Occupational scene simulation system and method for promoting employment and entrepreneurship

Abstract

The invention relates to the technical field of informatization systems, in particular to an occupational scene simulation system and method for promoting employment and entrepreneurship, and the system comprises a scene generation module, a multi-modal interaction module, a multi-modal fusion module, a dynamic evaluation module, a dynamic evaluation module, a dynamic evaluation module and a dynamic evaluation module, the system comprises a dynamic evaluation module which is in communication connection with the multi-modal fusion module, a decision optimization module which is in communication connection with the dynamic evaluation module, a path planning module which is in communication connection with the decision optimization module, and a visualization module which is in communication connection with the path planning module and receives the optimal occupational development path. Based on a three-dimensional situation projection technology, generating a visual representation of a user capability development track; and sending the visual representation to the multi-modal interaction module for presentation in a virtual reality environment. The problems existing in an existing occupational planning and ability evaluation system are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of information systems, and in particular to a career scenario simulation system and method for promoting employment and entrepreneurship. Background Art

[0002] With the increasing complexity of the job market and the diversification of career development paths, traditional career planning and ability assessment methods have been unable to meet the needs of modern society. In recent years, the application of virtual reality (VR) technology and artificial intelligence (AI) algorithms in the fields of vocational training and talent assessment has gradually increased, but the existing systems still have many limitations.

[0003] At present, there are some career simulation systems based on VR technology on the market. These systems usually provide preset work scenarios, allowing users to experience specific work content in a virtual environment. However, such systems often lack personalization and dynamic adaptability, and cannot adjust the simulation content according to the user's actual performance and ability level. In addition, they usually only focus on single-dimensional skill assessment, such as operational skills or knowledge reserves, while ignoring the assessment of comprehensive capabilities and long-term career development planning.

[0004] On the other hand, AI-based career planning systems are also beginning to emerge. These systems recommend potentially suitable career directions for users by analyzing their resumes, assessment results, and career interests. However, such systems often rely on static data and simple matching algorithms, making it difficult to capture the dynamic changes in user capabilities and potential development trends. At the same time, due to the lack of simulation of actual work scenarios, it is difficult for users to truly experience the work content and challenges of different professions, resulting in a large deviation between the recommended results and the actual situation.

[0005] There is also a common problem with existing technologies, namely the limitations of data acquisition and analysis. Most systems rely only on users’ self-reports or simple online tests, which cannot comprehensively and objectively assess users’ actual abilities. This assessment method is not only easily affected by subjective factors, but also difficult to capture users’ performance and potential in actual work environments.

[0006] In addition, existing career planning systems usually lack long-term tracking and dynamic adjustment mechanisms. They often provide one-time advice or planning and cannot be continuously optimized according to changes in user capabilities and changes in the external environment. This makes it difficult for users to obtain continuous and effective career development guidance, especially when facing a rapidly changing job market and emerging professions.

[0007] Finally, existing systems also have shortcomings in data presentation and user interaction. Most systems use traditional two-dimensional charts or text reports to present evaluation results and suggestions. This approach is not only lacking in intuitiveness, but also difficult to stimulate user enthusiasm and willingness to continue using.

[0008] In view of the above problems, there is an urgent need for a career scenario simulation system that can provide immersive experience, dynamic assessment, personalized planning and intuitive visualization to better meet the needs of modern society for career development and talent training. Summary of the invention

[0009] The present invention aims to solve the problems of insufficient personalization, lack of dynamic adaptability, limited data acquisition, lack of long-term planning, and poor user experience in existing career planning and ability assessment systems. By organically combining virtual reality technology, multimodal data analysis, reinforcement learning algorithms, and dynamic assessment models, the present invention provides a new career scenario simulation system and method for promoting employment and entrepreneurship.

[0010] The present invention proposes a career scenario simulation system for promoting employment and entrepreneurship, comprising:

[0011] Scenario generation module for:

[0012] Generate virtual career scenarios based on the user's initial ability vector and career goals;

[0013] Sending the virtual career scenario to a multimodal interaction module;

[0014] The multimodal interaction module is communicatively connected with the scenario generation module and is used to:

[0015] receiving the virtual career scenario and presenting it in a virtual reality environment;

[0016] Collecting behavioral data of the user in the virtual reality environment, wherein the behavioral data includes visual data, voice data, and biosensor data;

[0017] Sending the behavior data to a multimodal fusion module;

[0018] The multimodal fusion module is communicatively connected with the multimodal interaction module and is used to:

[0019] receiving the behavior data;

[0020] Based on a spatiotemporal synchronized cross-modal attention mechanism, the behavior data is fused to obtain fusion features;

[0021] Sending the fused features to a dynamic evaluation module;

[0022] A dynamic evaluation module is communicatively connected with the multimodal fusion module and is used to:

[0023] receiving the fused feature;

[0024] Based on the nonlinear capability evolution model, the user's professional capabilities are dynamically evaluated to obtain an updated capability vector;

[0025] Sending the updated capability vector to a decision optimization module;

[0026] A decision optimization module is connected in communication with the dynamic evaluation module and is used to:

[0027] receiving the updated capability vector;

[0028] Generate career development decisions based on reinforcement learning algorithms;

[0029] sending the career development decision to a path planning module;

[0030] A path planning module is in communication with the decision optimization module and is used to:

[0031] receiving said career development decision;

[0032] Build a dynamic career development graph and generate the optimal career development path based on the Monte Carlo tree search algorithm;

[0033] Sending the optimal career development path to a visualization module;

[0034] A visualization module is in communication with the path planning module and is used to:

[0035] Receive the optimal career development path;

[0036] Generate a visual representation of the user's capability development trajectory based on three-dimensional situation projection technology;

[0037] The visual representation is sent to the multimodal interaction module for presentation in a virtual reality environment.

