Operation management system based on reinforcement learning

Abstract

The invention relates to the technical field of operation management, in particular to an operation management system based on reinforcement learning, and the system comprises an operation analysis unit which is used for determining an order state according to an order development coefficient and an order danger index, and determining an operation analysis mode according to the order state; the multi-stage optimization unit is used for determining an optimization mode according to the complaint homogeneity in multi-stage optimization analysis; the racking adjusting unit is used for determining a racking adjusting mode according to the defect mapping coefficient; the agent optimization unit is used for determining an agent adjustment mode according to the yield correlation coefficient; the recommendation optimization unit is used for determining an operation recommendation mode according to the agency influence coefficient and the development difficulty coefficient in recommendation optimization analysis; the operation management efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of operation management, and in particular, to an operation management system based on reinforcement learning. Background Art

[0002] Driven by the current wave of digitalization and intelligentization, the operation environment of the operator industry is undergoing unprecedented changes. In traditional operator industries, there are generally problems such as low efficiency, scattered data, and cumbersome operations in business promotion, channel management, commission settlement, etc. These problems not only restrict the development speed of operator services, but also increase operation costs and reduce customer satisfaction. In order to break through these bottlenecks and achieve intelligent upgrading of operation management, it has become an inevitable choice to integrate reinforcement learning technology into the operation management system. With its powerful autonomous learning and decision-making capabilities, reinforcement learning technology can automatically adjust operation strategies and optimize business processes according to the changing market environment and business requirements. Therefore, how to give full play to the advantages of reinforcement learning technology to improve operation management efficiency is a technical problem that needs to be solved urgently by those skilled in the art.

[0003] Chinese Patent Publication No. CN113537826A discloses a manufacturing operation management system, including: a business terminal for receiving customer order information, a main control terminal for processing customer order information, and a management terminal for allocating customer order information. The business terminal, the main control terminal, and the management terminal are all connected in sequence through Ethernet; the business terminal includes a receiving module and a payment module connected to the receiving module; the main control terminal includes a user interface, a database, and an analysis module; the database includes a recording unit for receiving order information sent by the business terminal and a storage unit for storing order information; the analysis module performs segmentation processing on the order information in the storage unit; the management terminal receives the analysis result of the analysis module and allocates the tasks in the order to the corresponding workshops according to the analysis result. It can be seen that the above technical solution has the following problems: only based on the analysis result for task allocation, the operation analysis method is single, and it is impossible to dynamically adjust the operation analysis method according to the actual application scenario, resulting in poor operation management efficiency. Summary of the Invention

[0004] Therefore, the present invention provides an operation management system based on reinforcement learning to overcome the problems in the prior art that the operation analysis method is single and it is impossible to dynamically adjust the operation analysis method according to the actual application scenario, resulting in poor operation management efficiency.

[0005] To achieve the above object, the present invention provides an operation management system based on reinforcement learning, including:

[0006] An operation analysis unit for determining the order status according to the order development coefficient and the order risk index, and determining the operation analysis methods as multi-level optimization analysis and recommended optimization analysis according to the order status;

[0007] A multi-level optimization unit, which is connected to the operation analysis unit and is used to determine the optimization method as determining the shelf adjustment method according to the defect mapping coefficient or determining the agent adjustment method according to the yield correlation coefficient in the multi-level optimization analysis;

[0008] A shelf adjustment unit, which is connected to the multi-level optimization unit and is used to determine the shelf adjustment method as updating the function description according to the topological correlation degree or the correlation mapping value according to the defect mapping coefficient;

[0009] An agent optimization unit, which is connected to the multi-level optimization unit and is used to determine the agent adjustment method as adjusting the agent level according to the agent evaluation index or determining the commodity distribution method according to the commodity influence coefficient according to the yield correlation coefficient;

[0010] A recommendation optimization unit, which is connected to the operation analysis unit and is used to determine the operation recommendation method as determining the associated recommendation method according to the commodity richness or performing commodity recommendation according to the label effectiveness of the commodity label in the recommendation optimization analysis according to the agent influence coefficient and the development difficulty coefficient.

[0011] Furthermore, when the order status is that the order development coefficient is less than the preset order development coefficient or the order risk index is greater than or equal to the preset order risk index, the operation analysis unit determines that the operation analysis method is multi-level optimization analysis.

[0012] Furthermore, when the order status is that the order development coefficient is greater than or equal to the preset order development coefficient and the order risk index is less than the preset order risk index, the operation analysis unit recommends optimization analysis for the operation analysis method.

[0013] Furthermore, the multi-level optimization unit determines the optimization method according to the complaint homogeneity, including:

[0014] If the complaint homogeneity is greater than or equal to the preset complaint homogeneity, it is determined that the optimization method is to determine the shelf adjustment method according to the defect mapping coefficient;

[0015] If the complaint homogeneity is less than the preset complaint homogeneity, it is determined that the optimization method is to determine the agent adjustment method according to the yield correlation coefficient.

