Internet right and interest product agent management system

Abstract

The invention provides an internet right product agent management system, and relates to the technical field of product agent management. The system comprises a product management module, an order management module, an agent management module, a database and a data analysis module. The system processes on-off and stock of products through the product management module, the order management module calculates commission, and the agent management module achieves credit evaluation. The data analysis module predicts requirements through a multi-layer architecture; a data input layer integrates internal and external data; the analysis engine layer adopts feature coding and a time attention mechanism to model space-time correlation, and outputs a demand prediction formula; and the decision output layer generates an agent personalized report through inventory calibration, credit grading and regional weighting strategies, recommends products and early warns risks. The system realizes intelligent acquisition and sales decision making, and improves the operation capability and market influence of the system.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of product agency management, and particularly relates to an internet rights product agency management system. BACKGROUND

[0002] The internet rights product agency management refers to a mode in which the owner of rights products, such as a copyright party and a patent party, entrusts an agent to manage and operate the rights products by means of authorization.

[0003] With the popularization of the Internet and the development of the digital economy, the types and quantities of internet rights products are increasing, and the market scale of the agency management is also expanding. In the fields of digital music, online video and e-books, more and more copyright parties choose to promote and manage their products through agents to expand market coverage and improve operation efficiency. The subjects participating in the internet rights product agency management are increasingly diversified, and different subjects compete in the market by virtue of their respective resources and advantages.

[0004] With the development of the Internet industry and the continuous expansion of the rights product market, the demand for agency management systems is also increasing. According to the market scale data of related agency management software, the market scale of China's agency management software has reached a certain scale, and the global market scale has also grown synchronously. It is expected that the global market scale of agency management software will grow at a certain compound annual growth rate until 2030. The internet rights product agency management system as a subfield also benefits from the overall market growth trend.

[0005] At present, most of the agency management systems on the market still rely on traditional database management and simple statistical analysis. With the popularization of artificial intelligence in various industries, the existing technology has the problems that in the face of massive user data and transaction records, only simple data screening and statistics can be performed, and the sales trend of products cannot be accurately predicted in combination with data; there is a lack of ability to guide business decisions based on user behavior data analysis results, and the needs of agents cannot be met. Therefore, a more excellent internet rights product agency management system is urgently needed to solve the above problems. SUMMARY

[0006] The purpose of the present application is to provide an internet rights product agency management system to solve the problem that the data analysis results of large amounts of data in the prior art are difficult to directly guide business decisions.

[0007] To achieve the above-mentioned purpose, the present application provides an internet rights product agency management system, which comprises:

[0008] a product management module, an order management module, an agent management module, a database module and a data analysis module.

[0009] The product management module is used to publish the classification, quantity, commission rate of Internet equity products, and manage the listing, purchase, sales, and delisting of Internet equity products, as well as inventory registration;

[0010] The order management module is used to generate purchase and sales orders, calculate agency commissions and commission payment records, and query commission details;

[0011] The agent management module is used for agent account registration, real-name authentication, sales license registration, assigning the agent account with the call authority authentication of the product management module and the order management module, recording the agent's region, historical purchase volume and credit score. The credit score calculation rule is s j =0.7×fulfillment rate+0.3×(1-refund rate), fulfillment rate=number of completed orders / total number of orders;

[0012] The database module stores data information of the product management module, the order management module and the agent management module;

[0013] The data analysis module is provided with a data input layer, an analysis engine layer and a decision output layer. The data input layer accesses the database module to obtain internal data, external market data, and accesses the industry report API to obtain competitor prices and hot search terms. The analysis engine layer processes and analyzes the data, predicts future product demand content, and recommends popular Internet equity product procurement plans to agents. The decision output layer forms a personalized report for the agent based on the analysis of the analysis engine layer and feeds it back to the agent management module.

