Method and device for improving accuracy of intelligent interview evaluation

Abstract

The invention discloses a method and device for improving the accuracy of intelligent interview evaluation, and relates to the technical field of intelligent interview, and the method comprises the steps: obtaining the interview session data of candidates; comparing the answer of the candidate to the interview question with the plurality of typical answers corresponding to the question to obtain a comparison result of the answer of the candidate and each typical answer and an ability score corresponding to each typical answer; estimating the original ability score of the candidate under each investigation dimension based on the comparison result of the answer of the candidate and each typical answer and the ability score corresponding to each typical answer; calculating the ranking of the candidates in the norm population based on the original ability scores of the candidates under each investigation dimension; and converting the ranking into a standard ability score of the candidate under each investigation dimension. According to the invention, the overall accuracy of intelligent interview evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of online interview based on artificial intelligence, and particularly to a method and device for improving the accuracy of intelligent interview evaluation. Background Art

[0002] With the increasing activity of the talent market, the limitations of video interviews have begun to be continuously exposed, and it has not solved the problem that HR needs to spend a large amount of time on batch initial screening of talents in the traditional sense. Thanks to the continuous development of AI technology, the method of "AI + video interview" for initial screening and selection of talents has emerged and gradually become the mainstream in the market. A large number of employers have started to adopt the method of intelligent interviews, creating interview robots, and presetting interview questions to conduct talent evaluation and judgment on the interview videos of candidates.

[0003] However, there are still a series of problems with AI technology in talent evaluation. The hallucination problem of large models is still relatively serious. If a scoring system such as a large model is directly used to score the answers of candidates, there will be relatively large differences in the scoring results produced by multiple calls for the same answer, and the results will be inconsistent each time. For example, the scores of candidates may fluctuate between 5 points, 6 points, and 7 points. This score fluctuation will bring a series of subsequent problems. For example, the true score of candidate A is 5 points, but the large model fluctuates to 7 points; the true score of candidate B is 7 points, but the large model fluctuates to 5 points. In a real recruitment scenario, candidate A may be wrongly recruited instead of candidate B because of this.

[0004] Generally, there are the following several optimization methods for the above phenomena: One is to use the few-shot learning method, that is, to increase examples so that the large model can better understand human intentions. The other is to call multiple times and take the average value to reduce the probability of the large model making mistakes. These several traditional optimization methods all treat the scoring system such as the large model as a black box, and can only reduce the probability of the scoring system making mistakes to a certain extent. In fact, technical personnel do not understand in which cases the scoring system such as the large model is more likely to make mistakes, so it is very difficult to optimize it targeted. Summary of the Invention

[0005] In view of the above defects or deficiencies in the prior art, the present invention provides a method and device for improving the accuracy of intelligent interview evaluation, which predicts the probability of a scoring system such as a large model making mistakes in various situations through prior knowledge and basic assumptions in the talent evaluation scenario, so as to improve the overall accuracy.

[0006] One aspect of the present invention provides a method for improving the accuracy of intelligent interview evaluation, including the following steps: obtaining the interview session data of a candidate, where the interview session data includes interview questions under each evaluation dimension and the candidate's answers to each question; comparing the candidate's answers to the interview questions with multiple typical answers corresponding to each question to obtain the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer, where the value range of the comparison results is 0 or 1; estimating the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer; calculating the ranking of the candidate in the norm group based on the original ability scores of the candidate under each evaluation dimension; and converting the ranking into the standard ability scores of the candidate under each evaluation dimension.

[0007] Another aspect of the present invention further provides a device for improving the accuracy of intelligent interview evaluation, including: a data acquisition module configured to obtain the interview session data of a candidate, where the interview session data includes interview questions under each evaluation dimension and the candidate's answers to each question; a comparison module configured to compare the candidate's answers to the interview questions with multiple typical answers corresponding to each question to obtain the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer, where the value range of the comparison results is 0 or 1; an estimation module configured to estimate the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer; a ranking calculation module configured to calculate the ranking of the candidate in the norm group based on the original ability scores of the candidate under each evaluation dimension; and a scoring module configured to convert the ranking into the standard ability scores of the candidate under each evaluation dimension.