[0016] Furthermore, the shelf adjustment unit determines the shelf adjustment method according to the defect mapping coefficient, including:

[0017] If the defect mapping coefficient is greater than or equal to the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the topological correlation degree;

[0018] If the defect mapping coefficient is less than the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the correlation mapping value.

[0019] Further, the agent optimization unit determines the agent adjustment method according to the yield correlation coefficient, including:

[0020] If the yield correlation coefficient is greater than or equal to the preset yield correlation coefficient, it is determined that the agent adjustment method is to adjust the agent level according to the agent evaluation index;

[0021] If the yield correlation coefficient is less than the preset yield correlation coefficient, it is determined that the agent adjustment method is to determine the commodity distribution method according to the commodity influence coefficient.

[0022] Further, the agent optimization unit determines the commodity distribution method according to the commodity influence coefficient, including:

[0023] If the commodity influence coefficient is greater than or equal to the preset commodity influence coefficient, it is determined that the commodity distribution method is to distribute according to the agent evaluation index;

[0024] If the commodity influence coefficient is less than the preset commodity influence coefficient, it is determined that the commodity distribution method is to distribute according to the correlation influence coefficient.

[0025] Further, when the agent influence coefficient is greater than or equal to the preset agent influence coefficient or the development difficulty coefficient is less than the preset development difficulty coefficient, the recommendation optimization unit determines that the operation recommendation method is to determine the associated recommendation method according to the commodity richness;

[0026] If the commodity richness is greater than or equal to the preset commodity richness, it is determined that the associated recommendation method is to make an associated recommendation according to the matching influence degree;

[0027] If the commodity richness is less than the preset commodity richness, it is determined that the associated recommendation method is to make an associated recommendation according to the commodity synchronization coefficient.

[0028] Further, when the agent influence coefficient is less than the preset agent influence coefficient and the development difficulty coefficient is greater than or equal to the preset development difficulty coefficient, the recommendation optimization unit determines that the operation recommendation method is to recommend commodities according to the label validity of the commodity label.

[0029] Further, the generation method of the commodity label includes:

[0030] If the commodity dislocation index is greater than or equal to the preset commodity dislocation index, the label generation method is to generate a commodity label according to the sales matching degree;

[0031] If the commodity dislocation index is less than the preset commodity dislocation index, the label generation method is to generate a commodity label according to the search matching degree.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows. In the technical solution of the present invention, the order status is determined based on the order development coefficient and the order risk index. The order problems can be accurately located through the order development coefficient and the order risk index. Then, different operation analysis methods are adaptively selected according to the order status, making the selection of operation analysis methods more in line with the actual application scenarios. The operation efficiency and decision-making accuracy of the operator industry can be improved by accurately positioning the order status, adapting to the rapid changes in the market and customer demands, enabling the promotion industry to deeply explore the market trends, optimize the resource allocation, and thus significantly enhance the overall operation efficiency.

[0033] Further, in the present invention, the potential risks of the products are effectively reflected by the complaint homogeneity degree, and then different shelf adjustment methods are adaptively selected according to the defect mapping coefficient, so that the selected shelf adjustment methods can improve the customer satisfaction, reduce the workload of customer service in handling complaints at the same time, and optimize the resource allocation.

[0034] Further, in the present invention, the correlation of valid orders in each agent account is effectively reflected by the yield correlation coefficient, and then different agent adjustment methods are adaptively selected according to the yield correlation coefficient, making the selected agent adjustment methods more in line with the actual operation scenarios. Adjusting the agent level according to the agent evaluation index can accurately match the agent capabilities, improve the operation efficiency. At the same time, determining the product allocation method according to the product impact coefficient can optimize the product allocation logic and reduce the operation risks.

[0035] Further, in the present invention, the sales situations of different products are effectively reflected by the product impact coefficient, and then different product allocation methods are selected according to the product impact coefficient, so that the selected product allocation methods can accurately recommend to improve the recommendation conversion rate, avoid blindly expanding the recommendation scope in the complex market environment, reduce the waste of resources caused by ineffective recommendations, and thus improve the operation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the unit connection diagram of the operation management system based on reinforcement learning of the present invention;

[0037] Figure 2 It is the flowchart of determining the operation analysis method according to the order status of the present invention;

[0038] Figure 3 It is the flowchart of determining the optimization method according to the complaint homogeneity degree of the present invention;

[0039] Figure 4 It is the flowchart of determining the shelf adjustment method according to the defect mapping coefficient of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0040] To make the objectives and advantages of the present invention more clear and understandable, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0042] It should be noted that in the description of the present invention, the terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0043] In addition, it should also be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0044] Please refer to Figures 1 to 4 as shown, the present invention provides an operation management system based on reinforcement learning, including:

[0045] An operation analysis unit for determining the order status according to the order development coefficient and the order risk index, and determining the operation analysis methods as multi-level optimization analysis and recommended optimization analysis according to the order status;

[0046] A multi-level optimization unit connected to the operation analysis unit for determining, in the multi-level optimization analysis, the optimization method as determining the shelf adjustment method according to the defect mapping coefficient or determining the agent adjustment method according to the yield correlation coefficient;

[0047] A shelf adjustment unit connected to the multi-level optimization unit for determining the shelf adjustment method as updating the function description according to the topological correlation degree or the correlation mapping value according to the defect mapping coefficient;

[0048] An agent optimization unit connected to the multi-level optimization unit for determining the agent adjustment method as adjusting the agent level according to the agent evaluation index or determining the commodity distribution method according to the commodity influence coefficient according to the yield correlation coefficient;

[0049] A recommended optimization unit, which is connected to the operation analysis unit, is used to determine the operation recommendation method as the associated recommendation method determined according to the product richness or the product recommendation according to the label effectiveness of the product label in the recommended optimization analysis based on the proxy influence coefficient and the development difficulty coefficient.

[0050] The application scenario of the present invention is the operation optimization of the agent operation industry. In the present invention, a number of historical records are correspondingly set. Any historical record records at least one of the order development coefficient, order risk index, association mapping value, yield correlation coefficient, agent evaluation index, and product influence coefficient in the historical process of the operation optimization of the agent operation industry, and each historical record corresponds to a qualified mark, which records whether the operation optimization of the agent operation industry meets the user's needs. The qualified mark can be recorded manually. It can be understood that the user can determine whether the operation optimization process of the agent operation industry meets the requirements according to the self-set index. The self-set index can be, but is not limited to, the complaint rate, which will not be elaborated here. Among them, the complaint rate = the number of customers who file complaints in a single monitoring period / the total number of customers who purchase products in a single monitoring period;

[0051] In the present invention, a continuously circulating monitoring period is set. The order status is determined once at the end of each monitoring period. The duration of the monitoring period can be set according to the user's needs. The greater the user's demand for monitoring accuracy, the smaller the duration of the monitoring period. A value of the monitoring period is provided. The monitoring period is 10d.

[0052] The present invention includes a number of products and a number of agents. Each product corresponds to product information, which includes a number of function names and function descriptions corresponding to each function name. The function names include, but are not limited to, the basic monthly rent and the general traffic. When the function name is the basic monthly rent, the corresponding function description can be, but is not limited to, 0 yuan and 1 yuan. When the function name is the general traffic, the corresponding function description can be, but is not limited to, 20GB and 100GB. A single agent can represent a number of products, and a single agent can only sell the products represented by the agent. Customers can browse the product information and purchase products. This is easy for those skilled in the art to understand and will not be elaborated here.

[0053] Specifically, when the order status is that the order development coefficient is less than the preset order development coefficient or the order risk index is greater than or equal to the preset order risk index, the operation analysis unit determines that the operation analysis method is a multi-level optimization analysis.

[0054] Among them, the order status includes a first order status and a second order status. The first order status is that the order development coefficient is less than the preset order development coefficient or the order risk index is greater than or equal to the preset order risk index. The second order status is that the order development coefficient is greater than or equal to the preset order development coefficient and the order risk index is less than the preset order risk index;

[0055] The order development coefficient = the customer behavior threshold × the behavior coefficient + the order growth rate × the growth coefficient,

[0056] The customer behavior threshold = the number of times of adding to cart but not paying × the customer browsing duration. The number of times of adding to cart but not paying is the sum of the quantities of the unpaid added-to-cart items corresponding to each customer within the target monitoring period. The customer browsing duration is the sum of the durations of browsing items corresponding to the customers who have not purchased the items within the target monitoring period. Denote the current monitoring period as the target monitoring period;

[0057] The order growth rate = the number of successfully delivered orders in the target monitoring period / (the sum of the numbers of successfully delivered orders corresponding to each monitoring period before the target monitoring period / the number of monitoring periods before the target monitoring period);

[0058] For the values of the behavior coefficient and the growth coefficient, the user can obtain them by learning from historical records through a reinforcement learning algorithm. The selected reinforcement learning algorithms include, but are not limited to, DQN and PPO. The user can select according to the actual application scenario. Both DQN and PPO are technical means that are easy to understand by those skilled in the art and will not be elaborated in detail here. It can be understood that the present invention reflects the order development difficulty through the customer behavior threshold and the order growth rate. The user can use the reinforcement learning algorithm to obtain the influence degrees of the behavior coefficient and the growth coefficient on order development respectively according to historical records, and then select the values of the behavior coefficient and the growth coefficient correspondingly according to the influence degrees on order development. The greater the values of the behavior coefficient and the growth coefficient, the greater the influence degree on order development. Provide a set of values for the behavior coefficient and the growth coefficient, where the behavior coefficient is 0.6 and the growth coefficient is 0.4;