[0038] Preferably, the scenario generation module comprises:

[0039] A conditional generative adversarial network unit, which is used to generate virtual career scenarios based on the user's current ability vector and career goals;

[0040] A career constraint unit, connected to the conditional generative adversarial network unit, is used to impose career-related constraints on the generated virtual career scenarios to ensure the relevance of the generated scenarios to the user's abilities and goals;

[0041] The scenario library unit is connected to the conditional generative adversarial network unit and the occupational constraint unit, and is used to store and manage the generated virtual occupational scenarios.

[0042] Preferably, the multimodal interaction module comprises:

[0043] A virtual reality rendering unit, used for presenting the virtual career scenario on a virtual reality head display device;

[0044] The multimodal data collection unit is used to collect the user's behavior data in the virtual reality environment, including:

[0045] A visual collection subunit, used to collect the user's visual behavior data;

[0046] A voice collection subunit, used to collect the user's voice data;

[0047] A biosensor collection subunit, used to collect physiological data of the user;

[0048] The data preprocessing unit is connected to the multimodal data acquisition unit and is used to preprocess the collected behavior data, including data cleaning, format conversion and preliminary feature extraction.

[0049] Preferably, the multimodal fusion module comprises:

[0050] A feature extraction unit for extracting modality-specific features from the preprocessed behavioral data;

[0051] an attention calculation unit, connected to the feature extraction unit, for calculating cross-modal attention weights;

[0052] A feature fusion unit is connected to the attention calculation unit and is used to perform weighted fusion on features of different modalities based on the attention weight to obtain a fused feature.

[0053] Preferably, the dynamic evaluation module comprises:

[0054] A Bayesian inference unit, used to make probabilistic inferences about the user's current capabilities based on fused features;

[0055] The nonlinear evolution unit is connected to the Bayesian inference unit and is used to simulate the dynamic change process of the user's ability, including:

[0056] A differential equation solving subunit, used to solve the nonlinear differential equations describing the evolution of capabilities;

[0057] A parameter prediction subunit, used to dynamically predict the parameters of the capability evolution equation;

[0058] The capability updating unit is connected to the Bayesian inference unit and the nonlinear evolution unit, and is used to integrate the probability inference result and the nonlinear evolution result to generate an updated capability vector.

[0059] Preferably, the decision optimization module comprises:

[0060] A state encoding unit, used for encoding the user capability vector, the economic environment parameter and the industry feature vector into a state representation;

[0061] An action generation unit, connected to the state encoding unit, for generating possible career decision actions based on the current state;

[0062] a value evaluation unit, connected to the state encoding unit and the action generating unit, for evaluating the value of a state-action pair;

[0063] The strategy optimization unit is connected to the value evaluation unit and is used to optimize the decision-making strategy based on the TD3 algorithm, including a dynamic weight factor for adaptively adjusting the weights of rewards of different dimensions.

[0064] Preferably, the path planning module comprises:

[0065] A graph construction unit, used to construct a dynamic career development graph based on career development decisions and user capability vectors;

[0066] an edge weight calculation unit, connected to the graph construction unit, for calculating the weights of the edges in the graph, wherein the weights are based on the similarity between the user's current capabilities and the requirements of the target position;

[0067] a path search unit, connected to the graph construction unit and the edge weight calculation unit, for searching for an optimal path in the dynamic career development graph based on a Monte Carlo tree search algorithm;

[0068] The path optimization unit is connected to the path search unit and is used to perform smoothing and local optimization on the searched path.

[0069] Preferably, the visualization module comprises:

[0070] A data dimension reduction unit, used to project the high-dimensional capability vector into a three-dimensional space using a principal component analysis method;

[0071] A trajectory generation unit, connected to the data dimension reduction unit, for generating a continuous capability development trajectory based on the projected data points;

[0072] A dynamic updating unit, connected to the trajectory generating unit, for updating the visual trajectory in real time, including calculation of velocity and acceleration;

[0073] The interactive design unit is used to design how users interact with the visual interface, including zooming, rotating, and timeline control functions.

[0074] As a preferred feature, it also includes:

[0075] The edge computing module is deployed on the virtual reality headset to perform real-time rendering, data collection and preliminary processing tasks;

[0076] A cloud computing module, connected to the edge computing module via a network, for executing complex artificial intelligence algorithms and large-scale data processing tasks;

[0077] The task scheduling module is connected to the edge computing module and the cloud computing module, and is used to dynamically allocate computing tasks to the edge or the cloud according to task complexity, real-time requirements and network conditions.

[0078] The career scenario simulation method for promoting employment and entrepreneurship includes the following steps:

[0079] S1, generating virtual career scenarios based on the user's initial ability vector and career goals;

[0080] S2, presenting the virtual career scenario in a virtual reality environment, and collecting multimodal behavior data of the user in the virtual reality environment;

[0081] S3, based on the spatiotemporal synchronized cross-modal attention mechanism, fusing the multimodal behavior data to obtain fusion features;

[0082] S4, based on the nonlinear capability evolution model, using the fusion features to dynamically evaluate the user's professional capabilities to obtain an updated capability vector;

[0083] S5. generating a career development decision based on the reinforcement learning algorithm and the updated capability vector;

[0084] S6. Build a dynamic career development graph and generate the optimal career development path based on the Monte Carlo tree search algorithm;

[0085] S7. Generate a visual representation of the user's ability development trajectory based on three-dimensional situation projection technology and present it in a virtual reality environment;

[0086] S8. Repeat steps S1 to S7 until the user reaches a preset career development goal or the simulation time ends.

