Dynamic evaluation method, system and equipment for business travel resource supplier performance and medium

Abstract

The invention discloses a dynamic evaluation method, system and device for business travel resource supplier performance, and a medium, mainly relates to the technical field of dynamic evaluation, and is used for solving the problems that a static weight cannot respond to market fluctuation, negative index processing is rough, and multi-target collaborative optimization cannot be performed on an order result and a price factor in an existing scheme. Comprising the steps of obtaining business travel order features provided by each supplier in a recent preset time period, inputting the business travel order features into a trained Q-Learning algorithm, and obtaining a Q value for selecting each supplier as an action; determining index weight distribution corresponding to each Q value; min-Max normalization is adopted, and the normalized value of the forward index feature is calculated; calculating a normalized value of the negative index feature by adopting reciprocal normalization; calculating the comprehensive performance score of each supplier according to the normalized value and the index weight distribution corresponding to the supplier; and according to the comprehensive performance score, determining the ranking sequence of the suppliers in the resource provider dynamic evaluation system.

Description

Technical Field

[0001] This application relates to the field of dynamic evaluation technology, and in particular to a method, system, device and medium for dynamic evaluation of the performance of business travel resource providers. Background Technology

[0002] Currently, performance evaluation of business travel resource providers primarily employs static weighted scoring methods or statistical analysis based on historical data. Common approaches include: 1) manually setting fixed weights and calculating the total score through linear weighting; 2) using the analytic hierarchy process (AHP) to construct a hierarchical structure of indicators and determining weights through expert scoring; and 3) analyzing historical order data based on regression models to predict future supplier performance. These methods rely on human experience or static models and require periodic weight adjustments to adapt to business changes. Some companies have attempted to introduce basic machine learning algorithms for feature importance analysis, but issues such as lagging model updates and insufficient dynamic adaptability remain. Existing technologies generally focus on single-dimensional indicators (such as lowest price priority) and lack quantitative assessments of the dynamic correlation between supplier service stability (such as temporary cancellation rates) and business scenarios.

[0003] Traditional evaluation methods have the following drawbacks: static weights cannot respond to market fluctuations; negative indicators (such as complaint rates) are handled crudely, and conventional linear normalization dilutes the impact of extreme values, leading to the underestimation of high-risk suppliers; current technical solutions also have limitations in the design of reward functions, failing to perform multi-objective collaborative optimization of order results (such as customer satisfaction) and price factors. Summary of the Invention

[0004] This application provides a dynamic evaluation method, system, equipment, and medium for the performance of business travel resource suppliers, in order to solve the problems of existing solutions, such as static weights failing to respond to market fluctuations, poor handling of negative indicators (such as complaint rates), conventional linear normalization diluting the impact of extreme values, leading to the underestimation of high-risk suppliers, and failure to perform multi-objective collaborative optimization of order results and price factors.

[0005] Firstly, this application provides a dynamic evaluation method for the performance of business travel resource providers, the method comprising: Obtain the characteristics of business travel orders provided by each supplier within a preset historical time period; among which, the characteristics of business travel orders include positive indicator characteristics and negative indicator characteristics; Take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy strategy as the policy, and calculate the reward using the order results and order prices in the business travel order features; use the state, action, policy, reward and initial Q value to train the Q-Learning algorithm and obtain the trained Q-Learning algorithm; Obtain the characteristics of business travel orders provided by each supplier within the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm, obtain the Q value for selecting each supplier as an action, and determine the indicator weight distribution corresponding to each Q value; Min-Max normalization is used to calculate the normalized value of positive indicator features; reciprocal normalization is used to calculate the normalized value of negative indicator features; based on the normalized value and the indicator weight distribution corresponding to the supplier, the comprehensive performance score of each supplier is calculated. The ranking order of suppliers in the resource provider dynamic evaluation system is determined based on the comprehensive performance score.

[0006] In one implementation of this application, the business travel order characteristics include at least: business travel order characteristic data, service quality dimension data, stability dimension data, financial dimension data, and risk control dimension data; Furthermore, the business travel order characteristic data includes: booking price and handling fee; service quality data includes: order confirmation rate, average confirmation time, and customer complaint rate; stability data includes: interface availability rate; financial data includes: settlement accuracy rate; and risk control data includes: number of violation records.

[0007] In one implementation of this application, the reward is calculated using the order result and order price in the characteristics of a business travel order, specifically including: Through the formula; R = W_base*R_base + W_saving *R_saving + W_complaint*R_complaint, calculate the reward R; Here, W_base, W_saving, and W_complain are preset weights; R_base represents the order result, where success is a positive value and failure is a negative value; R_saving represents the savings reward, which is the highest price of the same hotel order among different suppliers minus the price of the current supplier; R_complaint represents the complaint penalty, which is 0 when there is no complaint and a preset negative value when there is a complaint.