[0059] The order risk index = the number of returned orders in the target monitoring period / the total number of successfully delivered orders and returned orders in the target monitoring period;

[0060] The values of the preset order development coefficient and the preset order risk index can be determined by the user according to the actual application scenario. The larger the value of the preset order development coefficient and the smaller the value of the preset order risk index, the greater the user's need for multi-level optimization analysis. Provide a method to determine the values of the preset order development coefficient and the preset order risk index. Detect the historical records of the user's multi-level optimization analysis, and record the average value of the order development coefficients corresponding to the historical records that can meet the user's needs as the preset order development coefficient, and record the average value of the order risk indices corresponding to the historical records that can meet the user's needs as the preset order risk index.

[0061] Specifically, when the order status is that the order development coefficient is greater than or equal to the preset order development coefficient and the order risk index is less than the preset order risk index, the operation analysis unit recommends optimization analysis for the operation analysis method.

[0062] Specifically, the multi-level optimization unit determines the optimization method according to the complaint homogeneity, including:

[0063] If the complaint homogeneity is greater than or equal to the preset complaint homogeneity, it is determined that the optimization method is to determine the shelf adjustment method according to the defect mapping coefficient;

[0064] If the complaint homogeneity is less than the preset complaint homogeneity, it is determined that the optimization method is to determine the agency adjustment method according to the yield correlation coefficient.

[0065] Among them, the complaint homogeneity = the average value of the complaint reference values corresponding to each agency / the standard deviation of the complaint reference values corresponding to each agency, and the complaint reference value corresponding to a single agency = the number of orders complained against by the agency during the target monitoring period / the number of orders successfully delivered by the agency during the target monitoring period;

[0066] The value of the preset complaint homogeneity can be determined by the user according to the actual application scenario. The smaller the value of the preset complaint homogeneity, the greater the user's need to determine the shelf adjustment method according to the defect mapping coefficient. Provide a method to determine the value of the preset complaint homogeneity. Detect the historical records of the user's determination of the shelf adjustment method according to the defect mapping coefficient, and record the average value of the complaint homogeneities corresponding to the historical records that can meet the user's needs as the preset complaint homogeneity.

[0067] Specifically, the shelf adjustment unit determines the shelf adjustment method according to the defect mapping coefficient, including:

[0068] If the defect mapping coefficient is greater than or equal to the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the topological correlation degree;

[0069] If the defect mapping coefficient is less than the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the associated mapping value.

[0070] Among them, the way to confirm the defect mapping coefficient is as follows: for a single product, the defect mapping coefficient corresponding to the product = the total number of function names that appear in each complaint information corresponding to the product during the target monitoring period / the total number of keywords that appear in each complaint information corresponding to the product during the target monitoring period; the complaint information is associated with the order through the order ID, and the order ID is a digital or string code. The complaint information contains several words, which is easy to understand for those skilled in the art and will not be elaborated here.

[0071] For the value of the preset defect mapping coefficient, the user can determine it according to the actual application scenario. The larger the value of the preset defect mapping coefficient, the greater the user's need to update the product description according to the associated mapping value. A value of the preset defect mapping coefficient is provided, and the preset defect mapping coefficient is 30%.

[0072] Updating the function description according to the topological correlation degree includes: updating the function description corresponding to the function name with a topological correlation degree greater than the preset topological correlation degree.

[0073] Updating the function description according to the associated mapping value includes: updating the function description corresponding to the function name with an associated mapping value greater than the preset associated mapping value.

[0074] When updating the function description corresponding to the function name, the user can change the data or make relevant supplements to the function description according to the actual needs, which will not be elaborated here.

[0075] The way to confirm the topological correlation degree is as follows: for a single function name corresponding to a single product, the topological correlation degree is the number of times the function name appears in each complaint information corresponding to the product during the target monitoring period.

[0076] The way to confirm the associated mapping value is as follows: for a single function name corresponding to a single product, mark the product as the target product, mark the function name as the target function name, mark the complaint information corresponding to the target function name during the target monitoring period as the target complaint information, detect the historical record of supplementing the function description for the target function name of the target product, and record the number of the same keywords in the complaint information corresponding to the target product in each historical record that can meet the user's needs as the associated mapping value of the target function name corresponding to the target product.