[0087] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0088] First, the present invention realizes highly personalized and dynamically adaptable career scenario simulation. Through the collaborative work of the scenario generation module and the multimodal interaction module, the system can dynamically generate virtual career scenarios that meet personal characteristics based on the user's initial abilities and career goals. This approach not only provides a more realistic and relevant experience, but also can be continuously adjusted as the user's abilities change, ensuring that the simulation content always remains challenging and targeted.

[0089] Secondly, the present invention breaks through the limitations of traditional evaluation methods and realizes a comprehensive and objective ability evaluation. The innovative design of the multimodal fusion module and the dynamic evaluation module enables the system to simultaneously analyze the user's visual, voice and physiological data, thereby fully capturing the user's performance in the virtual work environment. This multi-dimensional data collection and analysis method greatly improves the accuracy and objectivity of the evaluation and provides users with a more realistic and reliable ability portrait.

[0090] Furthermore, the present invention realizes long-term and dynamic career development planning through the innovative algorithms of the decision optimization module and the path planning module. The decision optimization algorithm based on reinforcement learning can consider the user's ability status, career goals and external environmental factors to generate the optimal career development strategy. At the same time, the construction of a dynamic career development graph and the application of the Monte Carlo tree search algorithm enable the system to plan a feasible and personalized long-term development path for users. This method not only helps users to set clear career goals, but also can make dynamic adjustments according to actual progress and external changes, providing users with continuous and effective career guidance.

[0091] In addition, the visualization module of the present invention greatly improves the user experience through three-dimensional situation projection technology and interactive design. The intuitive visualization of the ability development trajectory enables users to clearly understand their ability changes and development direction. The immersive experience combined with virtual reality technology not only enhances the user's sense of participation, but also improves the system's stickiness, which is conducive to users' long-term tracking and improvement of their own career development.

[0092] Finally, the present invention adopts an edge-cloud computing collaborative architecture to effectively balance system performance and user experience. The edge computing module is responsible for real-time rendering and data acquisition, ensuring the smooth response of the virtual reality environment; while the cloud computing module undertakes complex AI algorithms and large-scale data processing tasks, ensuring the powerful analysis capabilities of the system. This architectural design not only improves the overall performance of the system, but also provides flexibility for future expansion and upgrades.

[0093] In summary, the present invention effectively solves many problems existing in the existing career planning and ability assessment systems through innovative technical solutions. It not only provides a comprehensive, dynamic and personalized career development tool for job seekers and employees, but also provides advanced talent assessment and training methods for enterprises and educational institutions. This innovative solution is expected to play an important role in promoting employment, promoting entrepreneurship and optimizing the talent market, and has a positive impact on career development and talent training in modern society. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is the main flow chart of the system of the present invention.

[0095] Figure 2 This is an internal structure diagram of the scenario generation module of the present invention.

[0096] Figure 3 This is a diagram of the internal structure of the multimodal interaction module of the present invention.

[0097] Figure 4 This is a diagram of the internal structure of the multimodal fusion module of the present invention.

[0098] Figure 5 It is the internal structure diagram of the multi-dynamic evaluation module of the present invention.

[0099] Figure 6 It is the internal structure diagram of the decision optimization module of the present invention.

[0100] Figure 7 It is the internal structure diagram of the path planning module of the present invention.

[0101] Figure 8 It is the internal structure diagram of the visualization module of the present invention. DETAILED DESCRIPTION

[0102] Please refer to Figure 1-8 The present invention provides a career scenario simulation system and method for promoting employment and entrepreneurship, aiming to provide users with personalized career development planning and ability training programs through virtual reality technology and artificial intelligence algorithms. The specific implementation methods of the present invention will be described in detail below.

[0103] A career scenario simulation system for promoting employment and entrepreneurship includes a scenario generation module 1, a multimodal interaction module 2, a multimodal fusion module 3, a dynamic evaluation module 4, a decision optimization module 5, a path planning module 6 and a visualization module 7. These modules work together through a specific communication connection method to jointly achieve the technical objectives of the present invention.

[0104] The scenario generation module 1 is used to generate a virtual career scenario based on the user's initial ability vector and career goals. Specifically, the scenario generation module 1 can use the conditional generative adversarial network (cGAN) technology to generate a realistic career scenario based on the user's ability characteristics and desired career direction. For example, for a computer science graduate who intends to engage in artificial intelligence research and development, the system may generate a scenario for a simulated interview for an AI engineer position in a technology company. The generated virtual career scenario will be sent to the multimodal interaction module 2 for presentation in a virtual reality environment.

[0105] The multimodal interaction module 2 is communicatively connected to the scenario generation module 1, and is used to receive virtual occupation scenarios and present them in a virtual reality environment. The present invention preferably uses high-performance VR devices, such as Oculus Quest 3, to provide an immersive experience. The multimodal interaction module 2 is also responsible for collecting the user's behavioral data in the virtual reality environment, including visual data, voice data, and biosensor data. For example, the system can capture the user's eye movements and facial expressions through the built-in camera of the VR headset, collect voice through a microphone, and collect physiological indicators such as heart rate and skin conductance through worn biosensors. These multimodal data provide a comprehensive basis for subsequent ability assessment. The collected behavioral data will be sent to the multimodal fusion module 3 for processing.

[0106] The multimodal fusion module 3 is connected to the multimodal interaction module 2 for receiving the behavior data and performing fusion processing. The present invention innovatively adopts a spatiotemporal synchronized cross-modal attention mechanism to fully utilize the correlation between different modal data. The specific fusion algorithm can be expressed as:

[0107]

[0108] Among them, F t is the fused feature, w i is the weight of each mode, Trans i is a modality-specific transformation network, X i is the original feature of each mode. Weight w i The calculation method is:

[0109]

[0110] Among them, MLP is a multi-layer perceptron, Pool is a pooling operation, and h t-1 is the hidden state of LSTM, which is used to capture timing information.

