Personalized soft skill training method and system based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a personalized soft skill training method and system based on artificial intelligence. The method comprises the following steps of performing evaluation by using a multi-dimensional questionnaire, and analyzing semantics of user answers in combination with a large language model to generate an initial user portrait; updating the initial user portrait; constructing a skill tree by using a word embedding technology and an improved Woltzz method; constructing and optimizing a global model based on the skill tree; and dynamically generating a personalized learning path according to the global model. By introducing an artificial intelligence large language model technology, a neural network word embedding algorithm and a graph theory algorithm and combining a'questionnaire-learning-practice 'closed-loop training mode, individuation, systematization and intelligentization of soft skill training are realized, the technical bottleneck of a traditional method in the aspect of subjective question type intelligent feedback is broken through, and the intelligent feedback of the subjective question type is realized. And the soft skill training effect and the user experience are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly relates to a personalized soft skill training method and system based on artificial intelligence. Background Art

[0002] With the increasing demand for soft skills in society, personal growth courses have gradually become an important way to enhance workplace competitiveness. The traditional course model has obvious deficiencies in personalized training, real-time feedback, and quantitative evaluation, and it is difficult to meet the needs of users for efficient and accurate soft skill improvement. The training of soft skills usually involves complex psychological characteristics, behavioral patterns, and skill correlations, and it is difficult for traditional methods to systematically depict the user state and provide targeted training programs.

[0003] In the prior art, soft skill training mainly relies on the live teaching mode and the recorded teaching mode, including live classes and recorded classes, lacking a systematic user practice session and unable to provide real-time and accurate learning feedback. In addition, the description of soft skills is usually relatively general, lacking clear quantitative evaluation criteria, and there are often similarities between different soft skills, further increasing the difficulty of clear definition and evaluation. Although some learning software systems have the functions of practice and feedback, their application scope is limited to objective question types and there are obvious deficiencies in dealing with subjective question types. The prior art has the following specific defects: First, there is a lack of systematic description of the user's psychological characteristics and behavioral patterns; second, it is impossible to achieve quantitative evaluation of soft skills and generation of personalized training paths; third, there are technical bottlenecks in the intelligent evaluation of subjective question types. Therefore, there is an urgent need to propose an innovative method to solve the above defects and improve the effect and efficiency of soft skill training.

[0004] At present, there is not much research work on personalized soft skill training, and there is no specific personalized soft skill training method based on artificial intelligence. Summary of the Invention

[0005] In view of the defects in the prior art, the present invention provides a personalized soft skill training method and system based on artificial intelligence.

[0006] First aspect, a personalized soft skill training method based on artificial intelligence provided by the present invention includes the following steps: using a questionnaire for evaluation, and combining with a large language model to analyze the semantics of the user's answers, generating an initial user profile. The questionnaire includes objective questions, essay questions, and table questions. The initial user profile includes a psychological feature vector, a behavior pattern matrix, and a soft skill mastery distribution; updating the initial user profile and obtaining an update result; based on the update result, using word embedding technology and an improved Ward method to construct a hierarchical skill tree; based on the skill tree, constructing and optimizing a global model for predicting the user's cognitive load and emotional state; based on the global model, improving the user's soft skill mastery and training experience. The present invention uses a multi-dimensional questionnaire combined with a large language model to deeply analyze the semantics of the user's answers, generating an initial user profile including a psychological feature vector, a behavior pattern matrix, and a soft skill mastery distribution, realizing the accurate characterization of the user's soft skills and psychological state, breaking through the limitations of traditional evaluation methods in terms of depth and breadth; by updating the user profile, ensuring the real-time and accuracy of the user profile, overcoming the defect that the static profile cannot reflect the dynamic changes of the user; by using word embedding technology and an improved Ward method to construct a hierarchical skill tree, realizing the modularization and systematization of soft skill training, overcoming the disadvantages of isolated skill points and lack of association in traditional training; by constructing and optimizing the global model, dynamically generating personalized learning paths, improving the accuracy of predicting the user's cognitive load and emotional state, balancing the path length and cognitive load, significantly improving the user's soft skill mastery and training experience, breaking through the bottleneck of traditional training methods in terms of personalization and efficiency.

[0007] Optionally, the updating the initial user profile and obtaining an update result includes: establishing a weight adjustment model, which is used to dynamically adjust the weights in the user profile according to the user's real-time feedback and behavior data. The weight adjustment model satisfies the following expression: , where, represents the weight of the th feature dimension at time, represents the weight of the th feature dimension at time, represents the learning rate, represents the loss function under the weight at time, represents the real-time feedback value of the th feature dimension at time, represents the weight of the th feature dimension at time, Indicates the real-time feedback value of the th feature dimension at time; indicates the total number of feature dimensions; according to the weight adjustment model, the initial user profile is dynamically updated and an update result is obtained. By establishing a weight adjustment model and dynamically adjusting the weights in the user profile according to the user's real-time feedback and behavior data, the present invention realizes the real-time update and precise optimization of the user profile, breaking through the limitation that the traditional static profile cannot reflect the dynamic changes of users; by introducing a loss function and a learning rate to iteratively optimize the weights, the scientificity and stability of the weight adjustment are ensured, overcoming the subjectivity and arbitrariness of the traditional method in weight allocation; by comprehensively considering the real-time feedback values of multi-dimensional feature dimensions and their weight relationships, the user profile is dynamically updated, realizing a comprehensive portrayal of the user's psychological characteristics, behavior patterns, and soft skill mastery, overcoming the deficiencies of the traditional method in feature correlation and dynamics; through the mathematical expression of the weight adjustment model, the calculation logic and iterative process of weight update are clarified, improving the transparency and interpretability of the algorithm, and solving the problems of complexity and operability of the traditional method; by dynamically updating the user profile and obtaining an update result, real-time and accurate data support is provided for the generation of personalized learning paths, significantly improving the personalization and effectiveness of soft skill training.

[0008] Optionally, constructing a hierarchical skill tree based on the update result by using word embedding technology and an improved Ward's method includes: based on the update result, using a large language model to perform semantic parsing on the user's target soft skills and extract high-order soft skill labels; according to the high-order soft skill labels, combining with the basic soft skill set to generate an initial soft skill node set; based on the initial soft skill node set, combining with word embedding technology to establish a calculation model for the semantic similarity matrix between soft skill nodes; according to the calculation model, obtaining a distance matrix; based on the distance matrix, using the improved Ward's method for hierarchical clustering to obtain a hierarchical skill tree. By using a large language model to perform semantic parsing on the user's target soft skills and extract high-order soft skill labels, the present invention realizes a deep understanding and accurate characterization of the user's soft skill requirements, breaking through the limitations of traditional methods in semantic parsing accuracy and depth; by combining with the basic soft skill set to generate an initial soft skill node set, the comprehensiveness and systematicness of the skill tree are ensured, overcoming the one-sidedness and isolation of traditional methods in skill point coverage; by using word embedding technology to establish a calculation model for the semantic similarity matrix between soft skill nodes, a quantitative analysis of the semantic relationship between soft skill nodes is realized, avoiding the ambiguity and subjectivity of traditional methods in semantic relevance; by using the improved Ward's method for hierarchical clustering to construct a hierarchical skill tree, the modularization and hierarchicalization of soft skill training are realized, solving the chaos and inefficiency of traditional methods in skill organizational structure; through the hierarchical skill tree, a clear and systematic soft skill training framework is provided for users, significantly improving the pertinence and effect of training, and breaking through the singularity and inefficiency of traditional methods in training path design.