[0014] Furthermore, the data input layer acquires data at time t including: product set P t , P t ={p1,p2,...,p n}, each product p i Include attributes: Category c i , inventory q i 、Commission ratio θ i ;Agent Set A t ,A t ={a1,a2,...,a m}, each agent a j Included attributes: Region j , credit score j , historical purchase vector h j =(h j1 ,h j2 ,...,h jn ); order set O t , O t ={O k ||O k=(a k ,p k ,t k ,d k )}, each product O k Included attributes:Agency a k 、Products k , order generation time t k 、sales volume k ; External market data D t , Each external market data D t Included attributes: hot search word frequency w t , Competitive product prices w t =Keyword search volume / all category search volume; the data input layer pre-processes the above data and then the analysis engine layer analyzes the data.

[0015] Furthermore, the analysis engine layer predicts the demand for each product in the future period (t+1) by integrating time series characteristics, regional attributes and market sentiment. To guide and purchase agents; the analysis engine layer consists of a feature encoding layer, a time attention mechanism layer, and a prediction output layer. The demand forecasting formula is designed as follows:

[0016]

[0017] Among them, ELU(W p C i +b p ) is the basic prediction item, ∈ is the random fluctuation item, simulating the unpredictable fluctuation in the market, ∈~N(0,σ 2 )(σ=0.1),β·Ι 促销 is the artificial intervention item, β is the promotion gain coefficient, which is 0.5 according to experience, and 促销 It is an indicator variable, which takes 1 to indicate promotion and 0 to indicate no promotion;

[0018] The feature encoding layer converts product sets, agent sets, order sets, and external market data into high-dimensional learnable product features, regional features, and market sentiment feature vector representations to solve the problem of feature heterogeneity. The mathematical expression of its product feature sequence is:

[0019] X i =ReLU(W e [E i ||G l ||M t ]+b e ), [E i ||G l ||M t ]∈R5

[0020] E i For product characteristics, E i = [EMA(d i,t-7:t ), θ i , Norm(q i )], EMA(d i,t-7:t ) represents the exponential moving average of sales in the past 7 days, θ i is the commission ratio, and Norm(q i ) is the inventory pressure, which is calculated according to industry standards,

[0021] G l is a regional feature, represents the total historical purchase amount of all agents in region l for product p i , represents the competitive inventory pressure, represents the set of all products with category c i and geographical location matching l;

[0022] M t is the market heat of the category, M t = TF-IDF(w ci ), calculated by multiplying the term frequency by the inverse document frequency;

[0023] W e is the weight matrix, W e ∈ R 64×6 ;

[0024] b e is the bias term, b e ∈ R 64 ;

[0025] The time attention mechanism layer accepts the output X i of the feature encoding layer, stacks the feature vectors of the past 24 hours into a time series matrix, and calculates the periodicity, trend and burstiness of product demand in the time dimension through multi-head attention. Finally, a 64-dimensional context vector C i is output.

[0026] The input 64-dimensional product feature sequence is split into 8 heads and linearly mapped to the query, key and value spaces, with the calculation formula being: The projection matrices W Q , W K , and W V ∈ R 64×64 , and each head calculates attention independently:

[0027]

[0028] Q represents the time-of-day feature

[0029] K, V represent the past 24 hours feature

[0030]

[0031] The multi-head attention computation is: MultiHead(Q, K, V) = Concat(head1,..., head8) W O

[0032] W O ∈R 64×64 : output projection matrix;

[0033] The output context vector C i = MultiHeadAttention(Q, K, V);

[0034] The prediction output layer takes the context vector C i output by the temporal attention mechanism layer as input, and calculates the predicted demand by the demand prediction formula:

[0035]

[0036] The demand of each product in the future period is calculated, and the business rules are integrated for calibration, and then the predicted future product demand content is output;

[0037] The business rules of the prediction output layer include:

[0038] Inventory constraint calibration: when the predicted demand exceeds the current inventory, adjust according to the inventory calibration rule:

[0039] Inventory calibration rule:

[0040] Where, is the calibrated demand, is the original predicted demand output by the analysis engine layer, q i is the current inventory, and p is the inventory buffer coefficient, which is configured by the system and is empirically valued at 1.2;

[0041] When the predicted value exceeds 120% of the current inventory q i , the demand is forced to be limited to q i x 1.2 to avoid the risk of inventory overselling;