[0111] This fusion method can adaptively adjust the importance of different modal data. For example, in an interview scenario, voice and facial expressions may be given higher weights, while in a programming test scenario, operational behavior data may be more important. The fused features will be sent to the dynamic evaluation module 4.

[0112] The dynamic evaluation module 4 is connected to the multimodal fusion module 3 for receiving the fusion features and dynamically evaluating the user's professional ability. The present invention adopts a nonlinear ability evolution model based on the Bayesian probability graph model, which can be expressed as:

[0113]

[0114] Among them, c k represents the kth ability, O1:t Represents the observation sequence from time 1 to time t. In order to capture the dynamic change process of capability, the present invention introduces the capability evolution factor:

[0115]

[0116] Among them, μ m represents the mean of the mth observation, η is the learning rate, and J is the ability-observation matching function. This dynamic evaluation method can reflect the changes in user ability in real time. For example, if the user shows continuous improvement in communication ability in multiple simulated interviews, the system will increase the score of his communication ability accordingly. The updated ability vector obtained by the evaluation will be sent to the decision optimization module 5.

[0117] In this way, the system of the present invention realizes a closed loop of scenario generation, multimodal interaction and dynamic capability assessment, laying the foundation for subsequent decision optimization and path planning. By continuously iterating this process, the system can provide continuously optimized career development suggestions, helping users accumulate experience in a virtual environment, improve their capabilities, and ultimately achieve their career goals.

[0118] In a preferred embodiment of the present invention, the scenario generation module 1 includes a conditional generative adversarial network unit 11, a career constraint unit 12 and a scenario library unit 13. The conditional generative adversarial network unit 11 is used to generate a virtual career scenario based on the user's current ability vector and career goal. Specifically, WGA N-GP (Wasserstein GAN with GradientPenalty) can be used as the basic architecture, and its loss function can be expressed as:

[0119]

[0120] Among them, D is the discriminator, To generate the distribution, is the real distribution, and λ is the gradient penalty coefficient, which is usually set to 10. The occupation constraint unit 12 is connected to the conditional generative adversarial network unit 11 to impose occupation-related constraints on the generated virtual occupation scenarios to ensure the relevance of the generated scenarios to the user's abilities and goals. This can be achieved by adding additional constraints to the generator's loss function:

[0121]

[0122] in, For career-related losses, is the capability matching loss, α and β are trade-off coefficients, which can be adjusted according to the specific application scenario and can generally be initially set to 0.5 and 0.3.

[0123] The scenario library unit 13 is connected to the conditional generative adversarial network unit 11 and the occupational constraint unit 12, and is used to store and manage the generated virtual occupational scenarios. The present invention uses a distributed storage system (such as Ceph architecture) to manage a large amount of scenario data to ensure the stability and scalability of the system under high concurrency.

[0124] In another preferred embodiment of the present invention, the multimodal interaction module 2 includes a virtual reality rendering unit 21, a multimodal data acquisition unit 22 and a data preprocessing unit 23. The virtual reality rendering unit 21 is used to present a virtual career scenario on a virtual reality head display device. The present invention uses Unity 2022.3LTS as a rendering engine, combined with the high-resolution display (4K resolution per eye) and 120Hz refresh rate of Oculus Quest 3, to provide an ultimate immersive and low-latency experience.

[0125] The multimodal data acquisition unit 22 includes a visual acquisition subunit 221, a voice acquisition subunit 222, and a biosensor acquisition subunit 223. The visual acquisition subunit 221 uses the built-in binocular camera of the VR headset to collect the user's eye movement trajectory and facial expression data, with a sampling rate of up to 90Hz. The voice acquisition subunit 222 uses the built-in microphone array of the headset to collect 16kHz, 24bit high-quality audio. The biosensor acquisition subunit 223 collects physiological data such as heart rate, skin conductance, and body temperature through an external biosensor (such as the Empatica E4 wristband), with a sampling frequency of 64Hz.

[0126] The data preprocessing unit 23 is connected to the multimodal data acquisition unit 22 and is used to preprocess the collected behavior data. The preprocessing steps include:

[0127] 1) Data cleaning: remove outliers and noise, such as removing outliers exceeding 200 bpm in heart rate data.

[0128] 2) Format conversion: unify data from different sources into a standard format, such as converting audio data into MFCC features.

[0129] 3) Preliminary feature extraction: Calculate basic statistical features, such as the distribution of eye gaze points and changes in the fundamental frequency of speech.

[0130] Through the collection and preprocessing of this multimodal data, the system can fully capture the user's behavior in the virtual career scene, providing rich data support for subsequent ability assessment. For example, in a simulated interview scenario, the system can simultaneously analyze the user's voice content, tone changes, facial expressions and physiological reactions, thereby comprehensively evaluating their communication skills and stress resistance.

[0131] The present invention realizes highly intelligent and personalized career scenario simulation through the collaborative work of the above modules. The system can dynamically generate suitable virtual scenes according to the user's ability level and career goals, and provide an immersive experience through multimodal interaction. At the same time, the system can also collect and analyze the user's behavioral data in real time to provide a scientific basis for subsequent ability assessment and career planning. This innovative technical solution can not only help job seekers better understand their strengths and weaknesses, but also provide companies with more objective and comprehensive talent assessment methods, thereby promoting the healthy development of the employment market.

[0132] In a preferred embodiment of the present invention, the multimodal fusion module 3 includes a feature extraction unit 31, an attention calculation unit 32 and a feature fusion unit 33. These units work together to achieve efficient fusion processing of multimodal data.