[0008] A method and device for improving the accuracy of intelligent interview evaluation provided by the present invention predict the error probability of a scoring system such as a large model in various situations through prior knowledge and basic assumptions in the talent evaluation scenario, so as to clarify in which situations the scoring system is more likely to make mistakes, and optimize it targeted, ultimately improving the overall accuracy of intelligent interview evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes, and advantages of the present application will become more obvious: Figure 1 is a flowchart of a method for improving the accuracy of intelligent interview evaluation provided by an embodiment of the present application; Figure 2 is a structural diagram of a device for improving the accuracy of intelligent interview evaluation provided by another embodiment of the present application; Figure 3 It is a schematic structural diagram of an electronic device provided by another embodiment of the present application. Detailed implementation manners

[0010] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0011] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention are also intended to include the plural forms unless the context clearly indicates otherwise.

[0012] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present invention to describe the acquisition modules, these acquisition modules should not be limited to these terms. These terms are only used to distinguish the acquisition modules from each other.

[0013] Depending on the context, the word "if" as used herein can be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detected (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0014] It should be noted that the orientation terms such as "upper", "lower", "left" and "right" described in the embodiments of the present invention are described from the angles shown in the drawings, and should not be construed as limiting the embodiments of the present invention. In addition, in the context, it should also be understood that when it is mentioned that one element is formed "on" or "under" another element, it can not only be directly formed "on" or "under" another element, but also be indirectly formed "on" or "under" another element through an intermediate element.

[0015] In the context of intelligent interviews, it is difficult for experts to directly define the specific meaning of a particular score. For example, it is hard to clarify what a score of 6 or 7 represents respectively in the "persuasion and influence" assessment dimension, and what the differences are. Therefore, when experts score candidates' abilities, there are easily fluctuations in scores. For example, the same expert's score for the same candidate's ability may fluctuate between 5, 6, and 7. Relatively speaking, it is an easier task for experts to compare the abilities of two candidates. Further, when experts compare the abilities of two candidates, the smaller the actual ability gap between the two candidates, the higher the difficulty for experts to distinguish which is better or worse, and the larger the actual ability gap between the two candidates, the lower the difficulty for experts to distinguish which is better or worse.

[0016] Based on the above facts, this application assumes that the working principle of a scoring system based on software algorithms such as large models is consistent with the thinking logic of human experts. Therefore, this application adopts the following two methods to improve the accuracy of the scoring system's evaluation of interview data: (1) The scoring system does not directly score candidates, but ranks multiple candidates and then converts the rankings of the candidates into scores.

[0017] (2) The accuracy of the scoring system's comparison of the abilities of two candidates is related to the actual ability gap between the two candidates. If the actual ability gap between the two candidates is small, the accuracy of the scoring system in comparison is low. If the actual ability gap between the two candidates is large, the accuracy of the scoring system in comparison is high.

[0018] See Figure 1 , an embodiment of this application provides a method for improving the accuracy of intelligent interview evaluation, including the following steps: Step S101, obtain the interview session data of the candidate, where the interview session data includes interview questions under each assessment dimension and the candidate's answers to each question.

[0019] Specifically, the candidate accesses the intelligent interview system App through a mobile device. The intelligent interview system App collects the candidate's answering video through technical means such as intelligent questioning and intelligent follow-up questioning, then extracts the audio in the answering video, and further converts it into text content through speech recognition technology to form interview session data. The interview session data of a candidate in one assessment dimension includes the interviewer's questions, follow-up questions, and the candidate's answers to each question.

