A method for recognizing and controlling the matching results of a recommendation system

Abstract

The present invention provides a method for identifying and controlling the matching results of a recommendation system. The method includes setting the basic elements of the model, constructing a matching bias identification model, solving the model, evaluating the matching results, and simulating and verifying. By introducing the bias control factor of the recommendation platform, the influence of different bias degrees on the bilateral matching results is quantitatively evaluated, the fairness of the bilateral matching is optimized, and the bias control factor w is also optimized and adjusted, so as to optimize the matching results. The solution of the present invention realizes the differential measurement of the benefits of the bilateral matching subjects and the quantitative evaluation of the platform matching bias, so as to achieve better system recommendation and matching according to the evaluation results of the matching bias, provides a reference for the effective prevention and control of the bias matching phenomenon commonly existing on the recommendation platform, and the method has broad application prospects in many fields such as online shopping platforms, social platforms, and recruitment platforms.

Description

Technical Field

[0001] The present invention relates to the technical fields of bilateral matching algorithms and decision science, and particularly to a method for identifying and controlling matching results of a recommendation system. Background Art

[0002] As an important information dissemination medium, the network platform recommends various kinds of information to us every day. Essentially, the platform is a third-party intermediary that brings together different user groups and efficiently matches demand and supply. The intermediary matching process mainly includes three steps. The first step is the process in which the demand side tells the intermediary platform its needs. The second step is the process in which the supply side tells the intermediary platform its products and services. The third step is the process in which the intermediary platform uses its information advantage for matching.

[0003] In today's digital age, platforms often optimize the supply-demand matching through algorithmic recommendation systems. However, the profit-oriented design of these systems is prone to causing systematic matching biases, which not only weaken the fairness and efficiency of matching but may also trigger a series of problems.

[0004] The recommendation algorithms of platforms often focus on amplifying content that can attract users' attention, thus forming information cocoons. These algorithms will continuously push similar content based on data such as users' browsing history and interest preferences. Some platforms may also preferentially recommend paid content according to the payment situation of the supply side.

[0005] Existing models on the impact of intermediary platform preferences on the matching system are mainly developed based on a basic model in game theory, namely the Prisoner's Dilemma model. In this model, supplier A and demander B can choose between two strategies, namely cooperation and defection. On this basis, by introducing a penalty mechanism controlled by a third party (the intermediary platform), the defector is given the ability to influence the decision-making behavior of the intermediary, thereby constructing a model to evaluate the impact of intermediary preferences on the matching of the recommendation system.

[0006] However, using game models to explore the recommendation bias phenomenon of platforms has the following defects: on the one hand, game models assume that when both the supply side and the demand side choose cooperation (achieving bilateral matching), the benefits of the supply side and the demand side are the same. However, in actual matching problems, there are often significant differences in the benefit levels of the two matching sides. Therefore, this model has great limitations. On the other hand, the intermediary strategies introduced in the research of game models are discrete, and it is impossible to intuitively explore the dynamic impact of the intermediary on the matching system under different bias degrees. Therefore, existing models are difficult to accurately quantify the dynamic impact of the intermediary platform's matching results. Summary of the Invention

[0007] In view of the above problems, the present invention proposes a method for identifying and controlling the matching results of a recommendation system. By defining a continuously adjustable bias control factor w, it realizes the differential measurement of the benefits of bilateral matching entities and the quantitative evaluation of the intermediary matching results, thereby achieving better system recommendations and matches.

[0008] A method for identifying and controlling the matching results of a recommendation system, the method comprising the following steps:

[0009] Step S1, set the model elements. The model involves a recommendation platform, a supplier A, a demander B, and N candidate items for matching. Each time, the supplier A and the demander B perform personalized sorting on the N candidate items to form their respective preference lists. The matching energy of the supplier A related to the candidate item i is defined as The matching energy of the demander B related to the candidate item i is defined as where i = 1, 2,..., N. The energy value is equal to the ranking of the corresponding candidate item in the preference list. The higher the ranking, the smaller the energy value, indicating a more stable match.

[0010] Step S2, construct a matching bias recognition model, introduce a bias control factor w, and simulate the influence of platform bias on the matching results.

[0011] Step S3, model solution, calculate the corresponding relationship between the bilateral matching energy and the bias control factor w. The idea of the model solution is to first calculate the probability that the minimum total matching energy is equal to a specific value, then integrate all probabilities to obtain the average value of the minimum total energy, and calculate the average matching energy of the supplier A and the demander B.

