Probabilistic language multi-attribute man-post matching decision-making method and device based on three-way decision-making

Abstract

The invention discloses a probability language multi-attribute person and post matching decision method and device based on a three-way decision, and the method comprises the steps: calculating a combination weight according to a probability language-based evaluation information matrix of an employer to an applicant and a probability language-based evaluation information matrix of the applicant to the employer; the conditional probability that the employer and the applicant meet the corresponding attribute set is calculated; calculating a comprehensive relative utility function of the employer and the applicant under the corresponding attribute set, and respectively obtaining a three-branch decision classification of the post to the candidate and a three-branch decision classification of the candidate to the post by using a three-branch decision rule; meanwhile, a conservative intersection strategy is adopted to obtain an initial candidate set; based on the initial matching candidate set and the combined weight, respectively calculating the satisfaction degree and the fairness; and calculating by using the multi-objective optimization model to obtain an optimized matching result. By adopting the technical scheme of the invention, finer, robust and efficient bilateral matching is realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of information processing, and particularly relates to a probability language multi-attribute person-post matching decision method and device based on three-way decision. BACKGROUND

[0002] At present, the bilateral matching theory has shown strong vitality in the person-post matching scene; the quality of the matching scheme is directly related to the utility of the two parties and the resource utilization efficiency, and constitutes a key link of the optimization of organizational human capital allocation.

[0003] Although the existing research has made significant progress, it still has the following three limitations when dealing with large-scale and high-uncertainty person-post matching problems: firstly, in the information perception aspect, most of the existing probability language distance measures treat all language items equally, fail to effectively distinguish the confidence degree of decision makers on different options, ignore the "reliability" difference of information itself, and may lead to dilution or misjudgment of key information. Secondly, in the attribute weight determination aspect, the existing methods mostly rely on a single subjective or objective weighting method, and it is difficult to unify the decision maker's intention and the objective information of the data in complex person-post matching. Thirdly, and most importantly, in the decision mechanism: whether it is TOPSIS based on utility aggregation, TODIM considering psychological behavior, or ORESTE with unique preference relationship processing capability, their essence is all based on the "binary decision" paradigm of full ordering. When the matching information is fuzzy or missing, rigidly classifying it into the acceptance or rejection category will amplify the possibility of misjudgment; at the same time, in the case of a large number of candidates, if there is no pre-screening link, the calculation load will increase and the decision cost will rise. SUMMARY

[0004] To solve the problems existing in the prior art, the present application provides a probability language multi-attribute person-post matching decision method and device based on three-way decision.

[0005] To achieve the above-mentioned purpose, the present application provides the following solutions: A probability language multi-attribute person-post matching decision method based on three-way decision, comprising: Step S1, obtaining the evaluation information matrix of the employer on the job seeker based on probability language according to the evaluation index of the employer on the job seeker; obtaining the evaluation information matrix of the job seeker on the employer based on probability language according to the evaluation index of the job seeker on the employer; Step S2, calculating the combined weight according to the evaluation information matrix of the employer on the job seeker based on probability language and the evaluation information matrix of the job seeker on the employer based on probability language; Step S3, calculating the conditional probability of the employer and the job seeker satisfying the corresponding attribute set based on the ideal solution method according to the probability language evaluation information of the employer and the job seeker. Step S4, according to the condition probability that the employer and the job seeker meet the corresponding attribute set, the comprehensive relative utility function of the employer and the job seeker under the corresponding attribute set is calculated, and the three-branch decision rule is used to obtain the three-branch decision classification of the post to the candidate and the three-branch decision classification of the candidate to the post respectively; and a conservative intersection strategy is adopted to obtain an initial candidate set; Step S5, based on the initial matching candidate set and the combination weight, the satisfaction degree and the fairness degree are calculated respectively; Step S6, according to the satisfaction degree and the fairness degree, the multi-objective optimization model is used to calculate the optimized matching result.

