Personnel online auditing method and system based on large model

Abstract

The invention relates to the technical field of personnel online auditing, and discloses a personnel online auditing method and system based on a large model, and the method comprises the steps: carrying out the feature extraction of the pre-obtained multi-modal data of a target personnel, and constructing a multi-modal feature vector; respectively analyzing the target ability and behavior intention of the target person through an execution sandbox and a trajectory analysis sandbox in a pre-constructed double-layer sandbox architecture to obtain an ability analysis result; configuring a dynamic question-answering mechanism, and dynamically adjusting the question-answering content of the target person to obtain a dynamic question-answering result; evaluating the ability of the target person in combination with the ability analysis result and the dynamic question and answer result, and generating an ability evaluation report; according to the method, the ability of the target personnel is comprehensively evaluated from two dimensions of result and process by constructing a double-layer sandbox architecture, personalized evaluation contents are formulated for different personnel, online real-time examination and verification of the ability of the target personnel can be realized, and comprehensiveness and accuracy of an ability evaluation result are improved.

Description

Technical Field

[0001] This application relates to the field of online personnel verification technology, and more specifically to a method and system for online personnel verification based on a large model. Background Technology

[0002] Traditional personnel competency assessment and review methods rely on manual evaluation, paper-based document review, and standardized testing. The review process is complex, the amount of data reviewed by the reviewers is large, and it is difficult to grasp the review standards. It is difficult to develop different review methods for different personnel and conduct dynamic reviews. As a result, the review process is characterized by low efficiency, strong subjectivity, and limited scope.

[0003] Existing technologies suffer from the following problems: based on a single type of data, they cannot comprehensively capture the overall capabilities of the assessed individuals; using fixed assessment processes and standardized questions, they cannot adjust the assessment content and difficulty according to the real-time performance of the assessed individuals, making it difficult to effectively review the generalization and problem-solving abilities of personnel when facing unknown problems; assessing capabilities based on the final results neglects the analysis of the execution process, resulting in an inaccurate identification of personnel's thinking patterns, coding styles, and potential capability shortcomings, leading to inaccurate and incomplete capability analysis results; to address at least one of the above problems, this application proposes an online personnel review method and system based on a large model. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide an online personnel verification method and system based on a large-scale model, which can effectively solve the problems in the background technology. The specific technical solution of this application is as follows:

[0005] A method for online personnel verification based on a large model includes:

[0006] Feature extraction is performed on the multimodal data of the target personnel to be acquired, and a multimodal feature vector is constructed.

[0007] Based on the multimodal feature vectors, the target capabilities and behavioral intentions of the target personnel are analyzed through the execution sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, to obtain capability analysis results;

[0008] Based on the capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the question-and-answer content of the target personnel, thereby obtaining dynamic question-and-answer results;

[0009] Based on the capability analysis results and dynamic question-and-answer results, the capabilities of the target personnel are assessed using a preset capability assessment model, and a capability assessment report is generated for online review of the personnel.

[0010] Specifically, the step of extracting features from the multimodal data of the target personnel to construct a multimodal feature vector includes:

[0011] Acquire multimodal data of the target personnel, wherein the multimodal data includes text data, code data, and audio data;

[0012] The text data, code data, and audio data are respectively extracted using a pre-defined feature extraction model to obtain a first feature set, a second feature set, and a third feature set.

[0013] By combining the first feature set, the second feature set, and the third feature set, a multimodal feature vector is constructed.

[0014] Specifically, based on the multimodal feature vectors, the target personnel's capabilities and behavioral intentions are analyzed through the execution sandbox and trajectory analysis sandbox in a pre-constructed two-layer sandbox architecture, respectively, to obtain capability analysis results, including:

[0015] Based on multimodal feature vectors, a two-layer sandbox architecture including an execution sandbox and a trajectory analysis sandbox is constructed.

[0016] The code data from the multimodal data is input into the execution sandbox to simulate the execution of the code and analyze the target capabilities of the target personnel to obtain the first target capability;

[0017] Input the code data into the trajectory analysis sandbox, collect the code execution process trajectory, analyze the target person's behavioral intention, and obtain the first behavioral intention;

[0018] A comparative analysis of the first target capability and the first behavioral intention is conducted to identify the target personnel's strengths and weaknesses, and to obtain the capability analysis results.

[0019] Specifically, the step of inputting code data from multimodal data into the execution sandbox, simulating the execution of the code, and analyzing the target capabilities of the target personnel to obtain the first target capability, including:

[0020] The code data from the multimodal data is input into the execution sandbox to simulate the execution of the code, and the code execution metrics are collected during the execution process.

[0021] Based on the code execution metrics, the time and space complexity are calculated, and the target capabilities of the target personnel are analyzed to obtain the first target capability.

[0022] Specifically, the step of inputting code data into the trajectory analysis sandbox, collecting the code execution process trajectory, analyzing the target person's behavioral intent, and obtaining the first behavioral intent includes:

[0023] The code data is input into the trajectory analysis sandbox for execution, and trajectory data is collected during the code execution process;

[0024] Based on the trajectory data, the coding style and logical correlation are identified through a preset trajectory analysis model, and a trajectory index vector is constructed.

[0025] Based on the trajectory indicator vector analysis, the target person's behavioral intention is obtained, and the first behavioral intention is obtained.

[0026] Specifically, based on the capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the question-and-answer content of the target personnel, resulting in dynamic question-and-answer results, including:

[0027] Based on the capability analysis results, identify the strengths and weaknesses of the capabilities, and set scene switching points and difficulty adjustment points at the corresponding locations.

[0028] At the scene switching point, the scene switching content in the question and answer content is identified, and a replacement scene is generated to replace the scene switching content, thereby obtaining the first adjustment result;

[0029] At the difficulty adjustment point, the matching result between the question and answer content and the target personnel's ability is analyzed, and a corresponding difficulty adjustment instruction is generated to adjust the difficulty in real time, thus obtaining the second adjustment result.

[0030] By combining the first and second adjustment results, the question-and-answer content of the target personnel is dynamically adjusted to obtain dynamic question-and-answer results.

