or information in text interactions, such as publishing documents or answering questions. Interactive behaviors can include three categories: questioning, supplementing, and correcting; these behaviors indicate the depth and effectiveness of knowledge flow. Micro-behavioral feedback can refer to the nonverbal behaviors of subjects receiving knowledge, extracted from video stream data, such as nodding, eye contact, note-taking, and expressions of confusion. The instantaneous cognitive absorption index can be an indicator used to characterize the degree to which knowledge is absorbed at the receiving end, calculated based on the temporal patterns of micro-behavioral feedback. The circulation efficiency coefficient can be a numerical value used to evaluate the efficiency of knowledge flow, derived by correlating heuristic value and the instantaneous cognitive absorption index.

[0047] Specifically, in industry-education collaboration scenarios, knowledge flow is a core element that directly impacts collaboration efficiency and educational outcomes. Traditional methods typically rely on post-event surveys or simple indicators (such as interaction frequency) to assess knowledge flow. However, this approach has many limitations. For example, a high interaction frequency does not necessarily indicate effective knowledge flow; it may merely be superficial interaction. Furthermore, the degree of knowledge absorption is often difficult to quantify, leading to inaccurate assessments. In addition, knowledge flow involves multimodal data, but existing technologies often analyze text or video data in isolation, failing to comprehensively capture the entire process of knowledge from supply to absorption. This may result in knowledge barriers in collaboration, such as poor knowledge transfer, insufficient absorption, or value blind spots, thereby affecting the overall collaborative effect. To address the above issues, this step first collects text stream interaction data, including emails, chat logs, and document comments, through the log system of a digital platform (such as collaborative office software). The text data undergoes preprocessing, including word segmentation, stop word removal, entity recognition, and intent analysis, to identify knowledge-related events. Simultaneously, dynamic video stream data, including meeting videos and classroom recordings, is collected via cameras deployed in the collaborative environment. This video data is preprocessed, including frame extraction, face detection, and behavior recognition, to extract micro-level behavioral feedback. Then, based on the preprocessed text stream data, a knowledge interaction network is constructed: identifying knowledge supply events (e.g., by detecting keywords such as "share" and "explain" in the text) and demand response events (e.g., by detecting direct responses to supply events such as "ask a question" and "comment"), mapping the subjects to supply and response nodes in the network, and establishing directed edges to represent the knowledge flow. Next, the heuristic value is quantified: for each knowledge supply event, analysis is performed... The corresponding high-quality interactive behaviors in the demand response events (including follow-up questions, supplements, and corrections) are calculated using natural language processing technology. The proportion (e.g., the percentage of follow-up questions in all responses) and depth (e.g., the number of levels of follow-up questions or the amount of supplementary information) of these behaviors are used to comprehensively derive an inspirational value index. Simultaneously, based on dynamic video stream data, an instantaneous cognitive absorption index is constructed: decoding the micro-behavioral feedback of the subject when receiving knowledge supply, and using computer vision technology to identify the temporal patterns of nodding, gazing, recording, and confused expressions (e.g., an increase in the frequency of nodding or the continuous appearance of confused expressions in a short period of time after knowledge supply). The instantaneous cognitive absorption index is calculated according to the behavior type, frequency, and temporal relationship. Finally, cross-modal association is performed: the inspirational value calculated from text stream data is associated with the instantaneous cognitive absorption index calculated from video stream data (e.g., through weighted combination or similarity calculation), assigning a circulation efficiency coefficient to each knowledge stream, and aggregating the circulation efficiency coefficients of all knowledge streams to generate a knowledge flow information set.

[0048] The method provided in this embodiment identifies knowledge supply and response events based on text stream interactive data, combines dynamic data decoding of video streams with micro-behavioral feedback, and generates a knowledge flow information set through cross-modal correlation analysis. This can identify efficient knowledge hubs and value blind spots in real time, helping managers to promptly identify obstacles to knowledge transfer and optimize the allocation of collaborative resources.

[0049] In some embodiments, physiological microstates based on facial action units and body posture vectors are extracted from dynamic video stream data, and the instantaneous consistency of microstate transition probabilities between different subjects is calculated to quantify unconscious, underlying behavioral synchronicity. Simultaneously, acoustic features of voice stream communication data and facial expressions of dynamic video stream data are fused to identify co-occurrence and transmission events of cross-modal emotional valence in dialogue rounds, and to construct an emotional resonance gradient describing the bidirectional flow of emotions between paired subjects. Based on the underlying connections formed by behavioral synchronicity, the emotional resonance gradient is driven to propagate and superimpose in the collaborative relationship network. By simulating this ripple effect, a group emotional tone representing the level of cohesion of the group or even the entire team emerges. Based on behavioral synchronicity, emotional resonance gradient, and group emotional tone, an emotional connection information set is constructed.

[0050] The immediate consistency of microstate transition probabilities can refer to the degree of synchronization of physiological microstate change sequences between different subjects, quantifying unconscious behavioral matching by calculating the similarity of microstate transition probabilities. Behavioral synchronicity can refer to the unconscious, underlying nonverbal behavioral consistency between subjects, such as simultaneous nodding or similar body posture changes, reflecting the degree of implicit coordination. Emotional valence can refer to the positive or negative degree of emotion, such as positive or negative emotional polarity. Cross-modal emotional valence co-occurrence and transmission events can refer to events in dialogue turns where emotional valence appears simultaneously in the voice and video modalities or is transmitted along the dialogue sequence, reflecting the dynamics of emotional interaction. Emotional resonance gradient can refer to the intensity change of the bidirectional flow of emotion between paired subjects, similar to a gradient field, representing the intensity and direction of emotional influence. The ripple effect can refer to the phenomenon of emotion propagating and amplifying in a collaborative relationship network, similar to ripples spreading on water, describing the networked transmission process of emotional influence. Group emotional tone can refer to the overall emotional atmosphere of the entire team or group, such as positive, negative, or neutral, reflecting the collective emotional state.