[0042] Credit ranking strategy: for credit score s j≥0.8's agent, open all product procurement authority; for 0.6≤s j <0.8's agent, limit the procurement of high order volume products; for s j <0.6's agent, trigger risk control for manual review;

[0043] Regional heat weighting: the hot search word frequency w t Implement regional weighting:

[0044]

[0045] Where, w t is the hot search word frequency, N l is the number of agents in the region l, N total is the total number of agents nationwide, λ l is the regional economic coefficient, default 1.0;

[0046] Through regional heat weighting, the nationwide hot search word frequency is dynamically adjusted according to the region where the agent is located, more accurately reflecting the demand differences in different regions;

[0047] The decision output layer outputs the agent's personalized recommendation report, including: product recommendation list, sorted by product comprehensive score sorting algorithm, descending recommended list;

[0048] Comprehensive score sorting algorithm:

[0049] Where, weight configuration: α=0.5, β=0.3, γ=0.15, δ=0.05

[0050] Inventory shortage warning: if marked as "shortage";

[0051] Dynamic commission adjustment strategy: according to product heat, implement ladder commission:

[0052]

[0053] If the predicted demand is 2.5 times the average sales in the near future, the commission will be raised by 20%, and if the predicted demand is only 30% of the average sales in the near future, the commission will be reduced by 10%;

[0054] Inventory prediction warning: when the following conditions are met, trigger the restocking warning:

[0055] And

[0056] Inventory days are insufficient: the current inventory can only cover the predicted demand for T threshold days in the future;

[0057] Market heat verification: category search heat Above historical average

[0058] The report structure is divided into three modules: core recommendation, which is the top five product recommendation list of products, including revenue, inventory, and commission; risk warning, which is a list of low inventory products; and action suggestion, which is the commission adjustment range and urgent replenishment.

[0059] The present application has the following beneficial effects due to the above technical solutions:

[0060] Compared with the prior art, the analysis engine layer of the present application converts product attributes, regional features and market heat into high-dimensional vectors through feature coding, captures periodic demand by combining a time attention mechanism, and introduces a random fluctuation and promotion intervention term in the prediction formula, which can realize accurate demand prediction of internet equity products; intelligent recommendation and dynamic commission, based on a product comprehensive scoring algorithm (combining predicted demand, commission rate, inventory and hot search word frequency), generate a priority ordered purchase list for agents, and a dynamic commission strategy adjusts the commission proportion according to the predicted demand to encourage agents to preferentially sell high-potential products; the agent management module divides the permissions through credit points: high-score agents are open to full-category procurement, low-score agents trigger manual review, and the credit points are updated in real time combined with order data, which reduces the bad debt risk while ensuring compliance sales; the prediction output layer avoids overselling through inventory calibration rules, and triggers a replenishment warning when the inventory coverage days are insufficient or the market heat is high, which can effectively improve the inventory turnover rate and reduce the proportion of unsold products; the data input layer automatically accesses internal and external data sources, the analysis engine layer realizes end-to-end prediction through multi-head attention, and the decision layer directly generates an agent performance report containing core recommendations and risk warnings, which optimizes the running ability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0061] The accompanying drawings, which are included to provide a further understanding of the application and constitute a part of this application, do not limit the application, and in the drawings:

[0062] Figure 1 The structure diagram of the internet equity product agent management system of the present application;

[0063] Figure 2 The structure diagram of the data analysis module of the present application;

[0064] Figure 3 The flowchart of data processing and analysis, product demand prediction and report generation of the present application. DETAILED DESCRIPTION

[0065] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments, which are used to explain the present application but not as a limitation thereof.