[0031] Specifically, at the scene switching point, the scene switching content in the question and answer content is identified, and a replacement scene is generated to replace the scene switching content to obtain the first adjustment result, including:

[0032] At the scene switching point, by analyzing the content corresponding to the strengths and weaknesses of the capabilities, the scene switching content in the question and answer content can be identified;

[0033] Based on the contextual association of the scene switching content, a corresponding replacement scene is generated;

[0034] The scene switching content is dynamically replaced according to the replacement scene, and the corresponding scene content is reconstructed to obtain the first adjustment result.

[0035] Specifically, at the difficulty adjustment point, the matching result between the question and answer content and the target person's ability is analyzed, and a corresponding difficulty adjustment instruction is generated to adjust the difficulty in real time, resulting in a second adjustment result, including:

[0036] At the difficulty adjustment point, the matching result between the question and answer content and the target personnel's ability is analyzed to obtain the first matching result;

[0037] Based on the first matching result, match the corresponding difficulty adjustment instruction in the pre-built difficulty instruction library;

[0038] In response to the difficulty adjustment command, the difficulty of the question and answer content is rewritten and adjusted in real time to obtain the second adjustment result.

[0039] Specifically, combining the ability analysis results and dynamic question-and-answer results, the abilities of the target personnel are assessed using a pre-set ability assessment model, generating an ability assessment report, including:

[0040] Contextual analysis is performed on the ability analysis results and dynamic question-and-answer results to extract the corresponding set of ability indicators;

[0041] Based on the set of competency indicators, the competency of the target personnel is assessed using a preset competency assessment model, and a competency assessment report is generated.

[0042] A large-scale model-based online personnel verification system, used to implement the aforementioned large-scale model-based online personnel verification method, includes:

[0043] The data analysis module extracts features from the multimodal data of the target personnel to be acquired and constructs multimodal feature vectors.

[0044] The capability analysis module, based on the multimodal feature vector, analyzes the target capability and behavioral intent of the target personnel through the execution sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, and obtains the capability analysis results;

[0045] The question-and-answer adjustment module configures a dynamic question-and-answer mechanism based on the capability analysis results, dynamically adjusting the question-and-answer content of the target personnel to obtain dynamic question-and-answer results;

[0046] The personnel review module combines the capability analysis results and dynamic question-and-answer results to assess the capabilities of target personnel through a preset capability assessment model, generating a capability assessment report for online personnel review.

[0047] The beneficial effects of this application are as follows: Feature extraction and fusion based on multimodal data are used to construct multimodal feature vectors. A two-layer sandbox architecture, including an execution sandbox and a trajectory analysis sandbox, is built. Simultaneously, the target personnel's capabilities and behavioral intentions are analyzed to assess their comprehensive abilities. Scene switching points and difficulty adjustment points are set based on the capability analysis results, and the question-and-answer content is dynamically adjusted. At scene switching points, replacement scenarios are generated by analyzing contextual relationships, and scene content is dynamically reconstructed. At difficulty adjustment points, the difficulty of questions and answers is adjusted in real time based on capability matching results. Combining capability analysis results and dynamic question-and-answer results, a comprehensive analysis of the target personnel's overall capabilities is conducted, generating a capability assessment report. By fusing multimodal data and constructing a two-layer sandbox architecture, the capabilities of target personnel can be comprehensively assessed from both result and process dimensions. The assessment content and difficulty are dynamically adjusted based on the target personnel's real-time performance, allowing for personalized assessment content tailored to different individuals. This more effectively uncovers the true capability level of the assessed individuals, avoiding misjudgments caused by assessment content that is too difficult or too easy. This enables online real-time review of target personnel's capabilities, improving the comprehensiveness and accuracy of capability assessment results. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating an online personnel review method based on a large model, as described in this application.

[0049] Figure 2 This is a schematic diagram of the two-layer sandbox architecture in the embodiments of this application;

[0050] Figure 3 This is a schematic diagram of scene switching points and difficulty adjustment points in the embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of an online personnel review system based on a large model, as described in an embodiment of this application. Detailed Implementation

[0052] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are indicated by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0053] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0054] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0055] refer to Figure 1 As shown, this application presents a method for online personnel verification based on a large model, comprising:

[0056] S101. Extract features from the multimodal data of the target personnel to be acquired, and construct a multimodal feature vector;

[0057] S102. Based on the multimodal feature vector, the target ability and behavioral intention of the target personnel are analyzed by executing the sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, to obtain the ability analysis results.

[0058] S103. Based on the capability analysis results, configure a dynamic question-and-answer mechanism to dynamically adjust the question-and-answer content of the target personnel and obtain dynamic question-and-answer results;

[0059] S104. Combining the capability analysis results and dynamic question-and-answer results, the capabilities of the target personnel are assessed using a preset capability assessment model, and a capability assessment report is generated for online review of the personnel.

[0060] With the development of remote work and online recruitment, enterprises have increasingly diversified and precise needs for talent assessment, and traditional review methods are no longer sufficient to meet the requirements of modern talent evaluation. This application provides a multimodal data fusion and dynamic capability assessment mechanism, which can effectively solve the problems of low efficiency and limited coverage of traditional personnel review methods, and meet the needs of large-scale, high-quality online review.

[0061] In this embodiment, feature extraction is performed on the multimodal data of the target personnel to construct a multimodal feature vector. The multimodal data includes text data, code data, and audio data. For each data type, a corresponding preset feature extraction model is used for feature extraction. The extracted features are then aligned, and feature weights are assigned based on their contribution to the personnel competency assessment. The features and their corresponding weights are then weighted and fused to obtain the multimodal feature vector. By combining multimodal data for feature extraction and fusion, the multidimensional performance of personnel competencies can be analyzed, avoiding the biased assessment caused by single data. Weighted fusion of different features effectively integrates complementary information from multiple features, avoiding feature redundancy or information loss. The constructed multimodal feature vector comprehensively and accurately reflects the overall competency characteristics of the target personnel, providing a high-quality data foundation for competency analysis.