[0020] Step S102, compare the candidate's answers to the interview questions with multiple typical answers corresponding to the questions, obtain the comparison results between the candidate's answers and each typical answer, and the ability scores corresponding to each typical answer. The value range of the comparison results is 0 or 1.

[0021] Specifically, for each evaluation dimension, such as "persuasion and influence", "team collaboration", and "adaptability", multiple typical answers are preset. These typical answers represent how candidates with low, medium, and high abilities would respond. Call a scoring system such as a large language model and compare the candidate's response with the typical answers in combination with the scoring criteria for the current evaluation dimension. The more typical answers involved in the comparison, the higher the accuracy of the talent evaluation. For each evaluation dimension, the number of typical answers is not limited. For example, it can be 3 or 100. The following uses the comparison results of three typical answers as an example for illustration. The three typical answers are respectively marked as , and the ability scores corresponding to each typical answer (some typical answers indicate low ability, some indicate mediocre ability, and some indicate excellent ability) are , and the comparison results are respectively . If the candidate's response is better than or equal to the typical answer, the comparison result for this time is recorded as 1. If the candidate's response is inferior to the typical answer, the comparison result for this time is recorded as 0. That is to say,[[]] takes values of 0 or 1. The content finally output by the scoring system is: .

[0022] Step S103: Estimate the candidate's original ability score under each evaluation dimension based on the comparison results between the candidate's response and each typical answer and the ability score corresponding to each typical answer.[[]]

[0023] In this step, the scoring system estimates the candidate's original ability score under each evaluation dimension based on the comparison results between the candidate's response and each typical answer and the ability score corresponding to each typical answer. The technical concept is as follows: First, establish a probability distribution hypothesis for a single comparison between the candidate's response and each typical answer, that is: establish a comparison probability model between the candidate's ability θ and the typical answer ability b to reflect the superiority or inferiority of the candidate's ability θ relative to the typical answer ability b ; Then, assume that each comparison is independent, and the joint probability is the product of the probabilities of each time. By maximizing the joint probability (i.e., maximum likelihood estimation), find the optimal θ value that can best explain all comparison results; Finally, use an iterative optimization algorithm to iteratively update the parameters to approximate the θ value that maximizes the objective function (i.e., log-likelihood), and the most accurate candidate ability score can be obtained.[[]]

[0024] The specific implementation process of the above technical concept is as follows: (1)Establish the probability distribution hypothesis for the single comparison between the candidate's answer and each typical answer: Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the probability that the candidate's answer is better than the and th typical answer under the condition of given and

[0025] The above probability distribution hypothesis makes it that when θ = b , the scoring system believes that the probability of θ > b is 50%; when θ is slightly greater than b , the scoring system believes that the probability of θ > b is slightly greater than 50%; when θ is much greater than b , the scoring system believes that the probability of θ > b approaches 1; when θ is slightly less than b , the scoring system believes that the probability of θ > b is slightly less than 50%; when θ is much less than b , the scoring system believes that the probability of θ > b approaches 0. That is to say, the accuracy rate of the scoring system in comparing the abilities of two candidates is related to the actual ability gap between the two candidates. If the actual ability gap between the two candidates is small, the accuracy rate of the scoring system in comparison is low; if the actual ability gap between the two candidates is large, the accuracy rate of the scoring system in comparison is high.

[0026] (2)According to the probability distribution hypothesis of the single comparison, establish the joint probability distribution hypothesis for the multiple comparisons between the candidate's answer and each typical answer: Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the number of typical answers; represents the probability that the candidate's answer is better than the and Under the condition of, the joint probability distribution of the comparison results between the candidate's answers and each typical answer.

[0027] (3) Calculate the original ability scores of the candidate under each inspection dimension when the joint probability distribution is the largest through an iterative optimization algorithm, including: a. Set the initial original ability scores of the candidate under each inspection dimension and the initial approximate matrix of the Hessian matrix , and set the iterative convergence threshold. Among them, is the initial guess value, which can be set as the average ability of the typical answers or any reasonable value; is the initial Hessian approximation matrix, which can be set as the identity matrix I to accelerate convergence. The convergence threshold is set to stop the iteration when the gradient is small enough.