[0006] The application also provides a probability language multi-attribute person-post matching decision device based on three-branch decision, comprising: The first processing module is used for obtaining the evaluation information matrix of the employer based on probability language to the job seeker according to the evaluation index of the employer to the job seeker, and obtaining the evaluation information matrix of the job seeker based on probability language to the employer according to the evaluation index of the job seeker to the employer; The second processing module is used for calculating the combination weight according to the evaluation information matrix of the employer based on probability language to the job seeker and the evaluation information matrix of the job seeker based on probability language to the employer; The third processing module is used for calculating the condition probability that the employer and the job seeker meet the corresponding attribute set based on the ideal solution method according to the probability language evaluation information of the employer and the job seeker; The fourth processing module is used for calculating the comprehensive relative utility function of the employer and the job seeker under the corresponding attribute set according to the condition probability that the employer and the job seeker meet the corresponding attribute set, and obtaining the three-branch decision classification of the post to the candidate and the three-branch decision classification of the candidate to the post respectively by using the three-branch decision rule; and an initial candidate set is obtained by adopting a conservative intersection strategy; The fifth processing module is used for calculating the satisfaction degree and the fairness degree based on the initial matching candidate set and the combination weight respectively; The sixth processing module is used for calculating the optimized matching result by using the multi-objective optimization model according to the satisfaction degree and the fairness degree.

[0007] Compared with the prior art, the application has the beneficial effects that: 1. A probability language distance measure based on reliability is proposed. By introducing the concept of reliability of language items, the traditional distance formula is improved, the information measurement is more in line with the real confidence level of the decision maker, and the foundation is laid for subsequent accurate decision.

[0008] 2. Construct a subjective and objective combined attribute weight calculation model. The combination of BWM subjective method and probability-semantic comprehensive entropy objective method entropy is optimized through game theory negotiation mechanism, ensuring the rationality and robustness of weight distribution.

[0009] 3. Introduce probability language multi-attribute three-way decision for preliminary screening, realizing the change of decision paradigm. Through threshold determination, all matching pairs are scientifically divided into three regions of "immediate acceptance (positive domain)", "delayed decision (boundary domain)" and "immediate rejection (negative domain)". This not only explicitly includes and utilizes uncertainty, avoids forced decision on boundary matching pairs when information is insufficient, but also realizes the optimization of decision resources - focuses on the core matching pairs in the positive domain and the boundary domain for subsequent complex multi-objective optimization, thereby significantly improving the processing efficiency and decision reliability of large-scale matching problems.

[0010] 4. Construct a final matching optimization model considering satisfaction and fairness. On the basis of preliminary screening, for matching pairs in the positive domain and the boundary domain, the matching satisfaction of both parties and the overall fairness of the matching result are considered to establish a multi-objective optimization model to obtain a stable and relatively satisfactory final matching scheme for both parties.

[0011] In summary, the present application forms a more refined, robust and efficient two-sided matching decision framework through the progressive process of "reliability distance measure, subjective and objective combined weighting, three-way decision preliminary screening, multi-objective optimization matching", providing a new effective way to solve the two-sided matching problem in complex uncertain environment. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings described in the following embodiments are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0013] Figure 1 The flow chart of the probability language multi-attribute person-post matching decision method based on three-way decision of the embodiments of the present application. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0015] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0016] Embodiment 1 As Figure 1 shown, the present application provides a probability language multi-attribute person-post matching decision-making method based on three-way decision, comprising: Step S1, obtaining the evaluation information matrix of the employer on the job seeker based on probability language according to the evaluation index of the employer on the job seeker; obtaining the evaluation information matrix of the job seeker on the employer based on probability language according to the evaluation index of the job seeker on the employer; Step S2, calculating the subjective weight by using the BWM method according to the evaluation information of the employer on the job seeker based on probability language and the evaluation information of the job seeker on the employer based on probability language, calculating the objective weight by using the probability-semantic comprehensive entropy, and calculating the combined weight by using the game theory negotiation mechanism; Step S3, calculating the conditional probability of the employer and the job seeker satisfying the corresponding attribute set based on the ideal solution method according to the probability language evaluation information of the employer and the job seeker; Step S4, calculating the comprehensive relative utility function of the employer and the job seeker under the corresponding attribute set according to the conditional probability of the employer and the job seeker satisfying the corresponding attribute set, respectively obtaining the three-way decision classification of the post on the candidate and the three-way decision classification of the candidate on the post by using the three-way decision rule, and obtaining the initial candidate set by adopting the conservative intersection strategy; Step S5, calculating the satisfaction degree and the fairness degree based on the initial matching candidate set and the combined weight; Step S6, calculating the optimized matching result by using the multi-objective optimization model according to the satisfaction degree and the fairness degree.