[0062] Specifically, based on multimodal feature vectors, a pre-constructed two-layer sandbox architecture is used to analyze the target personnel's target capabilities and behavioral intentions using an execution sandbox and a trajectory analysis sandbox, respectively, to obtain capability analysis results. The two-layer sandbox architecture, constructed based on multimodal feature vectors, includes an independent yet data-interactive execution sandbox and a trajectory analysis sandbox. The execution sandbox analyzes the target personnel's target capabilities by simulating code execution and monitoring operational metrics in real time, combining these metrics to analyze the target personnel's target capabilities. The trajectory analysis sandbox analyzes the target personnel's behavioral intentions by collecting complete trajectory data during code execution, identifying the target personnel's coding style and logical connections, and analyzing their behavioral intentions. Combining target capabilities and behavioral intentions, the architecture analyzes the target personnel's strengths and weaknesses to obtain capability analysis results. This two-layer sandbox architecture combines result evaluation and process analysis, capturing behavioral intention characteristics often overlooked in traditional auditing processes, improving the comprehensiveness of personnel capability assessment, accurately identifying core strengths and weaknesses, and obtaining accurate capability analysis results.

[0063] After analyzing the capabilities of target personnel and obtaining the capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the content of the questions and answers, resulting in dynamic question-and-answer results. Based on the capability analysis results, strengths and weaknesses are identified, and scenario switching points are set in the question-and-answer sessions corresponding to strengths, while difficulty adjustment points are set in the question-and-answer sessions corresponding to weaknesses. At the scenario switching points, scenario-switching content is generated based on contextual relationships, replacing the scenario in the question-and-answer content. At the difficulty adjustment points, the difficulty of the question-and-answer content is adjusted in real time. By combining scenario switching and difficulty adjustment, the question-and-answer content of the target personnel is dynamically adjusted, resulting in dynamic question-and-answer results. By setting scenario switching points and difficulty adjustment points to dynamically adjust the scenario and difficulty of the question-and-answer content, a deeper analysis of the target personnel's potential capabilities can be conducted in the dimension of capabilities strengths, while the problems can be accurately located in the dimension of weaknesses, making the dynamic question-and-answer results more targeted and comprehensive.

[0064] Specifically, by combining the results of competency analysis and dynamic question-and-answer sessions, a pre-set competency assessment model is used to evaluate the competencies of target personnel, generating a competency assessment report. Contextual correlation analysis is performed on the competency analysis results and dynamic question-and-answer results to extract corresponding competency indicators. These indicators are then used to evaluate the competencies of target personnel using the same pre-set competency assessment model, generating a competency assessment report for online review. Contextual correlation analysis can reveal the relationship between competency analysis results and dynamic question-and-answer results. Combining these two approaches to analyze the competencies of target personnel improves the accuracy and comprehensiveness of the analysis results.

[0065] This application extracts and fuses features from multimodal data to construct multimodal feature vectors. It then builds a two-layer sandbox architecture comprising an execution sandbox and a trajectory analysis sandbox. Simultaneously, it analyzes the target personnel's abilities and behavioral intentions to assess their comprehensive capabilities. Based on the capability analysis results, it sets scenario switching points and difficulty adjustment points, dynamically adjusting the question-and-answer content. At scenario switching points, it generates replacement scenarios by analyzing contextual relationships and dynamically reconstructs the scenario content. At difficulty adjustment points, it adjusts the question-and-answer difficulty in real time based on capability matching results. Combining capability analysis results and dynamic question-and-answer results, it comprehensively analyzes the target personnel's overall capabilities and generates a capability assessment report. By fusing multimodal data and constructing a two-layer sandbox architecture, it can comprehensively assess the target personnel's capabilities from both result and process dimensions. It dynamically adjusts the assessment content and difficulty based on the target personnel's real-time performance, allowing for customized assessment content for different individuals. This more effectively uncovers the true capability level of the assessed, avoiding misjudgments caused by assessment content that is too difficult or too easy. It enables online real-time verification of the target personnel's capabilities, improving the comprehensiveness and accuracy of capability assessment results.

[0066] Furthermore, feature extraction is performed on the multimodal data of the target personnel to be acquired, and a multimodal feature vector is constructed, including:

[0067] S201. Obtain multimodal data of the target personnel, wherein the multimodal data includes text data, code data, and audio data;

[0068] S202. Extract features from text data, code data, and audio data respectively using a preset feature extraction model to obtain a first feature set, a second feature set, and a third feature set;

[0069] S203. Combine the first feature set, the second feature set, and the third feature set to construct a multimodal feature vector.

[0070] In this embodiment, multimodal data of the target personnel is acquired. This multimodal data includes text data, code data, and audio data. Text data includes, but is not limited to, materials submitted voluntarily by the target personnel; code data includes, but is not limited to, code entered in real-time by the target personnel during the assessment process; and audio data includes, but is not limited to, recordings of technical interviews conducted by the target personnel. It should be noted that consent has been obtained regarding the personal privacy data involved in this application. The acquired data undergoes preprocessing such as cleaning and normalization to obtain preprocessed multimodal data. By acquiring and preprocessing multimodal data, the bias in assessment caused by single data sources can be avoided, providing rich and complete data for the analysis and assessment of the target personnel's capabilities.

[0071] Specifically, a pre-defined feature extraction model is used to extract features from text data, code data, and audio data, respectively, to obtain a first feature set, a second feature set, and a third feature set. For text data, the feature extraction model includes, but is not limited to, the BERT model based on the Transformer architecture. The BERT model is trained using a large amount of historical text data to obtain a pre-trained BERT model. The text data is then input into the pre-trained BERT model, which analyzes the semantics of the text data and extracts the corresponding semantic features, resulting in the first feature set. For code data, the feature extraction model includes, but is not limited to, the CodeBERT model adapted for code semantic understanding. The CodeBERT model is trained using a large amount of historical code data to obtain a pre-trained CodeBERT model. The code data is then input into the pre-trained CodeBERT model, which analyzes the syntactic logic and semantic relationships of the code data and extracts the corresponding features, resulting in the second feature set. For audio data, the feature extraction model includes, but is not limited to, the LSTM model. The LSTM model is trained using a large amount of historical audio data to obtain a pre-trained LSTM model. The audio data is then input into the pre-trained LSTM model, which analyzes the acoustic characteristics of the audio data and extracts the corresponding features, resulting in the third feature set.