[0009] Optionally, the calculation model for the semantic similarity matrix between soft skill nodes satisfies the following expression: , where, is the semantic similarity matrix between the th soft skill node and the th soft skill node, is the word embedding vector of the th soft skill node, is the word embedding vector of the th soft skill node; the distance matrix satisfies the following relational expression: , where, is the distance matrix between the th soft skill node and the th soft skill node; the distance update formula of the improved Ward's method is as follows: , Among them, represents the distance between new clusters, represents the cluster the number of soft skill nodes included, represents the cluster the number of soft skill nodes included, represents the total number of samples, represents the cluster and the cluster the distance between them, represents the cluster the central vector of, represents the cluster the central vector of. Through calculating the semantic similarity matrix between soft skill nodes by word embedding vectors, the present invention realizes the accurate quantification of the semantic relationship between soft skill nodes, breaking through the fuzziness and subjectivity in semantic similarity calculation of traditional methods; by converting the semantic similarity matrix into a distance matrix, it provides a scientific and reliable data basis for hierarchical clustering, overcoming the limitations and inaccuracies in distance measurement of traditional methods; by improving the distance update formula in the Ward's method, comprehensively considering the influence of cluster size, distance between clusters and central vectors, it realizes the dynamic optimization and precise division of hierarchical clustering, overcoming the deficiencies in complexity and stability of traditional clustering methods, and also solving the computational complexity and inefficiency problems in dealing with large-scale data of traditional methods; by constructing a skill tree with a hierarchical structure, it provides a clear and systematic soft skill training framework for users, significantly improving the pertinence and effect of training, and breaking through the monotony and inefficiency in training path design of traditional methods.

[0010] Optionally, constructing and optimizing a global model for predicting user cognitive load and emotional state based on the skill tree includes: collecting in real time the behavioral data, physiological data, and text sentiment data of the user during the learning process based on the skill tree; using the behavioral data, the physiological data, and the text sentiment data to construct mixed feature data; using the mixed feature data to construct and optimize a global model for predicting the user's cognitive load and emotional state on the premise of protecting user privacy. By collecting in real time the behavioral data, physiological data, and text sentiment data of the user, the present invention realizes the comprehensive monitoring of the multi-dimensional state of the user during the learning process, breaking through the limitations of the traditional single data source in information coverage; by constructing mixed feature data, the behavioral, physiological, and emotional data are organically combined, providing a richer and more accurate representation of the user state, overcoming the one-sidedness and inefficiency of the traditional method in feature extraction; by constructing and optimizing the global model on the premise of protecting user privacy, the balance between data sharing and privacy protection is achieved, overcoming the deficiencies of the traditional centralized learning in data security and privacy protection; using the mixed feature data to train the global model significantly improves the generalization ability and prediction accuracy of the model; through the optimized global model, the user's cognitive load and emotional state are accurately predicted, providing a scientific basis for the dynamic adjustment of the personalized learning path, and significantly improving the personalization and user experience of the soft skill training.

[0011] Optionally, according to the global model, obtain the prediction results of the user's cognitive load and emotional state; based on the prediction results, construct an objective function for balancing the path length and the cognitive load, and the objective function is as follows: , wherein, is the learning path, is the soft skill node and is the distance between them, is the cognitive load of the path , , is the weight coefficient; through the objective function, a personalized learning path is dynamically generated; through the personalized learning path, the user's soft skill mastery and training experience are improved. The present invention obtains the prediction results of the user's cognitive load and emotional state through the global model, realizing the accurate grasp of the user's learning state, breaking through the ambiguity and lag of the traditional method in state prediction; by comprehensively considering the balance between path length and cognitive load, a scientific and reasonable basis for learning path generation is provided, overcoming the monotony and inefficiency of the traditional method in path design; by dynamically generating a personalized learning path, ensuring the optimal matching of path length and cognitive load, significantly improving the user's learning efficiency and experience, overcoming the deficiencies of the traditional fixed path in personalization and adaptability; using the weight coefficient to flexibly adjust the priority of path length and cognitive load, realizing the dynamic optimization and precise control of path generation, solving the adaptability problem of the traditional method in complex scenarios; through the dynamic generation of the personalized learning path, a customized soft skill training plan is provided for the user, significantly improving the pertinence and effect of training, breaking through the limitations and inefficiency of the traditional method in training path design.

[0012] Optionally, the dynamically generating a personalized learning path through the objective function includes: dynamically adjusting the weights of the path length and cognitive load through the objective function, and optimizing the generated path; according to the changes in the user's cognitive load and emotional state, making real-time fine-tuning of the generated path, and dynamically generating a personalized learning path, which can not only meet the learning efficiency but also adapt to the user's cognitive load and emotional state. The present invention realizes the accurate positioning of the learning path and the personalized starting point design, breaking through the blindness and inefficiency of the traditional method in path starting point selection; by dynamically adjusting the weights of the path length and cognitive load and optimizing the generated path, ensuring the scientificity and efficiency of the path, overcoming the monotony and rigidity of the traditional method in path optimization; according to the changes in the user's cognitive load and emotional state, making real-time fine-tuning of the generated path, realizing the dynamic adjustment and real-time optimization of the learning path, overcoming the deficiencies of the traditional fixed path in adaptability and flexibility; through the dynamically generated personalized learning path, which can not only meet the learning efficiency but also adapt to the user's cognitive load and emotional state, significantly improving the user's learning experience and effect, solving the disconnection problem between the traditional method in path design and user state matching; through real-time fine-tuning and dynamic optimization, a highly personalized and adaptable soft skill training plan is provided for the user, breaking through the limitations and inefficiency of the traditional method in training path design.

