Confidence-driven adaptive interview evaluation method, device, medium and equipment

Abstract

The application discloses a confidence-driven adaptive interview evaluation method, device, medium and equipment, relates to the technical field of interview evaluation, and comprises the following steps: obtaining at least one knowledge tree matched with an interviewee based on a resume and a knowledge graph of the interviewee; determining a resume prior probability of each knowledge node in the knowledge tree according to a matching condition of a resume keyword and the preset knowledge graph, and determining a target branch of each knowledge tree; for each parent knowledge node in each knowledge tree and each child knowledge node in the target branch, generating a question corresponding to each knowledge node in a top-down order of the knowledge nodes, and calculating a comprehensive score confidence corresponding to each knowledge node based on the resume prior probability and an answer corresponding to the question; and updating the comprehensive score confidence of other knowledge nodes in the knowledge tree according to the comprehensive score confidence of the knowledge node at the lowermost layer in the target branch, and generating an interview evaluation report. The application improves the accuracy of interview evaluation.

Description

Technical Field

[0001] This invention relates to the field of interview assessment technology, and in particular to a confidence-driven adaptive interview assessment method, apparatus, medium, and device. Background Technology

[0002] With the increasing penetration of artificial intelligence technology into the field of human resources, intelligent interview systems based on large language models have gradually become a research hotspot.

[0003] Currently, intelligent interviews based on large language models mainly generate multiple questions by statically associating questions with job skills using knowledge graphs, and then use the large model to generate natural and fluent dialogues, and evaluate candidates based on their answers.

[0004] However, when applied to in-depth competency assessments for technical positions, the aforementioned existing technologies suffer from a disconnect between the questioning path and the domain knowledge structure. Current adaptive questioning typically relies solely on the dialogue context for generation, making it difficult to accurately probe a candidate's grasp of technical details. Summary of the Invention

[0005] In view of this, the present invention provides a confidence-driven adaptive interview assessment method, apparatus, medium and device, the main purpose of which is to solve the problem that current interview assessment methods based on large language models have difficulty in accurately detecting the candidate's mastery of technical details.

[0006] According to one aspect of this application, a confidence-driven adaptive interview evaluation method is provided, the method comprising: Obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, match the resume keywords with a preset knowledge graph to obtain candidate knowledge points that match the interviewee, wherein the candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and multiple child knowledge nodes on branches; Based on the matching of the resume keywords with the preset knowledge graph, determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree, and determine the target branch of each knowledge tree; For each parent knowledge node in the knowledge tree and each child knowledge node in the target branch, questions corresponding to each knowledge node are generated in the order of knowledge nodes from top to bottom. The confidence level is calculated based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node, so as to obtain the comprehensive score confidence level corresponding to each knowledge node. Based on the comprehensive score confidence level corresponding to the lowest-level knowledge node in the target branch, the comprehensive score confidence level of other knowledge nodes in the knowledge tree is updated. An interview evaluation report is generated based on the question-and-answer results of each knowledge node in the knowledge tree and the final comprehensive score confidence level.

[0007] Optionally, the prior probability of the resume includes prior probability values ​​at multiple probability levels; the calculation of confidence based on the answer to the question and the prior probability of the resume corresponding to each knowledge node, to obtain the comprehensive score confidence for each knowledge node, includes: The answer to the question is input into a preset scoring prediction model to obtain the observation score corresponding to the knowledge node; The Gaussian likelihood value for each probability level is calculated based on the observation score corresponding to the knowledge node. The posterior probability for each probability level is calculated based on the Gaussian likelihood value and the prior probability value. The maximum posterior probability is taken as the Bayesian confidence of the knowledge node. Calculate the node abstraction value corresponding to the knowledge node based on the number of knowledge nodes contained in the next layer of the knowledge node, and calculate the graph coverage confidence value corresponding to the knowledge node. The comprehensive score confidence of the knowledge node is calculated based on the Bayesian confidence, graph coverage confidence, and node abstraction value of the knowledge node.

[0008] Optionally, calculating the graph coverage confidence score corresponding to the knowledge node includes: When the knowledge node is a leaf node, the graph coverage confidence of the knowledge node is equal to the Bayes confidence. When the knowledge node is a non-leaf node, the graph coverage confidence of the knowledge node is zero.

[0009] Optionally, the comprehensive score confidence level corresponding to the sub-knowledge node can be calculated using the following method:

[0010] in, To assess the confidence level of the overall score, This represents the node abstraction level value. Bayesian confidence level, Spectrum coverage confidence level.

[0011] Optionally, updating the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch includes: When the confidence level of the comprehensive score corresponding to the knowledge node at the lowest level in the target branch is greater than or equal to the preset confidence level threshold, from bottom to top, for each level in the target branch other than the lowest level, the confidence level of the graph coverage corresponding to the knowledge node at the current level is updated based on the confidence level of the comprehensive score of the knowledge node at the next lower level, and the confidence level of the comprehensive score corresponding to the knowledge node at the current level is updated based on the updated confidence level of the graph coverage.

[0012] Optionally, updating the graph coverage confidence of the knowledge nodes in the current layer based on the comprehensive score confidence of the knowledge nodes in the next layer includes: When the knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average value as the graph coverage confidence of the knowledge node.

[0013] Optionally, updating the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch further includes: When the confidence level of the comprehensive score corresponding to the lowest level knowledge node in the target branch is less than the preset confidence level threshold, the number of questions and answers corresponding to the lowest level knowledge node is obtained. When the number of questions and answers corresponding to the lowest level knowledge node is less than the maximum number of questions and answers threshold, a new question is generated based on the lowest level knowledge node and the historical question and answer situation. Based on the answers to the new questions and the prior probabilities of the resumes, the confidence level is calculated to obtain the new comprehensive score confidence level corresponding to the knowledge node, and it is determined whether the new comprehensive score confidence level is greater than or equal to the preset confidence level threshold. If so, then for each layer of the target branch except the bottom layer, from bottom to top, the comprehensive score confidence of the knowledge node in the current layer is updated based on the comprehensive score confidence of the knowledge node in the next layer; If not, return to the step of obtaining the number of questions and answers corresponding to the lowest level knowledge node and regenerate a new comprehensive score confidence score until the number of questions and answers equals the maximum number threshold. Then, from bottom to top, for each level in the target branch except the lowest level, update the comprehensive score confidence score corresponding to the knowledge node of the current level based on the comprehensive score confidence score of the knowledge node of the next level.