[0008] In one implementation of this application, determining the index weight distribution corresponding to each Q value specifically includes: The relationship between the Q value and the indicator weight distribution can be obtained through the preset interface. The magnitude of the Q value is directly proportional to the weight of positive indicator features in the indicator weight distribution and inversely proportional to the weight of negative indicator features.

[0009] In one implementation of this application, Min-Max normalization is used to calculate the normalized value of the positive indicator feature, specifically including: The normalized value of the positive indicator feature, Score_i, is calculated using the formula: Score_i = (X_i - X_min) / (X_max - X_min). Where X_i represents the i-th positive indicator feature, X_min represents the preset minimum value of the i-th positive indicator feature, X_max represents the preset maximum value of the i-th positive indicator feature, the value range of i is [1,N], and N represents the number of positive indicator features.

[0010] In one implementation of this application, reciprocal normalization is used to calculate the normalized value of the negative index feature, specifically including: Through the formula: Score_j = (X_transformed_j - X_transformed_min) / (X_transformed_max - X_transformed_min), calculates the normalized value of the negative indicator feature, Score_j; Where X_transformed_j represents the j-th negative indicator feature, X_transformed_min represents the preset minimum value of the j-th negative indicator feature, and X_transformed_max represents the preset maximum value of the j-th negative indicator feature. The value range of j is [1, M], and M represents the number of negative indicator features.

[0011] In one implementation of this application, the comprehensive performance score of each supplier is calculated based on the normalized value and the corresponding indicator weight distribution of the supplier, specifically including: Through the formula: Calculate the supplier's overall performance score ; Where N represents the number of positive indicator features, and M represents the number of negative indicator features. This represents the normalized value of the i-th positive indicator feature. This represents the weight value of the i-th positive indicator feature in the indicator weight distribution. Indicates the first The normalized value of each negative indicator feature. This represents the weight value of the j-th negative indicator feature in the indicator weight distribution.

[0012] Secondly, this application provides a dynamic evaluation system for the performance of business travel resource providers, the system comprising: The acquisition module is used to acquire the characteristics of business travel orders provided by each supplier within a preset historical time period; among which, the characteristics of business travel orders include positive indicator characteristics and negative indicator characteristics; The training module is used to take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy policy as the policy, and calculate the reward using the order results and order prices in the business travel order features; using the state, action, policy, reward and initial Q value, the Q-Learning algorithm is trained to obtain the trained Q-Learning algorithm; The weighting module is used to obtain the characteristics of business travel orders provided by each supplier in the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm to obtain the Q value for selecting each supplier as an action; and determine the indicator weight distribution corresponding to each Q value. The ranking module is used to calculate the normalized values ​​of positive indicator features using Min-Max normalization; to calculate the normalized values ​​of negative indicator features using reciprocal normalization; to calculate the comprehensive performance score of each supplier based on the normalized values ​​and the corresponding indicator weight distribution of the supplier; and to determine the ranking order of the suppliers in the resource provider dynamic evaluation system based on the comprehensive performance score.

[0013] Thirdly, this application provides a dynamic evaluation device for the performance of business travel resource providers, the device comprising: processor; And a memory that stores executable code, which, when executed, causes the processor to perform a dynamic evaluation method for the performance of a business travel resource provider, as described above.

[0014] Fourthly, this application provides a non-volatile computer storage medium storing computer instructions, which, when executed, implement a dynamic evaluation method for the performance of a business travel resource provider as described above.

[0015] As can be seen from the above technical solutions, this application has the following advantages: I. The real-time advantage of dynamic weighting in adapting to market fluctuations: By constructing a dynamic weight allocation mechanism using the Q-Learning algorithm, the lag problem of traditional static weight models is fundamentally solved. The technical solution explicitly uses order results and price as core parameters of the reward function, enabling the weight distribution to adjust in real time according to market supply and demand, price fluctuations, and other variables. For example, when a supplier suddenly experiences a decline in service quality (a change in negative indicator characteristics), the algorithm can automatically reduce its weight percentage through iterative feedback of state-action-reward, without manual intervention. This dynamic response capability based on reinforcement learning significantly improves the evaluation system's agility in responding to supply chain anomalies.