[0066] Embodiment one:

[0067] Please refer to Figure 1 The embodiment provides an internet equity product agent management system, comprising:

[0068] Product management module, order management module, agent management module, database module, data analysis module;

[0069] The product management module is used to publish the classification, quantity, commission rate of internet equity products, manage the on-shelf, purchase and sale, off-shelf and inventory registration of internet equity products;

[0070] The order management module is used to generate purchase and sale orders, calculate agent commission and commission payment records, and query commission details;

[0071] The agent management module is used for agent account registration, real-name authentication, sales permission registration, and assigns product management module and order management module calling permission authentication to agent accounts, records agent region, historical purchase quantity and credit score, and the credit score calculation rule is s j = 0.7 * performance rate + 0.3 * (1-refund rate), performance rate = completed order number / total order number;

[0072] The database module stores the data information of the product management module, the order management module and the agent management module;

[0073] Please refer to Figure 2 The data analysis module is provided with a data input layer, an analysis engine layer and a decision output layer, the data input layer accesses the database module to obtain internal data, also obtains external market data, accesses industry report API to obtain competitor prices and hot search word frequency; the analysis engine layer processes and analyzes data, predicts future product demand content, and recommends high-heat internet equity product purchase scheme for agents; the decision output layer forms an agent personalized report according to the analysis of the analysis engine layer, and feeds back to the agent management module.

[0074] Please refer to Figure 3 S1 The data input layer obtains product set, agent set, order set and external market data, and pre-processes the above data;

[0075] Specifically, the data input layer obtains data at time t, including: product set P t P t ={p1, p2,..., pn} each product p i including attribute: category c i , inventory q i , commission ratio θ i ; agent set A t , A t ={a1, a2,..., a m} each agent a j including attribute: region l j , credit score s j , historical purchase vector h j =(h j1 , h j2 ,..., h jn ); order set O t , O t ={O k ||O k =(a k , p k , t k , d k )} each product O k including attribute: agent a k , product p k , order generation time t k , sales d k ; external market data D t , each external market data D t including attribute: hot search frequency w t , competitor price w t =key word search volume / all category search volume; the data input layer pre-processes the above data, and the analysis engine layer analyzes the data after pre-processing.

[0076] The analysis engine layer predicts the demand of each product in the future period (t+1) by fusing time sequence features, regional attributes and market public opinion to guide and agent purchase; the analysis engine layer is composed of a feature encoding layer, a time attention mechanism layer and a prediction output layer, and the demand prediction calculation formula is designed as:

[0077]

[0078] wherein, ELU(W p C i +b p ) is a basic prediction term, ∈ is a random fluctuation term, simulating unpredictable fluctuations in the market, ∈~N(0, σ 2 )(σ=0.1), β·Ι 促销is the artificial intervention item, β is the promotion gain coefficient, which is 0.5 according to experience, and 促销 It is an indicator variable, which takes 1 to indicate promotion and 0 to indicate no promotion.

[0079] The S21 feature encoding layer converts product sets, agent sets, order sets, and external market data into high-dimensional learnable product features, regional features, and market sentiment feature vector representations;

[0080] Specifically, the feature encoding layer converts product sets, agent sets, order sets, and external market data into high-dimensional learnable product features, regional features, and market sentiment feature vector representations to solve the problem of feature heterogeneity. The mathematical expression of its product feature sequence is:

[0081] X i =ReLU(W e [E i ||G l ||M t ]+b e ), [E i ||G l ||M t ]∈R 5

[0082] E i is the product’s own characteristics, E i =[EMA(d i,t-7:t ),θ i ,Norm(q i )], EMA(d i,t-7:t ) represents the exponential moving average of sales over the past 7 days, θ i is the commission ratio, Norm(q i ) is the inventory pressure, its calculation is based on industry standards,

[0083] G l For regional characteristics, represents the sales of product p by all agents in region l i Total historical purchases, Indicates the inventory pressure of competing products, Indicates category c i The set of all products whose geographical location matches l;

[0084] M t For the category market popularity, M t =TF-IDF(w ci ), calculated by multiplying the term frequency by the inverse document frequency;

[0085] W e is the weight matrix, We ∈R 64×6 ;

[0086] b e is a bias term, b e ∈R 64 ;

[0087] The S22 time attention mechanism layer accepts the output of the feature encoding layer, and identifies the periodicity, trend and burst of product demand in the time dimension through multi-head attention calculation;