[0014] Optionally, the confidence-driven adaptive interview assessment method further includes: When the confidence level of the final comprehensive score of the parent knowledge node is less than the preset confidence level threshold, a new target branch is determined among multiple branches, and the confidence level of the final comprehensive score of the parent knowledge node is recalculated until the confidence level of the final updated comprehensive score of the parent knowledge node is greater than or equal to the preset confidence level threshold or the question and answer of the top-level knowledge nodes of all branches are completed.

[0015] According to another aspect of this application, a confidence-driven adaptive interview assessment apparatus is provided, comprising: The knowledge matching module is used to obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, and match the resume keywords with a preset knowledge graph to obtain candidate knowledge points that match the interviewee. The candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and child knowledge nodes on multiple branches. The prior probability determination module is used to determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree based on the matching of the resume keywords with the preset knowledge graph, and to determine the target branch of each knowledge tree. The confidence calculation module is used to generate questions corresponding to each knowledge node in the parent knowledge node and each child knowledge node in the target branch in the order of knowledge nodes from top to bottom, and calculate the confidence based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node to obtain the comprehensive score confidence for each knowledge node. The confidence update module is used to update the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch. The report generation module generates an interview evaluation report based on the question-and-answer results of each knowledge node in the knowledge tree and the final comprehensive score confidence level.

[0016] Optionally, the confidence calculation module is further used for: The answer to the question is input into a preset scoring prediction model to obtain the observation score corresponding to the knowledge node; The Gaussian likelihood value for each probability level is calculated based on the observation score corresponding to the knowledge node. The posterior probability for each probability level is calculated based on the Gaussian likelihood value and the prior probability value. The maximum posterior probability is taken as the Bayesian confidence of the knowledge node. Calculate the node abstraction value corresponding to the knowledge node based on the number of knowledge nodes contained in the next layer of the knowledge node, and calculate the graph coverage confidence value corresponding to the knowledge node. The comprehensive score confidence of the knowledge node is calculated based on the Bayesian confidence, graph coverage confidence, and node abstraction value of the knowledge node.

[0017] Optionally, the confidence calculation module is further used for: When the knowledge node is a leaf node, the graph coverage confidence of the knowledge node is equal to the Bayes confidence. When the knowledge node is a non-leaf node, the graph coverage confidence of the knowledge node is zero.

[0018] Optionally, the comprehensive score confidence level corresponding to the sub-knowledge node can be calculated using the following method:

[0019] in, To assess the confidence level of the overall score, This represents the node abstraction level value. Bayesian confidence level, Spectrum coverage confidence level.

[0020] Optionally, the confidence update module is further configured to: When the confidence level of the comprehensive score corresponding to the knowledge node at the lowest level in the target branch is greater than or equal to the preset confidence level threshold, from bottom to top, for each level in the target branch other than the lowest level, the confidence level of the graph coverage corresponding to the knowledge node at the current level is updated based on the confidence level of the comprehensive score of the knowledge node at the next lower level, and the confidence level of the comprehensive score corresponding to the knowledge node at the current level is updated based on the updated confidence level of the graph coverage.

[0021] Optionally, the confidence update module is further configured to: when a knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average as the graph coverage confidence of the knowledge node.

[0022] Optionally, the confidence update module is further configured to: When the confidence level of the comprehensive score corresponding to the lowest level knowledge node in the target branch is less than the preset confidence level threshold, the number of questions and answers corresponding to the lowest level knowledge node is obtained. When the number of questions and answers corresponding to the lowest level knowledge node is less than the maximum number of questions and answers threshold, a new question is generated based on the lowest level knowledge node and the historical question and answer situation. Based on the answers to the new questions and the prior probabilities of the resumes, the confidence level is calculated to obtain the new comprehensive score confidence level corresponding to the knowledge node, and it is determined whether the new comprehensive score confidence level is greater than or equal to the preset confidence level threshold. If so, then for each layer of the target branch except the bottom layer, from bottom to top, the comprehensive score confidence of the knowledge node in the current layer is updated based on the comprehensive score confidence of the knowledge node in the next layer; If not, return to the step of obtaining the number of questions and answers corresponding to the lowest level knowledge node and regenerate a new comprehensive score confidence score until the number of questions and answers equals the maximum number threshold. Then, from bottom to top, for each level in the target branch except the lowest level, update the comprehensive score confidence score corresponding to the knowledge node of the current level based on the comprehensive score confidence score of the knowledge node of the next level.

[0023] Optionally, the confidence update module is further configured to: When the confidence level of the final comprehensive score of the parent knowledge node is less than the preset confidence level threshold, a new target branch is determined among multiple branches, and the confidence level of the final comprehensive score of the parent knowledge node is recalculated until the confidence level of the final updated comprehensive score of the parent knowledge node is greater than or equal to the preset confidence level threshold or the question and answer of the top-level knowledge nodes of all branches are completed.

[0024] According to another aspect of this application, a storage medium is provided that stores at least one executable instruction, which causes a processor to perform operations corresponding to the confidence-driven adaptive interview evaluation method described above.

[0025] According to another aspect of this application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the confidence-driven adaptive interview evaluation method described above.