[0072] It is important to emphasize that by extracting features from text data, code data, and audio data using corresponding feature extraction models, we can obtain corresponding feature sets. This avoids the limitation that a single feature cannot reflect the comprehensive ability of the target person. The three feature sets correspond to the target person's written expression ability, programming ability, and verbal expression ability, respectively, providing rich feature data for the analysis of the target person's ability and improving the accuracy of the analysis process.

[0073] Specifically, feature alignment is performed on the features in the first, second, and third feature sets. Weights are assigned to the features based on their importance to the assessment of personnel capabilities. The features are then weighted and summed according to their respective weights to obtain a multimodal feature vector. By fusing the features extracted from the multimodal data, complementary information among the target personnel's textual, programming, and language ability features can be integrated to reflect the target personnel's comprehensive capabilities and improve the accuracy of the target personnel capability analysis process.

[0074] Furthermore, based on the multimodal feature vectors, the target personnel's capabilities and behavioral intentions are analyzed through the execution sandbox and trajectory analysis sandbox in a pre-constructed two-layer sandbox architecture, respectively, to obtain capability analysis results, including:

[0075] S301. Based on multimodal feature vectors, construct a two-layer sandbox architecture including an execution sandbox and a trajectory analysis sandbox;

[0076] S302. Input the code data from the multimodal data into the execution sandbox, simulate the execution of the code, and analyze the target capabilities of the target personnel to obtain the first target capability;

[0077] S303. Input the code data into the trajectory analysis sandbox, collect the code execution process trajectory, analyze the target person's behavioral intention, and obtain the first behavioral intention;

[0078] S304. Conduct a comparative analysis of the first target capability and the first behavioral intention to identify the target personnel's capability strengths and weaknesses, and obtain capability analysis results.

[0079] like Figure 2As shown, a two-layer sandbox architecture, comprising an execution sandbox and a trajectory analysis sandbox, is constructed based on multimodal feature vectors. A virtual machine is used to build the two-layer sandbox architecture, ensuring resource independence between the two sandboxes. The execution sandbox environment is configured according to the programming language type of the code data extracted from the multimodal feature vectors, ensuring smooth code execution. The execution sandbox includes a compiler library, a test case library, and a resource monitoring tool. The compiler ensures the code can be compiled correctly, the test cases perform process testing on the code, and the resource monitoring tool collects runtime metrics in real time. A trajectory analysis sandbox is configured, including a trajectory recording tool, a log analysis tool, and a trajectory analysis tool. The trajectory recording tool embeds tracking probes during code compilation or execution to record the complete execution trajectory of the code. The log analysis tool extracts and analyzes key events in the logs, and the trajectory analysis tool performs real-time analysis of the code execution trajectory.

[0080] Simultaneously, a connection is established between the execution sandbox and the trajectory analysis sandbox via an encrypted API interface, enabling bidirectional data synchronization between the two sandboxes and ensuring real-time sharing of code data and runtime status. This two-layer sandbox architecture guarantees isolation between the two sandboxes. The execution sandbox tests and analyzes the code execution results, while the trajectory analysis sandbox analyzes the trajectory during code execution, providing data support for the analysis of the target personnel's capabilities and intentions. Combining code capabilities with the comprehensive capabilities of the target personnel in intent analysis improves the comprehensiveness and accuracy of the analysis process.

[0081] Specifically, the code data from the multimodal data is input into the execution sandbox, the code is simulated and run, and the target personnel's target capabilities are analyzed to obtain the first target capability. Through code simulation, the code execution results and compilation error information can be obtained, realizing the quantitative analysis of the target personnel's target capabilities, calculating and analyzing the target personnel's code capabilities, and obtaining accurate target capability assessment results.

[0082] Specifically, code data is input into the trajectory analysis sandbox, and the trajectory of code execution is collected to analyze the behavioral intent of the target personnel and obtain the first behavioral intent. By collecting the trajectory of code execution and analyzing behavioral intent, the logical connections during code execution can be captured, and accurate behavioral intent analysis results can be obtained.

[0083] A comparative analysis of the primary target capability and the primary behavioral intention is conducted. The primary target capability is matched with the primary behavioral intention according to the corresponding relationship. If the target capability performs well and the corresponding behavioral intention performs well, it is judged as a capability advantage; if the target capability performs poorly and the corresponding behavioral intention performs poorly, it is judged as a capability weakness. For capability dimensions judged as advantages, the specific manifestations of the advantages are described; for capability dimensions judged as weaknesses, the specific problems and impacts of the weaknesses are identified. The analysis results of capability advantages and weaknesses are integrated to obtain structured capability analysis results. By conducting a correlation analysis between the primary target capability and the primary behavioral intention, combining the outcome performance of the capability with the process characteristics of the intention, the limitations of traditional assessments that only look at the results are overcome. This allows for a more accurate identification of advantageous and disadvantageous capabilities, improving the accuracy and comprehensiveness of capability analysis results.

[0084] Furthermore, the code data from the multimodal data is input into the execution sandbox to simulate the execution of the code and analyze the target personnel's target capabilities to obtain the first target capability, including:

[0085] S401. Input the code data from the multimodal data into the execution sandbox, simulate the execution of the code, and collect code execution metrics during the execution process;

[0086] S402. Calculate the time complexity and space complexity based on the code execution indicators, analyze the target personnel's target capabilities, and obtain the first target capability.

[0087] In this embodiment, code data from multimodal data is input into an execution sandbox to simulate code execution, and code execution metrics are collected during the process. When extracting code data from multimodal data, code segments directly related to the target capability assessment are located by code type and input into the execution sandbox. The execution sandbox automatically activates the corresponding compilation environment based on the code language to obtain the compilation result. If a compilation error occurs, the error type and data are recorded. The code's functionality, boundary conditions, and performance stress are tested according to the test case library in the execution sandbox. During the testing process, code execution metrics are collected through monitoring tools in the execution sandbox, including but not limited to total execution time, code block-level time consumption, peak memory usage, and resource consumption. By simulating code execution and collecting execution metrics, comprehensive data support can be provided for the target capability analysis process of the target personnel.