[0088] Specifically, the time attention mechanism layer accepts the output X i of the feature encoding layer, stacks the feature vectors of the continuous 24 hours into a time sequence matrix, identifies the periodicity, trend and burst of product demand in the time dimension through multi-head attention calculation, and finally outputs a 64-dimensional context vector C i ;

[0089] The input 64-dimensional product feature sequence is split into 8 heads, linearly mapped to the query (Query), key (Key) and value (Value) space, and the calculation formula is: Projection matrix W Q , W K , W V ∈R 64×64 , and each head independently calculates attention:

[0090]

[0091] Q represents the feature at the time

[0092] K, V represent the features in the past 24 hours

[0093]

[0094] Concatenation multi-head attention calculation is: MultiHead(Q, K, V) = Concat(head1, …, head8)W O

[0095] W O ∈R 64×64 : output projection matrix;

[0096] Output context vector C i = MultiHeadAttention(Q, K, V);

[0097] The prediction output layer accepts the context vector mapping output by the time attention mechanism layer, calculates the demand of each product in the future period through a demand prediction calculation formula, calibrates by integrating business rules, and then outputs the predicted fence product demand content;

[0098] Specifically, the prediction output layer maps the context vector C i As the input of the prediction output layer, the demand prediction calculation formula is:

[0099]

[0100] The demand of each product in the future period is calculated, calibrated by integrating business rules, and then the predicted future product demand content is output.

[0101] The business rules of the prediction output layer include:

[0102] Inventory constraint calibration: when the predicted demand exceeds the current inventory, adjust according to the inventory calibration rule:

[0103] Inventory calibration rule:

[0104] wherein, is the calibrated demand, is the original predicted demand output by the analysis engine layer, q i is the current inventory, and p is the inventory buffer coefficient, which is configured by the system and is valued at 1.2 according to experience;

[0105] When the predicted value exceeds 120% of the current inventory q i , the demand is forced to be limited to q i ×1.2 to avoid the risk of inventory overselling;

[0106] Credit ranking strategy: for agents with credit score s j ≥0.8, open all product procurement permissions; for agents with 0.6≤s j <0.8, limit the procurement of high-order products; for agents with s j <0.6, trigger risk control for manual review;

[0107] Regional heat weighting: for the regional hot search frequency w t Implement regional weighting:

[0108]

[0109] wherein, w t is the hot search frequency, N l is the number of agents in region l, N total is the total number of agents nationwide, and ll Region economic coefficient, default 1.0;

[0110] The nationwide hot search word frequency is dynamically adjusted according to the region of the agent by regional heat weighting, so as to more accurately reflect the demand difference in different regions;

[0111] In an example, if the nationwide hot search word frequency w t = 0.15 (15%), the proportion of agents in region A is and λ l = 1.2, then:

[0112] w l,t = 0.15 x (1+0.3) x 1.2 = 0.234

[0113] The interest heat of region A for the product is corrected to 23.4%, which is significantly higher than the national average level of 15%, and more inventory can be preferentially allocated to the region.

[0114] The S3 decision output layer generates an agent personalized recommendation report;

[0115] Specifically, the decision output layer outputs an agent personalized recommendation report, including: a product recommendation list, which is generated by a product comprehensive score sorting algorithm,

[0116] in descending order;

[0117] Comprehensive score sorting algorithm:

[0118] Wherein, the weight configuration: α = 0.5, β = 0.3, γ = 0.15, δ = 0.05

[0119] Inventory shortage warning: if it is marked as "shortage";

[0120] Dynamic commission adjustment strategy: according to product heat, implement step-by-step commission:

[0121]

[0122] If the predicted demand is 2.5 times the average sales in the near future, the commission will be increased by 20% (to encourage agents to preferentially sell), and if the predicted demand is only 30% of the average sales in the near future, the commission will be reduced by 10% (to reduce the risk of slow sales);

[0123] Inventory prediction warning: when the following conditions are met at the same time, trigger the replenishment warning:

[0124] and

[0125] Insufficient inventory days: the current inventory can only cover future Tthreshold Forecasting needs of the day;

[0126] Market heat verification: category search heat Above historical average

[0127] The report structure is divided into three modules: core recommendations, which include Top 5 products (including revenue, inventory, and commission); risk warnings, which include a list of low inventory products; and action recommendations, which include commission adjustment range and urgent replenishment.

