Artificial intelligence-based training evaluation and examination system

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a training evaluation and examination system based on artificial intelligence, and the system comprises a data collection and processing unit which collects user behavior data and environment state data; the task dynamic weight distribution unit combines a space-time cross attention mechanism with a lightweight space-time gating mechanism, introduces an assessment item type impact factor to calculate a dynamic distribution weight item, and constructs a dynamic assessment loss function; the task feedback association adjustment unit divides user feedback information into structured scores and unstructured text suggestions, updates dynamic allocation weight items in combination with a feedback memory and attenuation mechanism, and constructs a dynamic assessment feedback loss function; and the visual interface display unit displays the task score. According to the artificial intelligence-based training evaluation assessment system, a dynamic assessment loss function is constructed based on combination of a space-time cross attention mechanism and a lightweight time-space gating mechanism and is applied to artificial intelligence training evaluation assessment, so that the evaluation assessment system with dynamic feedback is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based training evaluation and assessment system. Background Art

[0002] The artificial intelligence-based training evaluation and assessment system aims to optimize the dynamic weight adjustment of multi-dimensional task allocation and improve the adaptability of the system to personalized user needs. Through spatio-temporal feature fusion and dynamic feedback memory mechanism, it controls the priority of task processing and resource scheduling to achieve more accurate task allocation, personalized evaluation criteria, and adaptive assessment result feedback.

[0003] Existing evaluation and assessment systems usually have difficulty in flexibly coping with the diverse needs of different users and environmental changes. Moreover, due to the traditional static scoring mechanism being unable to adjust task weights in real time and lacking dynamic perception of user behavior and environmental factors, it will lead to inaccurate task priorities and assessment criteria that do not adapt to the actual performance of users. Therefore, an artificial intelligence-based training evaluation and assessment system is designed. Summary of the Invention

[0004] The purpose of the present invention is to provide an artificial intelligence-based training evaluation and assessment system to solve the problems mentioned in the above background art, that is, due to the traditional static scoring mechanism being unable to adjust task weights in real time and lacking dynamic perception of user behavior and environmental factors, it will lead to inaccurate task priorities and assessment criteria that do not adapt to the actual performance of users.

[0005] To achieve the above object, the present invention provides an artificial intelligence-based training evaluation and assessment system, including: A data acquisition and processing unit, which is used to collect user behavior data and environmental status data and perform preliminary processing; It also includes a task dynamic weight allocation unit, which is used to use a spatio-temporal cross-attention mechanism combined with a lightweight spatio-temporal gating mechanism when fusing data spatio-temporal features, introduce an assessment item type influence factor to calculate the dynamic allocation weight term of each assessment item, and construct a dynamic assessment loss function; It also includes a task feedback correlation adjustment unit, which, based on user feedback information, uses a multi-modal feedback encoder to classify user feedback information into structured scores and unstructured text suggestions, combines feedback memory and attenuation mechanism to update the dynamic allocation weight term, and constructs a dynamic assessment feedback loss function; It also includes a visualization interface display unit, which is used to visually display the user training progress, task scores, and spatio-temporal dynamic weighting status.

[0006] As a further improvement of this technical solution, the user behavior data includes historical task completion status, scores, and behavior trajectories; the environmental status data includes device status, network performance, and environmental temperature and humidity.

[0007] As a further improvement of this technical solution, the task dynamic weight allocation unit includes a cross-attention module, a lightweight spatio-temporal gating module, and a loss function construction module; Among them, the cross-attention module uses a long short-term memory network to extract the temporal features of user behavior data, uses a convolutional neural network to extract the spatial features of environmental status data, and uses a spatio-temporal cross-attention mechanism to calculate the dynamic correlation between the temporal features and the spatial features to obtain spatio-temporal feature correlation values; The lightweight spatio-temporal gating module uses a gating mechanism to control the contribution of different features to the task weight based on the temporal features and the spatial features, and calculates the gating parameters; The loss function construction module constructs a dynamic assessment loss function based on the spatio-temporal feature attention mechanism and the gating parameters, and introduces an assessment item type influence factor to calculate the dynamic allocation weight item of the assessment item.