[0017] As an embodiment of the present application, in step S1, the set of the person-post matching parties is respectively and ; wherein, x i represents the i th post, y i represents the j th job seeker, X and Y the attribute set under the subject is respectively and ; wherein, E represents the factors concerned by the post side, such as profession, education, work experience, ability, etc.; C represents the factors concerned by the job seeker, such as work intensity, salary level, welfare treatment. In the probability language environment, the subject X iRegarding the subject Y j To the attribute C k Evaluation as .

[0018] Assuming S = {s0: very poor, s1: poor, s2: poor, s3: medium, s4: good, s5: very good, s6: very good}, then The probability of medium is 0.6, and the probability of good is 0.4.

[0019] As an embodiment of an embodiment of the present application, in step S2, in order to coordinate the subjective preference of the evaluation subject and the internal contradiction of the objective information of the probability language data, a set of subjective and objective coupled attribute weight solving framework is constructed. First, the improved BWM method is used to quantify the subjective weight, then the probability-semantic comprehensive entropy model is designed to determine the objective weight, and finally the game theory negotiation mechanism is used to realize the collaborative optimization of the two types of weights. Specifically, it includes: In step S2, it is assumed that there is evaluation information as shown in Table 1 and Table 2: Table 1 Table 2 Step 21, the improved BWM method is used to calculate the subjective weight The implementation process of the improved BWM is as follows: Step 211: Identify the best and worst attributes. In the attribute set Select the most critical attribute and the least important attribute .

[0020] Step 212: Construct a probability language comparison vector. Using the nine-quantile scale method, the preference intensity of the best attribute relative to other attributes is , and similarly the preference intensity of the worst attribute relative to other attributes is .

[0021] Step 213: Build a subjective weight optimization model. Take minimizing the maximum deviation as the criterion, and build a mathematical programming model to determine the optimal subjective weight : (9) Step 4: Consistency test. Calculate the consistency ratio , is a random consistency reference threshold. If CR <0.10, then accept the current comparison vector; otherwise, the preference needs to be adjusted.

[0022] Step 22: Calculate the objective weights using the probability-semantic comprehensive entropy. To deeply extract the implicit information structure of probabilistic language data, a comprehensive entropy weight model that integrates the disorder of probability distribution and semantic dispersion is constructed based on the characteristics of probabilistic language term sets.

[0023] Step 221: Calculate the semantic discrete entropy, which reflects the level of semantic differentiation of linguistic terms within an attribute. (10) Step 222: Calculate the probability distribution entropy, which characterizes the degree of disorder in the distribution of attribute values. For attributes... C j Its probability entropy is defined as: (11) Step 223: Construct probabilistic-semantic comprehensive entropy; integrate dual entropy information to obtain attributes. C j Comprehensive entropy measure: (12) Step 224: Determine the objective weights. Derive the objective weight vector based on the comprehensive entropy value. (13) Step 23: Game theory negotiation mechanism to achieve weight combination Subjective weight With objective weight They often exhibit inconsistencies. Game theory methods are used to explore the Nash equilibrium point between the two to achieve the optimal compromise in weight allocation.