[0133] The feature extraction unit 31 is used to extract modality-specific features from the preprocessed behavioral data. For visual data, the present invention uses a pre-trained ResNet-50 model to extract spatial features; for speech data, 1D-CNN combined with a self-attention mechanism is used to extract temporal features; for biosensor data, an LSTM network is used to capture long-term dependencies. This feature extraction method can make full use of the characteristics of each modality data and lay the foundation for subsequent fusion processing.

[0134] The attention calculation unit 32 is connected to the feature extraction unit 31 and is used to calculate the cross-modal attention weight. The present invention innovatively proposes a spatiotemporal synchronized cross-modal attention mechanism, and its core formula can be expressed as:

[0135]

[0136] Where Q, K, and V represent query, key, and value matrices, respectively. k is the dimension of the key, and M is the spatiotemporal synchronization matrix. The spatiotemporal synchronization matrix M is calculated as:

[0137]

[0138] Among them, t i and p i Respectively represent the timestamp and spatial position of the i-th feature, σ t and σ s are learnable time and space scale parameters. This attention mechanism can effectively capture the spatiotemporal correlation between different modal data and improve the fusion effect. The feature fusion unit 33 is connected to the attention calculation unit 32 to perform weighted fusion on the features of different modalities based on the attention weights to obtain fusion features. The specific fusion process can be expressed as:

[0139]

[0140] Among them, F is the final fusion feature, α i The attention weight of the i-th modality, Transform i is a modality-specific transformation network, X i is the original feature of the i-th mode. Transform network Transform i It is implemented using a fully connected layer, and its parameters are optimized through end-to-end training.

[0141] The multimodal fusion method of the present invention can adaptively adjust the importance of different modal data, improving the system's ability to understand user behavior. For example, in a simulated interview scenario, when the user is answering questions, voice and facial expression data may receive higher weights; while in a simulated programming test scenario, operational behavior data may become more important. This flexible fusion strategy enables the system to accurately capture the user's key performance in different professional scenarios.

[0142] In another preferred embodiment of the present invention, the dynamic evaluation module 4 includes a Bayesian inference unit 41, a nonlinear evolution unit 42 and a capability update unit 43. These units together realize an accurate dynamic evaluation of the user's professional capability.

[0143] The Bayesian inference unit 41 is used to perform probabilistic inference on the user's current ability based on the fusion features. The present invention adopts a variational inference method to approximate the posterior distribution by minimizing the evidence lower bound (ELBO):

[0144]

[0145] Among them, c represents the capability vector, represents the observed data (i.e., fusion features), is the variational posterior distribution, and p(c) is the prior distribution. By optimizing ELBO, the system can obtain a probability distribution estimate of the user's current ability. The nonlinear evolution unit 42 is connected to the Bayesian inference unit 41 to simulate the dynamic change process of the user's ability. The unit includes a differential equation solving subunit 421 and a parameter prediction subunit 422. The differential equation solving subunit 421 is responsible for solving the nonlinear differential equation describing the evolution of ability:

[0146]

[0147] Wherein, o represents the Hadamard product, α, β and γ are dynamic parameters. The parameter prediction subunit 422 uses the LSTM network to dynamically predict these parameters:

[0148]

[0149] Among them, ht is the hidden state of LSTM, and c t are the observation and capability estimation at the current moment, respectively. The capability updating unit 43 is connected to the Bayesian inference unit 41 and the nonlinear evolution unit 42, and is used to integrate the probability inference results and the nonlinear evolution results to generate an updated capability vector. The updating process can be expressed as:

[0150]

[0151] in, and They represent the results of Bayesian inference and nonlinear evolution respectively. λ is the balance factor, which can be determined by cross-validation and can generally be set to 0.6-0.8.

[0152] The dynamic evaluation method of the present invention combines the uncertainty modeling of Bayesian inference with the dynamic characteristics of nonlinear evolution, and can more accurately capture the changing trend of user capabilities. For example, the system can identify the rapid improvement or slow decline of certain abilities (such as communication skills and problem-solving skills) of users during continuous simulation training, thereby providing more accurate career development suggestions.

[0153] In another preferred embodiment of the present invention, the decision optimization module 5 includes a state encoding unit 51, an action generation unit 52, a value evaluation unit 53 and a strategy optimization unit 54. These units work together to achieve career development decision optimization based on reinforcement learning.

[0154] The state encoding unit 51 is used to encode the user capability vector, economic environment parameters and industry feature vector into a state representation. The present invention adopts an autoencoder structure for state encoding, wherein both the encoder and the decoder are three-layer fully connected networks, and the hidden layer uses a LeakyReLU activation function. The encoding process can be expressed as:

[0155] s = Encoder([c,e,i]),

[0156] Wherein, c is the capability vector, e is the economic environment parameter, i is the industry characteristic vector, and s is the encoded state representation. The action generation unit 52 is connected to the state encoding unit 51 and is used to generate possible career decision actions based on the current state. The present invention adopts a continuous action generation method based on Gaussian process, in which the probability distribution of the action is:

[0157]

[0158] Among them, μ(s) and Σ(s) are the mean and covariance functions, respectively, parameterized by a deep neural network:

[0159] μ(s)=W μφ(s)+b μ ,∑(s)=diag(exp(W σ φ(s)+b σ )),

[0160] Among them, φ(s) is the nonlinear feature map of the state, W μ 、b μ , W σ and b σ is a learnable parameter. The value evaluation unit 53 is connected to the state encoding unit 51 and the action generation unit 52, and is used to evaluate the value of the state-action pair. The present invention adopts a dual Q learning method, using two independent Q networks to reduce over-estimation bias:

[0161]

[0162] Among them, ψ(s,a) is the joint feature representation of the state-action pair, W 1 、b 1 , W 2 and b 2 is the parameter of the Q network. The strategy optimization unit 54 is connected to the value evaluation unit 53 and is used to optimize the decision strategy based on the TD3 algorithm. The core update formula of the TD3 algorithm is as follows:

[0163] Policy Update:

[0164]

[0165] Q value update:

[0166]

[0167] Among them, θ and θ′ represent the parameters of the current policy network and the target policy network respectively, ρ is the state distribution, γ is the discount factor, σ is the noise standard deviation, and c is the clipping range. The present invention introduces a dynamic weight factor α t Adaptively adjust the weights of rewards in different dimensions:

[0168] α t =σ(W α ·[c k ,e m ] T +b α ),

[0169] Among them, σ is the sigmoid function, W α and b α are learnable parameters.

[0170] The decision optimization method of the present invention can generate the best career development strategy according to the user's ability status and external environment. For example, in a recession, the system may recommend that users invest more in self-skill improvement; while when the industry is developing rapidly, it may recommend that users actively seek new career opportunities. This dynamic optimization strategy can help users make wise decisions at different career stages and economic environments.

[0171] In another preferred embodiment of the present invention, the path planning module 6 includes a graph construction unit 61, an edge weight calculation unit 62, a path search unit 63 and a path optimization unit 64. These units work together to achieve optimal development path planning based on user capabilities and career goals.

[0172] The graph construction unit 61 is used to construct a dynamic career development graph based on career development decisions and user capability vectors. The graph can be represented as G = (V, E), where V is a node set representing different career states; E is an edge set representing the transition between career states. Preferably, the present invention adopts a hierarchical graph structure to divide the career development process into three levels: short-term, medium-term and long-term, and each level contains a number of nodes. The graph construction process can be represented as:

[0173] V={v i,j |i∈{1,2,3},j∈{1,…,N i}},

[0174] Among them, i represents the level, j represents the node number in the level, and N i is the number of nodes in the i-th layer. The edge weight calculation unit 62 is connected to the graph construction unit 61 and is used to calculate the weight of the edge in the graph. The present invention innovatively proposes an edge weight calculation method based on capability similarity and conversion difficulty:

[0175] w(v i,j ,v i+1,k )=α·sin(c(v i,j ),c(v i+1,k ))-(1-α)·diff(v i,j ,v i+1,k ),

[0176] Among them, c(v) represents the capability vector corresponding to node v, sim(·,·) is the capability similarity function, diff(·,·) is the conversion difficulty function, and α is the balance factor, which can be determined by cross-validation and can generally be set to 0.6-0.8. The capability similarity function uses cosine similarity:

[0177]

[0178] The conversion difficulty function takes into account time, economic cost and risk factors:

[0179] diff(v 1 ,v 2 )=β 1 T(v 1 ,v 2 )+β 2 C(v 1 ,v 2 )+β 3 R(v 1 ,v 2 ),

[0180] Among them, T, C and R represent time, economic cost and risk function respectively, β 1 , β 2 and β 3 is a weight coefficient, which can be adjusted according to user preferences. The path search unit 63 is connected to the graph construction unit 61 and the edge weight calculation unit 62, and is used to search for the optimal path in the dynamic career development graph based on the Monte Carlo Tree Search (MCTS) algorithm. The core steps of the MCTS algorithm include selection, expansion, simulation and backtracking. In the selection stage, the present invention adopts the UCB1 (Upper Confidence Bound 1) formula to balance exploration and utilization:

[0181]

[0182] Among them, w i is the cumulative reward of node i, n i is the number of visits to node i, N is the number of visits to the parent node, and c is the exploration coefficient, which is usually set to In the simulation stage, the present invention adopts a fast random simulation strategy and uses a lightweight neural network to evaluate the potential value of the path:

[0183] V(p)=f θ (φ(p)),

[0184] Among them, p is the path, φ(p) is the characteristic representation of the path, and f θ is the parameterized value function.

[0185] The path optimization unit 64 is connected to the path search unit 63 and is used to smooth and locally optimize the searched path. The present invention uses a dynamic programming method to smooth the path, and the objective function is:

[0186]

[0187] Where p and p′ represent the original path and the smoothed path respectively, and λ is the smoothing coefficient, which is usually set between 0.1 and 0.5.

[0188] The path planning method of the present invention can provide users with personalized and feasible career development paths. For example, for a computer science graduate who aspires to become an AI researcher, the system may plan a development path such as "junior development engineer → algorithm engineer → senior algorithm engineer → AI researcher" and provide specific ability improvement suggestions and time planning for each stage.

[0189] In another preferred embodiment of the present invention, the visualization module 7 includes a data dimension reduction unit 71, a trajectory generation unit 72, a dynamic update unit 73 and an interaction design unit 74. These units together realize an intuitive visualization display of the user's ability development trajectory.

[0190] The data dimension reduction unit 71 is used to project the high-dimensional capability vector into a three-dimensional space using the principal component analysis (PCA) method. The specific dimension reduction process can be expressed as:

[0191] X 3D =X·W PCA ,

[0192] Among them, X is the original high-dimensional capability matrix, W PCA is the PCA transformation matrix, which is obtained by solving the following optimization problem:

[0193]

[0194] The trajectory generation unit 72 is connected to the data dimension reduction unit 71 and is used to generate a continuous capability development trajectory based on the projected data points. The present invention uses a cubic spline interpolation method to generate a smooth trajectory:

[0195] S(t)=a i +b i (tt i )+c i (tt i ) 2 +d i (tt i ) 3 ,t i ≤t <t i+1 ,

[0196] Among them, a i 、b i 、c i and d i are the spline coefficients, obtained by solving the linear equations.