[0008] As a further improvement of this technical solution, the cross-attention module uses a long short-term memory network to extract the temporal features of user behavior data, uses a convolutional neural network to extract the spatial features of environmental status data, and uses a spatio-temporal cross-attention mechanism to calculate the dynamic correlation between the temporal features and the spatial features to obtain spatio-temporal feature correlation values. The specific method steps are as follows: S2.1.1. Use a long short-term memory network to extract the temporal features of user behavior data , and use a convolutional neural network to extract the spatial features of environmental status data ; S2.1.2. Fuse the temporal features and the spatial features, and introduce a spatio-temporal cross-attention mechanism to analyze the dynamic correlation between the temporal features and the spatial features, and calculate to obtain spatio-temporal feature correlation values: ; Among them, is the spatio-temporal feature correlation value; is the Softmax function; is the temporal feature, serving as the query in the spatio-temporal cross-attention mechanism; is the spatial feature, serving as the key and value in the spatio-temporal cross-attention mechanism; is the transpose of the spatial feature; is the scaling factor; is the assessment item index.

[0009] As a further improvement of this technical solution, the lightweight spatio-temporal gating module is based on temporal features and spatial features, uses a gating mechanism to control the contribution of different features to the task weight, and calculates the gating parameters. The specific method is as follows: ; Wherein, is the gating parameter; is the Sigmoid activation function; is the weight matrix; is the concatenation matrix of temporal features and spatial features.

[0010] As a further improvement of this technical solution, the loss function construction module is based on the spatio-temporal feature attention mechanism and the gating parameter, and introduces the assessment item type influence factor to calculate the dynamic allocation weight item of the assessment item, and constructs a dynamic assessment loss function. The specific method steps are as follows: S2.3.1. Analyze the proportion of each task in the practical task type through the text classification model, analyze the assessment item type, and calculate the assessment item type influence factor: ; ; Wherein, is the assessment item index; is the th assessment item type; indicates that the th assessment item is of the practical task type; indicates that the th assessment item is of the theoretical task type; is the proportion of the th assessment item in the practical task type; is the proportion of the th assessment item in the theoretical task type; is the assessment task type threshold; is the th assessment item type influence factor; is the weight of the th assessment item type being of the theoretical task type; is the weight of the th assessment item type being of the practical task type; S2.3.2. Based on the spatio-temporal feature attention mechanism and the gating parameter, and introduce the assessment item type influence factor to calculate the dynamic allocation weight item of the assessment item: ; Wherein, is the dynamic allocation weight item of the th assessment item; is the assessment project type; is the influence factor of the assessment project type; S2.3.3. Construct a dynamic assessment loss function based on dynamically allocated weight terms: ; Among them, is the dynamic assessment loss function; is the total number of assessment projects; is the th loss of the assessment project.

[0011] As a further improvement of this technical solution, the task feedback correlation adjustment unit includes a dynamic weight update module and a feedback loss function construction module; Among them, the dynamic weight update module uses a multi-modal feedback encoder to divide user feedback information into structured scores and unstructured text suggestions, constructs a user feedback adjustment function, and updates the dynamically allocated weight terms in combination with the feedback memory and decay mechanism; The feedback loss function construction module constructs a dynamic assessment feedback loss function based on the updated dynamically allocated weight terms.