[0026] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages: This application provides a confidence-driven adaptive interview assessment method, apparatus, medium, and device. Based on the matching results of the interviewee's resume and knowledge graph, at least one knowledge tree matching the interviewee is obtained. According to the matching of resume keywords with the preset knowledge graph, the prior probability of each knowledge node in the knowledge tree is determined, and the target branch of each knowledge tree is determined. For each knowledge node in the knowledge tree, questions corresponding to each knowledge node are generated in a top-down order. Based on the prior probability of the resume and the answers to the questions, the comprehensive score confidence of each knowledge node is calculated. Based on the comprehensive score confidence of the lowest-level knowledge node in the target branch, the comprehensive score confidence of other knowledge nodes in the knowledge tree is updated, and an interview assessment report is generated. This maps the candidate's interview process to a structured knowledge graph, ensuring that interview questions, answer recognition, scoring, and reporting all revolve around knowledge points. Questions and answers are conducted layer by layer according to the knowledge graph structure, accurately probing the candidate's mastery of technical details and improving the accuracy of the interview assessment. Using prior probability of the resume and answer results to calculate the score confidence reduces reliance on the self-reported confidence of large language models, further improving the accuracy of the interview assessment.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0028] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application is shown; Figure 2 Another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application is shown; Figure 3 This paper illustrates another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application; Figure 4 This paper illustrates another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application; Figure 5 Another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application is shown; Figure 6 This paper illustrates another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application; Figure 7 This paper illustrates another flowchart of a confidence-driven adaptive interview assessment method provided in an embodiment of this application; Figure 8 This illustration shows a block diagram of a confidence-driven adaptive interview assessment device provided in an embodiment of this application. Figure 9 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is shown.

[0029] in, Figure 8 Chinese module: 802 - Knowledge Matching Module; 804 - Prior Probability Determination Module; 806 - Confidence Calculation Module; 808 - Confidence Update Module; 810 - Report Generation Module; Figure 9 In Chinese: 902 - Processor; 904 - Communication interface; 906 - Memory; 908 - Communication bus; 910 - Program. Detailed Implementation

[0030] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0031] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0032] To address the problem that current interview assessment methods based on large language models struggle to accurately assess a candidate's grasp of technical details, this application provides a confidence-driven adaptive interview assessment method, such as... Figure 1 As shown, the method includes: 102: Obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, match the resume keywords with the preset knowledge graph to obtain candidate knowledge points that match the interviewee. Among them, the candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and child knowledge nodes on multiple branches. 104: Based on the matching results between resume keywords and the preset knowledge graph, determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree, and determine the target branch of each knowledge tree; 106: For each parent knowledge node in the knowledge tree and each child knowledge node in the target branch, generate questions corresponding to each knowledge node in the order of knowledge nodes from top to bottom. Calculate the confidence level based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node to obtain the comprehensive score confidence level corresponding to each knowledge node. 108: Update the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch; 110: Generate an interview evaluation report based on the question-and-answer results of each knowledge node in the knowledge tree and the final comprehensive score confidence level.

[0033] Specifically, the system receives candidate resume data and domain knowledge graph data. The knowledge graph consists of multiple knowledge points, each containing multiple knowledge trees. A knowledge tree consists of a parent knowledge node and multiple branches, with each branch including at least one level of child knowledge nodes. The system converts the knowledge point names or descriptions in the knowledge graph into vectors. When performing knowledge point matching, the input text is converted into a query vector, and the similarity between the query vector and the knowledge point vector is calculated.

[0034] Extract skills, projects, education, and work experience from the interviewee's resume, convert the resume content into semantic summary sentences, generate summary vectors from the semantic summary sentences, search for similar knowledge points in the knowledge graph vector based on the summary vectors, filter candidate knowledge points based on the search results, select several interview directions from the candidate knowledge points, usually each interview direction corresponds to a knowledge tree, and generate initial questions for each interview direction to form an interview plan.

[0035] Based on the matching degree between the identified resume content and the knowledge graph, the prior probability of the resume in the scoring confidence calculation is set, so that the knowledge points of the resume and its direct child nodes in the mission can obtain priors of different strengths.

[0036] For each defined knowledge tree, a preset number of target branches are selected from multiple branches. For each target branch, questions and answers are conducted sequentially on the knowledge nodes in the order from top to bottom. The comprehensive score confidence of the knowledge node is calculated based on the answers and the prior probability of the resume.

[0037] After obtaining the overall score confidence score of the lowest-level knowledge node, backtracking is performed. Based on the overall score confidence score of the lowest-level knowledge node, the overall score confidence score of the knowledge node above it is updated. After the overall score confidence score of the knowledge node above it is updated, the overall score confidence score of the knowledge node above it is updated again based on the updated overall score confidence score of the knowledge node above it, and so on, until the overall score confidence score of the parent knowledge node is obtained. The overall score confidence score of the knowledge node that is no longer updated is taken as the final overall score confidence score of that knowledge node.

[0038] Each knowledge tree repeats the above operations, generating an interview evaluation report based on the question-and-answer results of knowledge nodes in each knowledge tree and the final comprehensive score confidence level. Each interview question is tied to a knowledge point, and each answer is mapped to one or more knowledge points, transforming the interview process from natural language dialogue into structured graph exploration.

[0039] This application provides a confidence-driven adaptive interview assessment method. Compared with existing technologies, it obtains at least one knowledge tree matching the interviewee based on the matching results of the interviewee's resume and knowledge graph. Based on the matching of resume keywords with the preset knowledge graph, it determines the resume prior probability of each knowledge node in the knowledge tree and identifies the target branch of each knowledge tree. For each knowledge node in the knowledge tree, it generates corresponding questions in a top-down order. Based on the resume prior probability and the answers to the questions, it calculates the comprehensive score confidence of each knowledge node. Based on the comprehensive score confidence of the lowest-level knowledge node in the target branch, it updates the comprehensive score confidence of other knowledge nodes in the knowledge tree, generating an interview assessment report. This method maps the candidate's interview process to a structured knowledge graph, ensuring that interview questions, answer recognition, scoring, and reporting all revolve around knowledge points. Questions and answers are conducted layer by layer according to the knowledge graph structure, accurately probing the candidate's mastery of technical details and improving the accuracy of the interview assessment. Using resume prior probability and answering results to calculate the score confidence reduces reliance on the self-reported confidence of large language models, further improving the accuracy of the interview assessment.