[0016] II. The data fidelity advantage of multi-indicator differentiated normalization: A dual-track processing strategy of Min-Max and reciprocal normalization is adopted: for positive indicators (such as order completion rate), the comparability advantage of linear normalization is retained, while for negative indicators (such as complaint rate), the impact of extreme values ​​is amplified through reciprocal transformation. This design directly solves the problem of evaluation bias of high-risk suppliers by conventional linear normalization—when a supplier has an extremely high negative indicator value, reciprocal normalization will cause its normalization result to decay exponentially, thus accurately reflecting its potential risk in the calculation of the comprehensive performance score, avoiding the dilution of danger signals caused by data smoothing in traditional methods.

[0017] III. Decision support advantages of multi-objective collaborative optimization: By simultaneously incorporating order results (quality dimension) and order price (cost dimension) into the reward calculation module of Q-Learning, the technical solution achieves multi-objective collaborative optimization of supplier evaluation. During algorithm training, suppliers with high prices but low quality will naturally receive lower Q-values ​​due to the design of the reward function, while suppliers with the best cost-effectiveness will emerge through strategy iteration. This mechanism directly avoids the resource allocation imbalance caused by single-dimensional evaluation, ensuring that the ranking order reflects both service quality and procurement costs, providing a more comprehensive decision-making basis for business travel resource scheduling. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a dynamic evaluation method for the performance of a business travel resource supplier provided in an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of the internal structure of a dynamic evaluation system for the performance of business travel resource suppliers provided in an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the internal structure of a dynamic evaluation device for the performance of a business travel resource supplier provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Those skilled in the art should understand that the embodiments described below are merely preferred embodiments of this disclosure and do not imply that this disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely used to explain the technical principles of this disclosure and are not intended to limit the scope of protection of this disclosure. Based on the preferred embodiments provided by this disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of this disclosure.

[0024] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0025] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0026] The embodiment provides a dynamic evaluation method for the performance of business travel resource providers, such as... Figure 1 As shown in the embodiments of this application, the method mainly includes the following steps: Step 110: Obtain the characteristics of business travel orders provided by each supplier within a preset historical time period.

[0027] In some embodiments, business travel order characteristics include at least: business travel order characteristic data, service quality dimension data, stability dimension data, financial dimension data, and risk control dimension data; Furthermore, the business travel order characteristic data includes: booking price and handling fee; service quality data includes: order confirmation rate, average confirmation time, and customer complaint rate; stability data includes: interface availability rate; financial data includes: settlement accuracy rate; and risk control data includes: number of violation records.

[0028] It's important to note that the systematic collection of multi-dimensional characteristic data from suppliers' historical business travel orders provides a structured analytical foundation for subsequent supplier evaluation. Specifically, the solution clearly defines a data collection framework encompassing five core dimensions: business travel order characteristic data (booking price, handling fees), service quality data (order confirmation rate, average confirmation time, customer complaint rate), stability data (interface availability), financial data (settlement accuracy), and risk control data (number of violations). This standardized data collection method ensures that the evaluation system covers key indicators throughout the entire lifecycle of supplier services, avoiding potential dimension gaps in traditional evaluations. By integrating objective data fields such as price, service response, system stability, financial compliance, and risk control, quantifiable data support is provided for establishing a multi-dimensional supplier performance evaluation model.

[0029] Step 120: Take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy policy as the policy, and calculate the reward using the order results and order prices in the business travel order features; use the state, action, policy, reward and initial Q value to train the Q-Learning algorithm and obtain the trained Q-Learning algorithm.

[0030] The reward calculation, which utilizes order results and prices from business travel order characteristics, specifically includes: Through the formula; R = W_base*R_base + W_saving *R_saving + W_complaint*R_complaint, calculate the reward R; Here, W_base, W_saving, and W_complain are preset weights; R_base represents the order result, where success is a positive value and failure is a negative value; R_saving represents the savings reward, which is the highest price of the same hotel order among different suppliers minus the price of the current supplier; R_complaint represents the complaint penalty, which is 0 when there is no complaint and a preset negative value when there is a complaint.

[0031] It's worth noting that by establishing a dynamic supplier selection mechanism based on the Q-Learning algorithm, intelligent decision-making for business travel service procurement has been achieved. Its core advantages are: First, it employs a multi-dimensional reward function design, incorporating order results (R_base), price savings (R_saving), and complaint status (R_complaint) into a unified evaluation framework. Through flexible configuration of preset weights (W_base / W_saving / W_complaint), it can adapt to the priority requirements of different business scenarios. This design ensures the basic value of successful orders, incentivizes cost optimization through price differences, and constrains service quality through a complaint penalty mechanism. Second, the introduction of the ε-greedy strategy strikes a balance between exploring new suppliers and utilizing known high-quality suppliers, avoiding strategy rigidity caused by solely pursuing short-term gains. During training, the initial Q-value provides a learning starting point for the algorithm, while the closed-loop feedback mechanism of state-action-reward enables the system to continuously optimize its supplier selection strategy.