[0024] Step 231: Establish the combined weight expression. Let the combined weights be... Convex combination of subjective and objective weights: (14) Step 232: Design the game optimization objective. Minimize the sum of squared Euclidean distances between the combined weights and the subjective and objective weights: (15) Step 233: Solve for the equilibrium strategy coefficients. Substitute the combined weight expression into the objective function, and then... Elimination, simplification, removal ,right Taking the partial derivative and setting it to zero, we can obtain the dynamic equilibrium coefficients: (16) As one embodiment of the present invention, in step S3, the conditional probability estimation is based on the ideal solution. For any two probabilistic language term sets and ,but and Similarity is defined as: (17) For the evaluation decision matrix, let... For the positive ideal solution and It is a negative ideal solution. The formulas for calculating the similarity with positive and negative ideal solutions are as follows: Finally, each object With positive ideal solution The relative closeness between them is: (18) Among them, parameters Used to characterize the different risk preferences of decision-makers.

[0025] Relative closeness can reflect the object Attribute State Set The magnitude of the conditional probability, then the object Satisfy attribute set The conditional probability is: .

[0026] As one embodiment of the present invention, step S4 includes: Step 41: Estimation of the relative utility function object-based In attributes The following evaluation value The distance formula can be used to derive the object's... The relative loss function and the relative profit function.

[0027] Set attribute The minimum and maximum evaluation values ​​are respectively and . In attributes The loss for misclassification is , Since correct classification will not result in any loss, therefore Object The loss classified into the delayed decision region falls between that classified into the acceptance and rejection regions, therefore... , ,in The contribution avoidance coefficient, the value of which depends on the decision-maker's risk preference. The relative loss function under the attribute is shown in Table 3. Similarly, the relative gain function can be obtained as shown in Table 4.

[0028] Table 3 wherein: denotes the positive domain of the attribute shown, denotes the negative domain of the attribute shown. denotes the accept behavior, denotes the loss of correctly classifying an object that should be accepted as accepted. denotes the loss of incorrectly classifying an object that should be rejected as accepted. denotes the delay decision behavior. denotes the loss of classifying an object that should be accepted into the delay decision region. denotes the loss of classifying an object that should be rejected into the delay decision region. denotes the reject behavior. denotes the loss of incorrectly classifying an object that should be accepted as rejected. denotes the loss of correctly classifying an object that should be rejected as rejected.

[0029] Table 4 wherein: denotes the positive domain of the attribute shown, denotes the negative domain of the attribute shown. denotes the accept behavior, denotes the gain of correctly classifying an object that should be accepted as accepted. denotes the gain of incorrectly classifying an object that should be rejected as accepted. denotes the delay decision behavior. denotes the gain of classifying an object that should be accepted into the delay decision region. denotes the gain of classifying an object that should be rejected into the delay decision region. denotes the reject behavior. denotes the gain of incorrectly classifying an object that should be accepted as rejected. denotes the object that should be rejected correctly classified as rejected.

[0030] For any object , Under , the relative utility function is: (19) where is the set of states, and is the set of actions, representing accept, delay, and reject, respectively. is the object Under attribute , the gain for different actions (accept, delay, and reject) under different states (positive domain, negative domain). is the object Under attribute , the loss for different actions (accept, delay, and reject) under different states (positive domain, negative domain). The relative utility function is shown in Table 5.

[0031] Table 5 For any object , the comprehensive relative utility function of object under is: (20) where is the combined weight of attribute . is the relative utility of object under attribute for different actions (accept, delay, and reject) under different states (positive domain, negative domain).

[0032] Obviously, and . For any object , the conditional probability of and the relative utility function can be combined to obtain the expected utility function of object when taking different actions, as follows: where is the expected utility of object when taking the accept action, is the expected utility of object the expected utility of taking the accept action, for the object the expected utility of taking the reject action, for the object the conditional probability that the object belongs to set C, for the object the conditional probability that the object does not belong to set C, and .