[0128]

[0129] In the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, it can be fixed connection, or indirect connection through an intermediate medium, or internal communication of two elements or interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0130] In the description of the present application, the terms "upper", "lower", "front", "rear", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation on the present application.

[0131] The terms "first", "second", "third" (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the described embodiments of the present application can be implemented in an order other than those illustrated or described herein.

[0132] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or maintenance tool including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or maintenance tools.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that, for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application, and these will not affect the implementation effect and the practicality of the patent.

Claims

1. An Internet rights product agency management system, comprising: Product management module, order management module, agent management module, database module, data analysis module; The product management module is used to publish the classification, quantity, commission rate of Internet equity products, and manage the listing, purchase, sales, and delisting of Internet equity products, as well as inventory registration; The order management module is used to generate purchase and sales orders, calculate agent commissions and commission payment records, and query commission details; the agent management module is used to register agent accounts, perform real-name authentication, register sales licenses, assign call permissions for product management modules and order management modules to agent accounts, and record agent regions, historical purchase volumes, and credit scores. The credit score calculation rule is s j =0.7×fulfillment rate+0.3×(1-refund rate), fulfillment rate=number of completed orders / total number of orders; the database module stores data information of the product management module, the order management module and the agent management module; The data analysis module is characterized by: It comprises a data input layer, an analysis engine layer, and a decision output layer. The data input layer accesses the database module to obtain internal data, external market data, and accesses the industry report API to obtain competitor product prices and hot search terms. The analysis engine layer processes and analyzes the data, predicts future product demand, and recommends popular internet equity product procurement plans to agents. The decision output layer generates agent personalized reports based on the analysis of the analysis engine layer and feeds back to the agent management module.

2. The Internet rights product agency management system according to claim 1, characterized in that: The data input layer obtains product sets, agent sets, order sets and external market data, and pre-processes the above data; Specifically, the data input layer acquires data at time t including: product set P t , P t ={p1,p2,...,p n }, each product p i Include attributes: Category c i , inventory q i 、Commission ratio θ i ;Agent Set A t ,A t ={a1,a2,...,a m }, each agent a j Included attributes: Region j , credit score j , historical purchase vector h j =(h j1 ,h j2 ,...,h jn ); order set O t , O t ={O k |\O k =(a k ,p k ,t k ,d k )}, each product O k Included attributes:Agency a k 、Products k , order generation time t k 、sales volume k ; External market data D t , Each external market data D t Included attributes: hot search word frequency w t , Competitive product prices w t =Keyword search volume / all category search volume; the data input layer pre-processes the above data and then the analysis engine layer analyzes the data.

3. The Internet rights product agency management system according to claim 2, characterized in that: The analysis engine layer consists of a feature encoding layer, a time attention mechanism layer, and a prediction output layer. The analysis engine layer predicts the demand for each product in the future period (t+1) by integrating time series features, regional attributes, and market sentiment. To guide and purchase through agents; the demand forecasting formula is designed as follows: Among them, ELU(W p C i +b p ) is the basic prediction item, ∈ is the random fluctuation item, simulating the unpredictable fluctuation in the market, ∈~N(0,σ 2 )(σ=0.1),β·Ι 促销 is the artificial intervention item, β is the promotion gain coefficient, which is 0.5 according to experience, and 促销 It is an indicator variable, which takes 1 to indicate promotion and 0 to indicate no promotion.