[0088] Specifically, the time and space complexity are calculated based on code execution metrics to analyze the target personnel's target capabilities and determine the first target capability. The execution time of the code is analyzed based on the data volume trend using code execution metrics, and the time complexity is calculated in conjunction with the code's hierarchical structure. For example, if the execution time increases fourfold when the data volume increases to 2n, and there are nested loops, the time complexity is determined to be O(n²). Based on the memory usage of variables and data structures in the code, a static memory usage baseline is determined according to the number of bytes. The space complexity is calculated by analyzing the memory usage changes with the data volume using code execution metrics. For example, if the memory peak approximately doubles when the data volume increases from n to 2n, and there is no memory leak, the space complexity is determined to be O(n). Combining time and space complexity, the target personnel's target capability level is determined. If the time complexity is O(n²) or the space complexity is O(n²), it is determined to be intermediate; if the time complexity is greater than or equal to O(2n), it is determined to be intermediate. n If there is a memory leak, it is judged as weak; if the time complexity is less than or equal to O(n) and the space complexity is less than or equal to O(n), it is judged as strong and the first target capability is obtained.

[0089] It is important to emphasize that combining time complexity with code structure and runtime metrics can avoid the limitations of relying solely on code structure for judgment, and the calculated time complexity more closely reflects the actual performance of the code. Combining static and dynamic memory complexity avoids the shortcomings of traditional methods that only calculate static memory, and more comprehensively reflects the code's ability to utilize resources. Analyzing the target capability by combining the code's time and space complexity can improve the accuracy of the target capability analysis results.

[0090] Furthermore, the code data is input into the trajectory analysis sandbox, and the trajectory of the code execution process is collected to analyze the behavioral intent of the target person, thereby obtaining the first behavioral intent, including:

[0091] S501. Input the code data into the trajectory analysis sandbox and execute it to collect trajectory data during the code execution process;

[0092] S502. Based on the trajectory data, identify the coding style and logical correlation through a preset trajectory analysis model, and construct a trajectory index vector;

[0093] S503. Analyze the target person's behavioral intention based on the trajectory indicator vector to obtain the first behavioral intention.

[0094] In this embodiment, code data is input into the trajectory analysis sandbox for execution, and trajectory data is collected during the code execution process. Compiled code data from the execution sandbox is synchronized to the trajectory analysis sandbox. Based on the programming language type, the corresponding dynamic instrumentation tool is loaded to ensure compatibility between the instrumentation tool and the code runtime environment. The trajectory events to be collected include function call events, variable operation events, branch jump events, and exception handling events. The dynamic instrumentation tool begins collecting trajectory data based on these events. The collected complete trajectory data is then cleaned to obtain preprocessed trajectory data. By selecting an instrumentation tool, external interference can be reduced, and language-adaptive fine-grained trajectory collection can be achieved, avoiding the problem of traditional log printing only recording key nodes and incomplete trajectories, thus obtaining complete and accurate trajectory data.

[0095] Specifically, based on trajectory data, a pre-defined trajectory analysis model is used to identify coding style and logical correlation, constructing a trajectory indicator vector. Coding style correlation features and logical correlation features are extracted from the trajectory data. Coding style correlation features include, but are not limited to, variable naming features and annotation features; logical correlation features include, but are not limited to, function call features and exception handling features. The extracted features are input into the pre-defined trajectory analysis model to identify coding style and logical correlation. The trajectory analysis model includes, but is not limited to, a graph neural network model. A large amount of historical trajectory data is used to train the graph neural network model, resulting in a pre-trained model. The coding style correlation features and logical correlation features are then input into the pre-trained graph neural network model, which analyzes the coding style and logical correlation, outputting coding style scores and logical correlation scores. The coding style scores and logical correlation scores are weighted and summed to construct the trajectory indicator vector. By analyzing coding style and logical correlation, and combining the logical dependencies between events, the model's output scores are converted into indicator vectors, providing clear data input for behavioral intent analysis.

[0096] Specifically, based on trajectory indicator vectors, the behavioral intentions of the target personnel are analyzed through a pre-set intention analysis model to obtain the first behavioral intention. The intention analysis model includes, but is not limited to, a neural network model. The neural network model is trained using a large amount of historical trajectory indicator vector data to obtain a pre-trained neural network model. The trajectory indicator vectors are input into the pre-trained neural network model, and the model analyzes the behavioral intentions of the target personnel, calculates the corresponding indicator dimension intention score, and determines the intention level. The intention level includes high, medium, and low, thus obtaining the first behavioral intention. By calculating the intention score, the behavioral intentions of the target personnel are analyzed, thereby improving the accuracy and comprehensiveness of the behavioral intention analysis results.

[0097] Furthermore, based on the aforementioned capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the question-and-answer content of the target personnel, resulting in dynamic question-and-answer results, including:

[0098] S601. Based on the capability analysis results, identify the capability strengths and weaknesses, and set scene switching points and difficulty adjustment points at the corresponding locations.

[0099] S602. At the scene switching point, identify the scene switching content in the question and answer content, and generate a replacement scene to replace the scene switching content to obtain the first adjustment result;

[0100] S603. At the difficulty adjustment point, analyze the matching result between the question and answer content and the target personnel's ability, generate the corresponding difficulty adjustment instruction to adjust the difficulty in real time, and obtain the second adjustment result.

[0101] S604. Combine the first adjustment result and the second adjustment result to dynamically adjust the question and answer content of the target personnel to obtain dynamic question and answer results.

[0102] like Figure 3 As shown, based on the ability analysis results, strengths and weaknesses are identified, and scenario switching points and difficulty adjustment points are set at corresponding locations. Based on the strengths and weaknesses identified in the ability analysis, the corresponding strengths and weaknesses are located. For each strength, a scenario switching point is set in the corresponding assessment stage of the question-and-answer process, allowing for a more complex scenario and in-depth assessment of the target personnel's abilities. For each weakness, a difficulty adjustment point is set in the corresponding assessment stage of the question-and-answer process, dynamically adjusting the difficulty to avoid either being too difficult (resulting in an inability to answer) or too easy (failing to accurately analyze the personnel's abilities). By setting scenario switching points and difficulty adjustment points, the assessment scenario and difficulty can be adjusted in real time, ensuring that the question-and-answer content matches the target personnel's abilities, thus improving the accuracy and effectiveness of the assessment process.