[0040] In one embodiment of the present invention, such as Figure 2 As shown, the prior probability of a resume includes prior probability values ​​at multiple probability levels; based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node, the confidence level is calculated to obtain the comprehensive score confidence level corresponding to each knowledge node, including: 202: Input the answer to the question into the preset scoring prediction model to obtain the observed score corresponding to the knowledge node; 204: Calculate the Gaussian likelihood value for each probability level based on the observed scores corresponding to the knowledge node. Calculate the posterior probability for each probability level based on the Gaussian likelihood value and the prior probability value. Use the maximum posterior probability as the Bayesian confidence of the knowledge node. 206: Calculate the node abstraction value corresponding to the knowledge node based on the number of knowledge nodes contained in the next layer of the knowledge node, and calculate the graph coverage confidence corresponding to the knowledge node; 208: The comprehensive score confidence of a knowledge node is calculated based on the Bayesian confidence score, graph coverage confidence score, and node abstraction score of the knowledge node.

[0041] In this embodiment, calculating the graph coverage confidence level corresponding to the sub-knowledge node includes: When a knowledge node is a leaf node, the graph coverage confidence of the knowledge node is equal to the Bayes confidence. When a knowledge node is a non-leaf node, the confidence level of its graph coverage is zero.

[0042] In one embodiment, the comprehensive score confidence of other knowledge nodes in the knowledge tree is updated based on the comprehensive score confidence of the lowest-level knowledge node in the target branch, including: When a knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average as the graph coverage confidence of the knowledge node.

[0043] In this embodiment, the prior probability of the resume includes multiple probabilities, such as setting six levels, with score as the level, score∈{0, 1, 2, 3, 4, 5}. The meanings of the ability levels are as follows: 0: No relevant ability demonstrated; 1: Understands the concept; 2: Can describe the basic process; 3: Has practical experience; 4: Can handle complex problems; 5: Has in-depth design and optimization capabilities.

[0044] Based on the matching between the resume and the knowledge points, the following probability distribution is set for each knowledge point: P(score=0),P(score=1),P(score=2),P(score=3),P(score=4),P(score=5); Where P() is the resume prior probability. Knowledge points identified as relevant to the resume are assigned the following resume hit priors: score=0: 0.03; score=1: 0.07; score=2: 0.20; score=3: 0.35; score=4: 0.25; score=5: 0.10.

[0045] The direct child nodes of the knowledge points hit in the resume use weak resume hit priors: score=0: 0.06; score=1: 0.12; score=2: 0.25; score=3: 0.32; score=4: 0.18; score=5: 0.07.

[0046] For knowledge points not mentioned in the resume, use a uniform prior: score=0: 0.167; score=1: 0.167; score=2: 0.167; score=3: 0.167; score=4: 0.166; score=5: 0.166.

[0047] If a knowledge point satisfies multiple prior sources simultaneously, the prior sources are selected according to the following priority: Resume-based prior hit > Weak resume-based prior hit > Uniform prior hit

[0048] Based on the answer, determine the current main knowledge point and additional knowledge points. Input the answer corresponding to the question into the preset scoring prediction model to obtain the observation score corresponding to the knowledge node, including the observation score corresponding to the current main knowledge point and each additional knowledge point. Perform Bayesian update on the observation scores corresponding to the current main knowledge point and each additional knowledge point respectively.

[0049] First, calculate the Gaussian likelihood value for each knowledge point at each probability level using a fixed discrete Gaussian likelihood function:

[0050] Where: L i Let be the Gaussian likelihood value at probability level i. s obs Score the observations.

[0051] The posterior distribution is calculated using the following method:

[0052] Where E represents the evidence for the current response, Let represent the prior probability of the i-th probability level before the update. This represents the posterior probability of the i-th probability level after updating based on the current answer evidence.

[0053] Bayesian confidence scores are calculated using the maximum posterior probability.

[0054] in, Bayesian confidence level.

[0055] Calculate the knowledge graph coverage confidence score for each sub-knowledge node: Regarding the knowledge points about leaves ,in, Spectrum coverage confidence level.

[0056] For non-leaf knowledge points, the graph coverage confidence is calculated based on the comprehensive score confidence of the direct child nodes included in the knowledge node, avoiding assigning high confidence to abstract parent nodes based on only a few answers. Since the current knowledge graph node does not contain an explicit weight field, uniform weighting is used. :

[0057] The confidence level of the graph coverage of non-leaf knowledge points is:

[0058] In the formula, n represents the total number of direct child nodes included in the current knowledge node, and C i Let represent the overall score confidence level of the i-th direct child node; if the i-th child node is not evaluated, then let . .

[0059] For example, a knowledge node has 10 direct knowledge child nodes, and only two child nodes have been evaluated so far: the confidence scores of these two child nodes are 0.90 and 0.80 respectively, while the confidence scores of the other 8 child nodes are 0. Then:

[0060] Calculate the node abstraction level value based on the number of direct child nodes of the knowledge node:

[0061] in, is the node abstraction level value, and n is the number of direct child nodes.

[0062] The comprehensive confidence score is obtained by fusing the Bayesian confidence score and the knowledge graph coverage confidence score based on the node abstraction degree value:

[0063] in, This indicates the confidence level of the overall score. This represents the node's abstraction level. Indicates the Bayesian confidence level. This indicates the confidence level of the map coverage.

[0064] The confidence score is calculated using resume prior probability, Bayesian posterior distribution, and knowledge graph coverage, reducing the reliance on the self-reported confidence score of large language models.