[0012] As a further improvement of this technical solution, the dynamic weight update module uses a multi-modal feedback encoder to divide user feedback information into structured scores and unstructured text suggestions, constructs a user feedback adjustment function, and updates the dynamically allocated weight terms in combination with the feedback memory and decay mechanism. The specific method steps are as follows: S3.1.1. Use a multi-modal feedback encoder to divide user feedback information into structured scores and unstructured text suggestions; S3.1.2. Based on the structured scores and unstructured text suggestions, construct a user feedback adjustment function: User feedback adjustment function: ; Among them, is the user feedback adjustment function; is the user index; is the th feedback information of the user; is the structured score; is the learnable weight of the structured score; is the unstructured text suggestion; is the learnable weight of the unstructured text suggestion; is the Sigmoid activation function; is the BERT model; S3.1.3. Based on the user feedback adjustment function, update the dynamically allocated weight terms in combination with the feedback memory and decay mechanism: ; Among them, is the dynamic weight item for integrating user feedback information; is the th dynamic weight item for the assessment item; is the feedback adjustment coefficient; is the total number of feedbacks; is the feedback times index; is the attenuation coefficient; is the th feedback information of the

[0013] As a further improvement of this technical solution, the feedback loss function construction module constructs a dynamic assessment feedback loss function based on the updated dynamic weight item, specifically as follows: ; Among them, is the dynamic assessment feedback loss function; is the dynamic weight item for integrating user feedback information; is the total number of assessment items; is the th loss of the assessment item.

[0014] As a further improvement of this technical solution, the visualization interface display unit is used to visually display the user training progress, task score, and spatio-temporal dynamic weighting status, specifically as follows: S4.1. Display the dynamic weighting status of the user's current task in the spatio-temporal dimension, combined with the calculation results of the spatio-temporal cross-attention mechanism and the lightweight spatio-temporal gating mechanism; S4.2. Display that the dynamically adjusted task weights will be updated in real time; S4.3. Display the user's feedback on the current task and how it affects the task adjustment through a chart.

[0015] Compared with the prior art, the beneficial effects of the present invention: 1. In this artificial intelligence training evaluation and assessment system, the dynamic weight allocation algorithm based on spatio-temporal feature fusion can dynamically adjust the weight allocation of each task by combining the user's historical behavior data and the current environmental state, evaluate the impact of each task on different user performances, and construct a dynamic assessment loss function trained based on artificial intelligence, avoiding the inaccuracy brought by the traditional static weighting method, and improving the system's adaptability to personalized needs and response ability to complex environments.

[0016] 2. In the artificial intelligence-based training evaluation and assessment system, an adaptive task relevance adjustment algorithm based on user feedback is used to adjust the task weights in real time according to user feedback, and the long-term effect of the feedback is combined to optimize task allocation, and a dynamic assessment feedback loss function trained based on artificial intelligence is constructed to improve the accuracy of task allocation and the reliability of assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is the overall flow block diagram of the present invention; The meanings of the reference numerals in the figure are as follows: 1. Data acquisition and processing unit; 2. Task dynamic weight allocation unit; 21. Cross-attention module; 22. Lightweight spatio-temporal gating module; 23. Loss function construction module; 3. Task feedback correlation adjustment unit; 31. Dynamic weight update module; 32. Feedback loss function construction module; 4. Visualization interface display unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Please refer to Figure 1 as shown, a training evaluation and assessment system based on artificial intelligence is provided, including: A data acquisition and processing unit 1, which is used to collect user behavior data and environmental status data and perform preliminary processing; In this embodiment, the user behavior data includes historical task completion status, scores, and behavior trajectories; the environmental status data includes device status, network performance, and environmental temperature and humidity.