[0032] This is mainly reflected in two aspects: First, by using historical order characteristics (such as interface availability and settlement accuracy) as state input, the decision-making process can comprehensively reflect key attributes such as supplier stability and financial compliance, avoiding decision-making biases caused by relying on a single indicator. Second, the dynamic reward calculation mechanism enables the system to capture market changes in real time—when a supplier exhibits a price advantage or fluctuates in service quality, the algorithm automatically adjusts the selection probability through the reward function. This design is particularly suitable for scenarios in business travel services where hotel prices fluctuate frequently and supplier service quality is unstable, and it is more adaptable to actual business needs compared to static evaluation methods. All decisions are based on objective order data, without introducing any subjective evaluations or unverified assumptions, ensuring the interpretability of the algorithm's output.

[0033] Step 130: Obtain the characteristics of business travel orders provided by each supplier in the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm to obtain the Q value for selecting each supplier as an action; determine the indicator weight distribution corresponding to each Q value.

[0034] Determine the weight distribution of the indicators corresponding to each Q value, specifically including: The relationship between the Q value and the indicator weight distribution can be obtained through the preset interface. The magnitude of the Q value is directly proportional to the weight of positive indicator features in the indicator weight distribution and inversely proportional to the weight of negative indicator features.

[0035] It's important to note that by dynamically adjusting the weight distribution of indicators in the supplier selection strategy, refined decision-making in business travel service procurement is achieved. Its core value lies in two aspects: First, applying the trained Q-Learning algorithm to real-time order feature data (such as the latest interface availability and current price levels) allows supplier selection to promptly reflect market changes and service quality fluctuations. Second, through the pre-defined relationship between Q-values ​​and indicator weights (positive indicator weights increase with increasing Q-value, while negative indicator weights decrease), the system can automatically optimize evaluation dimensions—for example, when a supplier has a significant price advantage, its R_saving indicator weight will be strengthened; if a complaint record exists, the penalty weight of R_complaint will be increased accordingly. This dynamic weight allocation mechanism avoids the rigidity of static scoring systems, allowing the algorithm to flexibly adjust evaluation priorities according to actual business needs. All weight adjustments are based on objective Q-value calculations, without introducing human intervention or empirical assumptions, ensuring the transparency and traceability of the decision-making process.

[0036] Step 140: Calculate the normalized values ​​of positive indicator features using Min-Max normalization; calculate the normalized values ​​of negative indicator features using reciprocal normalization; calculate the comprehensive performance score of each supplier based on the normalized values ​​and the corresponding indicator weight distribution; determine the ranking order of suppliers in the resource provider dynamic evaluation system based on the comprehensive performance score.

[0037] In some embodiments, Min-Max normalization is used to calculate the normalized value of the positive indicator features, specifically including: The normalized value of the positive indicator feature, Score_i, is calculated using the formula: Score_i = (X_i - X_min) / (X_max - X_min). Where X_i represents the i-th positive indicator feature, X_min represents the preset minimum value of the i-th positive indicator feature, X_max represents the preset maximum value of the i-th positive indicator feature, the value range of i is [1,N], and N represents the number of positive indicator features.

[0038] In some embodiments, reciprocal normalization is used to calculate the normalized value of the negative indicator feature, specifically including: Through the formula: Score_j = (X_transformed_j - X_transformed_min) / (X_transformed_max - X_transformed_min), calculates the normalized value of the negative indicator feature, Score_j; Where X_transformed_j represents the j-th negative indicator feature, X_transformed_min represents the preset minimum value of the j-th negative indicator feature, and X_transformed_max represents the preset maximum value of the j-th negative indicator feature. The value range of j is [1, M], and M represents the number of negative indicator features.

[0039] In some embodiments, the comprehensive performance score of each supplier is calculated based on the normalized value and the corresponding indicator weight distribution of the supplier, specifically including: Through the formula: Calculate the supplier's overall performance score ; Where N represents the number of positive indicator features, and M represents the number of negative indicator features. This represents the normalized value of the i-th positive indicator feature. This represents the weight value of the i-th positive indicator feature in the indicator weight distribution. Indicates the first The normalized value of each negative indicator feature. This represents the weight value of the j-th negative indicator feature in the indicator weight distribution.