[0012] Step S4, matching result evaluation, evaluate the stability of the matching results through fairness indicators, construct a matching fairness indicator f, which is used to measure the change in the matching fairness degree between the supplier and the demander under different matching biases. Among them, represents the average matching energy of the supplier A, represents the average matching energy of the demander B. The value range of f is between 0 and 1. When f approaches 0, the energy gap between the two matching parties is larger, indicating extreme inequality and low fairness and stability of the match. When f approaches 1, it means that the energies of the two matching parties are completely equal, that is, high fairness and stability, and the bias control factor is dynamically adjusted according to the evaluation results.

[0013] Step S5: Simulation verification, verify the accuracy and effectiveness of the model solution through simulation.

[0014] Furthermore, the specific content of step S2 includes: set the bias control factor as w, and the energy threshold of the supplier A then the optimization objective is: , where the constraint is , is the minimum total matching energy. The smaller the degree of bias control w of the platform, the lower the upper limit of the energy threshold set for the supplier A, and the more favorable the matching result is for the supplier A. During the matching process, when the matching energy of the supplier A is less than the threshold n, the matching proceeds normally; otherwise, by removing the candidates that make the matching energy of the supplier A greater than n, it is ensured that the matching energy of the supplier A is within the energy threshold.

[0015] Furthermore, calculating the probability that the minimum total matching energy equals a specific value includes:

[0016] Set the minimum total matching energy found by the platform to equal z, and the corresponding matching energy of the supplier A is , and at this time the matching energy of the demander B is greater than or equal to . First, calculate the probability that z is the minimum total matching energy, that is, the probability that the matching energies corresponding to all candidates are greater than z, denoted as P(u≥z). Since each pairing is a randomly independent event, the joint probability is equal to the product of the marginal probabilities, and since P(u≥z)=1 - P(u<z), the probability that the minimum total matching energy u = z is:

[0017] ;

[0018] where P represents probability, represents the probability of u<z, represents choosing the minimum value between z and n. For , the probability converges to the exponential function exp form.

[0019] Furthermore, the step S3 specifically includes: integrating the probability to obtain the average value of the minimum total energy which is:

[0020] ;

[0021] If the matching process of the supplier A is consistent with the ideal matching, the calculation method of the average matching energy of the supplier A is: , and the average matching energy of the demander B is the difference between the average total energy and the average matching energy of the supplier A. The average matching energy of the demander B is: = .

[0022] Existing recommendation platforms (such as e-commerce recommendations, social networks, recruitment systems, etc.) generally have the problem of matching imbalance caused by algorithmic bias, making it difficult for the platform to accurately identify the intensity of its own bias. The present invention proposes a method for identifying and controlling the matching results of a recommendation system. By introducing the bias control factor of the platform, constructing a matching bias identification model, solving the model, and evaluating the matching results, it realizes the quantitative evaluation of the influence of different degrees of bias on the bilateral matching results, optimizes the fairness of bilateral matching, and also optimizes and adjusts the bias control factor w, thereby optimizing the matching results. The solution of the present invention realizes the differential measurement of the benefits of bilateral matching entities and the quantitative evaluation of platform matching bias, so as to achieve better system recommendations and matching according to the evaluation results of matching bias, providing a reference for the effective prevention and control of the common matching bias problem on recommendation platforms. This method has broad application prospects in many fields such as online shopping platforms, social platforms, and recruitment platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 It is a schematic flowchart of a method for identifying and controlling the matching results of a recommendation system provided by an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of an ideal matching process provided by an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of a biased matching process provided by an embodiment of the present invention;

[0027] Figure 4 It is a schematic diagram of optimizing and adjusting the bias control factor w provided by an embodiment of the present invention;

[0028] Figure 5 It is a schematic diagram of the simulation verification result provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0030] The present invention proposes a method for identifying and controlling the matching results of a recommendation system, as Figure 1 shown, the method includes the following steps:

[0031] Step S1: Set model elements. The model involves an intermediary platform, supplier A, demander B, and N candidates for matching. Set preference information, perform personalized ranking on the N candidates. In specific operations, a random ranking method can be used to simulate this process, and rank the N candidates according to the preference information to form their respective preference lists. The matching energy of supplier A related to candidate i is defined as , and the matching energy of demander B related to candidate i is defined as , where i = 1, 2,..., N. The energy value is equal to the ranking of the corresponding candidate in the preference list. The higher the ranking, the smaller the energy value, indicating a more stable match. In practical applications, the candidates are products or information waiting to be selected by both parties at the same time.