[0013] Optionally, the making real-time fine-tuning of the generated path according to the changes in the user's cognitive load and emotional state and dynamically generating a personalized learning path includes: determining a fine-tuning path objective function according to the changes in the user's cognitive load and emotional state, and the fine-tuning path objective function is as follows: , wherein, is the path weight vector at time [[ID=]], representing the weight distribution of each soft skill node in the current learning path, is the path weight vector at time [[ID=]], is the learning rate at time [[ID=]], used to control the step size of weight update, is the th soft skill node and the th soft skill node the distance between them, , are weight coefficients, is the cognitive load of the path [[ID=]] at time [[ID=]], is the cognitive load of the path [[ID=]] at time [[ID=]], is the number of soft skill nodes in the path [[ID=]]; Based on the fine-tuning path objective function, the generated path is fine-tuned in real time to obtain a fine-tuning result; Through the fine-tuning result, a personalized learning path is dynamically generated. The present invention determines the fine-tuning path objective function based on the changes in the user's cognitive load and emotional state, realizes the real-time dynamic adjustment of the learning path, and breaks through the lag and rigidity of the traditional method in path adjustment; By introducing the path weight vector and the learning rate, the weight distribution of each skill node in the path is dynamically optimized, ensuring the scientificity and accuracy of the path adjustment, and overcoming the subjectivity and inefficiency of the traditional method in weight distribution; By comprehensively considering the balance between the path length and the cognitive load, the generated path is fine-tuned in real time, significantly improving the adaptability of the path and the user's learning experience, and overcoming the deficiencies of the traditional fixed path in personalization and adaptability; By fine-tuning the weight coefficients and the learning rate in the fine-tuning path objective function, the amplitude and direction of the path adjustment are flexibly controlled, solving the adaptability problem of the traditional method in complex scenarios; By the dynamically generated personalized learning path, it can not only meet the learning efficiency, but also adapt to the user's cognitive load and emotional state, significantly improving the pertinence and effect of soft skill training, and breaking through the limitations and inefficiencies of the traditional method in training path design.

[0014] Optionally, the real-time fine-tuning of the generation path based on the fine-tuning path objective function and obtaining the fine-tuning result include: setting an initial path weight vector and an initial learning rate based on the fine-tuning path objective function; calculating the gradient of the fine-tuning path objective function according to the initial path weight vector and the initial learning rate; updating the learning rate using the gradient; updating the path weight vector according to the learning rate; and performing real-time fine-tuning on the generation path through the path weight vector and obtaining the fine-tuning result. Based on the fine-tuning path objective function, the present invention sets the initial path weight vector and the initial learning rate, providing a scientific and reasonable starting point for path fine-tuning, breaking through the blindness and inefficiency of traditional methods in initial parameter setting; by calculating the gradient of the fine-tuning path objective function, it realizes precise control of the path adjustment direction, ensuring the efficiency and stability of path optimization, and overcoming the complexity and inaccuracy of traditional methods in gradient calculation; using the gradient to update the learning rate and dynamically adjusting the step size of path optimization significantly improves the convergence speed and accuracy of path fine-tuning, overcoming the limitations and inefficiency of traditional fixed learning rates in path optimization; updating the path weight vector according to the learning rate realizes the dynamic optimization and real-time adjustment of path weight allocation, solving the lag and rigidity of traditional methods in weight update; performing real-time fine-tuning on the generation path through the path weight vector and obtaining the fine-tuning result provides users with a highly personalized and adaptable learning path, significantly improving the pertinence and effect of soft skill training, and breaking through the limitations and inefficiency of traditional methods in training path design.

[0015] Second aspect, a personalized soft skill training system based on artificial intelligence provided by the present invention includes an input device, a processor, an output device, and a memory. The input device, the processor, the output device, and the memory are interconnected. Among them, the memory is used to store computer programs, and the computer programs include program instructions. The processor is configured to call the program instructions, and the system uses the described personalized soft skill training method based on artificial intelligence. The system provided by the present invention has a high degree of integration, and the information transmission between each component is smooth. By using a multi-dimensional questionnaire combined with a large language model to deeply analyze the semantics of the user's answers, an initial user profile including a psychological feature vector, a behavior pattern matrix, and a soft skill mastery distribution is generated, realizing the accurate characterization of the user's soft skills and psychological state, and breaking through the limitations of traditional assessment methods in terms of depth and breadth; by dynamically updating the user profile and optimizing the global model, the real-time prediction and dynamic adjustment of the user's cognitive load and emotional state are realized, overcoming the lag and inefficiency of traditional training systems in user state monitoring and feedback; using word embedding technology and an improved Ward's method to construct a hierarchical skill tree and dynamically generate a personalized learning path, realizing the modularization, systematization, and high personalization of soft skill training, significantly improving the scientific nature of training and the user experience, and breaking through the singularity and inefficiency of traditional training systems in path design; by making real-time fine-tuning of the generated path, dynamically adjusting the path weight vector and learning rate, ensuring the real-time and accuracy of path optimization, solving the adaptability problem of traditional methods in complex scenarios, and providing a highly personalized and adaptable soft skill training plan for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the personalized soft skill training method based on artificial intelligence according to an embodiment of the present invention; Figure 2 It is a schematic structural diagram of the personalized soft skill training system based on artificial intelligence according to an embodiment of the present invention; Figure 3 It is a flowchart of the operation of the artificial intelligence large language model according to an embodiment of the present invention; Figure 4 It is a flowchart of building a skill tree according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described here are only for illustrative purposes and are not used to limit the present invention. In the following description, in order to provide a thorough understanding of the present invention, a large number of specific details are set forth. However, it is obvious to those of ordinary skill in the art that the present invention does not have to be implemented with these specific details. In other instances, well-known circuits, software, or methods are not specifically described to avoid obscuring the present invention.

[0018] Throughout the specification, references to "an embodiment", "embodiments", "an example" or "examples" mean that the specific features, structures, or characteristics described in connection with the embodiment or example are included in at least one embodiment of the present invention. Thus, the phrases "in an embodiment", "in embodiments", "an example" or "examples" that appear throughout the specification do not necessarily all refer to the same embodiment or example. Additionally, the specific features, structures, or characteristics may be combined in any suitable combination and / or sub-combination in one or more embodiments or examples. Further, those of ordinary skill in the art should understand that the diagrams provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0019] Please refer to Figure 1 , embodiments of the present invention provide a personalized soft skill training method based on artificial intelligence, and the method includes the following steps: S1. Conduct an assessment using a questionnaire and analyze the semantics of the user's answers in combination with a large language model to generate an initial user profile. The questionnaire includes objective questions, essay questions, and table questions, and the initial user profile includes a psychological feature vector, a behavior pattern matrix, and a soft skill mastery distribution.

[0020] Among them, S1 further includes the following steps: S11. Design a multi-dimensional questionnaire.

[0021] In one embodiment, first determine the assessment dimensions, where the assessment dimensions include psychological characteristics, behavior patterns, and soft skills; the psychological characteristics include personality type, emotional stability, and self-awareness, the behavior patterns include communication style, decision-making style, and work habits, and the soft skills include communication ability, empathy, teamwork ability, and time management.

[0022] Further, design questionnaire questions, where the questionnaire questions include but are not limited to objective questions, essay questions, and table questions.