[0040] It's important to note that the standardized scoring calculation process ensures the standardization and comparability of supplier performance evaluation. Its core value lies in two aspects: First, Min-Max normalization is used for positive indicators (such as order confirmation rate and price savings), converting indicators of different dimensions into standardized scores within the [0,1] range, eliminating the impact of differences in the magnitude of indicators on the evaluation results. For negative indicators (such as complaint rate and violation records), reciprocal normalization is used, maintaining the relative differences between indicators while ensuring the consistency of the evaluation direction—a higher score indicates better performance. This differentiated approach allows various indicators to coexist reasonably within the same evaluation system.

[0041] Secondly, by setting preset thresholds for X_min and X_max, the system can filter out abnormal data interference, ensuring the stability of the evaluation results. For example, when a supplier's interface availability fluctuates abnormally, the preset maximum and minimum values ​​can effectively control the impact of this indicator on the total score. Simultaneously, the comprehensive performance score calculation formula based on weight distribution allows the system to flexibly adjust the weight of each evaluation dimension according to the needs of different business scenarios. This design is particularly suitable for business travel procurement scenarios that require a balance between cost control and service quality.

[0042] Finally, the dynamic ranking mechanism based on comprehensive performance scores provides an intuitive basis for resource allocation decisions. The ranking results directly reflect the relative strengths and weaknesses of suppliers under a multi-dimensional evaluation system, avoiding decision-making biases that may result from single-indicator evaluations. The entire process is based entirely on objective data calculations, without introducing subjective judgment factors, ensuring the fairness and traceability of the evaluation results. System maintenance personnel can optimize the evaluation strategy by adjusting the weight distribution parameters without modifying the underlying algorithm structure, demonstrating high maintainability.

[0043] As described above, this embodiment constructs a dynamic weight allocation mechanism using the Q-Learning algorithm, fundamentally solving the lag problem of traditional static weight models. The technical solution explicitly uses order results and price as core parameters of the reward function, enabling the weight distribution to adjust in real time according to market supply and demand, price fluctuations, and other variables. For example, when a supplier suddenly experiences a decline in service quality (a change in negative indicator characteristics), the algorithm can automatically reduce its weight percentage through iterative feedback of state-action-reward, without manual intervention. This dynamic response capability based on reinforcement learning significantly improves the evaluation system's agility in responding to supply chain anomalies.

[0044] A dual-track processing strategy of Min-Max and reciprocal normalization is adopted: for positive indicators (such as order completion rate), the comparability advantage of linear normalization is retained, while for negative indicators (such as complaint rate), the impact of extreme values ​​is amplified through reciprocal transformation. This design directly solves the problem of evaluation bias of high-risk suppliers by conventional linear normalization—when a supplier has an extremely high negative indicator value, reciprocal normalization will cause its normalization result to decay exponentially, thus accurately reflecting its potential risk in the calculation of the comprehensive performance score, avoiding the dilution of danger signals caused by data smoothing in traditional methods.

[0045] By simultaneously incorporating order results (quality dimension) and order price (cost dimension) into the reward calculation module of Q-Learning, the technical solution achieves multi-objective collaborative optimization of supplier evaluation. During algorithm training, suppliers with high prices but low quality will naturally receive lower Q-values ​​due to the design of the reward function, while suppliers with the best cost-effectiveness will emerge through strategy iteration. This mechanism directly avoids the resource allocation imbalance caused by single-dimensional evaluation, ensuring that the ranking order reflects both service quality and procurement costs, providing a more comprehensive decision-making basis for business travel resource scheduling.

[0046] In addition, this application Figure 2 This application provides a dynamic evaluation system for the performance of business travel resource providers. For example... Figure 2 As shown in the embodiments of this application, the system mainly includes: The acquisition module 210 is used to acquire the characteristics of business travel orders provided by each supplier within a preset historical time period; wherein, the characteristics of business travel orders include positive indicator characteristics and negative indicator characteristics; Business travel order characteristics should include at least: business travel order characteristic data, service quality data, stability data, financial data, and risk control data; Furthermore, the business travel order characteristic data includes: booking price and handling fee; service quality data includes: order confirmation rate, average confirmation time, and customer complaint rate; stability data includes: interface availability rate; financial data includes: settlement accuracy rate; and risk control data includes: number of violation records.