[0033] Based on the Bayesian decision process and the principle of maximizing the expected utility, the following three decision rules can be obtained: (P1) and then (P2) and then (P3) and then where is the accept region, is the delay region, is the reject region. Rule (P1) indicates that for the object if the expected utility of taking the accept action is greater than taking the delay action and greater than taking the reject action, then the object is classified into the accept region; rule (P2) indicates that for the object if the expected utility of taking the delay action is greater than taking the accept action and greater than taking the reject action, then the object is classified into the delay region; rule (P3) indicates that for the object if the expected utility of taking the reject action is greater than taking the accept action and greater than taking the delay action, then the object is classified into the accept region. Let where represents the comprehensive relative utility of correctly classifying the object that should be accepted into the accept region. represents the comprehensive relative utility of incorrectly classifying the object that should be rejected into the accept region. represents the comprehensive relative utility of classifying the object that should be accepted into the delay decision region. represents the comprehensive relative utility of classifying the object The overall relative utility of the classification into the delay decision area. The overall relative utility of the classification into the delay decision area. The overall relative utility of the classification into the delay decision area. The overall relative utility of the classification into the delay decision area. The overall relative utility of the classification into the delay decision area.

[0034] The above decision rules can be simplified as: (P4) The above decision rules can be simplified as: (P5) The above decision rules can be simplified as: (P6) The above decision rules can be simplified as: Step 42, clustering of the initial matching pair set For each candidate matching pair , the two parties independently make a three-way decision classification.

[0035] (1) Based on the three-way decision classification of the post to the candidate, the following is obtained: : positive domain (the post side considers it to be highly matched) : boundary domain (the post side considers it to be further investigated) : negative domain (the post side explicitly rejects) (2) Based on the three-way decision classification of the candidate to the post, the following is obtained: : positive domain (the candidate side considers it to be highly matched) : boundary domain (the candidate side considers it to be further investigated) : negative domain (the candidate side explicitly rejects) The candidate set entering the optimization model Considering the bilateral will, the bilateral veto guarantee, the calculation complexity control, and the reduction of the conflict of stability constraints in the optimization model, a conservative intersection strategy is adopted, only the matching pair that neither falls into the negative domain is reserved, ensuring that the pairing entering the optimization stage has basic bilateral recognition, and the mathematical expression is: (21) As an embodiment of the embodiment of the present application, in step S5, the probability language evaluation information of the two parties is converted into matching satisfaction. Let the subject be The matching satisfaction of the subject is , and the calculation formula is: (22) wherein, is the combined weight of the j th evaluation attribute, g is the language scale function of ().

[0036] Let the matching satisfaction of the employer to the employee be , and its calculation formula is: (23) wherein, is the combined weight of the i th evaluation attribute, g is the language scale function of ().

[0037] From the perspective of dynamic game, the expansion of the satisfaction difference evolves into the relaxation of the bilateral incentive compatibility constraint, leading to the difficulty in meeting the fair matching condition, and ultimately reducing the transaction probability. Therefore, in the feasible matching scheme set, the fairness benchmark should be set to minimize the dispersion degree of the utility distribution of the counterpart, i.e., the high convergence of the bilateral perceived income. Based on this, the fairness degree of the employer to the employee is , and its calculation formula is: (24) As an embodiment of the embodiment of the present application, in step S6, considering the satisfaction and fairness of the person-post matching, a multi-objective optimization model can be established with the goal of maximizing the matching satisfaction and fairness of the bilateral subjects: (M-1) (25) wherein, p is the number of employers, q is the number of employees, =0 indicates that the i th employer and the j th employee are not matched, =1 indicates that the i th employer and the j th employee are matched. is the matching satisfaction of the employer to the employee , is the matching satisfaction of the employee to the employer , is the fairness degree between the employer and the employee .

[0038] To reflect the different optimization emphasis of satisfaction and fairness, the model (M-1) is improved as follows: (M-2) (26) In the model (M-2), since the two target dimensions are the same, let , the multi-objective is converted into a single objective by using the weighting method as follows: (M-3) (27) Wherein, , is the weight of measuring the importance of satisfaction and fairness.