[0051] Specifically, in industry-education collaboration scenarios, emotional connection is a core element in maintaining cooperative relationships. However, traditional methods often rely on explicit interaction data (such as text content) and neglect the quantitative analysis of nonverbal behavior and emotional dynamics, leading to incomplete collaborative assessments. For example, text analysis alone cannot capture unconscious behavioral synchronization (such as nodding and smiling) and emotional resonance between subjects. These implicit factors directly affect team trust, collaboration satisfaction, and project sustainability. If emotional connection is weak, even if knowledge flow is efficient, emotional conflicts may lead to a deadlock or premature termination of cooperation. Furthermore, emotional connection is an implicit dimension in the collaborative field, but existing technologies mostly focus on explicit indicators (such as interaction frequency or content quality) and ignore the emotional signals conveyed by nonverbal behavior (such as facial expressions and body posture). For example, in a collaborative meeting, if one party nods and smiles frequently while the other party is speaking, it indicates high emotional resonance; conversely, if one party looks confused or leans back, it may indicate emotional resistance.To address the above issues, this step first involves real-time acquisition of dynamic video stream data via cameras deployed in the physical space of the industry-education collaboration (e.g., a conference room). This data is then preprocessed, including frame extraction, face detection, facial landmark localization, and body pose estimation, to extract physiological microstates. For example, a facial motion coding system is used to identify facial motion units (e.g., AU12 represents a raised corner of the mouth), and a pose estimation algorithm generates body pose vectors (e.g., a shoulder tilt angle of 15 degrees). Simultaneously, a microphone array is used to acquire and preprocess audio stream communication data, including noise reduction and speech activity detection, to extract acoustic features (e.g., a fundamental frequency of 250Hz). Then, behavioral synchronicity is calculated based on the preprocessed video stream data: a time series of physiological microstates is constructed, and a probabilistic model (e.g., a Hidden Markov Model) is used to calculate the microstate transition probability matrix. The matrices between different subjects are compared, and unconscious behavioral synchronicity is quantified by calculating an instantaneous consistency index (e.g., a dynamic time warping similarity score of 0.85). Simultaneously, the audio and video stream data are fused to construct an emotional resonance gradient. Acoustic features are extracted from speech stream data and input into a pre-trained emotion recognition model to calculate emotion valence scores (e.g., a positive score of 0.9). Facial expression features are extracted from video stream data to calculate emotion valence. Timestamps are aligned to identify cross-modal emotion valence co-occurrence events (e.g., the frequency of an event where one party speaks positively and the other smiles is 3 times). The frequency and intensity of these events in each dialogue turn are statistically analyzed, and an emotion resonance gradient is generated through weighted aggregation (e.g., a gradient strength of 0.7). Next, the emotion resonance gradient is propagated in the collaborative relationship network based on behavioral synchronicity: the relationship network is modeled as a graph structure, with nodes representing subjects, edges using behavioral synchronicity scores (e.g., 0.6) as initial weights, and the emotion resonance gradient as input. The emotion impact value is calculated using a propagation model (e.g., a convolutional network as shown in the figure) by simulating the ripple effect, and the emotional state of all nodes emerges as a group emotional tone (e.g., an overall positive score of 0.8). Finally, behavioral synchronicity, emotion resonance gradient, and group emotional tone are integrated to construct an emotion connection information set, which is stored in a structured format for subsequent collaborative situational awareness.

[0052] The method provided in this embodiment identifies unconscious behavioral coordination and emotional resonance among subjects, providing emotional dimension data support for the assessment of collaborative health. Based on the quantitative analysis of emotional resonance gradient and group emotional tone, emotional risks can be detected early, providing managers with a basis for intervention, thereby enhancing team cohesion and improving the stability and sustainability of collaboration.

[0053] In some embodiments, a cross-dimensional coupling analysis is performed on the power situation information set, knowledge flow information set, and emotional connection information set. By identifying stable emerging collaborative behavior configurations with specific characteristics, a variety of typical collaborative patterns are deconstructed. Based on the configurational characteristics and collaborative effectiveness of the typical collaborative patterns, they are identified as benign collaboration patterns or malignant risk patterns. The identified malignant risk patterns are then integrated to generate a collaborative barrier risk pattern set.

[0054] Cross-dimensional coupling analysis refers to the comprehensive correlation and interaction analysis of information sets across three dimensions—power dynamics, knowledge flow, and emotional connection—to identify their mutual influences and overall synergistic dynamics. Collaborative behavioral configurations refer to specific behavioral patterns or structural combinations that stably emerge across the power, knowledge, and emotional dimensions during collaboration, such as the correlation between power concentration and knowledge blockage. Benign collaboration models refer to collaborative behavioral configurations that promote efficient industry-education collaboration and enhance collaborative outcomes, such as power balance, knowledge flow, and positive emotions. Malicious risk models refer to collaborative behavioral configurations that may lead to collaboration failure, conflict, or inefficiency, such as power monopoly, knowledge blockage, and negative emotions.

[0055] Specifically, in industry-education collaboration scenarios, the success of the collaboration process highly depends on the dynamic balance of multiple dimensions such as power, knowledge, and emotion. However, traditional evaluation methods are often limited to single-dimensional analysis and cannot capture the complex interactions between multiple dimensions, leading to incomplete situational awareness and delayed risk identification. For example, focusing solely on knowledge flow efficiency may overlook decision-making biases caused by power inequality or knowledge-sharing barriers caused by emotional resistance. These multi-dimensional coupling problems may lead to collaboration deadlock or project failure. To address these issues, this step first obtains pre-constructed sets of power situation information, knowledge flow information, and emotional connection information. These sets have been generated through multimodal data processing and stored in a structured database. For example, the power situation information set includes the dynamic influence weight of each subject (e.g., the enterprise's weight is 0.8), the knowledge flow information set includes the circulation efficiency coefficient of knowledge flow (e.g., the efficiency coefficient of a certain knowledge flow is 0.6), and the emotional connection information set includes behavioral synchronicity scores (e.g., inter-subject synchronicity is 0.7), emotional resonance gradient vectors (e.g., gradient strength is 0.5), and group emotional tone indices. The data is first quantified (e.g., a positive tone of 0.9); then, cross-dimensional coupling analysis is performed to align the data from the three information sets spatiotemporally. Features from different dimensions are associated based on timestamps or event identifiers of interactive sessions. For example, for the same collaborative meeting, the subject's influence weight in power dynamics, the circulation efficiency coefficient in knowledge transfer, and the emotional resonance gradient in emotional connection are extracted to form multi-dimensional feature vectors. Pattern recognition algorithms (such as cluster analysis) are then used to group these feature vectors to identify stable emerging collaborative behavior patterns. For example, K-means clustering can be used to divide the feature vectors into several clusters, where cluster A may correspond to... The configuration of "high power concentration - low knowledge flow - negative emotions" is followed by a configuration of "balanced power - high knowledge flow - positive emotions" for cluster B. Next, typical collaborative patterns are deconstructed, and detailed feature analysis is performed on each cluster. For example, analysis of cluster A shows that the enterprise's influence weight is greater than 0.9, the knowledge flow efficiency coefficient is less than 0.3, and the group emotional tone index is less than 0.4, indicating that power monopoly leads to knowledge blockage and emotional resistance. Simultaneously, based on historical collaborative efficiency data (such as project completion quality scores), the collaborative efficiency of each configuration is evaluated, and those configurations positively correlated with high efficiency (such as cluster B) are classified as good. The study identifies collaborative patterns and associates them with high-performance cases. Configurations negatively correlated with low efficiency or conflict (such as cluster A) are classified as malignant risk patterns and their risk characteristics are marked. Finally, all malignant risk patterns are integrated to generate a set of collaborative barrier risk patterns. The identified malignant risk patterns are then structured and summarized. Each pattern includes a configuration description, risk level (such as high risk), and potential impact description. For example, risk pattern 1 is described as "centralized power inhibits knowledge sharing," and risk pattern 2 is described as "negative emotions exacerbate communication barriers." This set of risk patterns is stored in a database or in JSON format for subsequent root cause analysis and strategy recommendations.