[0065] In one embodiment of the invention, the comprehensive score confidence of other knowledge nodes in the knowledge tree is updated based on the comprehensive score confidence of the lowest-level knowledge node in the target branch, including: When the confidence level of the comprehensive score corresponding to the knowledge node at the lowest level in the target branch is greater than or equal to the preset confidence level threshold, from bottom to top, for each level in the target branch except the lowest level, the confidence level of the graph coverage corresponding to the knowledge node at the current level is updated based on the confidence level of the comprehensive score of the knowledge node at the next lower level, and the confidence level of the comprehensive score corresponding to the knowledge node at the current level is updated based on the updated confidence level of the graph coverage.

[0066] In this embodiment, the graph coverage confidence of the knowledge node in the current layer is updated based on the comprehensive score confidence of the knowledge node in the next layer. This includes: when the knowledge node is a non-leaf node, calculating the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and using the average as the graph coverage confidence of the knowledge node.

[0067] Specifically, the confidence level of the overall score corresponding to the lowest-level knowledge node in the target branch is compared with a preset confidence threshold. If the confidence level of the overall score corresponding to that knowledge node is greater than or equal to the preset confidence threshold, it indicates that the interviewee has a good grasp of this knowledge point. At this point, follow-up questions on that knowledge node cease, and backtracking scoring begins. This involves updating the graph coverage confidence level of the knowledge node above the lowest-level knowledge node based on the confidence level of the overall score of the lowest-level knowledge node in the target branch. The updated graph coverage confidence level is then used to update the overall score confidence level of the knowledge node above the next higher level. For other knowledge nodes in the target branch besides the lowest-level knowledge node, the overall score confidence level of the current knowledge node is updated sequentially based on the confidence level of the knowledge node below the next higher level. Once the confidence level of a knowledge point reaches the threshold, follow-up questions on the current branch cease, reducing repetitive questions.

[0068] In one embodiment of the invention, updating the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch further includes: When the confidence level of the comprehensive score corresponding to the lowest level knowledge node in the target branch is less than the preset confidence level threshold, the number of questions and answers corresponding to the lowest level knowledge node is obtained. When the number of questions and answers corresponding to the lowest level knowledge node is less than the maximum number threshold, a new question is generated based on the lowest level knowledge node and the historical question and answer situation. The confidence level is calculated based on the answers to the new questions and the prior probabilities of the resumes to obtain the new comprehensive score confidence level for the knowledge nodes. It is then determined whether the new comprehensive score confidence level is greater than or equal to the preset confidence level threshold. If so, then for each layer in the target branch except the bottom layer, from bottom to top, update the comprehensive score confidence of the knowledge node corresponding to the current layer based on the comprehensive score confidence of the knowledge node in the next layer; If not, return to the step of obtaining the number of questions and answers corresponding to the knowledge node at the lowest level and regenerate a new comprehensive score confidence score until the number of questions and answers equals the maximum number threshold. Then, from bottom to top, for each level in the target branch except the lowest level, update the comprehensive score confidence score corresponding to the knowledge node at the current level based on the comprehensive score confidence score of the knowledge node at the next lower level.

[0069] Specifically, if the overall score confidence level of the lowest-level knowledge node in the target branch is less than a preset confidence level threshold, the number of questions and answers corresponding to the lowest-level knowledge node (also called the maximum depth) is obtained. When the number of questions and answers corresponding to this knowledge node is less than the maximum number threshold, a new question is generated based on the question and answer situation of this knowledge node, and the overall score confidence level is calculated based on the question and answer situation of the new question. If the number of questions and answers corresponding to this knowledge node is greater than or equal to the maximum number threshold, then follow-up questions for this knowledge node are stopped, and question and answer questions for other knowledge nodes or the overall score confidence level of other knowledge nodes are calculated.

[0070] If the number of questions and answers corresponding to a knowledge node is less than the maximum threshold, then the follow-up questions for that knowledge node are stopped and questions and answers for other knowledge nodes are conducted. The confidence level of the comprehensive score of the knowledge node above that knowledge node is updated.

[0071] For all knowledge nodes in the target branch except the lowest level knowledge node, the overall score confidence of the current knowledge node is updated sequentially based on the overall score confidence of the knowledge node in the next lower level. Note that, like the lowest level knowledge node, the other knowledge nodes in the target branch are also evaluated based on the number of answers and the overall score confidence to determine whether to continue asking follow-up questions. The overall score confidence of the knowledge node in the previous level is updated based on the overall score confidence of the current level knowledge node.

[0072] In one embodiment, the interviewee inputs their resume, receives questions, and inputs answers through a client. The server-side state management module manages the interview, including an interview state management module, an exploration stack, and a backtracking stack. The knowledge graph and index layer stores the knowledge graph, including a knowledge graph module and a knowledge graph vector index. The process agent layer is primarily responsible for question answering, including an interview planning agent, a knowledge matching agent, an observation and scoring agent, and a follow-up question generation agent. The scoring confidence update module calculates the comprehensive scoring confidence based on the question-and-answer situation. The comprehensive scoring confidence, also known as the final scoring confidence, determines whether to continue asking follow-up questions and whether to return to the backtracking scoring module. If returning to the backtracking scoring module, it updates the final confidence of the knowledge node's parent layer. If backtracking is not required, the report generation module generates an interview report. Specifically, as shown below... Figure 3 As shown, multiple intelligent agents are used to complete interview planning, knowledge point matching, observation scoring, and follow-up question generation, thereby improving the modularity and controllability of the system.

[0073] In one embodiment, such as Figure 4 As shown, for the lowest-level knowledge node or the knowledge node whose comprehensive score confidence has been updated, it is determined whether the comprehensive score confidence is greater than a preset threshold. If it is greater than or equal to the preset threshold, the system returns to the backtracking scoring module to update the comprehensive score confidence of the knowledge node at the next higher level. If it is less than the preset threshold, the system obtains the follow-up question depth, also known as the number of questions and answers. When the number of follow-up questions reaches the maximum, the system stops following up on the knowledge node and updates the comprehensive score confidence of the knowledge node at the next higher level based on the comprehensive score confidence of that knowledge node. If the number of follow-up questions has not reached the maximum, a new question is generated based on the historical question and answer situation, and the follow-up question is initiated.