[0020] In this embodiment, the historical task completion situation reflects the user's task execution efficiency, success rate, and task completion time, and is used to evaluate the user's capabilities and behavior patterns; the score is used to evaluate the user's satisfaction with the task and potential improvement points during task execution; the behavior trajectory includes the user's interaction records and operation paths with the system, and is used to optimize the interaction design and improve the user interface; the device status reflects the working conditions of the device, whether there are faults, and the usage of system resources, and is used to evaluate the reliability of device operation to ensure that there are no bottlenecks or faults in the hardware resources during task execution; the network performance is used to identify task delays or failures caused by network instability, and then take measures to optimize network resource allocation; by monitoring the environmental temperature and humidity, the system can predict and respond to device performance changes caused by environmental factors, and optimize task allocation and execution timing; In this embodiment, the user behavior data and environmental status data are preliminarily processed as follows: Check for duplicate records in the dataset and remove duplicate user behavior data to avoid affecting model training; exclude abnormal records with negative scores or task times of 0; for missing scores or task completion situations, use KNN imputation to fill in the missing values by converting the user behavior data to the standard range through Z-score normalization; use the moving average method to smooth the data; adopt the 3-sigma rule to detect and remove abnormal data in the device status and network performance; It also includes a task dynamic weight allocation unit 2, which is used to use the spatio-temporal cross-attention mechanism combined with the lightweight spatio-temporal gating mechanism and introduce the assessment item type influence factor to calculate the dynamic allocation weight term of each assessment item when fusing the spatio-temporal features of the data, and construct a dynamic assessment loss function; In this embodiment, the task dynamic weight allocation unit 2 includes a cross-attention module 21, a lightweight spatio-temporal gating module 22, and a loss function construction module 23; Among them, the cross-attention module 21 uses a long short-term memory network to extract the time features of the user behavior data, uses a convolutional neural network to extract the spatial features of the environmental status data, and uses the spatio-temporal cross-attention mechanism to calculate the dynamic correlation between the time features and the spatial features to obtain the spatio-temporal feature correlation value; The lightweight spatio-temporal gating module 22 uses a gating mechanism to control the contribution of different features to the task weight based on the time features and the spatial features, and calculates the gating parameters; The loss function construction module 23 constructs a dynamic assessment loss function based on the spatio-temporal feature attention mechanism and the gating parameters, and introduces the assessment item type influence factor to calculate the dynamic allocation weight term of the assessment item.

[0021] In this embodiment, the cross-attention module 21 uses a long short-term memory network to extract the temporal features of user behavior data, uses a convolutional neural network to extract the spatial features of environmental state data, and uses a spatio-temporal cross-attention mechanism to calculate the dynamic correlation between the temporal features and the spatial features, obtaining a spatio-temporal feature correlation value. The specific method steps are as follows: S2.1.1. Use a long short-term memory network to extract the temporal features of user behavior data , and use a convolutional neural network to extract the spatial features of environmental state data ; S2.1.2. Fuse the temporal features and the spatial features, and introduce a spatio-temporal cross-attention mechanism to analyze the dynamic correlation between the temporal features and the spatial features, and calculate to obtain a spatio-temporal feature correlation value: ; Among them, is the spatio-temporal feature correlation value; is the Softmax function; is the temporal feature, serving as the query in the spatio-temporal cross-attention mechanism; is the spatial feature, serving as the key and value in the spatio-temporal cross-attention mechanism; is the transpose of the spatial feature; is the scaling factor; is the assessment item index.

[0022] In this embodiment, in the original cross-attention mechanism, we have three basic inputs: query, key, and value. We first calculate the similarity between the query and the key through dot product, and the result is a similarity matrix. Since the value of the dot product may be very large, we perform scaling to avoid the problem of gradient vanishing or gradient explosion; perform a Softmax operation on the similarity matrix to obtain a normalized attention weight; then, use the calculated attention weight to perform weighted summation on the value to obtain the output result; In this unit, considering the temporal features and the spatial features, calculate their dynamic correlation through the cross-attention mechanism, and input the temporal features and the spatial features into the cross-attention mechanism as the query, key, and value.

[0023] In this embodiment, the lightweight spatio-temporal gating module 22 is based on the temporal features and the spatial features, uses a gating mechanism to control the contribution of different features to the task weight, and calculates the gating parameter. The specific method is as follows: ; Among them, is the gating parameter; is the Sigmoid activation function; is the weight matrix; is the concatenation matrix of the temporal features and the spatial features.