[0032] Step S2: Build a matching bias recognition model. Introduce a bias control factor w to simulate the influence of the intermediary platform's bias on the matching results. In an ideal situation, the intermediary platform treats both sides fairly, and its matching goal is to minimize the total system energy: ;

[0033] Figure 2 shows the ideal matching process, where the two letter lists represent the preferences of supplier A and demander B respectively, and the numbers represent the priority order of the preference lists, that is, the matching energy corresponding to the candidates. The ticked position is the optimal matching item given by the platform.

[0034] In actual situations, the recommendation platform may be biased towards one party due to its own interests, giving priority to matching options with lower energy values for it, while ignoring the change in the matching energy of the other party. To simulate this biased matching process, introduce a bias control factor w. Let the energy threshold of supplier A be , then the optimization goal is adjusted to: , where the constraint condition is , is the minimum total matching energy. The smaller the bias control degree w of the intermediary, the lower the upper limit of the energy threshold set for supplier A, and the more favorable the matching result is for supplier A. During the matching process, when the matching energy of supplier A is less than the threshold n, the matching proceeds normally; otherwise, remove the candidates that make the matching energy of supplier A greater than n, so as to ensure that the energy of supplier A is within the threshold.

[0035] Figure 3 shows the biased matching process of the platform. When the platform sets the energy threshold of supplier A to 5, 2, and 1 respectively, the set of candidates that supplier A can choose becomes , and , since the optimal matching item is g when the platform has no bias, and the matching energy of supplier A is 3 at this time. Therefore, when , the matching energy of supplier A is within this threshold range, and the matching result is not disturbed; while when is set to 2 and 1, the matching energy of supplier A exceeds the energy threshold, and the platform will intervene to make the optimal matching items become c and d in turn to ensure that the matching energy of supplier A is always within the energy threshold range.

[0036] Step S3: Model solution, calculate the corresponding relationship between the bilateral matching energy and the bias control factor w. The idea of model solution is to first calculate the probability that the minimum total matching energy is equal to a specific value, and then integrate all probabilities to obtain the average value of the minimum total energy. Specifically:

[0037] Set the minimum total matching energy found by the recommendation platform to z. The minimum total matching energy is the candidate item with the minimum total matching energy of both parties found by the recommendation platform by obtaining the preference information of both parties. At this time, the matching energy of the corresponding supplier A is , so the matching energy of the demander B is greater than or equal to . First, calculate the probability that z is the minimum total matching energy, that is, the probability that the matching energy corresponding to all candidate items is greater than z, denoted as P(u≥z). Since each pairing is a random independent event, the joint probability is equal to the product of the marginal probabilities. Since P(u≥z)=1 - P(u<z), and for a relatively small , the equation holds, and the multiplication problem can be converted into an addition problem of exponential functions, and the probability that the minimum total matching energy u = z is obtained as:

[0038]

[0039] where P represents probability, represents the probability that u<z, represents choosing the minimum value between z and n. For , the probability converges to the exponential function exp form.

[0040] Then calculate the average value of the minimum total matching energy, integrate the probability, and obtain the average value as:

[0041]

[0042] For supplier A, its matching process is consistent with the ideal matching, so it is easy to calculate the average matching energy of A. : The average matching energy of demander B is the difference between the average total energy and the average matching energy of A. The average matching energy of demander B is: = ;

[0043] Step S4: Matching result evaluation. Evaluate the stability of the matching result through fairness indicators, and construct a matching fairness indicator f to measure the change in the matching fairness degree between the supplier and the demander under different matching biases. The value range of f is between 0 and 1. When f approaches 0, the energy gap between the two matching parties is larger, indicating extreme inequality and low fairness and stability of the matching. When f approaches 1, it means that the energy of the two matching parties is completely equal, that is, high fairness and stability.

[0044] At the same time, the bias control factor w can also be optimized and adjusted through the matching evaluation result. For example, Figure 4 as shown, the process of optimization and adjustment is as follows: The platform pre-sets the initial value of the bias control factor w, and then outputs the bilateral matching energy and matching fairness evaluation corresponding to w based on the matching bias recognition model. Finally, the bias control factor is dynamically adjusted according to the index result to make it in a relatively reasonable range and optimize the matching result.