[0056] The method provided in this embodiment utilizes multi-dimensional coupling analysis to identify collaborative behavior configurations and generate a risk pattern set, thereby achieving a holistic perception of the industry-education collaboration situation and precise risk positioning. It can detect malignant risk patterns early, providing managers with early warning and intervention basis, thus preventing collaborative deadlock and project failure. At the same time, by distinguishing benign collaboration patterns, it promotes the application of best practices, enhances the stability and adaptability of collaboration, and improves cooperation efficiency and the quality of results.

[0057] In some embodiments, a positive collaboration model needs to meet the following configurational characteristics: Power dynamics level: The dynamic influence weights exhibit a decentralized spindle-shaped distribution, and high-influence subjects possess a higher-than-average level of topic resonance and proposition inheritance, indicating that influence stems from recognition rather than position; Knowledge flow level: The circulation efficiency coefficient of knowledge flow is generally high, and cross-subject re-creation links frequently occur in the inspirational value, and the inspirational value is strongly positively correlated with the instantaneous cognitive absorption index; Emotional connection level: Behavioral synchronicity and emotional resonance gradient form a highly connected positive emotional loop in the collaborative relationship network, and the group's emotional tone remains stable in the positive range.

[0058] A spindle-shaped distribution can be characterized by a high central position and low extremes in the dynamic influence weights among subjects, indicating decentralized power with most subjects possessing moderate influence, thus avoiding power monopolies. High-influence subjects are individuals or organizations with high dynamic influence weights during collaboration, typically exhibiting strong topic resonance and proposition inheritance indicators. The team average can be the arithmetic mean of the dynamic influence weights of all subjects, serving as a baseline for measuring power distribution. Highly connected positive emotional loops can be characterized by dense, bidirectional transmission paths of emotional resonance gradients within a collaborative relationship network, with predominantly positive emotional valence.

[0059] Specifically, in industry-education collaboration scenarios, improving collaboration effectiveness depends not only on optimizing a single dimension, but also on the positive interaction and balance of power, knowledge, and emotion dimensions. However, traditional evaluation methods often focus on identifying negative risks, such as power conflicts or knowledge blockages, while neglecting the systematic identification of positive collaboration patterns. This results in a lack of positive guidance in management strategies and the promotion of best practices. For example, focusing only on malignant risk patterns may cause the collaboration process to fall into a passive defense, making it impossible to proactively cultivate an efficient collaboration environment, thereby limiting the full realization of collaboration potential. To address the above issues, this step first retrieves pre-constructed sets of power dynamics information, knowledge flow information, and emotional connection information from the stored multimodal database. These sets include, for example, dynamic influence weights (e.g., 0.75), topic resonance (e.g., 0.8), and proposition inheritance (e.g., 0.85) from the power dynamics information set; circulation efficiency coefficients (e.g., 0.7), inspirational value (e.g., 0.6), and instantaneous cognitive absorption index (e.g., 0.75) from the knowledge flow information set; and behavioral synchronicity scores (e.g., 0.8), emotional resonance gradient vectors (e.g., 0.7), and group sentiment from the emotional connection information set. The initial assessment focuses on the tone index (e.g., 0.9). Next, at the power dynamics level, the team average of the dynamic influence weights of all subjects is calculated (e.g., 0.70). The analysis examines whether the weight distribution exhibits a decentralized spindle-shaped distribution (e.g., most weights are concentrated in the 0.65-0.75 range). Simultaneously, high-influence subjects are selected, and their topic resonance and proposition inheritance are verified to be higher than the team average (e.g., the average topic resonance of high-influence subjects is 0.85, while the team average is 0.75). If the distribution conforms to a spindle shape and the indicators of high-influence subjects are prominent, then the power dynamics level satisfies positive characteristics. Then, in the knowledge flow... The system is divided into several levels. First, at the knowledge flow level, the proportion of knowledge flows with high efficiency coefficients (e.g., over 80% with coefficients greater than 0.7) is examined. Second, data on heuristic value is analyzed to identify cross-subject re-creation stages (e.g., students further expanding upon corporate knowledge after it has been reinterpreted by universities). Third, the correlation coefficient between heuristic value and the instantaneous cognitive absorption index is calculated (e.g., a Pearson coefficient greater than 0.8). A strong positive correlation indicates a positive characteristic at the knowledge flow level. Finally, at the emotional connection level, data on behavioral synchronicity and emotional resonance gradients are extracted from the emotional connection information set to construct a collaborative relationship network diagram. The system analyzes whether network connectivity forms a highly connected positive sentiment loop (e.g., clustering coefficient greater than 0.6 and gradient vector magnitude greater than 0.5) and assesses whether the group sentiment tone index is consistently in the positive range (e.g., index greater than 0.5 at multiple consecutive time points). If the sentiment loop connectivity is high and the tone is positive, then the sentiment connection level satisfies the benign characteristics. Integrating the above judgment results, if all three levels of power situation, knowledge flow, and sentiment connection satisfy all characteristics, then the collaborative behavior configuration is judged as a benign collaboration mode, and its label and feature description are stored in the benign mode library for subsequent adaptive strategy recommendation and report generation.