[0024] In this embodiment, the gating mechanism is a mechanism that adjusts the contribution of each input to the final output of the model by controlling the transmission of different information flows; in this unit, the gating mechanism dynamically adjusts the weights of spatio-temporal features. Through the learned gating parameters, according to the current task requirements and environmental conditions, it adjusts the contributions of temporal features and spatial features to the task results.

[0025] In this embodiment, the loss function construction module 23 constructs a dynamic assessment loss function based on the spatio-temporal feature attention mechanism and gating parameters, and introduces the assessment item type influence factor to calculate the dynamic allocation weight term of the assessment item. The specific method steps are as follows: S2.3.1. Analyze the proportion of each task in the practical task type through the text classification model, analyze the assessment item type, and calculate the assessment item type influence factor: ; ; Among them, is the assessment item index; is the th assessment item type; indicates that the th assessment item is of the practical task type; indicates that the th assessment item is of the theoretical task type; is the proportion of the th assessment item in the practical task type; is the proportion of the th assessment item in the theoretical task type; is the assessment task type threshold; is the th assessment item type influence factor; is the weight of the th assessment item type being the theoretical task type; is the weight of the th assessment item type being the practical task type; S2.3.2. Based on the spatio-temporal feature attention mechanism and gating parameters, and introduce the assessment item type influence factor to calculate the dynamic allocation weight term of the assessment item: ; Among them, is the dynamic allocation weight term of the th assessment item; is the assessment item type; is the assessment item type influence factor; S2.3.3. Construct a dynamic assessment loss function based on dynamically allocated weight terms: ; wherein, is the dynamic assessment loss function; is the total number of assessment items; is the th loss of the assessment item.

[0026] In this embodiment, the traditional loss function is a simple weighted sum of the losses of all assessment items. Usually, these weights are fixed or manually set, where the weights are usually a fixed constant or a preset static weight, and the losses of all assessment items are weighted based on the same weight. The optimization strategy of the entire loss function is relatively fixed; In this unit, a dynamically allocated weight term is introduced and this weight term is calculated based on the spatio-temporal feature attention mechanism and the gating mechanism. The weight of each assessment item is not static, but is calculated according to the dynamic correlation of the time feature and the space feature, so that different weights can be assigned to each assessment item under different tasks and environmental conditions.

[0027] In this embodiment, a text classification model is used to analyze the proportion of each task in the practical task and the theoretical task type. This step extracts the features of the task content to judge the proportion of the practical task type and the proportion of the theoretical task type of the task; if the practical proportion of the task is greater than or equal to the preset threshold, the task is classified as a practical task type, that is ; otherwise, the task is classified as a theoretical task type, that is ; Based on the spatio-temporal feature attention mechanism combining the time feature and the space feature, these features are extracted from the historical user behavior data and the environmental state data, and combined with the gating parameter to perform weighted fusion on the spatio-temporal feature and the task type to calculate the final dynamically allocated weight term; introducing a task type weight factor to further adjust the final weight of the task according to the practical and theoretical proportions of the task, that is ; It further includes a task feedback correlation adjustment unit 3. The task feedback correlation adjustment unit 3 classifies the user feedback information into structured scores and unstructured text suggestions based on the user feedback information using a multi-modal feedback encoder, and updates the dynamically allocated weight term in combination with the feedback memory and decay mechanism to construct a dynamic assessment feedback loss function; In this embodiment, the task feedback correlation adjustment unit 3 includes a dynamic weight update module 31 and a feedback loss function construction module 32; Among them, the dynamic weight update module 31 uses a multi-modal feedback encoder to classify user feedback information into structured scores and unstructured text suggestions, constructs a user feedback adjustment function, and updates the dynamically allocated weight items in combination with the feedback memory and decay mechanism; The feedback loss function construction module 32 constructs a dynamic assessment feedback loss function based on the updated dynamically allocated weight items.