[0103] Specifically, at scene transition points, the system identifies scene transition content within the question-and-answer content and generates a replacement scene to replace the transition content, resulting in the first adjustment result. At difficulty adjustment points, the system analyzes the matching results between the question-and-answer content and the target personnel's abilities, generates corresponding difficulty adjustment instructions, and adjusts the difficulty in real time, resulting in the second adjustment result. By switching scenes and adjusting difficulty in real time, the system can combine the real-time ability analysis results of the target personnel to adjust the question-and-answer content in real time, ensuring that the question-and-answer content matches the target personnel's abilities and improving the assessment effect.

[0104] Specifically, the question-and-answer content of the target personnel is dynamically adjusted based on the first and second adjustment results to obtain dynamic question-and-answer results. The logical consistency between the first and second adjustment results is analyzed and judged. If the adjustment correlation dimensions of the first and second adjustment results are different, there is no conflict, and they can be directly integrated. If the adjustment correlation dimensions are the same, and scenario switching increases the difficulty while difficulty adjustment decreases it, this is considered a conflict. The priority for adjusting the difficulty of the capability weakness is set higher than the priority for scenario switching of the capability strength, and difficulty adjustment is performed first. The adjusted first and second adjustment results are then integrated to dynamically adjust the question-and-answer content of the target personnel, resulting in dynamic question-and-answer results. Through logical consistency verification and priority setting, the content confusion caused by conflicting adjustment results is avoided, ensuring the logic and coherence of the question-and-answer content, improving the adjustment effect, and thus enhancing the accuracy of the assessment process.

[0105] Furthermore, at the scene transition point, scene transition content in the question-and-answer content is identified, and a replacement scene is generated to replace the scene transition content, resulting in a first adjustment result, including:

[0106] S701. At the scene switching point, by analyzing the content corresponding to the strengths and weaknesses of the capabilities, the scene switching content in the question and answer content is identified.

[0107] S702. Based on the contextual association of the scene switching content, generate the corresponding replacement scene;

[0108] S703. Dynamically replace the scene switching content according to the replacement scene, reconstruct the corresponding scene content, and obtain the first adjustment result.

[0109] In this embodiment, at the scene switching point, the scene switching content in the question and answer content is identified by analyzing the content corresponding to the capability advantages and capabilities disadvantages; the content at the scene switching point is analyzed to extract the content corresponding to the capability advantages and capabilities disadvantages, which are used as the scene switching content in the question and answer content; by identifying the scene switching content, data basis is provided for scene switching, thereby improving the efficiency of scene switching.

[0110] Specifically, by combining the contextual relationships of the scene switching content, corresponding replacement scenes are generated. Contextual information about the scene switching content is extracted, including the preceding context, technical background, and interaction history. Scene switching templates are generated based on historical scene data, and a replacement scene resource library is constructed. Scene switching templates from this library that match the industry domain of the contextual information, have a matching technical background, and are more complex than the original scene switching content are selected. Corresponding replacement scenes are generated according to these templates, and the semantic coherence of the replacement scenes with respect to the context is analyzed and verified. By extracting contextual information, the problem of replacement scenes being out of place in the actual question-and-answer context can be avoided, ensuring a high degree of adaptation between the replacement scenes and the context. The construction of a scene resource library allows for rapid matching of corresponding scene switching templates, improving scene switching efficiency.

[0111] Specifically, the process involves dynamically replacing the content of the scene transition based on the replacement scenario, reconstructing the corresponding scene content, and obtaining the first adjustment result. The information in the replacement scenario is then matched with the content of the scene transition, and the core information of the replacement scenario is used to replace the scene transition content and integrated into the original Q&A content. This core information includes background, requirements, and difficulties. After integrating the replacement scenario information, semantic connection is verified, and transitional statements are added to avoid logical breaks during information fusion and splicing. The replacement content is then integrated to obtain the first adjustment result. By verifying and adjusting the logical consistency of the replacement content, issues of unnatural expression caused by direct replacement are avoided, ensuring the logical adaptation of replacements for different types of Q&A content. Through scene content reconstruction, scene upgrades are achieved while maintaining the interactive logic of the Q&A content, preventing disconnects between Q&A due to replacements and improving the quality of the Q&A content.

[0112] Furthermore, at the difficulty adjustment point, the matching result between the question and answer content and the target personnel's ability is analyzed, and corresponding difficulty adjustment instructions are generated for real-time difficulty adjustment, resulting in a second adjustment result, including:

[0113] S801. At the difficulty adjustment point, analyze the matching results between the question and answer content and the target personnel's abilities to obtain the first matching result;

[0114] S802. Based on the first matching result, match the corresponding difficulty adjustment instruction in the pre-built difficulty instruction library;

[0115] S803. In response to the difficulty adjustment command, the difficulty of the question and answer content is rewritten and adjusted in real time to obtain the second adjustment result.

[0116] In this embodiment, at the difficulty adjustment point, the matching result between the question and answer content and the target person's ability is analyzed to obtain the first matching result. At the difficulty adjustment point, the content of the ability deficiency at that position is analyzed, and the corresponding ability feature vector is extracted. The difficulty of the content at that position is analyzed, and features reflecting the difficulty of the content are extracted, including but not limited to technical depth and logical complexity. A content difficulty feature vector is constructed, and the cosine similarity between the ability feature vector and the content difficulty feature vector is calculated. A similarity threshold is set according to the matching accuracy requirements. If the cosine similarity is less than the similarity threshold, it indicates that the content difficulty at that position does not match the target person's ability, and the first matching result is obtained. By analyzing the person's ability features and content difficulty features, the accuracy of the matching result can be improved to ensure that the matching result matches the target person's true ability level.