[0060] The method provided in this embodiment, based on the systematic identification of benign collaboration models, achieves positive guidance and optimization of the health of industry-education collaboration. It can identify characteristics of efficient collaboration at an early stage, provide best practice benchmarks for managers, and thus proactively strengthen collaboration effectiveness. At the same time, by promoting benign models, it enhances team cohesion and innovation motivation, and promotes the sustainable development and quality improvement of industry-education collaboration.

[0061] In some embodiments, at least two malignant risk patterns are identified: Unipolar Dominance Risk Pattern: At the power dynamics level, the dynamic influence weight of a single entity is abnormally high, while the topic resonance and proposition inheritance of other entities remain consistently low, forming a decision-making black hole; at the knowledge flow level, the circulation efficiency coefficient issued by this dominant entity is greater than a preset efficiency threshold, while the inspirational value and circulation efficiency coefficient of the reverse knowledge flow that forms feedback are close to zero; at the emotional connection level, behavioral synchronicity is manifested in the unilateral catering of most entities to this single entity, while the emotional resonance gradient and group emotional tone are in a suppressed state; Suspended Collaboration Risk Pattern: At the power dynamics level, the topic resonance is higher than the team's historical average, but the proposition inheritance is close to zero, indicating active discussion but no consensus can be formed; at the knowledge flow level, the instantaneous cognitive absorption index of the knowledge flow remains at a median level due to social etiquette, but the associated inspirational value remains at a low level, revealing that knowledge interaction remains at a superficial level; at the emotional connection level, behavioral synchronicity is greater than a preset density threshold, but the emotional resonance gradient is weak, and the group emotional tone is characterized by distant politeness and a lack of deep resonance.

[0062] A unipolar-dominated risk model can be a vicious collaborative pattern where a single entity occupies absolute power, leading to unidirectional knowledge flow and emotional suppression. Its core characteristics are concentrated power, closed-loop knowledge, and emotional alienation. A suspended collaborative risk model can be a vicious collaborative pattern that appears active on the surface but lacks substantial consensus, deep knowledge integration, and profound emotional resonance. Its core characteristics are formalized interaction, superficial knowledge transfer, and polite emotional alienation. An abnormally high dynamic influence weight can occur when a single entity's dynamic influence weight significantly exceeds that of all other entities, exceeding a preset threshold (usually more than three times the average weight of other entities), thus forming an absolute power advantage. A persistently low topic resonance can occur when, apart from the dominant entity, the topic resonance index of other collaborating entities is consistently below 50% of the team average, with no significant recovery across multiple collaborative phases, indicating a lack of widespread acceptance of the topic. A persistently low proposition inheritance can occur when the inheritance index of other entities from the core textual propositions proposed by the dominant entity consistently approaches zero, indicating that the core propositions have not been collectively absorbed and developed, and consensus cannot be formed. Decision-making black holes can occur when a single entity holds absolute power, preventing other entities from contributing their opinions and suggestions. This results in a closed system of "one-way output and forced execution," lacking democracy and scientific rigor. Reverse knowledge flow, on the other hand, is the flow of knowledge from the original recipient to the recipient (e.g., students to university teachers, or university teachers to business representatives), representing a significant aspect of two-way knowledge integration. Weak emotional resonance gradients or a suppressed state can indicate extremely low intensity of two-way emotional flow between paired entities, or predominantly negative and passive emotional transmission (e.g., awe, resistance), lacking positive emotional resonance and creating a somber group atmosphere. Maintaining moderate social etiquette can mean the recipient's instantaneous cognitive absorption index is in the "moderate" range (typically 0.4-0.6), but this index is not based on substantive knowledge absorption but rather on superficial social etiquette feedback (e.g., polite nods, agreement). Superficial responses can mean that knowledge interaction only meets formal requirements (e.g., "received," "understood"), without in-depth discussion, questioning, or extension of the knowledge, preventing deep integration. Deep resonance can be a state of high emotional compatibility between subjects based on common goals and mutual trust, which is manifested in the frequent co-occurrence of positive emotions, natural synchronization of non-verbal behaviors, and low communication costs.