[0039] Example 2 The application also provides a probability language multi-attribute person-post matching decision device based on three-branch decision, comprising: A first processing module is configured to obtain an evaluation information matrix of a user unit on a job candidate based on probability language according to evaluation indexes of the user unit on the job candidate, and obtain an evaluation information matrix of the job candidate on the user unit based on probability language according to evaluation indexes of the job candidate on the user unit; A second processing module is configured to calculate a combination weight according to the evaluation information matrix of the user unit on the job candidate based on probability language and the evaluation information matrix of the job candidate on the user unit based on probability language; A third processing module is configured to calculate a conditional probability that the user unit and the job candidate meet a corresponding attribute set based on an ideal solution method according to probability language evaluation information of the user unit and the job candidate; A fourth processing module is configured to calculate a comprehensive relative utility function of the user unit and the job candidate under the corresponding attribute set according to the conditional probability that the user unit and the job candidate meet the corresponding attribute set, obtain a three-branch decision classification of a post on a candidate and a three-branch decision classification of a candidate on a post by using a three-branch decision rule, and obtain an initial candidate set by adopting a conservative intersection strategy; A fifth processing module is configured to calculate a satisfaction degree and a fairness degree of the initial matching candidate set and the combination weight respectively; A sixth processing module is configured to calculate an optimized matching result by using a multi-objective optimization model according to the satisfaction degree and the fairness degree.

[0040] The above-described embodiments are only descriptions of the preferred modes of the application, and do not limit the scope of the application. Without departing from the design spirit of the application, various modifications and improvements of the technical solutions of the application made by those skilled in the art shall fall within the protection scope of the claims of the application.

Claims

1. A probability language multi-attribute person-post matching decision method based on three-branch decision, characterized in that, The method comprises the following steps: Step S1, obtaining a probability language-based evaluation information matrix of the employer on the job seeker according to evaluation indexes of the employer on the job seeker; obtaining a probability language-based evaluation information matrix of the job seeker on the employer according to evaluation indexes of the job seeker on the employer; Step S2, calculating a combination weight according to the probability language-based evaluation information matrix of the employer on the job seeker and the probability language-based evaluation information matrix of the job seeker on the employer; Step S3, calculating condition probabilities of the employer and the job seeker satisfying corresponding attribute sets respectively according to probability language evaluation information of the employer and the job seeker based on an ideal solution method; Step S4, calculating comprehensive relative utility functions of the employer and the job seeker under corresponding attribute sets according to the condition probabilities of the employer and the job seeker satisfying the corresponding attribute sets, and obtaining three-way decision classifications of the job seeker on the post and the post on the job seeker respectively by using a three-way decision rule; meanwhile, an initial candidate set is obtained by adopting a conservative intersection strategy; Step S5, calculating satisfaction degrees and fairness degrees of the initial matching candidate set and the combination weight respectively; Step S6, obtaining an optimized matching result by using a multi-objective optimization model according to the satisfaction degrees and the fairness degrees.

2. A person-job matching device characterized by comprising: The method comprises the following steps: The first processing module is configured to obtain a probability language-based evaluation information matrix of the employer on the job seeker according to evaluation indexes of the employer on the job seeker; obtain a probability language-based evaluation information matrix of the job seeker on the employer according to evaluation indexes of the job seeker on the employer; The second processing module is configured to calculate a combination weight according to the probability language-based evaluation information matrix of the employer on the job seeker and the probability language-based evaluation information matrix of the job seeker on the employer; The third processing module is configured to calculate condition probabilities of the employer and the job seeker satisfying corresponding attribute sets respectively according to probability language evaluation information of the employer and the job seeker based on an ideal solution method; The fourth processing module is configured to calculate comprehensive relative utility functions of the employer and the job seeker under corresponding attribute sets according to the condition probabilities of the employer and the job seeker satisfying the corresponding attribute sets, and obtain three-way decision classifications of the job seeker on the post and the post on the job seeker respectively by using a three-way decision rule; meanwhile, an initial candidate set is obtained by adopting a conservative intersection strategy; The fifth processing module is configured to calculate satisfaction degrees and fairness degrees of the initial matching candidate set and the combination weight respectively; The sixth processing module is configured to obtain an optimized matching result by using a multi-objective optimization model according to the satisfaction degrees and the fairness degrees.

Images