[0023] Specifically, in the dimension of psychological characteristics, use the questions in the psychological assessment scale, including but not limited to the MBTI personality test and the Big Five personality scale, and present them in the form of multiple-choice questions or scale questions. For example: "When facing pressure, how do you usually react? (A. Analyze calmly B. Seek support C. Avoid problems D. Others)".

[0024] In the dimension of behavior patterns, design scenario questions that describe specific work or life scenarios and ask the user for their reactions or choices. For example: "In a team discussion, you are more inclined to: (A. Speak actively and put forward your own opinions B. Listen to others' opinions and rarely speak actively C. Wait to be invited to speak D. Others)" In the dimension of soft skills, the user's skill mastery is understood through self-assessment questions. For example: "Which soft skill do you hope to focus on cultivating?"; "Can you share the specific reasons for choosing this skill?"; "Can you describe a difficult scenario related to this skill that you encountered in your actual work or life?"

[0025] Furthermore, conduct a pre-test on the questionnaire, collect feedback, and adjust the questions to ensure its accuracy and understandability.

[0026] S12. Collect user responses.

[0027] In one embodiment, first determine the target user group and publish the questionnaire using an online questionnaire platform, where the online questionnaire platform includes Wenjuanxing; Furthermore, monitor the filling progress of the questionnaire to ensure that a sufficient number of valid responses are collected, and clean the data to remove invalid or duplicate responses.

[0028] S13. Perform semantic analysis using an artificial intelligence large language model.

[0029] In one embodiment, first select an artificial intelligence large language model, where the artificial intelligence large language model includes the GPT series and BERT. It is necessary to ensure that the model has been fully trained to understand natural language and process complex semantic information.

[0030] Furthermore, perform preprocessing operations on the user responses. The preprocessing operations include word segmentation and stop word removal, and convert the preprocessed responses into an input format acceptable to the model.

[0031] Furthermore, input the user responses into the large language model, analyze the results output by the model, and extract key information. The key information includes sentiment tendency, topic classification, and entity recognition.

[0032] S14. Generate an initial user profile.

[0033] In one embodiment, first construct a psychological feature vector.

[0034] Specifically, according to the analysis of the psychological feature-related questions in the user responses by the large language model, extract key features, and quantify the key features into numerical values to construct a psychological feature vector. For example, for personality types, map different personality types to specific numerical intervals.

[0035] Furthermore, construct a behavior pattern matrix.

[0036] Specifically, analyze the results of questions related to behavioral patterns in the user's answers to identify the user's behavioral choices in different scenarios; based on these choices, construct a behavioral pattern matrix to describe the user's behavioral characteristics in different dimensions. For example, construct a two-dimensional matrix where the horizontal axis represents different scenarios or tasks, and the vertical axis represents the user's behavioral choices or tendencies.

[0037] Furthermore, evaluate the mastery of soft skills.

[0038] Specifically, based on the self-assessment results of questions related to soft skills in the user's answers, evaluate the user's mastery of different soft skills, quantify the evaluation results as numerical values, and construct a distribution map or table of soft skill mastery. For example, use a five-point scale or a hundred-point scale to evaluate the user's soft skill mastery and draw a bar chart or a pie chart for visual display.

[0039] Furthermore, integrate the data information of the psychological feature vector, the behavioral pattern matrix, and the soft skill mastery to generate an initial user profile.

[0040] S2. Update the initial user profile and obtain the update result.

[0041] In one embodiment, first, according to the initial user profile, assign initial weights to each feature dimension, where the feature dimensions include the psychological feature dimension, the behavioral pattern dimension, and the soft skill dimension, and the features include user interest tags and behavioral frequencies.

[0042] Furthermore, construct a weight adjustment model for dynamically adjusting the weights, and the specific steps are as follows: First, calculate the loss function at the current moment, and the loss function is as follows:

[0043] where is the loss function under the current moment weight, represents the real-time feedback value of the th feature dimension at the moment, represents the weight of the th feature dimension at the moment, represents the real-time feedback value of the th feature dimension at the moment, represents the total number of feature dimensions. Furthermore, calculate the gradient of the loss function, that is, the partial derivative of the loss function with respect to the weight , as follows:

[0044] Further, update the weights according to the gradient of the loss function, and its weight adjustment model is: , where represents the weight of the -th feature dimension at time, represents the weight of the -th feature dimension at time, represents the learning rate, represents the loss function under the weights at represents the real-time feedback value of the -th feature dimension at time, represents the weight of the -th feature dimension at time, represents the real-time feedback value of the -th feature dimension at time, represents the total number of feature dimensions; the weight adjustment model is used to dynamically adjust the weights of the feature dimensions in the user profile according to the user's real-time feedback and behavior data.

[0045] Further, according to the updated weights, recalculate the user profile and store the updated user profile in the database for downstream tasks to use, where the downstream tasks include a recommendation system.

[0046] It should be noted that by continuously dynamically adjusting the weights, it is ensured that the user profile is always consistent with the user's latest state. In addition, this algorithm can also predict the future user profile according to the user's long-term behavior trend, providing forward-looking guidance for future personalized recommendations and services.

[0047] S3. According to the update result, use word embedding technology and an improved Ward method to construct a skill tree with a hierarchical structure.

[0048] Among them, S3 further includes the following steps: S31. Parse the semantics of the user's target soft skills and extract high-order soft skill labels.

[0049] In one embodiment, first, according to the update result, use a large language model to perform semantic parsing on the user's target soft skills and extract high-order soft skill labels.

[0050] Specifically, collect the descriptive text of the user's target soft skills, and the sources of the descriptive text include but are not limited to user questionnaires and job descriptions.

[0051] Furthermore, the large language model is used to semantically analyze the user's target soft skills. During the analysis process, the model will understand and extract the key features and connotations of the soft skills.

[0052] Furthermore, according to the analysis results, combined with the predefined high-order soft skill tag library, the high-order soft skill tags that best match the user's target soft skills are extracted. The high-order soft skill tag library includes widely recognized soft skill categories, such as but not limited to leadership, empathy, communication skills, teamwork, and innovation ability.

[0053] It should be noted that the prior art usually relies on manual definition or simple keyword matching to extract soft skill tags, lacking in-depth understanding of the semantics of the user's description. In contrast, the present invention uses the large language model for semantic analysis, which can automatically extract high-order soft skill tags, significantly improving the accuracy and coverage of the tags.

[0054] S32. Generate an initial set of soft skill nodes.

[0055] In one embodiment, a basic set of soft skills is first constructed. The basic set of soft skills includes a series of specific and learnable soft skills, such as but not limited to effective communication, time management, and conflict resolution.

[0056] Furthermore, according to the extracted high-order soft skill tags, the relevant skills are screened out from the basic set of soft skills to form an initial set of soft skill nodes.

[0057] S33. Establish a semantic similarity matrix calculation model between nodes and obtain a distance matrix.