[0077] The values of the preset topological correlation degree and the preset correlation mapping value can be determined by the user according to the actual application scenario. The greater the user's demand for improving operation efficiency, the smaller the values of the preset topological correlation degree and the preset correlation mapping value. Detect the historical records of the user's function description update according to the topological correlation degree, and record the average value of the topological correlation degree corresponding to the function name for which the function description is updated in the historical records that can meet the user's needs as the preset topological correlation degree. Detect the historical records of the user's function description update according to the correlation mapping value, and record the average value of the correlation mapping value corresponding to the function name for which the function description is updated in the historical records that can meet the user's needs as the preset correlation mapping value.

[0078] Specifically, the agent optimization unit determines the agent adjustment method according to the yield correlation coefficient, including:

[0079] If the yield correlation coefficient is greater than or equal to the preset yield correlation coefficient, it is determined that the agent adjustment method is to adjust the agent level according to the agent evaluation index;

[0080] If the yield correlation coefficient is less than the preset yield correlation coefficient, it is determined that the agent adjustment method is to determine the commodity distribution method according to the commodity influence coefficient.

[0081] Among them, the yield correlation coefficient = 1 / the standard deviation of the commodity permission coefficients corresponding to each agent, and the commodity permission coefficient corresponding to a single agent = the number of commodities that the agent can sell / the total amount of commodities;

[0082] The value of the preset yield correlation coefficient can be determined by the user according to the actual application scenario. The smaller the value of the preset yield correlation coefficient, the greater the user's demand for adjusting the agent level according to the agent evaluation index. Provide a value of the preset yield correlation coefficient, detect the historical records of the user's adjustment of the agent level according to the agent evaluation index, and record the average value of the yield correlation coefficients corresponding to the historical records that can meet the user's needs as the preset yield correlation coefficient;

[0083] Adjust the agent level according to the agent evaluation index. Among them, if the agent evaluation index is greater than or equal to the preset agent evaluation index, increase the adjustment for the agent level;

[0084] If the agent evaluation index is less than the preset agent evaluation index, decrease the adjustment for the agent level;

[0085] It should be noted that when increasing or decreasing the adjustment for the agent level, the adjustment amount is 1 level each time. When increasing the adjustment for the agent level, it is adjusted from a level with a larger number to a level with a smaller number. When decreasing the adjustment for the agent level, it is adjusted from a level with a smaller number to a level with a larger number. When increasing the adjustment for the agent level, if the agent level is 1, the agent level remains unchanged. When decreasing the adjustment for the agent level, if the agent level is 5, the agent level remains unchanged;

[0086] The agent level corresponding to a single agent is the level stated by the agent. The agent levels include level 1, level 2, ……, level 5;

[0087] The agent evaluation index corresponding to a single agent = (the total number of orders successfully delivered by the agent within the target monitoring period - the complaint reference value corresponding to the agent) / (6 - the agent level);

[0088] The value of the preset agent evaluation index can be determined by the user according to the actual application scenario. The larger the value of the preset agent evaluation index, the greater the user's need to decrease the adjustment for the agent level. Provide a value of the preset agent evaluation index, detect the historical records of the user's decreasing adjustment for the agent level, and record the average value of the agent evaluation indices corresponding to the historical records that can meet the user's needs as the preset agent evaluation index.

[0089] Specifically, the agent optimization unit determines the commodity allocation method according to the commodity influence coefficient, including:

[0090] If the commodity influence coefficient is greater than or equal to the preset commodity influence coefficient, it is determined that the commodity allocation method is to allocate according to the agent evaluation index;

[0091] If the commodity influence coefficient is less than the preset commodity influence coefficient, it is determined that the commodity allocation method is to allocate according to the correlation influence coefficient.

[0092] Among them, the commodity influence coefficient = the standard deviation of the order reference values corresponding to each commodity + the standard deviation of the agent reference values corresponding to each commodity. For a single commodity, this commodity is recorded as the target commodity. The order reference value corresponding to the target commodity is the number of orders successfully delivered by the target commodity within the target monitoring period. The agent reference value corresponding to the target commodity is the standard deviation of the sub-reference values corresponding to each agent. The sub-reference value corresponding to a single agent is the number of orders of the target commodity successfully delivered by the agent;

[0093] The value of the preset commodity influence coefficient can be determined by the user according to the actual application scenario. The smaller the value of the preset commodity influence coefficient, the greater the user's demand for associative allocation based on similar reference values. Provide a value of the preset commodity influence coefficient, detect the historical records of the user's associative allocation based on similar reference values, and record the average value of the commodity influence coefficients corresponding to the historical records that can meet the user's needs as the preset commodity influence coefficient;

[0094] Allocate according to the proxy evaluation index. Among them, for each proxy, when allocating commodities to a single proxy, record the proxy as the first target proxy. There is a positive correlation between the proxy evaluation index corresponding to the first target proxy and the quantity of allocated commodities corresponding to the first target proxy. Allocate commodities to the first target proxy in descending order of the associative influence coefficient between the commodity and the first target proxy until the quantity of allocated commodities corresponding to the first target proxy is reached;