[0197] The dynamic updating unit 73 is connected to the trajectory generating unit 72 for updating the visual trajectory in real time. The present invention innovatively introduces a trajectory updating method based on a physical model, and compares the capability development process to the motion of a damped spring:

[0198]

[0199] Among them, m is the mass parameter, c is the damping coefficient, k is the elastic coefficient, and x 0 is the target position, and F(t) is the external force function, which represents the influence of learning and environmental factors. By adjusting these parameters, the system can simulate different types of ability development curves, such as rapid improvement, slow progress, or fluctuating progress.

[0200] The interaction design unit 74 is used to design the interaction mode between the user and the visual interface. The present invention adopts a natural interaction mode based on gesture recognition and supports the following operations:

[0201] 1. Zoom: Open or close your hands to control the zoom of the view.

[0202] 2. Rotation: One-hand rotation gesture controls the rotation of the view.

[0203] 3. Timeline control: Horizontal sliding gesture controls the forward or backward movement of the timeline.

[0204] Preferably, the present invention also introduces a voice command control function, and the user can control the visual interface through simple voice instructions such as "enlarge", "rotate", "play", etc.

[0205] The visualization method of the present invention provides users with an intuitive and interactive display of their ability development trajectory. For example, users can clearly see their progress in different professional skills, understand the gap between their current position and their career goals, and explore specific ability development details through interactive operations.

[0206] Finally, the present invention also provides a career scenario simulation method for promoting employment and entrepreneurship, comprising the following steps:

[0207] S1, generating virtual career scenarios based on the user's initial ability vector and career goals;

[0208] S2, presenting the virtual career scenario in a virtual reality environment, and collecting multimodal behavior data of the user in the virtual reality environment;

[0209] S3, based on the spatiotemporal synchronized cross-modal attention mechanism, fusing the multimodal behavior data to obtain fusion features;

[0210] S4, based on the nonlinear capability evolution model, using the fusion features to dynamically evaluate the user's professional capabilities to obtain an updated capability vector;

[0211] S5. generating a career development decision based on the reinforcement learning algorithm and the updated capability vector;

[0212] S6. Build a dynamic career development graph and generate the optimal career development path based on the Monte Carlo tree search algorithm;

[0213] S7. Generate a visual representation of the user's ability development trajectory based on three-dimensional situation projection technology and present it in a virtual reality environment;

[0214] S8. Repeat steps S1 to S7 until the user reaches a preset career development goal or the simulation time ends.

[0215] In step S1, the generation of virtual career scenarios adopts the conditional generative adversarial network (cGAN) technology, and its discriminator loss function is:

[0216]

[0217] Among them, x is the real data, z is random noise, y is the conditional vector (containing user ability and career goal information), and λ is the gradient penalty coefficient, which is usually set to 10.

[0218] In step S3, the core formula of the spatiotemporal synchronized cross-modal attention mechanism is:

[0219]

[0220] Where Q, K, and V represent query, key, and value matrices, respectively. k is the dimension of the key, and M is the space-time synchronization matrix.

[0221] In step S4, the nonlinear capability evolution model is described by a differential equation:

[0222]

[0223] Among them, c is the capability vector, α, β and γ are dynamic parameters predicted by the LSTM network.

[0224] In step S5, the reinforcement learning algorithm adopts the improved TD3 algorithm, and its strategy update formula is:

[0225]

[0226] Among them, θ is the policy network parameter, ρ is the state distribution, Q 1 For the first Q network.

[0227] In step S6, the node selection strategy of the Monte Carlo tree search algorithm adopts the UCB1 formula:

[0228]

[0229] Among them, w i is the cumulative reward of node i, ni is the number of visits to node i, N is the number of visits to the parent node, and c is the exploration coefficient.

[0230] Through the above steps, the method of the present invention can provide users with a personalized, immersive career scenario simulation experience, help users understand their own abilities, explore career development paths, and thus promote employment and entrepreneurship. For example, a newly graduated computer science student can simulate different career options (such as software development, data analysis, artificial intelligence research, etc.) through this system, experience various work scenarios, assess their own ability gaps, and obtain targeted development suggestions. This can not only help job seekers make more informed career choices, but also provide companies with more comprehensive and objective talent assessment methods, thereby promoting the healthy development of the employment market.

[0231] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A career scenario simulation system for promoting employment and entrepreneurship, characterized in that: include: Scenario generation module for: Generate virtual career scenarios based on the user's initial ability vector and career goals; Sending the virtual career scenario to a multimodal interaction module; The multimodal interaction module is communicatively connected with the scenario generation module and is used to: receiving the virtual career scenario and presenting it in a virtual reality environment; Collecting behavioral data of the user in the virtual reality environment, wherein the behavioral data includes visual data, voice data, and biosensor data; Sending the behavior data to a multimodal fusion module; The multimodal fusion module is communicatively connected with the multimodal interaction module and is used to: receiving the behavior data; Based on a spatiotemporal synchronized cross-modal attention mechanism, the behavior data is fused to obtain fusion features; Sending the fused features to a dynamic evaluation module; A dynamic evaluation module is communicatively connected with the multimodal fusion module and is used to: receiving the fused feature; Based on the nonlinear capability evolution model, the user's professional capabilities are dynamically evaluated to obtain an updated capability vector; Sending the updated capability vector to a decision optimization module; A decision optimization module is connected in communication with the dynamic evaluation module and is used to: receiving the updated capability vector; Generate career development decisions based on reinforcement learning algorithms; sending the career development decision to a path planning module; A path planning module is in communication with the decision optimization module and is used to: receiving said career development decision; Build a dynamic career development graph and generate the optimal career development path based on the Monte Carlo tree search algorithm; Sending the optimal career development path to a visualization module; A visualization module is in communication with the path planning module and is used to: Receive the optimal career development path; Generate a visual representation of the user's capability development trajectory based on three-dimensional situation projection technology; The visual representation is sent to the multimodal interaction module for presentation in a virtual reality environment.