[0058] In one embodiment, first, the word embedding technology of the neural network pre-trained language model is used to convert each skill in the initial set of soft skill nodes into a word embedding vector. It should be noted that the word embedding vector is obtained by mapping each soft skill node to a high-dimensional vector space.

[0059] Furthermore, according to the semantic similarity matrix calculation model between soft skill nodes, the semantic similarity between each pair of skill nodes is calculated to form a semantic similarity matrix. The semantic similarity matrix calculation model between soft skill nodes satisfies the following expression: , where is the semantic similarity matrix between the th soft skill node and the th soft skill node, is the word embedding vector of the th soft skill node, is the The word embedding vectors of soft skill nodes.

[0060] Furthermore, based on the semantic similarity matrix, a distance matrix is obtained through a transformation algorithm, and the distance values of the distance matrix reflect the semantic differences between soft skill nodes. The expression corresponding to the transformation algorithm is as follows: , where is the th soft skill node and the th soft skill node

[0061] S34. Use the improved Ward's method for hierarchical clustering to construct a skill tree with a hierarchical structure.

[0062] In one embodiment, the clustering is initialized first, and each skill node is regarded as a separate cluster.

[0063] Furthermore, use the improved Ward's method for hierarchical clustering, and the distance update formula is: , where represents the distance between new clusters, represents the number of soft skill nodes included in cluster , represents the number of soft skill nodes included in cluster , represents the total number of samples, represents the distance between cluster and cluster , represents the center vector of cluster , represents the center vector of cluster .

[0064] Furthermore, iteratively merge the two closest clusters until the predetermined number of clusters is reached or other stopping conditions are met.

[0065] Furthermore, using graph theory algorithms, a tree diagram is generated, that is, a skill tree with a hierarchical structure is obtained. The skill tree intuitively shows the association relationships and development paths between different soft skills. The association relationships and the development paths are used to identify the basic soft skills with the highest scores of the user as their advantageous items, and find the new skills with the shortest distance from the advantageous skills in the skill tree as development suggestions, which ensures the continuity and accessibility of the learning path and helps the user gradually expand the scope of soft skills. In addition, in the soft skill tree, the graph theory algorithm refers to the method of applying graph theory concepts to solve practical problems or improve personal abilities. Graph theory is a branch of mathematics that studies graph structures composed of nodes and edges and is widely used in the fields of computer science, network analysis, and social networks.

[0066] It should be noted that for the Ward's method, the traditional Ward's method only considers the weights of the number of clustering samples , while ignoring the distances between the cluster center vectors. The improved Ward's method of the present invention can more accurately reflect the semantic similarity between clusters by introducing as an adjustment term, especially in high-dimensional data, the distances of the cluster center vectors can better capture the differences between clusters; in high-dimensional data, the traditional Ward's method is easily affected by noise and redundant information, resulting in unstable clustering results. The improved Ward's method of the present invention can effectively reduce the influence of noise and improve the robustness of clustering by introducing the distance adjustment term of the center vector; in the clustering of skill nodes, semantic similarity is a key factor. The traditional Ward's method is only based on the number of samples and distances and cannot fully capture semantic information. The improved Ward's method of the present invention can better reflect the semantic relationships between skill nodes by combining the distances of the center vectors, thereby improving the accuracy of clustering; when the traditional Ward's method merges clusters, it only depends on the number of samples and distances, which easily leads to premature merging of some semantically unrelated clusters. The improved Ward's method of the present invention can more reasonably merge clusters by dynamically adjusting the cluster merging strategy and avoid premature merging of semantically unrelated clusters.

[0067] In summary, the present invention makes an innovative improvement to the traditional Ward's method. By introducing the distance adjustment term of the cluster center vector, the accuracy, robustness, and semantic interpretability of clustering are significantly improved. The improved method can better adapt to high-dimensional data, such as word embedding vectors, and shows superior performance in the hierarchical clustering of skill nodes.

[0068] S4. Based on the skill tree, construct and optimize a global model for predicting the user's cognitive load and emotional state.

[0069] Among them, S4 further includes the following steps: S41. Based on the skill tree, collect the user's behavioral data, physiological data, and text sentiment data in real time during the learning process.

[0070] In one embodiment, the user's behavioral data is collected in real time through the user learning platform, including clickstream data, learning progress data, and interaction behavior data. The clickstream data includes the number of clicks, dwell time, and page jump data. The learning progress data includes the number of completed tasks and learning duration data. The interaction behavior data includes the answering accuracy rate and the number of repeated learning times. The log analysis tool is used to record the user's behavioral data.

[0071] Furthermore, the user's physiological data is collected in real time through wearable devices. The physiological data includes heart rate, skin conductance response, electroencephalogram, and eye movement data. The wearable devices include smart bracelets and electroencephalogram headsets. Sensor technology and data interfaces are used to transmit the physiological data to the data processing platform.

[0072] S42. Utilize the behavioral data, physiological data, and text sentiment data to construct mixed feature data.

[0073] In one embodiment, feature extraction is performed on the behavioral data, physiological data, and text sentiment data collected in step S41 respectively. Statistical features and temporal features are extracted from the behavioral data. Time domain features and frequency domain features are extracted from the physiological data. Sentiment scores and sentiment intensity features are extracted from the text sentiment data. The statistical features include mean, method, and maximum value. The temporal features include trend and periodicity. The time domain features include mean and standard deviation. The frequency domain features include power spectral density. The PCA feature engineering method is used to reduce the dimension and optimize the extracted features.

[0074] Furthermore, the behavioral features, physiological features, and text sentiment features are fused to construct mixed feature data. That is, features of different modalities are integrated into a unified feature dataset.

[0075] S43. Utilize the mixed feature data to construct and optimize a global model for predicting the user's cognitive load and emotional state on the premise of protecting user privacy.

[0076] In one embodiment, model training is performed on the premise of protecting user privacy.

[0077] Specifically, learning nodes are deployed on multiple learning platforms. Each node holds the mixed feature data of local users. That is, each user's device locally stores multiple feature data and trains a model locally to obtain a local model. Furthermore, a global model for predicting the user's cognitive load and emotional state is initialized on the central server.

[0078] Further, after the local model is updated, the model parameters are uploaded to the central server.

[0079] Further, the central server aggregates the model parameters of all users and updates the global model.

[0080] Further, the updated global model is sent to each user device for the next round of training.

[0081] Further, based on the deep learning algorithm of artificial intelligence, the global model is trained using hybrid feature data for predicting the cognitive load and emotional state of users. Among them, the prediction of the cognitive load refers to predicting the learning burden of users through behavioral data and physiological data, and the prediction of the emotional state refers to predicting the emotional state of users through text emotion data and physiological data; It should be noted that in the above process, by adopting the homomorphic encryption technology, the privacy and security of user data during transmission and aggregation are ensured. In addition, based on the prediction results of the global model, the cognitive load and emotional state of users can be fed back in real time. According to the cognitive load and emotional state of users, the learning path and content recommendation can be dynamically adjusted to provide a personalized learning experience.