[0095] The quantity of allocated commodities corresponding to a single proxy is the quantity of commodities that the proxy can act as an agent for in the next monitoring period adjacent to the target monitoring period;

[0096] Allocate according to the associative influence coefficient. Among them, for each proxy, when allocating commodities to a single proxy, record the proxy as the second target proxy. In the next monitoring period adjacent to the target monitoring period, allocate the commodities with an associative influence coefficient greater than the preset associative influence coefficient with the second target proxy to the second target proxy;

[0097] The confirmation method of the associative influence coefficient is as follows: for a single proxy, record the proxy as the target proxy, record the commodities that the target proxy can sell in the target monitoring period as the sold commodities, and record the commodities that the target proxy cannot sell in the target monitoring period as the unsold commodities. For a single sold commodity, the associative influence coefficient between the sold commodity and the target proxy is positively correlated with the number of orders of the sold commodity successfully delivered by the target proxy in the target monitoring period; for a single unsold commodity, the associative influence coefficient between the unsold commodity and the target proxy is positively correlated with the corresponding feedback sales value;

[0098] The confirmation method of the feedback sales value is as follows: for a single unsold commodity corresponding to the target proxy, record the unsold commodity as the target unsold commodity, and record the proxy with a proxy similarity greater than the preset proxy similarity with the target proxy as the reference proxy. Feedback sales value = the number of orders of the target unsold commodity successfully delivered by each reference proxy that sells the target unsold commodity in the target monitoring period / (the number of reference proxies that sell the target unsold commodity + 1);

[0099] The proxy similarity corresponding to any two proxies = 1 - (the absolute value of the difference between the proxy evaluation indices corresponding to the two proxies / the larger value of the proxy evaluation indices corresponding to the two proxies);

[0100] The values of the preset correlation influence coefficient and the preset proxy similarity can be determined by the user according to the actual application scenario. The greater the user's demand for improving operational efficiency, the greater the values of the preset correlation influence coefficient and the preset proxy similarity. Provide a set of values for the preset correlation influence coefficient and the preset proxy similarity. Detect the historical records of distribution according to the correlation influence coefficient, and record the average value of the correlation influence coefficients between the products assigned to each proxy in the historical records that can meet the user's needs and the corresponding proxies as the preset correlation influence coefficient. The preset proxy similarity is 70%.

[0101] Specifically, when the proxy influence coefficient is greater than or equal to the preset proxy influence coefficient or the development difficulty coefficient is less than the preset development difficulty coefficient, the recommendation optimization unit determines that the operation recommendation method is the correlation recommendation method determined according to the product richness;

[0102] If the product richness is greater than or equal to the preset product richness, it is determined that the correlation recommendation method is the correlation recommendation according to the matching influence degree;

[0103] If the product richness is less than the preset product richness, it is determined that the correlation recommendation method is the correlation recommendation according to the product synchronization coefficient.

[0104] Among them, the confirmation methods of the proxy influence coefficient and the development difficulty coefficient are as follows: for a single proxy, record this proxy as the target proxy. The proxy influence coefficient corresponding to the target proxy = the total commission obtained by the target proxy within the target monitoring period / (6 - proxy level). The development difficulty coefficient corresponding to the target proxy = 1 / the average value of the incremental coefficients corresponding to each monitoring period before the target monitoring period; The incremental coefficient corresponding to a single monitoring period = the number of orders successfully delivered in this monitoring period - the number of orders successfully delivered in the monitoring period adjacent to this monitoring period and before this monitoring period;

[0105] The values of the preset proxy influence coefficient and the preset development difficulty coefficient can be determined by the user according to the actual application scenario. The smaller the value of the preset proxy influence coefficient and the larger the value of the preset development difficulty coefficient, the greater the user's demand for determining the correlation recommendation method according to the product richness. Provide a set of values for the preset proxy influence coefficient and the preset development difficulty coefficient. Detect the historical records of the user determining the correlation recommendation method according to the product richness, and record the average value of the proxy influence coefficients corresponding to the historical records that can meet the user's needs as the preset proxy influence coefficient, and record the average value of the development difficulty coefficients corresponding to the historical records that can meet the user's needs as the preset development difficulty coefficient;

[0106] The product richness corresponding to a single agent = the number of products that the agent can represent during the target monitoring period / the total number of different products that each agent can represent during the target monitoring period;

[0107] The value of the preset product richness can be determined by the user according to the actual application scenario. The larger the value of the preset product richness, the greater the user's demand for associated recommendation based on the product synchronization coefficient. A value of the preset product richness is provided, and the preset product richness is 60%.