[0028] b. Calculate the objective function according to the following formula The - th iteration gradient value : ; ; Among them, the objective function is also the negative log-likelihood function, represents the first-order derivative of the objective function with respect to , reflecting the slope direction of the current point, that is, calculating the gradient value.

[0029] c. Determine the search direction of the - th iteration according to the following formula, that is: use the approximate inverse Hessian matrix to adjust the gradient direction: ; Among them, represents the approximate matrix of the Hessian matrix at the - th iteration.

[0030] d. Find the step size that minimizes the objective function , that is, find the optimal step size in the search direction to minimize the objective function.

[0031] e. Calculate the candidate's original ability score (i.e., parameter update) of the - th iteration according to the following formula: ; f. Calculate the - th iteration and the Gradient change of the current iteration : ; g. Calculate the approximate matrix of the Hessian matrix of the -th iteration according to the following formula : ; ; Iteratively loop the above calculation process until the gradient norm is less than the iteration convergence threshold, then stop the iteration. In this way, when the joint probability distribution is maximized, the original ability scores of the candidate in each evaluation dimension are obtained .

[0032] The above steps convert the subjective comparison results into a quantifiable probability model and implement automatic parameter estimation through an optimization algorithm

[0033] Step S104: Calculate the ranking of the candidate in the norm group based on the original ability scores of the candidate in each evaluation dimension

[0034] This step combines the original ability scores of the candidate with the norm database to further calculate the ranking in the norm group. For example: if the norm database includes data of 10,000 other candidates, and each candidate has an original ability score in the current evaluation dimension , where

[0035] if the candidate exceeds 8,000 people among 10,000 people, the ranking is 80%, that is, Top 80%

[0036] Step S105: Convert the ranking into the standard ability scores of the candidate in each evaluation dimension

[0037] Specifically (1) Convert the ranking to a percentile Input the absolute ranking (such as the 3rd) or relative ranking (such as Top 20%) of the candidate, and calculate the cumulative distribution probability (CDF) of the candidate For example: if ranked 1st among 100 people, then .

[0038] (2)Calculation of Quantiles of Normal Distribution To convert the percentile to the corresponding quantile of the standard normal distribution , the following formula is used: where is the inverse cumulative distribution function of the standard normal distribution. For example: = 50% corresponds to = 0, = 84.1% corresponds to = 1, = 99% corresponds to ≈ 2.326.

[0039] (3)Linear Transformation to Standard 10-Point Scale Map the value to a normal distribution with a mean of 5.5 points and a standard deviation of 1.5 points, i.e., satisfy: Standard score = 5.5 + Z × 1.5 For example: When = 0, the standard score is 5.5 points, When

[0040] (4)Truncation Processing We force the score to fall within the integer range of 0 - 10, i.e., if the calculated value is less than 0, output 0 points; if the calculated value is greater than 10, output 10 points.

[0041] The final standard scores obtained from the above steps are shown in Table 1: Table 1 The method for improving the accuracy of intelligent interview evaluation provided in this embodiment predicts the error probability of scoring systems such as large models under various circumstances through prior knowledge and basic assumptions, thereby clarifying in which circumstances the scoring system is more likely to make mistakes, so as to optimize it targeted, and finally improve the overall accuracy of intelligent interview evaluation.

[0042] See Figure 2 , another embodiment of the present invention also provides a device 200 for improving the accuracy of intelligent interview evaluation, including a data acquisition module 201, a comparison module 202, an estimation module 203, a ranking calculation module 204, and a scoring module 205. The device 200 can execute the method for improving the accuracy of intelligent interview evaluation in the method embodiment.