4. The Internet rights product agency management system according to claim 3, characterized in that: The feature encoding layer converts product sets, agent sets, order sets, and external market data into high-dimensional learnable product features, regional features, and market sentiment feature vector representations to solve the problem of feature heterogeneity. The mathematical expression of its product feature sequence is: X i JReLU(W e [E i ||G l ||M t ]+b e ),[E i ||G l ||M t ]∈R 5 E i is the product’s own characteristics, E i =[EMA(d i,t-7:t ),θ i ,Norm(q i )], EMA(d i,t-7:t ) represents the exponential moving average of sales over the past 7 days, θ i is the commission ratio, Norm(q i ) is the inventory pressure, its calculation is based on industry standards, G l For regional characteristics, represents the sales of product p by all agents in region l i The total historical purchase volume, Indicates the inventory pressure of competing products, Indicates category c i The set of all products whose geographical location matches l; M t For the category market popularity, M t =TF-IDF(w ci ), calculated by multiplying the term frequency by the inverse document frequency; W ee is the weight matrix, W e ∈R 64×6 ; b e is the bias term, b e ∈R 64 .

5. The Internet rights product agency management system according to claim 4, characterized in that: The temporal attention mechanism layer receives the output X of the feature encoding layer i , stack the feature vectors of 24 consecutive hours into a time series matrix, and identify the periodicity, trend and suddenness of product demand in the time dimension through multi-head attention calculation; finally output a 64-dimensional context vector C i ; Specifically, the input 64-dimensional product feature sequence is split into 8 heads and linearly mapped to the query, key, and value spaces. The calculation formula is: Projection matrix W Q 、W K 、W V ∈R 64×64 , each head calculates attention independently: Q represents the characteristics of the moment K and V represent the characteristics of the past 24 hours (8 dimensions per head) The calculation of splicing multi-head attention is: MultiHead(Q, K, V) = Concat(head1,…,head8)W O W O ∈R 64×64 : Output projection matrix; Output context vector C i =MultiHeadAttention(Q,K,V).

6. The Internet rights product agency management system according to claim 5, characterized in that: The prediction output layer outputs the context vector C of the temporal attention mechanism layer i As the input of the prediction output layer, the demand prediction formula is calculated: Estimate the demand for each product in the future period, integrate business rules for calibration, and then output the predicted future product demand content; The business rules of the prediction output layer include: Inventory constraint calibration: When the forecast demand exceeds the current inventory, adjustments are made according to the inventory calibration rules: Stock calibration rules: in, is the demand after calibration, is the original forecast demand, output by the analysis engine layer, q i is the current inventory level, ρ is the inventory buffer coefficient, which is configured by the system and is empirically set to 1.2; When the predicted value Exceeds current inventory i When the demand is 120%, the demand is forced to be limited to q i ×1.2, to avoid the risk of overselling inventory; Credit grading strategy: credit score j ≥0.8 agents, open all product purchasing rights; 0.6≤s j <0.8 agents, limit the purchase of high-order products; j For agents with a score <0.6, risk control will be triggered for manual review; Regional popularity weighting: the frequency w of hot search terms in region l t Implementing geographical weighting: Among them, w t is the frequency of hot search terms, N l is the number of agents in region l, N total is the number of general agents nationwide, λ l is the regional economic coefficient, the default value is 1.0; By weighting the regional popularity, the frequency of national hot search terms is dynamically adjusted according to the region where the agent is located, more accurately reflecting the differences in demand in different regions.

7. The Internet rights product agency management system according to claim 6, characterized in that: The decision output layer outputs personalized recommendation reports for agents, including: a product recommendation list, which is sorted in descending order using a product comprehensive score sorting algorithm; Comprehensive score sorting algorithm: Among them, the weight configuration is: α = 0.5, β = 0.3, γ = 0.15, δ = 0.05 Insufficient inventory warning: If It is marked as "in short supply"; Dynamic commission adjustment strategy: Implement tiered commissions based on product popularity: If the forecast demand is 2.5 times the recent average sales volume, the commission will be increased by 20%. If the forecast demand is only 30% of the recent average sales volume, the commission will be reduced by 10%. Inventory forecast warning: A replenishment warning is triggered when the following conditions are met simultaneously: and Insufficient inventory days: Current inventory can only cover the future T threshold forecast demand for the day; Market popularity verification: Category search popularity Higher than the historical average The structure of the personalized report is divided into three modules: core recommendations, which include the top five products in the product recommendation list, including revenue, inventory, and commission; risk warnings, which include a list of low-inventory products; and action recommendations, which include the commission adjustment range and replenishment urgency.

Images