[0108] Performing associated recommendation according to the matching influence degree includes: for a single customer corresponding to a single agent, marking the agent as the target agent, marking the customer as the target customer, and recommending the target agent products whose matching influence degree with the reference browsing products corresponding to the target customer is greater than the preset matching influence degree;

[0109] Performing associated recommendation according to the product synchronization coefficient includes: recommending the target agent products whose product synchronization coefficient corresponding to the target customer is greater than the preset product synchronization coefficient;

[0110] The reference browsing products corresponding to the target customer are the products with the longest browsing time of the target customer, and the target agent products are the products that the target customer can represent during the target monitoring period;

[0111] The matching influence degree = the number of keywords that exist simultaneously in the product information corresponding to the two products / the total number of different keywords that appear in the product information corresponding to the two products;

[0112] The confirmation method of the product synchronization coefficient is to mark the reference browsing products corresponding to the target customer as the target browsing products, mark a single target agent product as the first target agent product. The product synchronization coefficient between the first agent product and the target browsing products = the number of customers whose reference browsing products corresponding to the customer are the target browsing products and who purchase the first target agent product / the number of customers whose reference browsing products corresponding to the customer are the target browsing products and who have purchased products;

[0113] The values of the preset matching influence degree and the preset product synchronization coefficient can be determined by the user according to the actual application scenario. The greater the user's demand for improving the recommendation accuracy, the larger the values of the preset matching influence degree and the preset product synchronization coefficient. A value of the preset matching influence degree and the preset product synchronization coefficient is provided, the preset matching influence degree is 70%, and the preset product synchronization coefficient is 60%.

[0114] Specifically, when the agent influence coefficient is less than the preset agent influence coefficient and the development difficulty coefficient is greater than or equal to the preset development difficulty coefficient, the recommendation optimization unit determines that the operation recommendation method is to perform product recommendation according to the label validity of the product label.

[0115] Among them, product recommendations are made according to the label validity of product labels. When making product recommendations for a single customer corresponding to a single agent, the agent is denoted as the target agent, and the customer is denoted as the target customer. The target agent product with the largest label validity is selected for recommendation to the target customer;

[0116] The label validity corresponding to a single product is the average value of the label reference values corresponding to the product labels of the product. The label reference value corresponding to a single product label = the number of products with this label among the products browsed by the target customer / the total number of products browsed by the target customer.

[0117] Specifically, the generation method of the product label includes:

[0118] If the product dislocation index is greater than or equal to the preset product dislocation index, the label generation method is to generate product labels according to the sales matching degree;

[0119] If the product dislocation index is less than the preset product dislocation index, the label generation method is to generate product labels according to the search matching degree.

[0120] Among them, the product dislocation index is the standard deviation of the order reference values corresponding to each product. The value of the preset product dislocation index can be determined by the user according to the actual application scenario. The larger the value of the preset product dislocation index, the greater the user's demand for generating product labels according to the search matching degree. Provide a value of the preset product dislocation index, detect the historical records of the user generating product labels according to the search matching degree, and record the average value of the product dislocation indices corresponding to the historical records that can meet the user's needs as the preset product dislocation index;

[0121] Generating product labels according to the sales matching degree includes: for a single product, denoting the product as the target product, denoting the functional descriptions corresponding to the products browsed by the customers who have purchased the target product before purchasing the target product as reference information, denoting the keywords appearing in the reference information as the keywords to be selected, and using the keywords to be selected with a sales matching degree greater than the preset sales matching degree as the product labels of the target product;

[0122] The sales matching degree corresponding to a single keyword to be selected = the number of products corresponding to the functional descriptions that simultaneously appear this keyword to be selected and excellent keywords in the reference information / the total number of products browsed by the customers who have purchased the target product before purchasing the target product; excellent products are products with an order reference value greater than the preset order reference value, and the keywords in the functional descriptions corresponding to the excellent products are denoted as excellent keywords.

[0123] Generate product labels based on search matching degree, including: for a single product, record the product as the target product, record the functional descriptions of each product browsed by the customers who have purchased the target product before purchasing the target product as reference information, record the keywords appearing in the reference information as candidate keywords to be selected, and use the candidate keywords to be selected with a search matching degree greater than the preset search matching degree as the product label of the target product;

[0124] Search matching degree = the number of times a single candidate keyword to be selected appears in the reference information / the total number of keywords appearing in the reference information,

[0125] The values of the preset order reference value, the preset sales matching degree, and the preset search matching degree can be determined by the user according to the actual application scenario. The smaller the values of the preset order reference value, the preset sales matching degree, and the preset search matching degree, the greater the user's need to use the candidate keyword to be selected as the product label. Provide a set of values for the preset order reference value, the preset sales matching degree, and the preset search matching degree. Record the average value of the order reference values corresponding to each excellent product in the historical records that can meet the user's needs as the preset order reference value, the preset sales matching degree is 70%, and the preset search matching degree is 60%.