[0047] Training module 220 is used to take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy policy as the policy, calculate the reward using the order results and order prices in the business travel order features; and train the Q-Learning algorithm using the state, action, policy, reward and initial Q value to obtain the trained Q-Learning algorithm. Rewards are calculated using order results and order prices from business travel order characteristics, specifically including: Through the formula; R = W_base*R_base + W_saving *R_saving + W_complaint*R_complaint, calculate the reward R; Here, W_base, W_saving, and W_complain are preset weights; R_base represents the order result, where success is a positive value and failure is a negative value; R_saving represents the savings reward, which is the highest price of the same hotel order among different suppliers minus the price of the current supplier; R_complaint represents the complaint penalty, which is 0 when there is no complaint and a preset negative value when there is a complaint.

[0048] The weight module 230 is used to obtain the characteristics of business travel orders provided by each supplier in the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm to obtain the Q value for selecting each supplier as an action; and determine the indicator weight distribution corresponding to each Q value. The sorting module 240 is used to calculate the normalized value of positive indicator features using Min-Max normalization; to calculate the normalized value of negative indicator features using reciprocal normalization; to calculate the comprehensive performance score of each supplier based on the normalized value and the indicator weight distribution corresponding to the supplier; and to determine the ranking order of suppliers in the resource provider dynamic evaluation system based on the comprehensive performance score.

[0049] Min-Max normalization is used to calculate the normalized values ​​of positive indicator features, specifically including: The normalized value of the positive indicator feature, Score_i, is calculated using the formula: Score_i = (X_i - X_min) / (X_max - X_min). Where X_i represents the i-th positive indicator feature, X_min represents the preset minimum value of the i-th positive indicator feature, X_max represents the preset maximum value of the i-th positive indicator feature, the value range of i is [1,N], and N represents the number of positive indicator features.

[0050] Inverse normalization is used to calculate the normalized value of the negative indicator characteristics, specifically including: Through the formula: Score_j = (X_transformed_j - X_transformed_min) / (X_transformed_max - X_transformed_min), calculates the normalized value of the negative indicator feature, Score_j; Where X_transformed_j represents the j-th negative indicator feature, X_transformed_min represents the preset minimum value of the j-th negative indicator feature, and X_transformed_max represents the preset maximum value of the j-th negative indicator feature. The value range of j is [1, M], and M represents the number of negative indicator features.

[0051] Based on the normalized values ​​and the corresponding indicator weight distribution for each supplier, the comprehensive performance score for each supplier is calculated, specifically including: Through the formula: Calculate the supplier's overall performance score ; Where N represents the number of positive indicator features, and M represents the number of negative indicator features. This represents the normalized value of the i-th positive indicator feature. This represents the weight value of the i-th positive indicator feature in the indicator weight distribution. Indicates the first The normalized value of each negative indicator feature. This represents the weight value of the j-th negative indicator feature in the indicator weight distribution.

[0052] As described above, this embodiment constructs a dynamic weight allocation mechanism using the Q-Learning algorithm, fundamentally solving the lag problem of traditional static weight models. The technical solution explicitly uses order results and price as core parameters of the reward function, enabling the weight distribution to adjust in real time according to market supply and demand, price fluctuations, and other variables. For example, when a supplier suddenly experiences a decline in service quality (a change in negative indicator characteristics), the algorithm can automatically reduce its weight percentage through iterative feedback of state-action-reward, without manual intervention. This dynamic response capability based on reinforcement learning significantly improves the evaluation system's agility in responding to supply chain anomalies.

[0053] A dual-track processing strategy of Min-Max and reciprocal normalization is adopted: for positive indicators (such as order completion rate), the comparability advantage of linear normalization is retained, while for negative indicators (such as complaint rate), the impact of extreme values ​​is amplified through reciprocal transformation. This design directly solves the problem of evaluation bias of high-risk suppliers by conventional linear normalization—when a supplier has an extremely high negative indicator value, reciprocal normalization will cause its normalization result to decay exponentially, thus accurately reflecting its potential risk in the calculation of the comprehensive performance score, avoiding the dilution of danger signals caused by data smoothing in traditional methods.

[0054] By simultaneously incorporating order results (quality dimension) and order price (cost dimension) into the reward calculation module of Q-Learning, the technical solution achieves multi-objective collaborative optimization of supplier evaluation. During algorithm training, suppliers with high prices but low quality will naturally receive lower Q-values ​​due to the design of the reward function, while suppliers with the best cost-effectiveness will emerge through strategy iteration. This mechanism directly avoids the resource allocation imbalance caused by single-dimensional evaluation, ensuring that the ranking order reflects both service quality and procurement costs, providing a more comprehensive decision-making basis for business travel resource scheduling.