[0063] Specifically, in the context of industry-education collaboration, the stable improvement of collaborative effectiveness depends not only on the cultivation of a positive model, but also on the early identification and intervention of a malignant risk model. However, traditional evaluation methods often focus on macro indicators or anomalies in a single dimension, lacking a systematic judgment of risks coupled across multiple dimensions. This can lead to the collaboration process potentially harboring deep-seated problems such as power imbalances, knowledge blockages, or emotional resistance, without timely warnings or tracing the root causes. For example, focusing only on power conflicts may overlook the implicit interruption of knowledge flow, or analyzing only negative emotions may mask the root causes of power monopolies, thus making management strategies one-sided and lagging behind. To address the above issues, this step identifies malignant risk patterns based on power dynamics information sets, knowledge flow information sets, and emotional connection information sets. First, it retrieves these pre-constructed information sets from a stored multimodal database. For example, the power dynamics information set includes the dynamic influence weight, topic resonance, and proposition inheritance degree of each subject; the knowledge flow information set includes the circulation efficiency coefficient, heuristic value, and instantaneous cognitive absorption index of knowledge flow; and the emotional connection information set includes behavioral synchronicity scores, emotional resonance gradient vectors, and group emotional tone index. Next, it identifies unipolar dominance risk patterns: in the power dynamics... At the knowledge transfer level, the dynamic influence weights of all entities are calculated to identify any single entity with an abnormally high weight (e.g., more than twice the team average). Simultaneously, it is checked whether the topic resonance and proposition inheritance of other entities remain consistently low (e.g., the average is less than 30% of the team's historical average). If these conditions are met, an abnormal power dynamic is initially identified. At the knowledge transfer level, the knowledge flow emitted by the dominant entity is analyzed, and its circulation efficiency coefficient is checked to see if it exceeds a preset efficiency threshold (e.g., the threshold is set to 0.7). Furthermore, the inspiration from reverse knowledge flows (i.e., knowledge flows from other entities to the dominant entity) is examined. Whether the value and circulation efficiency coefficients approach zero (e.g., values ​​below 0.1), if the dominant subject's knowledge flow is efficient while the reverse knowledge flow is ineffective, then the knowledge flow level meets the abnormal characteristics. At the emotional connection level, analyze behavioral synchronicity data to check whether it manifests as a unidirectional catering to a single subject by a majority of subjects (e.g., synchronicity scores are significantly higher when pointing to that subject than in other directions). At the same time, assess whether the emotional resonance gradient and group emotional tone show a repressive state (e.g., gradient strength is consistently below 0.3 and tone index is stable in the negative range). If the emotional level meets these characteristics, it is judged as a unipolar dominant risk pattern. The characteristics and labels are stored in the risk pattern library. Then, the risk pattern of floating collaboration is judged: at the power situation level, the team's historical average of topic resonance and proposition inheritance is calculated, and it is checked whether the topic resonance is higher than the historical average (e.g., higher than the average by 10%), but whether the proposition inheritance is close to zero (e.g., value is lower than 0.1). If the discussion is active but consensus cannot be formed, the power situation level meets the abnormal characteristics. At the knowledge flow level, the instantaneous cognitive absorption index and heuristic value of knowledge flow are analyzed, and it is checked whether the instantaneous cognitive absorption index is maintained in the median due to social etiquette (e.g., index is between 0.4 and 0).Within a range of 6, if the associated heuristic value remains consistently low (e.g., below 0.2), and if knowledge interaction remains at a superficial level, then the knowledge flow level meets abnormal characteristics. At the emotional connection level, analyze whether behavioral synchronicity exceeds a preset density threshold (e.g., threshold set to 0.6), but whether the emotional resonance gradient is weak (e.g., gradient strength below 0.2). Simultaneously assess whether the group's emotional tone exhibits detached politeness (e.g., tone index within a neutral range but fluctuating frequently). If the emotional level meets these characteristics, it is determined to be a floating collaboration risk pattern, and its features and labels are stored in the risk pattern library. Finally, integrate all the judgment results to generate a set of malignant risk patterns for subsequent root cause tracing and adaptive strategy recommendation operations.

[0064] The method provided in this embodiment, based on the systematic identification of malignant risk patterns, enables early warning and precise intervention of risks in industry-education collaboration. It can comprehensively identify deep-seated problems such as power imbalance, knowledge flow obstruction, and emotional resistance, providing managers with targeted handling basis, thereby preventing the deterioration of collaborative relationships and project failure. At the same time, through multi-dimensional feature analysis, it enhances the accuracy and timeliness of risk perception, promoting the stable development and effectiveness improvement of industry-education collaboration.

[0065] In some embodiments, based on the risk pattern set of collaborative barriers, the abnormal data configurations in the power situation information set, knowledge flow information set, and emotional connection information set are traced in reverse to identify the core dimensions and key subject nodes that lead to collaborative barriers; a set of adaptive strategy intervention vectors that are precisely matched with the root cause dimensions are generated, and the intervention vectors include, but are not limited to: a dynamic discourse power rebalancing mechanism designed to break unipolar dominance and a deep knowledge inquiry protocol constructed to eliminate floating collaboration; based on the root cause tracing path and the adaptive strategy intervention vectors, a smart education dynamic perception status report containing diagnostic conclusions, evolution predictions, and dynamic intervention schemes is generated.

[0066] Anomalous data configurations can be combinations of data features that significantly deviate from normal collaborative patterns in specific dimensions (such as power, knowledge, or sentiment), such as abnormally high dynamic influence weights or persistently low heuristic value. Adaptive strategy intervention vectors can be a set of quantifiable and dynamically adjustable intervention strategies, each designed for a specific root cause dimension, such as dynamic discourse power rebalancing mechanisms or deep knowledge inquiry protocols. Dynamic discourse power rebalancing mechanisms can be intervention strategies that adjust topic allocation, decision-making participation rules, or feedback weights to balance the power distribution among subjects. Deep knowledge inquiry protocols can be intervention strategies that promote knowledge flow from superficial responses to re-creation through structured questioning processes and cross-subject knowledge deepening requirements.

[0067] Specifically, in industry-education collaboration scenarios, identifying risk patterns is only the first step in risk prevention and control. If the root causes cannot be quickly located and targeted strategies recommended, collaborative management will fall into a passive situation of "knowing the risks but not knowing how to solve them." Traditional methods often rely on human experience for root cause analysis, resulting in a highly subjective and inefficient tracing process. Furthermore, strategy recommendations lack dynamic adaptability and cannot cope with the complex changes in collaborative scenarios. For example, when the system detects a single dominant risk pattern, managers may only start from the surface phenomenon and take simple power decentralization measures, but ignore the deep connections in the dimensions of knowledge flow and emotional connection, resulting in limited intervention effects or even exacerbating conflicts. To address the above issues, this step involves root cause analysis and adaptive strategy recommendation based on a risk pattern set of collaborative barriers: First, the risk pattern set and associated multi-dimensional information set are acquired. Abnormal data configurations are identified through reverse tracing, such as abnormally high dynamic influence weights (e.g., more than twice the team average), excessive knowledge flow efficiency coefficients (e.g., greater than 0.7), and reverse flow indicators approaching zero (e.g., below 0.1), and persistently low emotional gradients (e.g., below 0.3) in the unipolar dominance pattern; and features such as proposition inheritance approaching zero (e.g., below 0.1) and low heuristic value (e.g., below 0.2) in the suspended collaboration pattern. Adaptive strategy intervention vectors are generated based on the root cause analysis results. For example, a dynamic discourse power rebalancing mechanism is initiated for power imbalance, and a deep questioning protocol is implemented for superficial knowledge. Finally, a smart education dynamic perception status report is generated, integrating diagnostic conclusions, evolutionary predictions (e.g., power imbalance may lead to team split), and dynamic intervention plans, achieving closed-loop management from risk identification to precise intervention.

[0068] The method provided in this embodiment, based on systematic root cause tracing and adaptive strategy recommendation, enables accurate diagnosis and efficient intervention of risks in industry-education collaboration. It can quickly identify the root causes of risks and generate dynamically matched solutions, improving the accuracy and timeliness of collaborative remediation. At the same time, by integrating prediction and solution guidance, it enhances the foresight and adaptability of collaborative management, promoting the stable development and efficiency optimization of industry-education relations.