[0126] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

[0127] The above are only the preferred embodiments of the present invention and are not used to limit the present invention; for those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An operation management system based on reinforcement learning, characterized in that Including: An operation analysis unit, which is used to determine the order status according to the order development coefficient and the order risk index, and determine the operation analysis methods as multi-level optimization analysis and recommended optimization analysis according to the order status; A multi-level optimization unit, which is connected to the operation analysis unit and is used to determine the optimization method as determining the shelf adjustment method according to the defect mapping coefficient or determining the agent adjustment method according to the yield correlation coefficient in the multi-level optimization analysis; A shelf adjustment unit, which is connected to the multi-level optimization unit and is used to determine the shelf adjustment method as updating the function description according to the topological correlation degree or the correlation mapping value according to the defect mapping coefficient; An agent optimization unit, which is connected to the multi-level optimization unit and is used to determine the agent adjustment method as adjusting the agent level according to the agent evaluation index or determining the commodity distribution method according to the commodity influence coefficient according to the yield correlation coefficient; A recommended optimization unit, which is connected to the operation analysis unit and is used to determine the operation recommendation method as determining the associated recommendation method according to the commodity richness or performing commodity recommendation according to the label validity of the commodity label in the recommended optimization analysis according to the agent influence coefficient and the development difficulty coefficient.

2. The operation management system based on reinforcement learning according to claim 1, wherein When the order status is that the order development coefficient is less than the preset order development coefficient or the order risk index is greater than or equal to the preset order risk index, the operation analysis unit determines that the operation analysis method is multi-level optimization analysis.

3. The operation management system based on reinforcement learning according to claim 2, wherein When the order status is that the order development coefficient is greater than or equal to the preset order development coefficient and the order risk index is less than the preset order risk index, the operation analysis unit recommends the optimization analysis method.

4. The operation management system based on reinforcement learning according to claim 2, wherein The multi-level optimization unit determines the optimization method according to the complaint homogeneity, including: If the complaint homogeneity is greater than or equal to the preset complaint homogeneity, it is determined that the optimization method is to determine the shelf adjustment method according to the defect mapping coefficient; If the complaint homogeneity is less than the preset complaint homogeneity, it is determined that the optimization method is to determine the agent adjustment method according to the yield correlation coefficient.

5. The operation management system based on reinforcement learning according to claim 4, characterized in that, The shelf adjustment unit determines the shelf adjustment method according to the defect mapping coefficient, including: If the defect mapping coefficient is greater than or equal to the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the topological correlation degree; If the defect mapping coefficient is less than the preset defect mapping coefficient, it is determined that the shelf adjustment method is to update the function description according to the correlation mapping value.

6. The operation management system based on reinforcement learning according to claim 4, wherein The agent optimization unit determines the agent adjustment method according to the yield correlation coefficient, including: If the yield correlation coefficient is greater than or equal to the preset yield correlation coefficient, it is determined that the agent adjustment method is to adjust the agent level according to the agent evaluation index; If the yield correlation coefficient is less than the preset yield correlation coefficient, it is determined that the agent adjustment method is to determine the commodity distribution method according to the commodity influence coefficient.

7. The operation management system based on reinforcement learning according to claim 6, wherein, The agent optimization unit determines the commodity distribution method according to the commodity influence coefficient, including: If the commodity influence coefficient is greater than or equal to the preset commodity influence coefficient, it is determined that the commodity distribution method is to distribute according to the agent evaluation index; If the commodity influence coefficient is less than the preset commodity influence coefficient, it is determined that the commodity distribution method is to distribute according to the associated influence coefficient.

8. The operation management system based on reinforcement learning according to claim 3, characterized in that, When the agent influence coefficient is greater than or equal to the preset agent influence coefficient or the development difficulty coefficient is less than the preset development difficulty coefficient, the recommended optimization unit determines that the operation recommendation method is the associated recommendation method determined according to the commodity richness; If the commodity richness is greater than or equal to the preset commodity richness, it is determined that the associated recommendation method is to make an associated recommendation according to the matching influence degree; If the commodity richness is less than the preset commodity richness, it is determined that the associated recommendation method is to make an associated recommendation according to the commodity synchronization coefficient.

9. The operation management system based on reinforcement learning according to claim 8, characterized in that, When the agent influence coefficient is less than the preset agent influence coefficient and the development difficulty coefficient is greater than or equal to the preset development difficulty coefficient, the recommended optimization unit determines that the operation recommendation method is to recommend commodities according to the label validity of the commodity labels.

10. The operation management system based on reinforcement learning according to claim 9, characterized in that, The generation method of the commodity labels includes: If the commodity dislocation index is greater than or equal to the preset commodity dislocation index, the label generation method is to generate commodity labels according to the sales matching degree; If the commodity dislocation index is less than the preset commodity dislocation index, the label generation method is to generate commodity labels according to the search matching degree.

Images