[0082] S5. Improve the user's soft skill mastery and training experience based on the global model.

[0083] In one embodiment, first, the optimized global model is used to predict the cognitive load and emotional state of users, and the prediction results of the cognitive load and emotional state are obtained. Among them, the cognitive load refers to the quantitative value of the burden degree of users during the learning process, the burden degree includes high load, medium load and low load, the emotional state refers to the quantitative value of the emotional tendency of users, and the emotional tendency includes positive, neutral and negative.

[0084] Further, based on the prediction results, an objective function for balancing the path length and cognitive load is constructed, and the objective function is as follows:

[0085] Among them, is the learning path, is the skill node and is the distance between them, is the cognitive load of the path , which comes from the prediction results of the global model, , are weight coefficients used to balance the importance of path length and cognitive load.

[0086] Furthermore, a personalized learning path is dynamically generated through the objective function.

[0087] Specifically, first, based on the user's current skill level, the initial node and the target node are determined.

[0088] Furthermore, the above objective function is used to search for the optimal learning path.

[0089] Furthermore, during the search process, the weights of the path length and the cognitive load are dynamically adjusted to ensure that the generated path can not only meet the learning efficiency but also adapt to the user's cognitive load and emotional state.

[0090] Furthermore, the generated path is evaluated, the objective function value is calculated, and the optimal path is selected according to the evaluation results.

[0091] Furthermore, according to the changes in the user's cognitive load and emotional state, the path is fine-tuned to ensure the real-time nature and adaptability of the path. Regarding the fine-tuning of the path, by adjusting the path weights in real time, it is ensured that the learning path can adapt to the user's cognitive load and emotional state while meeting the efficiency, and the goal is to minimize the following dynamically adjusted objective function: , where, is the path weight vector at time is the path weight vector at time is the learning rate at time for controlling the step size of weight update, is the and the soft skill nodes the distance between, , are weight coefficients for balancing the importance of the path length and the cognitive load, is the cognitive load of the path at time derived from the prediction results of the global model, is the path the number of soft skill nodes in. Among them, at each moment , according to the real-time feedback of the changes in the user's cognitive load and emotional state, the path weight vector is updated. The parameter update expression related to the path weight vector is as follows:

[0092] where, is the gradient of the objective function at , is the dynamically adjusted learning rate, and its calculation formula is:

[0093] where, is the initial learning rate, is the time step; the gradient of the objective function is calculated as follows:

[0094] where, is the partial derivative of the path length with respect to the weight, is the partial derivative of the cognitive load with respect to the weight; the corrected expression of the partial derivative of the path length with respect to the weight is as follows:

[0095] where, and are respectively the norm of the weight vector at time and the norm of the weight vector at time

[0096] It should be noted that the above calculations of partial derivatives are all based on the normalization of the weight vector to ensure the directionality and stability of gradient calculation.

[0097] The specific algorithm flow for the above parameter update is as follows: Set the initial weight vector and the initial learning rate ; For each moment , calculate the gradient of the objective function; Update the learning rate ; Update the weight vector ; According to the updated weight vector, combined with the termination condition, generate a new learning path ; The termination condition includes: when the change in the weight vector is less than the preset threshold or the maximum number of iterations is reached, stop the iteration.

[0098] It should be noted that the present invention ensures that users can efficiently master skills during the learning process without generating negative emotions due to excessive cognitive load. Traditional gradient calculation methods usually only consider a single objective, while the present invention significantly improves the accuracy and dynamics of gradient calculation by optimizing the path length and cognitive load. Existing technologies usually only optimize a single objective, while the present invention significantly improves the rationality and personalization of the learning path by jointly optimizing the path length and cognitive load. Existing technologies usually use fixed weight coefficients and cannot adapt to the real-time state changes of users, while the present invention realizes the real-time optimization and personalized recommendation of the learning path by dynamically adjusting the weight coefficients.

[0099] Therefore, the present invention realizes the personalized generation of the learning path through precise gradient calculation, joint modeling of objective optimization, and real-time and high-efficiency dynamic weight adjustment. Compared with existing technologies, the present invention has significant innovation and breakthroughs in objective optimization, dynamic adjustment, and real-time performance, and can provide users with a more accurate, efficient, and personalized learning experience.

[0100] Furthermore, by continuously collecting the real-time feedback data of users and continuously optimizing the learning path, a closed-loop system is formed, and the real-time feedback data includes the mixed feature data mentioned in step S4.

[0101] Furthermore, based on the continuously updated learning path, the mastery of users' soft skills and training experience are continuously improved.

[0102] Please refer to Figure 2 , Figure 2 which is a schematic structural diagram of the personalized soft skill training system based on artificial intelligence in the embodiment of the present invention. The system includes an input device, a processor, an output device, and a memory. The input device, the processor, the output device, and the memory are interconnected. Among them, the memory is used to store computer programs, and the computer programs include program instructions. The processor is configured to call the program instructions, and the system uses the described personalized soft skill training method based on artificial intelligence. It should be noted that the system focuses on breaking through the problem of automatic grading of subjective questions and provides a detailed feedback analysis report to provide users with accurate personalized growth suggestions and achieve true individualized teaching.

[0103] In this embodiment, the input device is used to collect user data, including a questionnaire collection module and a data collection module; Specifically, the questionnaire collection module includes a user interface and a data transmission interface, which collects the basic data of users' soft skills, psychological characteristics, and behavior pattern data through a multi-dimensional questionnaire, and transmits the data to the processor; the data collection module includes a camera, a microphone, integrated sensors of wearable devices, and a data preprocessing unit, which is used to collect the behavior data, physiological data, and text sentiment data of users during the learning process in real time, and transmit the data to the processor.

[0104] Further, the processor is the core computing unit of the system, which is used for data processing and model optimization, including a user portrait generation module, a skill tree construction module, a global model optimization module, and a path generation and optimization module.

[0105] Specifically, the user portrait generation module integrates a large language model and a dynamic weight adjustment algorithm, generates an initial user portrait based on the questionnaire data and mixed feature data, and dynamically updates the user portrait through the dynamic weight adjustment algorithm; the skill tree construction module integrates word embedding technology and an improved Ward's method, calculates the semantic similarity between skill nodes based on the user portrait using word embedding technology, and constructs a hierarchical skill tree through the improved Ward's method; the global model optimization module uses data fusion technology to optimize the global model, predicts the user's cognitive load and emotional state, and updates the model parameters on the premise of protecting user privacy; the path generation and optimization module generates an initial learning path based on the skill tree and the global model, and makes real-time fine-tuning of the path to dynamically generate a personalized learning path.