[0069] Figure 3 This is a schematic diagram of the structure of a multimodal smart education scene dynamic perception system based on field theory provided in an embodiment of this application, as shown below. Figure 3 As shown, the multimodal smart education scenario dynamic perception system 300 based on field theory in this embodiment includes: a health information module 301, a risk label module 302, and a report generation module 303; The health information module 301 is used to acquire a multimodal behavior dataset of industry-education collaboration, and based on the dataset, to perform collaborative role positioning and relationship network construction, analyze the dimensions of structural power, knowledge flow and emotional connection, and generate a health information set of school-enterprise relationship; the risk labeling module 302 is used to perform multi-dimensional coupled collaborative situational awareness and risk pattern recognition based on the health information set of school-enterprise relationship, and generate a collaborative barrier risk pattern set; the report generation module 303 is used to perform root cause tracing and adaptive strategy recommendation operations based on the collaborative barrier risk pattern set, and generate a smart education dynamic perception status report.

[0070] Optionally, when the health information module 301 generates a set of health information on school-enterprise relationships based on the acquired industry-education collaboration multimodal behavior dataset, performs collaborative role positioning and relationship network construction, analyzes structural power, knowledge flow, and emotional connection dimensions, and based on the industry-education collaboration multimodal behavior dataset, it is specifically used for: analyzing the role definitions and collaborative contractual relationships of multiple stakeholders, including enterprises, universities, and students, based on a preset industry-education collaboration project task list; and deploying multi-source heterogeneous data sensors at key physical and digital nodes of the project collaboration to form a panoramic behavior perception network for acquiring the industry-education collaboration multimodal behavior dataset; the industry-education collaboration... The multimodal behavior dataset includes voice stream communication data, text stream interaction data, and video stream dynamic data. Based on the voice stream communication data and the text stream interaction data, key features of discourse dominance and decision convergence are extracted to generate a power situation information set. Based on the text stream interaction data and the video stream dynamic data, a dynamic network of knowledge supply and demand response is constructed to generate a knowledge flow information set. Based on the video stream dynamic data and the voice stream communication data, the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence is calculated to generate an emotional connection information set. Based on the power situation information set, the knowledge flow information set, and the emotional connection information set, the health information set of the university-enterprise relationship is constructed.

[0071] Optionally, when the health information module 301 extracts key features of discourse dominance and decision convergence based on the voice stream communication data and the text stream interaction data to generate a power situation information set, it is specifically used for: constructing a topic consensus network based on topic initiation and consensus convergence to analyze the initiator's approval level; identifying new topics initiated by each subject based on the voice stream communication data and tracking the supporting voice feedback triggered by this topic in subsequent dialogues; the supporting voice feedback is quantified by calculating the aggregation degree of acoustic features of other subjects expressing agreement, elaborating in depth, or providing corroboration in subsequent turns, thereby generating a topic resonance index; simultaneously, based on the text stream interaction data, parsing the core text proposition proposed by the topic initiator and constructing a semantic diffusion tree of this proposition in subsequent text streams; quantifying the degree to which the core proposition is collectively absorbed and developed by analyzing the number, depth, and semantic fidelity of branch nodes in the semantic diffusion tree, generating a proposition inheritance index; and comprehensively calculating the dynamic influence weight of an individual in the collaborative field by combining the individual's topic resonance and proposition inheritance, thereby aggregating the weights of all subjects to generate the power situation information set.

[0072] Optionally, when the health information module 301 constructs a dynamic network of knowledge supply and demand response based on the text stream interaction data and the video stream dynamic data to generate a knowledge flow information set, it is specifically used for: constructing a knowledge interaction network based on direct interactive feedback to accurately characterize the effectiveness of knowledge flow; identifying knowledge supply events and corresponding demand response events based on the text stream interaction data, mapping the subject to the supply node and response node of the knowledge interaction network; quantifying the heuristic value of the supplied knowledge by analyzing the proportion and depth of three types of interactive behaviors—questioning, supplementing, and correcting—in the demand response events; simultaneously, decoding the micro-behavioral feedback of the subject when receiving knowledge supply from the video stream dynamic data, constructing an instantaneous cognitive absorption index by identifying the temporal patterns of nodding, gazing, recording, and confused expressions, to characterize the penetration and absorption state of knowledge at the receiving end; cross-modal correlation between the heuristic value and the instantaneous cognitive absorption index, assigning a circulation efficiency coefficient to each knowledge flow, thereby generating the knowledge flow information set that can identify knowledge hubs and value blind spots.

[0073] Optionally, when the health information module 301 calculates the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence based on the video stream dynamic data and the voice stream communication data to generate an emotional connection information set, it specifically performs the following: extracting physiological microstates based on facial action units and body posture vectors from the video stream dynamic data, and calculating the instantaneous consistency of the microstate transition probability between different subjects, thereby quantifying unconscious, underlying behavioral synchronicity; simultaneously, fusing the acoustic features of the voice stream communication data with the facial expressions of the video stream dynamic data, identifying the co-occurrence and transmission events of cross-modal emotional valence in dialogue rounds, and constructing an emotional resonance gradient describing the bidirectional flow of emotion between paired subjects; driving the emotional resonance gradient to propagate and superimpose in the collaborative relationship network based on the underlying connections formed by the behavioral synchronicity, and by simulating this ripple effect, emerging a group emotional tone that characterizes the level of cohesion of the group or even the entire team; and constructing the emotional connection information set according to the behavioral synchronicity, the emotional resonance gradient, and the group emotional tone.

[0074] Optionally, when the risk labeling module 302 generates a collaborative barrier risk pattern set by performing multi-dimensional coupled collaborative situational awareness and risk pattern identification based on the health information set of the school-enterprise relationship, it is specifically used to: perform cross-dimensional coupled analysis on the power situation information set, the knowledge flow information set, and the emotional connection information set; deconstruct various typical collaborative patterns by identifying stable emerging collaborative behavior configurations with specific characteristics; and, based on the configurational characteristics and collaborative effectiveness of the typical collaborative patterns, determine them as benign collaboration patterns or malignant risk patterns, and integrate the identified malignant risk patterns to generate the collaborative barrier risk pattern set.