[0028] In this embodiment, the adjustment of the task weight is based on the dynamic changes of spatio-temporal features and user feedback, enabling the weight to be adjusted in real time to reflect the current state of the task and user needs. The introduction of the feedback history and decay mechanism causes the impact of each feedback to gradually decay, thus avoiding the excessive influence of a single feedback on the task weight; not only using structured scores, but also combining unstructured text suggestions, enabling the system to more comprehensively understand user feedback and adjust the task weight according to the feedback content, thereby improving the accuracy of task allocation.

[0029] In this embodiment, the dynamic weight update module 31 uses a multi-modal feedback encoder to classify user feedback information into structured scores and unstructured text suggestions, constructs a user feedback adjustment function, and updates the dynamically allocated weight items in combination with the feedback memory and decay mechanism. The specific method steps are as follows: S3.1.1. Use a multi-modal feedback encoder to classify user feedback information into structured scores and unstructured text suggestions; S3.1.2. Based on the structured scores and unstructured text suggestions, construct a user feedback adjustment function: User feedback adjustment function: ; Among them, is the user feedback adjustment function; is the user index; is the th user's feedback information; is the structured score; is the learnable weight of the structured score; is the unstructured text suggestion; is the learnable weight of the unstructured text suggestion; is the Sigmoid activation function; is the BERT model; S3.1.3. Based on the user feedback adjustment function, update the dynamically allocated weight items in combination with the feedback memory and decay mechanism: ; Among them, is the dynamically allocated weight item for comprehensive user feedback information; is the Dynamic allocation weight items for each assessment item; is the feedback adjustment coefficient; is the total number of feedbacks; is the feedback count index; is the attenuation coefficient; is the th feedback information of the th user.

[0030] In this embodiment, the feedback loss function construction module 32 constructs a dynamic assessment feedback loss function based on the updated dynamic allocation weight items, specifically as follows: ; where is the dynamic assessment feedback loss function; is the dynamic allocation weight item integrating user feedback information; is the total number of assessment items; is the th loss of the assessment item.

[0031] In this embodiment, the multi-modal feedback encoder classifies user feedback information into structured scores and unstructured text suggestions to make full use of the multi-dimensional feedback information of users, rather than relying solely on a single feedback source; structured scores provide an intuitive assessment of task completion and are suitable for quantitative analysis; unstructured text suggestions can capture more context information, details, and potential emotional information; the BERT model extracts semantic information from unstructured text and converts it into features that can be used to adjust task weights, that is: is the structured score; is the unstructured text suggestion; The introduced feedback memory and attenuation mechanism can consider the continuous impact of historical feedback on task weights: the attenuation mechanism makes the impact of feedback with a longer time on task weights gradually decrease, which reflects the time sensitivity, that is, the impact of more recent feedback on task optimization should be greater; since the impact of users' long-term behavior and feedback over time is not constant but gradually weakens over time, the user feedback adjustment function can be obtained ; The dynamic allocation weight of the task can more accurately reflect the user's recent behavior without being affected by outdated feedback; based on the updated dynamic allocation weight items, that is: , a dynamic assessment feedback loss function is constructed, that is: ; thus ensuring that the loss of each assessment item can be adjusted according to real-time feedback, rather than just static weight weighting; this loss function can optimize task allocation in real time according to changes in user feedback, enabling the training process to flexibly adjust the optimization goal at different time points and different user states; In this way, the loss function not only takes into account the execution effect of the task, but also integrates the user's historical feedback information to ensure that the artificial intelligence is more in line with the actual evaluation during training.

[0032] It further includes a visualization interface display unit 4, which is used to visually display the user's training progress, task score, and spatio-temporal dynamic weighting status; The visualization interface display unit 4 is used to visually display the user's training progress, task score, and spatio-temporal dynamic weighting status, specifically as follows: S4.1. Display the dynamic weighting status of the user's current task in the spatio-temporal dimension, combining the calculation results of the spatio-temporal cross-attention mechanism and the lightweight spatio-temporal gating mechanism; S4.2. Display that the dynamically adjusted task weights will be updated in real time; S4.3. Display the user's feedback on the current task and how it affects the task adjustment through a chart.