[0117] Specifically, based on the first matching result, a corresponding difficulty adjustment instruction is matched in a pre-built difficulty instruction library; difficulty adjustment instructions are generated according to the matching result. For example, for adjustment needs where the operation rule is to simplify the problem conditions while retaining the core test points, a first difficulty adjustment instruction is generated to slightly reduce the difficulty; for adjustment needs where the operation rule is to break down complex problems and reduce operation requirements, a second difficulty adjustment instruction is generated to significantly reduce the difficulty; a difficulty instruction library is built based on the generated difficulty adjustment instructions, and the operation rules are analyzed according to the first matching result to match the corresponding difficulty adjustment instruction in the difficulty instruction library; by building a difficulty instruction library to provide a reference for instruction matching, the corresponding difficulty adjustment instruction is quickly matched in the difficulty instruction library based on the first matching result, thereby improving the efficiency of difficulty adjustment.

[0118] Specifically, in response to difficulty adjustment commands, the question-and-answer content is rewritten and its difficulty adjusted in real time to obtain a second adjustment result. Based on the matched difficulty adjustment command, the question-and-answer content is rewritten in real time. During the rewriting process, the original expression style is preserved, and the contextual coherence is checked and analyzed in real time. After rewriting, the similarity between the difficulty feature vectors before and after rewriting is calculated to check whether the rewritten content meets the difficulty level required by the command, thus verifying the difficulty of the rewritten content. Finally, the rewritten content is integrated to obtain the second adjustment result. Real-time rewriting of content through difficulty adjustment commands ensures that the rewritten content maintains a consistent style with the original content, adapts to the difficulty requirements of the command, ensures logical coherence, and improves the quality of the rewritten content.

[0119] Furthermore, combining the capability analysis results and dynamic question-and-answer results, the capabilities of the target personnel are assessed using a pre-set capability assessment model, generating a capability assessment report, including:

[0120] S901. Perform contextual analysis on the capability analysis results and dynamic question-and-answer results to extract the corresponding capability indicator set;

[0121] S902. Based on the set of capability indicators, assess the capabilities of the target personnel using a preset capability assessment model and generate a capability assessment report.

[0122] In this embodiment, contextual correlation analysis is performed on the capability analysis results and dynamic question-and-answer results to extract the corresponding capability indicator set. A Transformer-based contextual correlation analysis model is constructed and trained using a large amount of historical capability analysis results and dynamic question-and-answer results to obtain a pre-trained contextual correlation analysis model. The capability analysis results and dynamic question-and-answer results are input into the pre-trained contextual correlation analysis model, which analyzes the correlation between the results and extracts capability indicators, including but not limited to technical depth, logical completeness, and practical adaptability. The capability indicators are then integrated to obtain a capability indicator set. The model can quickly analyze the correlation between results and extract corresponding capability indicators, providing a data foundation for the capability analysis of target personnel. Combining capability analysis results with dynamic question-and-answer results to analyze the capabilities of target personnel can improve the accuracy and comprehensiveness of the analysis results.

[0123] Specifically, based on a set of competency indicators, the competencies of target personnel are assessed using a pre-defined competency assessment model, generating a competency assessment report. This model includes, but is not limited to, a multi-layer fully connected neural network (MLN) model. The MNN model is trained using a large amount of competency indicator data to obtain a pre-trained MNN model. The competency indicator set is then input into this pre-trained model, which combines multiple competency indicators to calculate the target personnel's comprehensive competency score. Based on this score, a detailed competency assessment report is generated. By integrating the competency indicators from the competency indicator set through the competency assessment model, the accuracy and comprehensiveness of the competency analysis results can be improved, and the generated competency assessment report can comprehensively reflect the competencies of the target personnel.

[0124] like Figure 4 As shown, an online personnel verification system based on a large model is used to implement an online personnel verification method based on a large model, including:

[0125] The data analysis module extracts features from the multimodal data of the target personnel to be acquired and constructs multimodal feature vectors.

[0126] The capability analysis module, based on the multimodal feature vector, analyzes the target capability and behavioral intent of the target personnel through the execution sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, and obtains the capability analysis results;

[0127] The question-and-answer adjustment module configures a dynamic question-and-answer mechanism based on the capability analysis results, dynamically adjusting the question-and-answer content of the target personnel to obtain dynamic question-and-answer results;

[0128] The personnel review module combines the capability analysis results and dynamic question-and-answer results to assess the capabilities of target personnel through a preset capability assessment model, generating a capability assessment report for online personnel review.

[0129] In this embodiment, the data analysis module analyzes and extracts features from the acquired multimodal raw data of the target personnel. Using a feature extraction model, features are extracted from text, code, and audio data, and the feature vectors from different modalities are fused into a multimodal feature vector. Feature fusion avoids the limitations of single-modal features, and the complementary fusion of multi-source data improves the comprehensiveness and accuracy of the description of the target personnel, thereby enhancing the accuracy and comprehensiveness of the personnel review process. Based on the output of the data analysis module, the capability analysis module performs parallel analysis in a two-layer sandbox architecture. The execution sandbox is responsible for verifying the actual runtime performance of the code in a secure, isolated environment and evaluating the personnel's technical implementation capabilities. The trajectory analysis sandbox analyzes the corresponding coding style and potential behavioral intentions by monitoring the internal logical path of the code execution process. This two-layer sandbox architecture provides dual verification of the target personnel's capabilities and intentions, deepening the analysis of the target personnel's comprehensive capabilities and improving the comprehensiveness and accuracy of the personnel capability review process.

[0130] Specifically, the Q&A adjustment module configures corresponding dynamic Q&A adjustment strategies based on the output of the capability analysis module. During the Q&A process, it identifies scenario switching points and difficulty adjustment points, and generates new Q&A content in real time that matches the target personnel's capabilities, taking into account the context. By dynamically adjusting the personnel assessment process, in-depth assessment and verification can be conducted focusing on the target personnel's strengths and weaknesses, thereby improving the accuracy and efficiency of the personnel capability review process. The personnel review module receives capability analysis results and dynamic Q&A results. Through a pre-set, trained capability assessment model, it integrates, weights, and calculates capability review indicators to generate a structured capability assessment report. By combining capability analysis results and dynamic Q&A results, a comprehensive capability assessment report reflecting the target personnel's overall capabilities can be obtained, thereby improving the rationality, fairness, and reliability of personnel review decisions.