[0074] In one embodiment, when a candidate's answer is insufficient, dynamic follow-up questions are asked along the knowledge graph's sub-nodes. The follow-up questioning path is controlled by a DFS (Depth-First Search) exploration stack. The dynamic follow-up questioning process based on the DFS exploration stack is as follows: Figure 5 As shown, the calculation process for rating confidence is as follows: Figure 6 As shown, the backtracking scoring process is as follows: Figure 7 As shown.

[0075] Once all planned knowledge points have been completed, the system generates a structured interview report. The interview report should include at least the following: total number of questions, number of planned nodes covered, number of knowledge points evaluated, average score, score for each knowledge point, final confidence level for each knowledge point, posterior distribution, contribution question and answer records, and candidate competency graph.

[0076] In one embodiment of the present invention, the confidence-driven adaptive interview assessment method further includes: When the confidence level of the final comprehensive score of the parent knowledge node is less than the preset confidence level threshold, a new target branch is determined among multiple branches, and the confidence level of the final comprehensive score of the parent knowledge node is recalculated until the confidence level of the final updated comprehensive score of the parent knowledge node is greater than or equal to the preset confidence level threshold or the question and answer of the top-level knowledge nodes of all branches are completed.

[0077] Specifically, if the overall score confidence of the parent knowledge node is lower than the preset confidence threshold, it indicates that the current evaluation result is not reliable enough. Among the existing multiple branch structures, a new target branch is selected as the next evaluation object. For the new target branch, the same method as the original target branch is used, calculating the overall score confidence of each knowledge node in the new target branch in a top-down order. After the overall score confidence of the top-level child knowledge node is calculated, the graph confidence of the parent knowledge node is updated. Based on the updated graph confidence, the overall score confidence of the parent knowledge node is updated. It is then determined whether the updated overall score confidence of the parent knowledge node is greater than the preset confidence threshold. If the updated overall score confidence of the parent knowledge node is still less than the preset confidence threshold, a new target branch is found, and the above steps are repeated until the updated overall score confidence of the parent knowledge node is greater than the preset threshold.

[0078] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides a confidence-driven adaptive interview assessment device, such as... Figure 8 As shown, the device includes: The knowledge matching module 802 is used to obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, match the resume keywords with a preset knowledge graph to obtain candidate knowledge points that match the interviewee. The candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and child knowledge nodes on multiple branches. The prior probability determination module 804 is used to determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree based on the matching of resume keywords with the preset knowledge graph, and to determine the target branch of each knowledge tree. The confidence calculation module 806 is used to generate questions corresponding to each knowledge node in the order of top to bottom for each parent knowledge node in the knowledge tree and each child knowledge node in the target branch. Based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node, the confidence is calculated to obtain the comprehensive score confidence of each knowledge node. The confidence update module 808 is used to update the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch. The report generation module 810 is used to generate an interview evaluation report based on the question-and-answer situation of each knowledge node in the knowledge tree and the final comprehensive score confidence level.

[0079] This application provides a confidence-driven adaptive interview assessment device. Compared with existing technologies, it obtains at least one knowledge tree matching the interviewee based on the matching results of the interviewee's resume and knowledge graph. Based on the matching of resume keywords with the preset knowledge graph, it determines the resume prior probability of each knowledge node in the knowledge tree and identifies the target branch of each knowledge tree. For each knowledge node in the knowledge tree, it generates corresponding questions in a top-down order. Based on the resume prior probability and the answers to the questions, it calculates the comprehensive score confidence of each knowledge node. Based on the comprehensive score confidence of the lowest-level knowledge node in the target branch, it updates the comprehensive score confidence of other knowledge nodes in the knowledge tree, generating an interview assessment report. This maps the candidate's interview process to a structured knowledge graph, ensuring that interview questions, answer recognition, scoring, and reporting all revolve around knowledge points. This accurately detects the candidate's mastery of technical details, improving the accuracy of the interview assessment. Using resume prior probability and answer information to calculate the score confidence reduces reliance on the self-reported confidence of large language models, further enhancing the accuracy of the interview assessment.

[0080] In one embodiment, the confidence calculation module is further configured to: The answers to the questions are input into a pre-set scoring prediction model to obtain the observed scores for the knowledge nodes. The Gaussian likelihood value for each probability level is calculated based on the observation score corresponding to the knowledge node. The posterior probability for each probability level is calculated based on the Gaussian likelihood value and the prior probability value. The maximum posterior probability is taken as the Bayesian confidence of the knowledge node. The node abstraction degree value corresponding to the knowledge node is calculated based on the number of knowledge nodes contained in the next layer of the knowledge node, and the graph coverage confidence degree corresponding to the knowledge node is also calculated. The comprehensive score confidence of a knowledge node is calculated based on its Bayes confidence, graph coverage confidence, and node abstraction value.

[0081] In one embodiment, the confidence calculation module is further configured to: When a child knowledge node is a leaf node, the graph coverage confidence of the child knowledge node is equal to the Bayes confidence. When a child knowledge node is a non-leaf node, the graph coverage confidence of the knowledge node is zero.

[0082] In one embodiment, the confidence update module is further configured to: When a knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average as the graph coverage confidence of the knowledge node.

[0083] In one embodiment, the comprehensive score confidence level corresponding to the sub-knowledge node is calculated using the following method:

[0084] in, To assess the confidence level of the overall score, This represents the node abstraction level value. Bayesian confidence level, Spectrum coverage confidence level.

[0085] In one embodiment, the confidence update module is further configured to: When the confidence level of the comprehensive score corresponding to the knowledge node at the lowest level in the target branch is greater than or equal to the preset confidence level threshold, from bottom to top, for each level in the target branch except the lowest level, the confidence level of the graph coverage corresponding to the knowledge node at the current level is updated based on the confidence level of the comprehensive score of the knowledge node at the next lower level, and the confidence level of the comprehensive score corresponding to the knowledge node at the current level is updated based on the updated confidence level of the graph coverage.