[0055] The above are method embodiments of this application. Based on the same inventive concept, embodiments of this application also provide a dynamic evaluation device for the performance of business travel resource providers. Figure 3 As shown, the device includes: a processor; and a memory storing executable code thereon, which, when executed, causes the processor to perform a dynamic evaluation method for the performance of a business travel resource provider as described in the above embodiments.

[0056] Specifically, the server retrieves the characteristics of business travel orders provided by each supplier within a preset historical time period. These business travel order characteristics include both positive and negative indicator features. The server uses the business travel order characteristics of all suppliers as the state, selects the i-th supplier as the action, applies the ε-greedy strategy as the policy, and calculates the reward using the order results and prices from the business travel order characteristics. Using the state, action, policy, reward, and initial Q-value, the server trains the Q-Learning algorithm to obtain a trained Q-Learning algorithm. The server then retrieves the business travel order characteristics of each supplier within the most recent preset time period, inputs these characteristics into the trained Q-Learning algorithm, and obtains the Q-values ​​for selecting each supplier as an action. The server determines the indicator weight distribution corresponding to each Q-value. Min-Max normalization is used to calculate the normalized value of the positive indicator features. Reciprocal normalization is used to calculate the normalized value of the negative indicator features. Based on the normalized values ​​and the indicator weight distribution corresponding to the suppliers, the server calculates the comprehensive performance score for each supplier. Based on the comprehensive performance score, the server determines the ranking order of the suppliers in the resource provider dynamic evaluation system.

[0057] By constructing a complete intelligent supplier evaluation closed loop, the system achieves continuous optimization capabilities for business travel procurement decisions. First, the system establishes a state space based on historical order characteristics (including positive indicators such as interface response time and order success rate, and negative indicators such as complaint rate and price fluctuations). Combined with the exploration-utilization balancing mechanism of the ε-greedy strategy, the Q-Learning algorithm can continuously discover potential high-quality suppliers while ensuring basic service quality. This design avoids the path dependence problem caused by traditional fixed supplier lists. When new suppliers appear in the market or the service quality of existing suppliers changes, the system can automatically adjust its selection strategy through a reward function.

[0058] Secondly, the dynamic weight allocation mechanism solves the multi-objective optimization problem. By establishing a mathematical relationship between the Q-value and the indicator weights (e.g., the weight of the price-saving indicator increases as the Q-value increases), the system can automatically adjust the evaluation focus according to actual business needs. For example, it automatically strengthens the weight of price indicators during peak travel seasons and increases the penalty coefficient for complaint indicators in service quality-sensitive scenarios. The combined application of Min-Max normalization and reciprocal normalization ensures the comparability of indicators of different natures (such as percentage scores and count data) in the calculation of comprehensive performance scores, avoiding evaluation bias caused by differences in indicator dimensions.

[0059] The resulting dynamic supplier ranking has three key benefits: First, it intuitively reflects a supplier's relative position within a multi-dimensional evaluation system through comprehensive performance scores; second, the ranking results are generated entirely based on objective order data, eliminating the subjectivity of human evaluation; and third, the ranking update mechanism is synchronized with the business cycle (e.g., monthly / quarterly updates), which not only reflects changes in supplier performance promptly but also avoids increased decision-making costs due to frequent changes. The entire system, through standardized data processing procedures and configurable weight parameters, maintains the stability of the core algorithm logic while providing flexibility to adapt to different business scenarios.

[0060] In addition, this application embodiment also provides a non-volatile computer storage medium storing executable instructions, which, when executed, implement a dynamic evaluation method for the performance of a business travel resource supplier as described above.

[0061] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dynamic evaluation method for the performance of business travel resource providers, characterized in that, The method includes: Obtain the characteristics of business travel orders provided by each supplier within a preset historical time period; among which, the characteristics of business travel orders include positive indicator characteristics and negative indicator characteristics; Take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy strategy as the policy, and calculate the reward using the order results and order prices in the business travel order features; use the state, action, policy, reward and initial Q value to train the Q-Learning algorithm and obtain the trained Q-Learning algorithm; Obtain the characteristics of business travel orders provided by each supplier within the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm, obtain the Q value for selecting each supplier as an action, and determine the indicator weight distribution corresponding to each Q value; Min-Max normalization is used to calculate the normalized values ​​of positive indicator features; reciprocal normalization is used to calculate the normalized values ​​of negative indicator features; based on the normalized values ​​and the corresponding indicator weight distribution of suppliers, the comprehensive performance score of each supplier is calculated; based on the comprehensive performance score, the ranking order of suppliers in the resource provider dynamic evaluation system is determined.

2. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Business travel order characteristics should include at least: business travel order characteristic data, service quality data, stability data, financial data, and risk control data; Furthermore, the business travel order characteristic data includes: booking price and handling fee; service quality data includes: order confirmation rate, average confirmation time, and customer complaint rate; stability data includes: interface availability rate; financial data includes: settlement accuracy rate; and risk control data includes: number of violation records.

3. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Rewards are calculated using order results and order prices from business travel order characteristics, specifically including: Through the formula; R = W_base*R_base + W_saving *R_saving + W_complaint*R_complaint, calculate the reward R; Here, W_base, W_saving, and W_complain are preset weights; R_base represents the order result, where success is a positive value and failure is a negative value; R_saving represents the savings reward, which is the highest price of the same hotel order among different suppliers minus the price of the current supplier; R_complaint represents the complaint penalty, which is 0 when there is no complaint and a preset negative value when there is a complaint.

4. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Determine the weight distribution of the indicators corresponding to each Q value, specifically including: The relationship between the Q value and the indicator weight distribution can be obtained through the preset interface. The magnitude of the Q value is directly proportional to the weight of positive indicator features in the indicator weight distribution and inversely proportional to the weight of negative indicator features.

5. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Min-Max normalization is used to calculate the normalized values ​​of positive indicator features, specifically including: The normalized value of the positive indicator feature, Score_i, is calculated using the formula: Score_i = (X_i - X_min) / (X_max - X_min). Where X_i represents the i-th positive indicator feature, X_min represents the preset minimum value of the i-th positive indicator feature, X_max represents the preset maximum value of the i-th positive indicator feature, the value range of i is [1,N], and N represents the number of positive indicator features.

6. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Inverse normalization is used to calculate the normalized value of the negative indicator characteristics, specifically including: Through the formula: Score_j = (X_transformed_j - X_transformed_min) / (X_transformed_max - X_transformed_min), calculates the normalized value of the negative indicator feature, Score_j; Where X_transformed_j represents the j-th negative indicator feature, X_transformed_min represents the preset minimum value of the j-th negative indicator feature, and X_transformed_max represents the preset maximum value of the j-th negative indicator feature. The value range of j is [1, M], and M represents the number of negative indicator features.

7. The dynamic evaluation method for the performance of business travel resource suppliers according to claim 1, characterized in that, Based on the normalized values ​​and the corresponding indicator weight distribution for each supplier, the comprehensive performance score for each supplier is calculated, specifically including: Through the formula: Calculate the supplier's overall performance score ; Where N represents the number of positive indicator features, and M represents the number of negative indicator features. This represents the normalized value of the i-th positive indicator feature. This represents the weight value of the i-th positive indicator feature in the indicator weight distribution. Indicates the first The normalized value of each negative indicator feature. This represents the weight value of the j-th negative indicator feature in the indicator weight distribution.

8. A dynamic evaluation system for the performance of business travel resource providers, characterized in that, The system includes: The acquisition module is used to acquire the characteristics of business travel orders provided by each supplier within a preset historical time period; among which, the characteristics of business travel orders include positive indicator characteristics and negative indicator characteristics; The training module is used to take the business travel order features of all suppliers as the state, select the i-th supplier as the action, take the ε-greedy policy as the policy, and calculate the reward using the order results and order prices in the business travel order features; using the state, action, policy, reward and initial Q value, the Q-Learning algorithm is trained to obtain the trained Q-Learning algorithm; The weighting module is used to obtain the characteristics of business travel orders provided by each supplier in the most recent preset time period, input the business travel order characteristics into the trained Q-Learning algorithm to obtain the Q value for selecting each supplier as an action; and determine the indicator weight distribution corresponding to each Q value. The ranking module is used to calculate the normalized values ​​of positive indicator features using Min-Max normalization; to calculate the normalized values ​​of negative indicator features using reciprocal normalization; to calculate the comprehensive performance score of each supplier based on the normalized values ​​and the corresponding indicator weight distribution of the supplier; and to determine the ranking order of the suppliers in the resource provider dynamic evaluation system based on the comprehensive performance score.

9. A dynamic evaluation device for the performance of business travel resource suppliers, characterized in that, The device includes: processor; And a memory having executable code stored thereon, which, when executed, causes the processor to perform a dynamic evaluation method for the performance of a business travel resource provider as described in any one of claims 1-7.

10. A non-volatile computer storage medium, characterized in that, It stores computer instructions, which, when executed, implement a dynamic evaluation method for the performance of a business travel resource supplier as described in any one of claims 1-7.

Images