[0131] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A method for online personnel verification based on a large model, characterized in that, include: Feature extraction is performed on the multimodal data of the target personnel to be acquired, and a multimodal feature vector is constructed. Based on the multimodal feature vectors, the target capabilities and behavioral intentions of the target personnel are analyzed through the execution sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, to obtain capability analysis results; Based on the capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the question-and-answer content of the target personnel, thereby obtaining dynamic question-and-answer results; Based on the capability analysis results and dynamic question-and-answer results, the capabilities of the target personnel are assessed using a preset capability assessment model, and a capability assessment report is generated for online review of the personnel.

2. The online personnel verification method based on a large model according to claim 1, characterized in that, The step of extracting features from the multimodal data of the pre-acquired target personnel and constructing a multimodal feature vector includes: Acquire multimodal data of the target personnel, wherein the multimodal data includes text data, code data, and audio data; The text data, code data, and audio data are respectively extracted using a pre-defined feature extraction model to obtain a first feature set, a second feature set, and a third feature set. By combining the first feature set, the second feature set, and the third feature set, a multimodal feature vector is constructed.

3. The online personnel verification method based on a large model according to claim 2, characterized in that, Based on the multimodal feature vectors, the target personnel's capabilities and behavioral intentions are analyzed through the execution sandbox and trajectory analysis sandbox in a pre-constructed two-layer sandbox architecture, respectively, to obtain capability analysis results, including: Based on multimodal feature vectors, a two-layer sandbox architecture including an execution sandbox and a trajectory analysis sandbox is constructed. The code data from the multimodal data is input into the execution sandbox to simulate the execution of the code and analyze the target capabilities of the target personnel to obtain the first target capability; Input the code data into the trajectory analysis sandbox, collect the code execution process trajectory, analyze the target person's behavioral intention, and obtain the first behavioral intention; A comparative analysis of the first target capability and the first behavioral intention is conducted to identify the target personnel's strengths and weaknesses, and to obtain the capability analysis results.

4. The online personnel verification method based on a large model according to claim 3, characterized in that, The step involves inputting code data from the multimodal data into the execution sandbox, simulating the execution of the code, and analyzing the target personnel's target capabilities to obtain the first target capability, including: The code data from the multimodal data is input into the execution sandbox to simulate the execution of the code, and the code execution metrics are collected during the execution process. Based on the code execution metrics, the time and space complexity are calculated, and the target capabilities of the target personnel are analyzed to obtain the first target capability.

5. The online personnel verification method based on a large model according to claim 3, characterized in that, The step of inputting code data into the trajectory analysis sandbox, collecting the code execution process trajectory, analyzing the target person's behavioral intent, and obtaining the first behavioral intent includes: The code data is input into the trajectory analysis sandbox for execution, and trajectory data is collected during the code execution process; Based on the trajectory data, the coding style and logical correlation are identified through a preset trajectory analysis model, and a trajectory index vector is constructed. Based on the trajectory indicator vector analysis, the target person's behavioral intention is obtained, and the first behavioral intention is obtained.

6. The online personnel verification method based on a large model according to claim 1, characterized in that, Based on the capability analysis results, a dynamic question-and-answer mechanism is configured to dynamically adjust the question-and-answer content of the target personnel, resulting in dynamic question-and-answer results, including: Based on the capability analysis results, identify the capability strengths and weaknesses, and set scene switching points and difficulty adjustment points at the corresponding locations. At the scene switching point, the scene switching content in the question and answer content is identified, and a replacement scene is generated to replace the scene switching content, thereby obtaining the first adjustment result; At the difficulty adjustment point, the matching result between the question and answer content and the target personnel's ability is analyzed, and a corresponding difficulty adjustment instruction is generated to adjust the difficulty in real time, thus obtaining the second adjustment result. By combining the first and second adjustment results, the question-and-answer content of the target personnel is dynamically adjusted to obtain dynamic question-and-answer results.

7. The online personnel verification method based on a large model according to claim 6, characterized in that, At the scene transition point, the scene transition content in the question and answer content is identified, and a replacement scene is generated to replace the scene transition content, resulting in a first adjustment result, including: At the scene switching point, by analyzing the content corresponding to the strengths and weaknesses of the capabilities, the scene switching content in the question and answer content can be identified; Based on the contextual association of the scene switching content, a corresponding replacement scene is generated; The scene switching content is dynamically replaced according to the replacement scene, and the corresponding scene content is reconstructed to obtain the first adjustment result.

8. The online personnel verification method based on a large model according to claim 6, characterized in that, At the difficulty adjustment point, the matching result between the question and answer content and the target person's ability is analyzed, and a corresponding difficulty adjustment instruction is generated for real-time difficulty adjustment to obtain a second adjustment result, including: At the difficulty adjustment point, the matching result between the question and answer content and the target personnel's ability is analyzed to obtain the first matching result; Based on the first matching result, match the corresponding difficulty adjustment instruction in the pre-built difficulty instruction library; In response to the difficulty adjustment command, the difficulty of the question and answer content is rewritten and adjusted in real time to obtain the second adjustment result.

9. The online personnel verification method based on a large model according to claim 1, characterized in that, Based on the aforementioned ability analysis results and dynamic question-and-answer results, the abilities of the target personnel are assessed using a pre-set ability assessment model, generating an ability assessment report, including: Contextual analysis is performed on the ability analysis results and dynamic question-and-answer results to extract the corresponding set of ability indicators; Based on the set of competency indicators, the competency of the target personnel is assessed using a preset competency assessment model, and a competency assessment report is generated.

10. An online personnel verification system based on a large model, characterized in that, A method for implementing an online personnel verification method based on a large model as described in any one of claims 1 to 9, comprising: The data analysis module extracts features from the multimodal data of the target personnel to be acquired and constructs multimodal feature vectors. The capability analysis module, based on the multimodal feature vector, analyzes the target capability and behavioral intent of the target personnel through the execution sandbox and trajectory analysis sandbox in the pre-constructed two-layer sandbox architecture, respectively, and obtains the capability analysis results; The question-and-answer adjustment module configures a dynamic question-and-answer mechanism based on the capability analysis results, dynamically adjusting the question-and-answer content of the target personnel to obtain dynamic question-and-answer results; The personnel review module combines the capability analysis results and dynamic question-and-answer results to assess the capabilities of target personnel through a preset capability assessment model, generating a capability assessment report for online personnel review.

Images