[0106] Further, the output device is used to display the training results and feedback to the user, including a learning path display module and an intelligent feedback module; Specifically, the learning path display module includes a graphical user interface and a visualization engine, displays the generated personalized learning path to the user in a graphical manner, and provides path adjustment suggestions; the intelligent feedback module integrates natural language generation technology and an emotion analysis engine, generates personalized intelligent feedback, such as improvement suggestions and emotional support, according to the user's answers to the practice questions and the prediction results of the global model, and outputs them in text or voice form.

[0107] Further, the memory is used to store the data and programs required for the operation of the system, including a user data storage module, a model parameter storage module, and a program instruction storage module; Specifically, the user data storage module includes a database management system, which is used to store user portraits, skill trees, learning paths, and mixed feature data; the model parameter storage module includes a distributed file system, which is used to store the parameters of the global model and the intermediate data during the optimization process; the program instruction storage module includes a non-volatile memory, which is used to store computer programs and their instructions for the processor to call.

[0108] Regarding the said system, the following is a specific implementation case: Please refer to Figure 3 and Figure 4 , Figure 3 , which is the operation flowchart of the artificial intelligence large language model according to the embodiment of the present invention; Figure 4 , which is the flowchart for building the skill tree according to the embodiment of the present invention. First, the system issued an initial questionnaire to the user. The initial questionnaire includes 2 main directions: loving oneself and loving others, and 8 dimensions: positive situations of loving oneself, negative situations of loving oneself, personal growth and goals, boundary setting and self - protection, positive situations of loving others, negative situations of loving others, actions of loving others, respect and inclusion. It should be noted that the dimension of "negative situations of loving oneself" includes such a question: "When you make mistakes or fail to meet your own expectations, how do you usually treat yourself?" The user's answer is: "I will feel deeply guilty, think that my performance is not perfect enough, and need to constantly improve." Furthermore, for the user's answer, the artificial intelligence gave a full score to this question. The scoring reasons are as follows: "Positive reflection: The user can realize their own mistakes or failure to meet expectations and show a desire for improvement, which reflects a positive attitude. High self - requirements: The user sets high standards for themselves, which will help promote personal development. Artificial intelligence improvement suggestions: Although self - reflection is crucial, excessive self - blame may have adverse effects. It is recommended that the user, while reflecting, also learn to be tolerant of their own imperfections, focus on how to learn lessons from mistakes and achieve growth. Be more tolerant of yourself and give yourself understanding and support as you would to a friend." Furthermore, according to the dimensions of each question, intelligent scoring is carried out and the scores are accumulated.

[0109] Furthermore, the user is asked: "Which soft skills do you hope to focus on cultivating? Can you share the specific reasons for choosing this skill? Can you describe a difficult scenario related to this skill that you encountered in actual work or life?" The user answered this question.

[0110] Furthermore, the large language model is used to perform semantic analysis on the answer content and extract the keywords of the target soft skills. For example: listening ability, teamwork.

[0111] Furthermore, combining the above - mentioned 8 soft - skill dimensions and the soft skills answered by the user above, an initial node set of the skill tree is formed. Then, using the word embedding technology of the neural network pre - training model, semantic similarity scoring is carried out on the nodes of the skill tree to form a similarity matrix, and using the similarity matrix, a hierarchical clustering algorithm is used to construct a personalized soft - skill tree.

[0112] Further, based on the soft skill tree, identify the dominant basic soft skills, i.e., the basic soft skills with the highest scores, and recommend the optimal development path.

[0113] Further, combine the questionnaire scores and the skill tree to form a user profile, which is stored in long-term memory as an important reference for subsequent AI feedback. Specifically, the evaluation results, dominant soft skills, and target soft skills of the questionnaire will be added to the prompt words to guide the subsequent AI interaction process.

[0114] Further, the user selects "empathy" from several soft skill modules as the object of learning and practice.

[0115] Further, the user studies and introduces articles on empathy.

[0116] Further, the user answers a series of practice questions on "empathy". One of the practice questions is: "During a team meeting, member A proposed a new project plan, elaborating on their ideas and plans in detail. However, member B started refuting before A could finish, listing various reasons why it was infeasible, without considering A's thinking and efforts at all. Please analyze which empathy elements are lacking in member B's behavior, the possible negative impacts of this behavior on team communication and project progress, and how to improve it." The user replied: "B lacks listening skills. This behavior is impolite. On the other hand, it will also make the team atmosphere rather bad, forming a situation where only one person has the say, which is not conducive to everyone pooling their wisdom. B should patiently wait for A to finish, summarize the key points of A's speech, and then state their own views." Further, based on the reinforcement learning algorithm, store the learning results of the practice questions in long-term memory.

[0117] Further, according to the user profile, the user has relatively high standards for themselves in the dimension of "loving oneself - self-acceptance - in the face of negative situations". On this basis, the AI will not only evaluate and provide feedback on the user's answers, but also combine the user's individual psychological characteristics to provide more considerate suggestions: "Regarding the relatively low score of the user in self-acceptance - in the face of negative situations, here are some suggestions: Self-affirmation: The user has demonstrated relatively high analytical abilities when answering questions. They can try to affirm their own advantages and efforts more often to improve their self-acceptance." The above AI-based personalized soft skill training system has the following effects: (1) Precise generation of user profiles. Based on advanced AI algorithms, the system can conduct multi-dimensional in-depth analysis of user questionnaire assessment data, accurately identify the user's basic psychological characteristics and behavior patterns, and store them in long-term memory; by continuously updating the user profile, it provides data support for the user's personalized practice, evaluation, and feedback.

[0118] (2) Personalized construction of the soft skill tree. First, through in-depth semantic analysis of the user's answers to the target soft skills by the large language model, key soft skill tags are extracted. These tags are combined with the basic soft skills already mastered in the user questionnaire to form the initial node set of the skill tree. Then, using the word embedding technology of the neural network pre-trained model, semantic similarity scores are given to the nodes of the skill tree to form a similarity matrix. Using this similarity matrix, a hierarchical clustering algorithm is applied to construct a personalized skill tree. This data-driven method not only reveals the internal connections between soft skills but also provides a clear ability progression path for users.

[0119] (3) Precise identification of advantageous skills and recommendation of the optimal development direction. The system will intelligently identify the user's advantageous soft skills and find the optimal development direction in the skill tree. By calculating the shortest path between skills, the most suitable next learning goal is recommended for the user to ensure the coherence and feasibility of the learning process. This method can effectively support the continuous growth of users and achieve the spiral ascent of soft skills.

[0120] (4) Diversified artificial intelligence practice evaluation and feedback system. Innovatively applying the large language model to semantically understand and intelligently analyze the user's answers, including objective questions, essay questions, and table questions. The system can not only provide accurate quantitative scores but also generate detailed evaluation reports containing dimensions of ability analysis and improvement suggestions, and combined with the user profile, provide all-round feedback guidance for users.