[0043] Specifically, the device 200 for improving the accuracy of intelligent interview evaluation includes: A data acquisition module 201, configured to acquire the interview session data of a candidate, where the interview session data includes interview questions under each evaluation dimension and the candidate's answers to each question; A comparison module 202, configured to compare the candidate's answers to the interview questions with multiple typical answers corresponding to the questions, obtain the comparison results between the candidate's answers and each typical answer, and the ability scores corresponding to each typical answer, where the value range of the comparison results is 0 or 1; An estimation module 203, configured to estimate the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer; A ranking calculation module 204, configured to calculate the ranking of the candidate in the norm population based on the original ability scores of the candidate under each evaluation dimension; A scoring module 205, configured to convert the ranking into the standard ability scores of the candidate under each evaluation dimension.

[0044] It should be noted that the device 200 for improving the accuracy of intelligent interview evaluation provided in this embodiment can be used to execute the technical solutions of the method embodiments. Its implementation principle and technical effects are similar to those of the method, and will not be elaborated here.

[0045] See Figure 3 , another embodiment of the present application also provides an electronic device 300 for implementing the method for improving the accuracy of intelligent interview evaluation in the method embodiments. The electronic device 300 may include, but is not limited to, any device capable of conducting online recruitment intelligent interviews, such as a smart phone, a tablet computer, a PDA, a PC, and a server. Figure 3 The illustrated electronic device 300 is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present invention.

[0046] As Figure 3 shown, the electronic device 300 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage device 308 into the random access memory (RAM) 303 to implement the method of the embodiments as described in the present invention. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 305. The input / output (I / O) interface 304 is also connected to the bus 305.

[0047] Typically, the following devices can be connected to the I / O interface 304: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 can allow the electronic device 300 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 3 the electronic device 300 with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. More or fewer devices can be alternatively implemented or had.

[0048] The above description is only a preferred embodiment of the present invention. Those skilled in the art should understand that the disclosed scope in the present invention is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the technical solution formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present invention.

Claims

1. A method for improving the accuracy of intelligent interview evaluation, characterized in that, It includes the following steps: Obtain the interview session data of the candidate, where the interview session data includes the interview questions under each evaluation dimension and the candidate's answers to each question; Compare the candidate's answers to the interview questions with multiple typical answers corresponding to each question to obtain the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer, where the value range of the comparison results is 0 or 1; Estimate the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer; Calculate the ranking of the candidate in the norm group based on the original ability scores of the candidate under each evaluation dimension; Convert the ranking into the standard ability scores of the candidate under each evaluation dimension.

2. The method for improving the accuracy of intelligent interview evaluation according to claim 1, wherein The step of estimating the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer includes: Establish a probability distribution hypothesis for a single comparison between the candidate's answers and each typical answer; Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the probability that the candidate's answer is better than the and th typical answer under the given conditions; Based on the probability distribution hypothesis for a single comparison, establish a joint probability distribution hypothesis for multiple comparisons between the candidate's answers and each typical answer; Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the number of typical answers; represents the joint probability distribution of the comparison results between the candidate's answer and each typical answer under the given and conditions; Calculate the original ability scores of the candidate under each evaluation dimension when the joint probability distribution is the largest through an iterative optimization algorithm.

3. The method for improving the accuracy of intelligent interview evaluation according to claim 2, wherein, The iterative optimization algorithm includes: Set the initial raw ability scores of candidates under each evaluation dimension and the initial approximation matrix of the Hessian matrix , and set the iterative convergence threshold; Calculate the objective function according to the following formula The gradient value of the th iteration ； ； Determine the search direction for the th iteration according to the following formula : ； Among them, represents an approximation matrix of the Hessian matrix for the Find the step size that minimizes the objective function ; ; Calculate the original ability score of the candidate for the th iteration according to the following formula : ； Calculate according to the following formula for the -th iteration and the -th iteration of the gradient change : ； Calculate the approximation matrix of the Hessian matrix for the th iteration according to the following formula : ； ； Iterate the above calculation process until the gradient norm is less than the iteration convergence threshold, and then stop the iteration to obtain the original ability scores of the candidates when the joint probability distribution is maximized in each evaluation dimension .