[0075] Optionally, the risk labeling module 302, in the judgment process based on the benign collaboration mode, is specifically used to: ensure that the benign collaboration mode meets the following configuration characteristics: power dynamic level: the dynamic influence weights exhibit a decentralized spindle-shaped distribution, and high-influence subjects possess topic resonance and proposition inheritance levels higher than the team average, indicating that influence stems from recognition rather than position; knowledge flow level: the circulation efficiency coefficient of knowledge flow is generally high, cross-subject re-creation links frequently occur in the inspirational value, and the inspirational value is strongly positively correlated with the instantaneous cognitive absorption index; emotional connection level: the behavioral synchronicity and the emotional resonance gradient form a highly connected positive emotional loop in the collaborative relationship network, and the group emotional tone is stably in the positive range.

[0076] Optionally, the risk labeling module 302, in the process of judging the malignant risk patterns, is specifically used to: identify at least two malignant risk patterns: a unipolar dominant risk pattern: at the power dynamic level, the dynamic influence weight of a single entity is abnormally high, while the topic resonance and proposition inheritance of other entities remain low, forming a decision-making black hole; at the knowledge flow level, the circulation efficiency coefficient issued by this dominant entity is greater than a preset efficiency threshold, and the inspirational value and circulation efficiency coefficient of the reverse knowledge flow that forms feedback are close to zero; at the emotional connection level, behavioral synchronicity is manifested in the fact that most entities are concerned about this single entity. The one-way appeasement, while the emotional resonance gradient and the group's emotional tone are in a suppressed state; the floating collaboration risk mode: at the power situation level, the topic resonance degree is higher than the team's historical average, but the proposition inheritance degree is close to zero, indicating that the discussion is active but consensus cannot be formed; at the knowledge flow level, the instantaneous cognitive absorption index of knowledge flow remains at a medium level due to social etiquette, but the related inspirational value remains at a low level, revealing that knowledge interaction remains at a shallow response level; at the emotional connection level, the behavioral synchronicity is greater than the preset density threshold, but the emotional resonance gradient is weak, and the group's emotional tone is characterized by distant politeness and lack of deep resonance.

[0077] Optionally, the report generation module 303 is specifically used to: based on the collaborative barrier risk pattern set, trace back its abnormal data configuration in the power situation information set, knowledge flow information set, and emotional connection information set, and identify the core dimensions and key subject nodes that lead to the collaborative barrier; generate a set of adaptive strategy intervention vectors that are precisely matched with the root cause dimensions, wherein the intervention vectors include, but are not limited to: a dynamic discourse power rebalancing mechanism designed to break unipolar dominance, and a deep knowledge inquiry protocol constructed to eliminate floating collaboration; and generate the smart education dynamic perception status report containing diagnostic conclusions, evolution predictions, and dynamic intervention schemes based on the root cause tracing path and the adaptive strategy intervention vectors.

[0078] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A dynamic perception method for multimodal smart education scenarios based on field theory, characterized in that, include: Obtain a multimodal behavior dataset of industry-education collaboration; based on the dataset, perform collaborative role positioning and relationship network construction; analyze the dimensions of structural power, knowledge flow and emotional connection; and generate a health information set of school-enterprise relationship. Based on the aforementioned information set on the health of school-enterprise relationships, multi-dimensional coupled collaborative situational awareness and risk pattern identification are performed to generate a set of collaborative barrier risk patterns. Based on the aforementioned collaborative barrier risk pattern set, root cause tracing and adaptive strategy recommendation operations are performed to generate a smart education dynamic perception status report.

2. The method according to claim 1, characterized in that, The process involves acquiring a multimodal behavior dataset of industry-education collaboration, defining collaborative roles and constructing relationship networks based on this dataset, analyzing dimensions of structural power, knowledge flow, and emotional connection, and generating a set of information on the health of school-enterprise relationships, including: Based on a pre-set list of tasks for industry-education collaboration projects, the roles and collaborative contractual relationships of multiple stakeholders, including enterprises, universities, and students, are analyzed. Based on the defined roles and collaborative contractual relationships, multi-source heterogeneous data sensors are deployed at key nodes in the physical and digital spaces of project collaboration to form a panoramic behavior perception network, which is used to acquire the multimodal behavior dataset of industry-education collaboration. The industry-education collaboration multimodal behavior dataset includes voice stream communication data, text stream interaction data, and video stream dynamic data; Based on the voice stream communication data and the text stream interaction data, key features of discourse dominance and decision convergence are extracted to generate a power situation information set. Based on the text stream interaction data and the video stream dynamic data, a dynamic network of knowledge supply and demand response is constructed to generate a knowledge flow information set. Based on the video stream dynamic data and the voice stream communication data, the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence is calculated to generate an emotional connection information set. Based on the power dynamics information set, the knowledge flow information set, and the emotional connection information set, the health information set of the school-enterprise relationship is constructed.

3. The method according to claim 2, characterized in that, Based on the speech stream communication data and the text stream interaction data, key features of discourse dominance and decision convergence are extracted to generate a power situation information set, including: Construct a topic consensus network based on topic initiation and consensus convergence to analyze the initiator's approval rating: Based on the voice stream communication data, identify new topics initiated by each subject and track the supporting voice feedback triggered by this topic in subsequent dialogues; The supportive voice feedback is quantified by calculating the degree of acoustic feature aggregation of other subjects in subsequent turns, expressing agreement, elaborating, or providing corroborating evidence, thereby generating a topic resonance index. Simultaneously, based on the text stream interaction data, the core text proposition proposed by the topic initiator is parsed, and a semantic diffusion tree of this proposition in the subsequent text stream is constructed; By analyzing the number, depth, and semantic fidelity of the branch nodes in the semantic diffusion tree, the degree to which the core proposition is collectively absorbed and developed is quantified, and a proposition inheritance index is generated. By combining the topic resonance and proposition inheritance of individuals, the dynamic influence weight of each individual in the collaborative field is calculated, and then the weights of all subjects are aggregated to generate the power situation information set.