2. The system according to claim 1, characterized in that The scenario generation module comprises: A conditional generative adversarial network unit, which is used to generate virtual career scenarios based on the user's current ability vector and career goals; A career constraint unit, connected to the conditional generative adversarial network unit, is used to impose career-related constraints on the generated virtual career scenarios to ensure the relevance of the generated scenarios to the user's abilities and goals; The scenario library unit is connected to the conditional generative adversarial network unit and the occupational constraint unit, and is used to store and manage the generated virtual occupational scenarios.

3. The system according to claim 1, characterized in that The multimodal interaction module includes: A virtual reality rendering unit, used for presenting the virtual career scenario on a virtual reality head display device; The multimodal data collection unit is used to collect the user's behavior data in the virtual reality environment, including: A visual collection subunit, used to collect the user's visual behavior data; A voice collection subunit, used to collect the user's voice data; A biosensor collection subunit, used to collect physiological data of the user; The data preprocessing unit is connected to the multimodal data acquisition unit and is used to preprocess the collected behavior data, including data cleaning, format conversion and preliminary feature extraction.

4. The system according to claim 1, characterized in that The multimodal fusion module comprises: A feature extraction unit for extracting modality-specific features from the preprocessed behavioral data; an attention calculation unit, connected to the feature extraction unit, for calculating cross-modal attention weights; A feature fusion unit is connected to the attention calculation unit and is used to perform weighted fusion on features of different modalities based on the attention weight to obtain a fused feature.

5. The system according to claim 1, characterized in that The dynamic evaluation module includes: A Bayesian inference unit, used to make probabilistic inferences about the user's current capabilities based on fused features; The nonlinear evolution unit is connected to the Bayesian inference unit and is used to simulate the dynamic change process of the user's ability, including: A differential equation solving subunit, used to solve the nonlinear differential equations describing the evolution of capabilities; A parameter prediction subunit, used to dynamically predict the parameters of the capability evolution equation; The capability updating unit is connected to the Bayesian inference unit and the nonlinear evolution unit, and is used to integrate the probability inference result and the nonlinear evolution result to generate an updated capability vector.

6. The system according to claim 1, characterized in that The decision optimization module includes: A state encoding unit, used for encoding the user capability vector, the economic environment parameter and the industry feature vector into a state representation; An action generation unit, connected to the state encoding unit, for generating possible career decision actions based on the current state; a value evaluation unit, connected to the state encoding unit and the action generating unit, for evaluating the value of a state-action pair; The strategy optimization unit is connected to the value evaluation unit and is used to optimize the decision-making strategy based on the TD3 algorithm, including a dynamic weight factor for adaptively adjusting the weights of rewards of different dimensions.

7. The system according to claim 1, characterized in that The path planning module includes: A graph construction unit, used to construct a dynamic career development graph based on career development decisions and user capability vectors; an edge weight calculation unit, connected to the graph construction unit, for calculating the weights of the edges in the graph, wherein the weights are based on the similarity between the user's current capabilities and the requirements of the target position; a path search unit, connected to the graph construction unit and the edge weight calculation unit, for searching for an optimal path in the dynamic career development graph based on a Monte Carlo tree search algorithm; The path optimization unit is connected to the path search unit and is used to perform smoothing and local optimization on the searched path.

8. The system according to claim 1, characterized in that The visualization module comprises: A data dimension reduction unit, used to project the high-dimensional capability vector into a three-dimensional space using a principal component analysis method; A trajectory generation unit, connected to the data dimension reduction unit, for generating a continuous capability development trajectory based on the projected data points; A dynamic updating unit, connected to the trajectory generating unit, for updating the visual trajectory in real time, including calculation of velocity and acceleration; The interactive design unit is used to design how users interact with the visual interface, including zooming, rotating, and timeline control functions.

9. The system according to any one of claims 1 to 8, characterized in that: Also includes: The edge computing module is deployed on the virtual reality headset to perform real-time rendering, data collection and preliminary processing tasks; A cloud computing module, connected to the edge computing module via a network, for executing complex artificial intelligence algorithms and large-scale data processing tasks; The task scheduling module is connected to the edge computing module and the cloud computing module, and is used to dynamically allocate computing tasks to the edge or the cloud according to task complexity, real-time requirements and network conditions.

10. A method for simulating career scenarios to promote employment and entrepreneurship, using the system of claim 9, characterized in that: The following steps are involved: S1, generating virtual career scenarios based on the user's initial ability vector and career goals; S2, presenting the virtual career scenario in a virtual reality environment, and collecting multimodal behavior data of the user in the virtual reality environment; S3, based on the spatiotemporal synchronized cross-modal attention mechanism, fusing the multimodal behavior data to obtain fusion features; S4, based on the nonlinear capability evolution model, using the fusion features to dynamically evaluate the user's professional capabilities to obtain an updated capability vector; S5. generating a career development decision based on the reinforcement learning algorithm and the updated capability vector; S6. Build a dynamic career development graph and generate the optimal career development path based on the Monte Carlo tree search algorithm; S7. Generate a visual representation of the user's ability development trajectory based on three-dimensional situation projection technology and present it in a virtual reality environment; S8. Repeat steps S1 to S7 until the user reaches a preset career development goal or the simulation time ends.

Images