4. A method for improving the accuracy of intelligent interview evaluation according to claim 1, characterized in that, The step of calculating the ranking of the candidate in the norm group based on the original ability scores of the candidate under each evaluation dimension includes: If the original ability score of a candidate in each evaluation dimension exceeds the predetermined number of people in the norm population among , then the ratio of and is used as the ranking of the candidate in the norm population.

5. A method for improving the accuracy of intelligent interview evaluation according to claim 1, characterized in that, The step of converting the ranking into the standard ability scores of the candidate under each evaluation dimension includes: Convert the ranking of the candidate in the norm group into standard ability scores according to the normal distribution.

6. A device for improving the accuracy of intelligent interview evaluation, characterized in that, It includes: A data acquisition module configured to obtain the interview session data of the candidate, where the interview session data includes the interview questions under each evaluation dimension and the candidate's answers to each question; A comparison module configured to compare the candidate's answers to the interview questions with multiple typical answers corresponding to each question to obtain the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer, where the value range of the comparison results is 0 or 1; An estimation module configured to estimate the original ability scores of the candidate under each evaluation dimension based on the comparison results between the candidate's answers and each typical answer and the ability scores corresponding to each typical answer; A ranking calculation module configured to calculate the ranking of the candidate in the norm group based on the original ability scores of the candidate under each evaluation dimension; A scoring module configured to convert the ranking into the standard ability scores of the candidate under each evaluation dimension.

7. The device for improving the accuracy of intelligent interview evaluation according to claim 6, characterized in that, The estimation module is further configured to: Establish a probability distribution hypothesis for a single comparison between the candidate's answers and each typical answer; Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the probability that the candidate's answer is better than the and th typical answer under the given conditions; Based on the probability distribution hypothesis for a single comparison, establish a joint probability distribution hypothesis for multiple comparisons between the candidate's answers and each typical answer; Among them, represents the original ability score of the candidate under each evaluation dimension; represents the ability score corresponding to the th typical answer; represents the comparison result between the candidate's answer and the th typical answer; represents the number of typical answers; represents the joint probability distribution of the comparison results between the candidate's answer and each typical answer under the given and conditions; Calculate the original ability scores of the candidate under each evaluation dimension when the joint probability distribution is the largest through an iterative optimization algorithm.

8. A device for improving the accuracy of intelligent interview evaluation according to claim 7, characterized in that, The estimation module is further configured to: Set the initial raw ability scores of candidates under each evaluation dimension and the initial approximation matrix of the Hessian matrix , and set the iterative convergence threshold; Calculate the objective function according to the following formula The gradient value at the th iteration is ； ； Determine the search direction for the th iteration according to the following formula : ； Among them, represents an approximation matrix of the Hessian matrix for the Find the step size that minimizes the objective function ; ; Calculate the original ability score of the candidate for the th iteration according to the following formula : ； Calculate according to the following formula for the -th iteration and the -th iteration of the gradient change : ； Calculate the approximate matrix of the Hessian matrix for the th iteration according to the following formula : ； ； Iterate the above calculation process until the gradient norm is less than the iteration convergence threshold, and then stop the iteration to obtain the original ability scores of the candidates at each evaluation dimension when the joint probability distribution is maximized .

9. The device for improving the accuracy of intelligent interview evaluation according to claim 6, characterized in that The ranking calculation module is further configured to: If the original ability score of a candidate in each assessment dimension exceeds the predetermined number of people in the norm population among , then the ratio of and is used as the ranking of the candidate in the norm population.

10. The device for improving the accuracy of intelligent interview evaluation according to claim 6, characterized in that, The scoring module is further configured to convert the ranking of the candidate among the norm group into a standard ability score according to the normal distribution.

Images