[0033] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. An artificial intelligence training evaluation system, characterized in that: include: A data collection and processing unit (1), the data collection and processing unit (1) is used to collect user behavior data and environmental status data, and perform preliminary processing; The task dynamic weight allocation unit (2) is used to use the spatiotemporal cross attention mechanism combined with the lightweight spatiotemporal gating mechanism when fusing the spatiotemporal features of the data, and introduce the assessment item type influence factor to calculate the dynamic allocation weight item of each assessment item, and construct a dynamic assessment loss function; A task feedback association adjustment unit (3), wherein the task feedback association adjustment unit (3) is based on user feedback information and uses a multimodal feedback encoder to divide the user feedback information into structured scores and unstructured text suggestions, combines feedback memory and decay mechanisms to update dynamic allocation weight items, and constructs a dynamic assessment feedback loss function; A visual interface display unit (4), wherein the visual interface display unit (4) is used to visually display user training progress, task scoring, and spatiotemporal dynamic weighted status.

2. The artificial intelligence training evaluation and assessment system according to claim 1 is characterized in that: The user behavior data includes historical task completion status, scores, and behavior trajectories; the environmental status data includes device status, network performance, and environmental temperature and humidity.

3. The artificial intelligence training evaluation and assessment system according to claim 2 is characterized in that: The task dynamic weight allocation unit (2) includes a cross attention module (21), a lightweight spatiotemporal gating module (22) and a loss function construction module (23); The cross attention module (21) uses a long short-term memory network to extract the temporal features of the user behavior data, uses a convolutional neural network to extract the spatial features of the environmental state data, and uses a spatiotemporal cross attention mechanism to calculate the dynamic correlation between the temporal features and the spatial features to obtain a spatiotemporal feature correlation value; The lightweight spatiotemporal gating module (22) uses a gating mechanism to control the contribution of different features to the task weight based on the temporal features and the spatial features, and calculates the gating parameters; The loss function construction module (23) is based on the spatiotemporal feature attention mechanism and gating parameters, and introduces the assessment item type impact factor to calculate the dynamic allocation weight item of the assessment item, thereby constructing a dynamic assessment loss function.

4. The artificial intelligence training evaluation and assessment system according to claim 3 is characterized in that: The cross attention module (21) uses a long short-term memory network to extract the temporal features of the user behavior data, uses a convolutional neural network to extract the spatial features of the environmental state data, and uses a spatiotemporal cross attention mechanism to calculate the dynamic correlation between the temporal features and the spatial features to obtain the spatiotemporal feature correlation value. The specific method steps are as follows: S2.1.

1. Extracting temporal features of user behavior data using long short-term memory networks , using convolutional neural networks to extract spatial features of environmental state data ; S2.1.

2. The temporal features are integrated with the spatial features, and the spatiotemporal cross attention mechanism is introduced to analyze the dynamic correlation between the temporal features and the spatial features, and the spatiotemporal feature correlation value is calculated: ； in, is the spatial-temporal feature correlation value; is the Softmax function; is the temporal feature, which serves as the query in the spatiotemporal cross attention mechanism; is the spatial feature, which serves as the key and value in the spatiotemporal cross attention mechanism; is the transpose of the spatial features; is the scaling factor; Index of assessment items.

5. The artificial intelligence training evaluation and assessment system according to claim 4 is characterized in that: The lightweight spatiotemporal gating module (22) uses a gating mechanism to control the contribution of different features to the task weight based on temporal features and spatial features, and calculates the gating parameters. The specific method is as follows: ； in, is the gating parameter; is the Sigmoid activation function; is the weight matrix; It is the concatenation matrix of temporal features and spatial features.