4. The method according to claim 3, characterized in that, The process of constructing a dynamic network for knowledge supply and demand response based on the text stream interaction data and the video stream dynamic data, and generating a knowledge flow information set, includes: Construct a knowledge interaction network based on direct interactive feedback to accurately characterize the effectiveness of knowledge flow: Based on the text flow interaction data, identify knowledge supply events and corresponding demand response events, and map the subjects as supply nodes and response nodes of the knowledge interaction network; By analyzing the proportion and depth of the three types of interactive behaviors—questioning, supplementing, and correcting—in the aforementioned demand response events, the heuristic value of supply knowledge is quantified. Meanwhile, from the dynamic data of the video stream, the micro-behavioral feedback of the decoding subject when receiving knowledge is analyzed. By identifying the temporal patterns of nodding, staring, recording and confused expressions, an instantaneous cognitive absorption index is constructed to characterize the penetration and absorption status of knowledge at the receiving end. By cross-modal correlation between the heuristic value and the instantaneous cognitive absorption index, a circulation efficiency coefficient is assigned to each knowledge flow, thereby generating the knowledge flow information set that can identify knowledge hubs and value blind spots.

5. The method according to claim 4, characterized in that, The step of calculating the co-occurrence frequency of nonverbal behavior synchronicity and emotional valence based on the dynamic data of the video stream and the communication data of the voice stream, and generating an emotional connectivity information set, includes: Physiological microstates, based on facial action units and body posture vectors, are extracted from the dynamic data of the video stream, and the instantaneous consistency of the microstate transition probability between different subjects is calculated to quantify the unconscious, underlying behavioral synchronization. Simultaneously, by fusing the acoustic features of the voice stream communication data with the facial expressions of the video stream dynamic data, the co-occurrence and transmission events of cross-modal emotional valence in dialogue rounds are identified, and an emotional resonance gradient describing the bidirectional flow of emotions between paired subjects is constructed. Based on the underlying connections formed by the aforementioned behavioral synchronicity, the emotional resonance gradient is driven to propagate and superimpose in the collaborative relationship network. By simulating this ripple effect, a group emotional tone that represents the level of cohesion of the group or even the entire team emerges. The emotional connection information set is constructed based on the behavioral synchronicity, the emotional resonance gradient, and the group emotional tone.

6. The method according to claim 5, characterized in that, The step of performing multi-dimensional coupled collaborative situational awareness and risk pattern identification based on the school-enterprise relationship health information set, and generating a collaborative barrier risk pattern set, includes: A cross-dimensional coupling analysis is performed on the power situation information set, the knowledge flow information set, and the emotional connection information set. By identifying the stable emerging collaborative behavior configurations with specific characteristics, a variety of typical collaborative patterns are deconstructed. Based on the configurational characteristics and collaborative effectiveness of the typical collaborative modes, they are identified as benign collaborative modes or malignant risk modes, and the identified malignant risk modes are integrated to generate the collaborative barrier risk mode set.

7. The method according to claim 6, characterized in that, The process for identifying the benign collaboration mode includes: The benign cooperation mode must satisfy the following configuration characteristics: At the power dynamics level: the dynamic influence weights exhibit a decentralized spindle-shaped distribution, and high-influence entities possess a higher-than-average level of topic resonance and proposition inheritance, indicating that influence stems from recognition rather than position. At the knowledge flow level: the circulation efficiency coefficient of knowledge flow is generally high, the inspirational value frequently involves cross-subject re-creation, and the inspirational value is strongly positively correlated with the instantaneous cognitive absorption index; At the level of emotional connection: the behavioral synchronicity and the emotional resonance gradient form a highly connected positive emotional loop in the collaborative relationship network, and the group's emotional tone remains stable in the positive range.

8. The method according to claim 7, characterized in that, The process for identifying the malignant risk pattern includes: At least two of the aforementioned malign risk patterns were identified: Unipolar-dominated risk model: At the power dynamic level, the dynamic influence weight of a single entity is abnormally high, while the resonance of topics and the inheritance of propositions among other entities remain low, forming a decision-making black hole; At the knowledge flow level, the circulation efficiency coefficient issued by this leading entity is greater than the preset efficiency threshold, while the heuristic value and circulation efficiency coefficient of the reverse knowledge flow that forms feedback are close to zero. At the level of emotional connection, behavioral synchronicity manifests as a one-way conformity of the majority of subjects to this single subject, while the emotional resonance gradient and the emotional tone of the group are in a state of suppression. Suspended Collaboration Risk Model: At the power dynamic level, the resonance of the topic is higher than the team's historical average, but the inheritance of the proposition is close to zero, indicating that the discussion is active but consensus cannot be formed. At the level of knowledge flow, the instantaneous cognitive absorption index of knowledge flow remains at a median level due to social etiquette, but the associated inspirational value remains at a low level, indicating that knowledge interaction remains at a superficial level. At the level of emotional connection, the behavioral synchronicity is greater than a preset density threshold, but the emotional resonance gradient is weak. The emotional tone of the group is characterized by detached politeness and a lack of deep resonance.

9. The method according to claim 8, characterized in that, The step of performing root cause analysis and adaptive strategy recommendation operations based on the collaborative barrier risk pattern set to generate a smart education dynamic perception status report includes: Based on the aforementioned risk pattern set of collaborative barriers, we trace back to the abnormal data configurations in the power situation information set, knowledge flow information set, and emotional connection information set to identify the core dimensions and key subject nodes that lead to collaborative barriers. Generate a set of adaptive strategy intervention vectors that precisely match the root cause dimension. The intervention vectors include, but are not limited to: a dynamic discourse power rebalancing mechanism designed to break unipolar dominance and a deep knowledge inquiry protocol constructed to eliminate floating collaboration. Based on the root cause tracing path and the adaptive strategy intervention vector, a smart education dynamic perception status report is generated, which includes diagnostic conclusions, evolutionary predictions, and dynamic intervention plans.

10. A multimodal intelligent education scenario dynamic perception system based on field theory, characterized in that, The method applied to any one of claims 1-9 includes: The health information module is used to acquire a multimodal behavior dataset of industry-education collaboration. Based on the multimodal behavior dataset of industry-education collaboration, it performs collaborative role positioning and relationship network construction, analyzes the dimensions of structural power, knowledge flow and emotional connection, and generates a health information set of school-enterprise relationship. The risk labeling module is used to perform multi-dimensional coupled collaborative situational awareness and risk pattern recognition based on the health information set of the school-enterprise relationship, and generate a collaborative barrier risk pattern set. The report generation module is used to perform root cause analysis and adaptive strategy recommendation operations based on the collaborative barrier risk pattern set, and generate a smart education dynamic perception status report.

Images