[0086] In one embodiment, the confidence update module is further configured to: when the knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average as the graph coverage confidence of the knowledge node.

[0087] In one embodiment, the confidence update module is further configured to: When the confidence level of the comprehensive score corresponding to the lowest level knowledge node in the target branch is less than the preset confidence level threshold, the number of questions and answers corresponding to the lowest level knowledge node is obtained. When the number of questions and answers corresponding to the lowest level knowledge node is less than the maximum number threshold, a new question is generated based on the lowest level knowledge node and the historical question and answer situation. The confidence level is calculated based on the answers to the new questions and the prior probabilities of the resumes to obtain the new comprehensive score confidence level for the knowledge nodes. It is then determined whether the new comprehensive score confidence level is greater than or equal to the preset confidence level threshold. If so, then for each layer in the target branch except the bottom layer, from bottom to top, update the comprehensive score confidence of the knowledge node corresponding to the current layer based on the comprehensive score confidence of the knowledge node in the next layer; If not, return to the step of obtaining the number of questions and answers corresponding to the knowledge node at the lowest level and regenerate a new comprehensive score confidence score until the number of questions and answers equals the maximum number threshold. Then, from bottom to top, for each level in the target branch except the lowest level, update the comprehensive score confidence score corresponding to the knowledge node at the current level based on the comprehensive score confidence score of the knowledge node at the next lower level.

[0088] In one embodiment, the confidence update module is further configured to: When the confidence level of the final comprehensive score of the parent knowledge node is less than the preset confidence level threshold, a new target branch is determined among multiple branches, and the confidence level of the final comprehensive score of the parent knowledge node is recalculated until the confidence level of the final updated comprehensive score of the parent knowledge node is greater than or equal to the preset confidence level threshold or the question and answer of the top-level knowledge nodes of all branches are completed.

[0089] According to one embodiment of the present invention, a storage medium is provided, the storage medium storing at least one executable instruction, the computer executable instruction being able to execute the confidence-driven adaptive interview assessment method in any of the above method embodiments.

[0090] Figure 9 The diagram illustrates a structural schematic of a computer device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0091] like Figure 9 As shown, the computer device may include: a processor 902, a communication interface 904, a memory 906, and a communication bus 908.

[0092] The processor 902, communication interface 904, and memory 906 communicate with each other via communication bus 908.

[0093] The communication interface 904 is used to communicate with other network elements such as clients or other servers.

[0094] The processor 902 is used to execute program 910, which can specifically execute the relevant steps in the above-described confidence-driven adaptive interview assessment method embodiment.

[0095] Specifically, program 910 may include program code that includes computer operation instructions.

[0096] Processor 902 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0097] Memory 906 is used to store program 910. Memory 906 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0098] Specifically, program 910 can be used to cause processor 902 to perform the following operations: The process involves obtaining the interviewee's resume, extracting keywords from the resume, matching the resume keywords with a pre-defined knowledge graph, and obtaining candidate knowledge points that match the interviewee. Each candidate knowledge point includes at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and child knowledge nodes on multiple branches. Based on the matching results between resume keywords and the preset knowledge graph, determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree, and determine the target branch of each knowledge tree. For each parent knowledge node in the knowledge tree and each child knowledge node in the target branch, questions corresponding to each knowledge node are generated in the order of knowledge nodes from top to bottom. The confidence level is calculated based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node, so as to obtain the comprehensive score confidence level corresponding to each knowledge node. Based on the comprehensive score confidence level of the lowest-level knowledge node in the target branch, update the comprehensive score confidence level of other knowledge nodes in the knowledge tree. An interview evaluation report is generated based on the question-and-answer results of each knowledge node in the knowledge tree and the final comprehensive score confidence level.

[0099] It will be apparent to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. In one embodiment, they can be implemented using device-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0100] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A confidence-driven adaptive interview assessment method, characterized in that, include: Obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, match the resume keywords with a preset knowledge graph to obtain candidate knowledge points that match the interviewee, wherein the candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and multiple child knowledge nodes on branches; Based on the matching of the resume keywords with the preset knowledge graph, determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree, and determine the target branch of each knowledge tree; For each parent knowledge node in the knowledge tree and each child knowledge node in the target branch, questions corresponding to each knowledge node are generated in the order of knowledge nodes from top to bottom. The confidence level is calculated based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node, so as to obtain the comprehensive score confidence level corresponding to each knowledge node. Based on the comprehensive score confidence level corresponding to the lowest-level knowledge node in the target branch, the comprehensive score confidence level of other knowledge nodes in the knowledge tree is updated. An interview evaluation report is generated based on the question-and-answer results of each knowledge node in the knowledge tree and the final comprehensive score confidence level. The prior probability of the resume includes prior probability values ​​under multiple ability levels; the confidence level is calculated based on the answers to the questions and the prior probability of the resumes corresponding to each knowledge node to obtain the comprehensive score confidence level corresponding to each knowledge node, including: The answer to the question is input into a preset scoring prediction model to obtain the observation score corresponding to the knowledge node; The Gaussian likelihood value for each ability level is calculated based on the observed score corresponding to the knowledge node. The posterior probability for each ability level is calculated based on the Gaussian likelihood value and the prior probability value. The maximum posterior probability is taken as the Bayesian confidence of the knowledge node. Calculate the node abstraction value corresponding to the knowledge node based on the number of knowledge nodes contained in the next layer of the knowledge node, and calculate the graph coverage confidence value corresponding to the knowledge node. The comprehensive score confidence of the knowledge node is calculated based on the Bayes confidence, graph coverage confidence, and node abstraction value of the knowledge node. Calculating the graph coverage confidence score corresponding to the knowledge node includes: When the knowledge node is a leaf node, the graph coverage confidence of the knowledge node is equal to the Bayes confidence. When the knowledge node is a non-leaf node, the graph coverage confidence is calculated based on the comprehensive score confidence of the direct child nodes included in the knowledge node.