[0045] In the application scenario of the shopping platform, at this time, supplier A is a merchant and demander B is a consumer. The calculation method of the bias control factor can also be w = 1 - M / (C + M), where M is the marketing service fee of the platform as an intermediary role, and C is the commission of the platform as an intermediary role. The smaller w is, the greater the bias of the platform.

[0046] Step S5: Simulation verification to verify the accuracy and effectiveness of the model solution;

[0047] To verify the accuracy of the solution result, when is calculated, the energy proportion of supplier A and demander B and the change of the matching fairness indicator f with the bias control factor are shown in Figure 5 as follows. The numerical calculation solution of the model (the dotted line in the figure, the part with subscript t) is consistent with the simulation experiment result (the solid line in the figure, the part with subscript s). The experimental result shows that the bias control factor can significantly affect the matching result, and different degrees of bias have significant differences in the impact on the matching result. By introducing the bias control factor w, the present invention realizes the differential measurement of the benefits of the bilateral matching entities and the quantitative evaluation of the platform bias, and optimizes the fairness of the bilateral matching.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying and controlling the matching results of a recommendation system, characterized in that, The method includes: Step S1, set model elements. The model involves a recommendation platform, supplier A, demander B, and N candidate items for matching. Each time, supplier A and demander B perform personalized ranking on the N candidate items to form their respective preference lists. The matching energy between supplier A and candidate item i is defined as , and the matching energy between demander B and candidate item i is defined as , where i = 1, 2, ……, N. The energy value is equal to the ranking of the corresponding candidate item in the preference list. The higher the ranking, the smaller the energy value, indicating a more stable match; Step S2: Build a matching bias recognition model, introduce a bias control factor w to simulate the impact of platform bias on the matching result, and set the energy threshold of supplier A , then the optimization objective is: ； Among them, the constraint condition is , is the minimum total matching energy. The smaller the bias control degree w of the platform, the lower the upper limit of the energy threshold set for supplier A, and the more favorable the matching result is for supplier A. During the matching process, when the matching energy of supplier A is less than the threshold n, the matching proceeds normally; otherwise, by removing the candidates that make the matching energy of supplier A greater than n, it is ensured that the matching energy of supplier A is within the energy threshold; Step S3, model solution, calculating the corresponding relationship between the bilateral matching energy and the bias control factor w. The idea of the model solution is to first calculate the probability that the minimum total matching energy is equal to a specific value, then integrate all probabilities to obtain the average value of the minimum total energy, and calculate the average matching energy of the supplier A and the demander B. The calculation of the probability that the minimum total matching energy is equal to a specific value includes: Set the minimum total matching energy found by the platform to be equal to z, and the matching energy of the corresponding supplier A is , and at this time the matching energy of the demander B is greater than or equal to . First, calculate the probability that z is the minimum total matching energy, that is, the probability that the matching energies corresponding to all candidates are greater than z, denoted as P(u≥z). Since each pairing is a random independent event, the joint probability is equal to the product of the marginal probabilities. And because P(u≥z)=1-P(u<z), the probability that the minimum total matching energy u = z is obtained as follows: ； where P represents probability, represents the probability that u < z, represents choosing the minimum of z and n, for , the probability converges to the exponential function exp form; Step S4, matching result evaluation. Evaluate the stability of the matching result through fairness indicators, and construct a matching fairness indicator f, , which is used to measure the change in the matching fairness degree between the supplier and the demander under different matching biases. Among them, represents the average matching energy of supplier A, represents the average matching energy of demander B. The value range of f is between 0 and 1. When f approaches 0, the energy gap between the two matching parties is larger, indicating extreme inequality and low fairness and stability of the matching. When f approaches 1, it means that the energies of the two matching parties are completely equal, that is, high fairness and stability, and dynamically adjust the bias control factor according to the evaluation result; Step S5: Simulation verification, verifying the accuracy and effectiveness of the model solution through simulation.

2. The method according to claim 1, wherein The said step S3 further includes: Integrate the probability to obtain the average value of the minimum total energy It is as follows: ； If the matching process of Supplier A is consistent with the ideal matching, the average matching energy of Supplier A is calculated as follows: , and the average matching energy of Demander B is the average total energy minus the average matching energy of Supplier A. The average matching energy of Demander B is: = .

Images