6. The artificial intelligence training evaluation and assessment system according to claim 5 is characterized in that: The loss function construction module (23) is based on the spatiotemporal feature attention mechanism and the gating parameters, and introduces the assessment item type impact factor to calculate the dynamic allocation weight item of the assessment item, and constructs a dynamic assessment loss function. The specific method steps are as follows: S2.3.

1. Analyze the proportion of each task in the practical task type through the text classification model, analyze the assessment project type and calculate the assessment project type influencing factor: ; ； in, Index for assessment items; For the Type of assessment items; Indicates The assessment items are of practical task type; Indicates The assessment items are of theoretical task type; For the The proportion of assessment items in the practical task type; For the The proportion of assessment items in the theoretical task type; It is the assessment task type threshold; For the The impact factor of the assessment project type; For the The weight of each assessment item type being a theoretical task type; For the The weight of each assessment item type as a practical task type; S2.3.

2. Based on the spatiotemporal feature attention mechanism and gating parameters, the dynamic allocation weight items of the assessment items are calculated by introducing the assessment item type impact factor: ； in, For the Dynamically allocate weights for each assessment item; The type of assessment item; To assess the impact factor of the project type; S2.3.

3. Construct a dynamic assessment loss function based on the dynamic allocation weight item: ； in, is the dynamic assessment loss function; is the total number of assessment items; For the The loss of an assessment item.

7. The artificial intelligence training evaluation and assessment system according to claim 6 is characterized in that: The task feedback association adjustment unit (3) includes a dynamic weight updating module (31) and a feedback loss function construction module (32); The dynamic weight updating module (31) uses a multimodal feedback encoder to divide user feedback information into structured ratings and unstructured text suggestions, constructs a user feedback adjustment function, and combines feedback memory and attenuation mechanisms to update dynamic allocation weight items; The feedback loss function construction module (32) constructs a dynamic assessment feedback loss function based on the updated dynamic allocation weight item.

8. The artificial intelligence training evaluation and assessment system according to claim 7 is characterized in that: The dynamic weight updating module (31) uses a multimodal feedback encoder to divide the user feedback information into structured ratings and unstructured text suggestions, constructs a user feedback adjustment function, and combines feedback memory and attenuation mechanisms to update the dynamic allocation weight items. The specific method steps are as follows: S3.1.

1. Use a multimodal feedback encoder to separate user feedback information into structured ratings and unstructured text suggestions. S3.1.

2. Construct a user feedback adjustment function based on structured ratings and unstructured text suggestions: User feedback adjustment function: ； in, Adjust functions for user feedback; Indexing for users; For the Feedback from individual users; Score structured; Learnable weights for structured scoring; Suggestions for unstructured text; Proposing learnable weights for unstructured text; is the Sigmoid activation function; It is the BERT model; S3.1.3, based on user feedback adjustment function, combined with feedback memory and decay mechanism to update the dynamic allocation weight item: ； in, Dynamically assign weight items for comprehensive user feedback information; For the Dynamically allocate weights for each assessment item; is the feedback adjustment coefficient; is the total number of feedbacks; is the index of feedback times; is the attenuation coefficient; For the User Feedback information.

9. The artificial intelligence training evaluation and assessment system according to claim 8 is characterized in that: The feedback loss function construction module (32) constructs a dynamic assessment feedback loss function based on the updated dynamic allocation weight term, which is as follows: ； in, Feedback loss function for dynamic assessment; Dynamically assign weight items for comprehensive user feedback information; is the total number of assessment items; For the The loss of an assessment item.

10. The artificial intelligence training evaluation and assessment system according to claim 1 is characterized in that: The visualization interface display unit (4) is used to visualize the user's training progress, task score and time-space dynamic weighted status, as follows: S4.

1. Display the dynamic weighted state of the user's current task in the spatiotemporal dimension, combining the calculation results of the spatiotemporal cross attention mechanism and the lightweight spatiotemporal gating mechanism; S4.2, the task weights showing dynamic adjustment will be updated in real time; S4.

3. Use charts to display user feedback on the current task and how it affects task adjustments.

Images