2. The confidence-driven adaptive interview assessment method as described in claim 1, characterized in that, The confidence level of the comprehensive score corresponding to the knowledge node is calculated using the following method: in, To assess the confidence level of the overall score, This represents the node abstraction level value. Bayesian confidence level, Spectrum coverage confidence level.

3. The confidence-driven adaptive interview assessment method as described in claim 1, characterized in that, The step of updating the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch includes: When the confidence level of the comprehensive score corresponding to the knowledge node at the lowest level in the target branch is greater than or equal to the preset confidence level threshold, from bottom to top, for each level in the target branch other than the lowest level, the confidence level of the graph coverage corresponding to the knowledge node at the current level is updated based on the confidence level of the comprehensive score of the knowledge node at the next lower level, and the confidence level of the comprehensive score corresponding to the knowledge node at the current level is updated based on the updated confidence level of the graph coverage. The update of the graph coverage confidence of the knowledge nodes in the current layer based on the comprehensive score confidence of the knowledge nodes in the next layer includes: When the knowledge node is a non-leaf node, calculate the average of the comprehensive score confidence of the knowledge nodes contained in the next layer of the knowledge node, and use the average value as the graph coverage confidence of the knowledge node.

4. The confidence-driven adaptive interview assessment method as described in claim 1, characterized in that, The step of updating the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch further includes: When the confidence level of the comprehensive score corresponding to the lowest level knowledge node in the target branch is less than the preset confidence level threshold, the number of questions and answers corresponding to the lowest level knowledge node is obtained. When the number of questions and answers corresponding to the lowest level knowledge node is less than the maximum number of questions and answers threshold, a new question is generated based on the lowest level knowledge node and the historical question and answer situation. Based on the answers to the new questions and the prior probabilities of the resumes, the confidence level is calculated to obtain the new comprehensive score confidence level corresponding to the knowledge node, and it is determined whether the new comprehensive score confidence level is greater than or equal to the preset confidence level threshold. If so, then for each layer of the target branch except the bottom layer, from bottom to top, the comprehensive score confidence of the knowledge node in the current layer is updated based on the comprehensive score confidence of the knowledge node in the next layer; If not, return to the step of obtaining the number of questions and answers corresponding to the lowest level knowledge node and regenerate a new comprehensive score confidence score until the number of questions and answers equals the maximum number threshold. Then, from bottom to top, for each level in the target branch except the lowest level, update the comprehensive score confidence score corresponding to the knowledge node of the current level based on the comprehensive score confidence score of the knowledge node of the next level.

5. The confidence-driven adaptive interview assessment method as described in any one of claims 1-4, characterized in that, The confidence-driven adaptive interview assessment method also includes: When the confidence level of the final comprehensive score of the parent knowledge node is less than the preset confidence level threshold, a new target branch is determined among multiple branches, and the confidence level of the final comprehensive score of the parent knowledge node is recalculated until the confidence level of the final updated comprehensive score of the parent knowledge node is greater than or equal to the preset confidence level threshold or the question and answer of the top-level knowledge nodes of all branches are completed.

6. A confidence-driven adaptive interview assessment device, characterized in that, include: The knowledge matching module is used to obtain the interviewee's resume, extract keywords from the resume to obtain resume keywords, and match the resume keywords with a preset knowledge graph to obtain candidate knowledge points that match the interviewee. The candidate knowledge points include at least one knowledge tree, and each knowledge node in the knowledge tree includes a parent knowledge node and child knowledge nodes on multiple branches. The prior probability determination module is used to determine the prior probability of the resume corresponding to each knowledge node in each knowledge tree based on the matching of the resume keywords with the preset knowledge graph, and to determine the target branch of each knowledge tree. The confidence calculation module is used to generate questions corresponding to each knowledge node in the parent knowledge node and each child knowledge node in the target branch in the order of knowledge nodes from top to bottom, and calculate the confidence based on the answers to the questions and the prior probability of the resume corresponding to each knowledge node to obtain the comprehensive score confidence for each knowledge node. The confidence update module is used to update the comprehensive score confidence of other knowledge nodes in the knowledge tree based on the comprehensive score confidence of the lowest-level knowledge node in the target branch. The report generation module is used to generate an interview evaluation report based on the question-and-answer situation of each knowledge node in the knowledge tree and the final comprehensive score confidence level. The prior probability of the resume includes prior probability values ​​under multiple ability levels; the confidence calculation module is also used for: The answer to the question is input into a preset scoring prediction model to obtain the observation score corresponding to the knowledge node; The Gaussian likelihood value for each ability level is calculated based on the observed score corresponding to the knowledge node. The posterior probability for each ability level is calculated based on the Gaussian likelihood value and the prior probability value. The maximum posterior probability is taken as the Bayesian confidence of the knowledge node. Calculate the node abstraction value corresponding to the knowledge node based on the number of knowledge nodes contained in the next layer of the knowledge node, and calculate the graph coverage confidence value corresponding to the knowledge node. The comprehensive score confidence of the knowledge node is calculated based on the Bayes confidence, graph coverage confidence, and node abstraction value of the knowledge node. The confidence calculation module is also used for: When the knowledge node is a leaf node, the graph coverage confidence of the knowledge node is equal to the Bayes confidence. When the knowledge node is a non-leaf node, the graph coverage confidence is calculated based on the comprehensive score confidence of the direct child nodes included in the knowledge node.

7. A storage medium storing at least one executable instruction, characterized in that, The executable instructions cause the processor to perform the operations corresponding to the confidence-driven adaptive interview evaluation method as described in any one of claims 1-5.

8. A computer device, comprising: The system comprises a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; characterized in that the memory stores at least one executable instruction, which causes the processor to perform an operation corresponding to the confidence-driven adaptive interview assessment method as described in any one of claims 1-5.

Images