[0121] In summary, the present invention proposes a personalized soft skill training method based on artificial intelligence by introducing the new generation of artificial intelligence large language model technology, neural network word embedding algorithm, and graph theory algorithm. Through the closed-loop training mode of "questionnaire - learning - practice", this method realizes the personalization, systematization, and intelligence of soft skill training. First, using the multi-dimensional questionnaire combined with the large language model to generate the initial user profile and dynamically update the user profile to accurately depict the user's psychological characteristics, behavior patterns, and soft skill mastery. Second, based on the word embedding technology and the improved Ward's method, a hierarchical skill tree is constructed and a personalized learning path is dynamically generated to balance the path length and cognitive load. Finally, by constructing and optimizing the global model, real-time prediction of the user's cognitive load and emotional state is achieved, and the learning path is fine-tuned in real time, thereby realizing efficient soft skill training. The present invention breaks through the technical bottlenecks of traditional methods in personalized training, quantitative evaluation, and intelligent feedback for subjective question types, and significantly improves the effect of soft skill training and the user experience.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A personalized soft skill training method based on artificial intelligence, characterized in that, The method includes the following steps: Conduct an assessment using a questionnaire, and analyze the semantics of the user's answers in combination with a large language model to generate an initial user profile. The questionnaire includes objective questions, essay questions, and table questions. The initial user profile includes a psychological feature vector, a behavior pattern matrix, and a soft skill mastery distribution; Update the initial user profile and obtain an update result; According to the update result, use word embedding technology and an improved Ward's method to construct a hierarchical skill tree; Based on the skill tree, construct and optimize a global model for predicting the user's cognitive load and emotional state; According to the global model, improve the user's soft skill mastery and training experience.

2. The personalized soft skill training method based on artificial intelligence according to claim 1, wherein, The step of updating the initial user profile and obtaining an update result includes: Establish a weight adjustment model, which is used to dynamically adjust the weights in the user profile according to the user's real-time feedback and behavior data. The weight adjustment model satisfies the following expression: ， Among them, represents the weight of the th feature dimension at moment, represents the weight of the th feature dimension at moment, represents the learning rate, represents the loss function under the weight at represents the real-time feedback value of the th feature dimension at moment, represents the weight of the th feature dimension at moment, represents the weight of the th feature dimension at moment, represents the total number of feature dimensions; According to the weight adjustment model, dynamically update the initial user profile and obtain an update result.

3. The personalized soft skill training method based on artificial intelligence according to claim 1, characterized in that, The step of constructing a hierarchical skill tree by using word embedding technology and an improved Ward's method according to the update result includes: According to the update result, use a large language model to semantically analyze the user's target soft skills and extract high-order soft skill labels; According to the high-order soft skill labels, combine with the basic soft skill set to generate an initial soft skill node set; Based on the initial soft skill node set, combine with word embedding technology to establish a calculation model for the semantic similarity matrix between soft skill nodes; According to the calculation model, obtain a distance matrix; Based on the distance matrix, use the improved Ward's method for hierarchical clustering to obtain a hierarchical skill tree.

4. The personalized soft skill training method based on artificial intelligence according to claim 3, characterized in that, The calculation model for the semantic similarity matrix between soft skill nodes satisfies the following expression: ， Among them, is the semantic similarity matrix between the th soft skill node and the th soft skill node, is the word embedding vector of the th soft skill node, is the word embedding vector of the th soft skill node; the distance matrix satisfies the following relational expression: ， Among them, is the distance matrix between the th soft skill node and the th soft skill node; the distance update formula of the improved Watts method is as follows: ， Among them, represents the distance between new clusters, represents the cluster the number of soft skill nodes included in it, represents the cluster the number of soft skill nodes included in it, represents the total number of samples, represents the cluster and the cluster the distance between them, represents the cluster the central vector of, represents the cluster the central vector of.

5. A personalized soft skill training method based on artificial intelligence according to claim 1, characterized in that The step of constructing and optimizing a global model for predicting the user's cognitive load and emotional state based on the skill tree includes: Based on the skill tree, collect the user's behavior data, physiological data, and text emotion data in real time during the learning process; Use the behavior data, the physiological data, and the text emotion data to construct mixed feature data; Use the mixed feature data to construct and optimize a global model for predicting the user's cognitive load and emotional state on the premise of protecting the user's privacy.

6. The personalized soft skill training method based on artificial intelligence according to claim 1, wherein The step of improving the user's soft skill mastery and training experience according to the global model includes: According to the global model, obtain the prediction results of the user's cognitive load and emotional state; Based on the prediction results, construct an objective function for balancing the path length and cognitive load. The objective function is as follows: ， Among them, is the learning path, is the soft skill node and is the distance between them, is the cognitive load of the path and 、 are the weight coefficients; Through the objective function, dynamically generate a personalized learning path; Through the personalized learning path, improve the user's soft skill mastery and training experience.

7. The personalized soft skill training method based on artificial intelligence according to claim 6, characterized in that, The step of dynamically generating a personalized learning path through the objective function includes: Through the objective function, dynamically adjust the weights of the path length and cognitive load to optimize the generated path; According to the changes in the user's cognitive load and emotional state, the generation path is fine-tuned in real time to dynamically generate a personalized learning path that can not only meet the learning efficiency but also adapt to the user's cognitive load and emotional state.

8. The personalized soft skill training method based on artificial intelligence according to claim 7, characterized in that The fine-tuning in real time of the generation path according to the changes in the user's cognitive load and emotional state to dynamically generate a personalized learning path includes: Determining a fine-tuning path objective function according to the changes in the user's cognitive load and emotional state, and the fine-tuning path objective function is as follows: ， Among them, is the path weight vector at a moment, representing the weight distribution of each soft skill node in the current learning path. is the path weight vector at a moment. is the learning rate at a moment, used to control the step size of weight update. is the th soft skill node and the th soft skill node the distance between them. , are weight coefficients. is the cognitive load of the path at a moment. is the number of soft skill nodes in the path; Based on the fine-tuning path objective function, fine-tuning the generation path in real time and obtaining a fine-tuning result; Dynamically generating a personalized learning path through the fine-tuning result.

9. The personalized soft skill training method based on artificial intelligence according to claim 8, wherein The fine-tuning in real time of the generation path based on the fine-tuning path objective function and obtaining a fine-tuning result includes: Setting an initial path weight vector and an initial learning rate based on the fine-tuning path objective function; Calculating the gradient of the fine-tuning path objective function according to the initial path weight vector and the initial learning rate; Updating the learning rate using the gradient; Updating the path weight vector according to the learning rate; Fine-tuning the generation path in real time and obtaining a fine-tuning result through the path weight vector.

10. A personalized soft skill training system based on artificial intelligence, the system uses a personalized soft skill training method based on artificial intelligence according to any one of claims 1 to 9, characterized in that, The system includes an input device, a processor, an output device, and a memory. The input device, the processor, the output device, and the memory are interconnected. Among them, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions.

Images