Content delivery account zeroing processing method, content delivery account management system, content delivery system

Abstract

The embodiment of the application provides a content delivery account zero clearing processing method, a content delivery account management system and a content delivery system. The content delivery account zero clearing processing method comprises the following steps: receiving a content delivery account zero clearing request, analyzing the content delivery account zero clearing request to determine a target content delivery account; performing a zero clearing operation on the target content delivery account and adding a zero clearing work order; in response to the result of the zero clearing operation indicating that the target content delivery account is in a non-zero clearing state, adding the target content delivery account to a monitoring pool and setting a zero clearing work order amount in the added zero clearing work order; performing a regular scan on the state of the target content delivery account in the monitoring pool, so as to initiate a zero clearing request and perform a zero clearing operation on the target content delivery account in response to the target content delivery account changing from the non-zero clearing state to a zero clearing state; and updating the zero clearing work order amount in the zero clearing work order in response to the zero clearing operation being successfully performed.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method for clearing content delivery accounts, a content delivery account management system, and a content delivery system. Background Technology

[0002] In the field of digital information dissemination, various media platforms widely undertake the management and data processing of content distribution accounts. Currently, due to limitations in the technical architecture and data interaction mechanisms of media platforms, some content distribution accounts cannot be reset under certain conditions. For example, when there are data synchronization delays, system interface compatibility issues, or locked underlying data structures, the reset command cannot be executed smoothly.

[0003] For users of content delivery accounts, when faced with this non-resettable state, the resources occupied by the account cannot be released in a timely manner. Due to the limited total system resources, users cannot quickly allocate these resources to other content delivery accounts to achieve optimized allocation and efficient utilization of information dissemination resources. This situation leads to inefficient resource management during the information dissemination process, and the problems of idle resources and unreasonable allocation have persisted for a long time, seriously restricting the improvement of the overall performance and functionality of the information dissemination system. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method for clearing content delivery accounts, a content delivery account management system, and a content delivery system to at least partially solve the above problems.

[0005] According to a first aspect of the present invention, a method for clearing content delivery accounts is provided, comprising:

[0006] Receive a request to clear the content delivery account and parse the request to determine the target content delivery account;

[0007] Perform a zeroing operation on the target content delivery account and add a zeroing work order;

[0008] In response to the result of the clearing operation, the target content delivery account is in a non-clearable state, so it is added to the monitoring pool and the clearing work order amount is set in the new clearing work order.

[0009] The status of the target content delivery account in the monitoring pool is scanned periodically. In response to the target content delivery account changing from a non-clearable state to a clearable state, a clearing request is initiated and the target content delivery account is cleared.

[0010] In response to the successful execution of the clearing operation, the amount in the clearing work order is updated.

[0011] According to a second aspect of the present invention, a content delivery account management system is provided, comprising: a monitoring terminal, a resource execution terminal, and an interaction terminal, wherein:

[0012] The interactive end is used to generate a request to clear the content delivery account and send it to the monitoring end;

[0013] The monitoring terminal is used to perform the following steps:

[0014] Receive a request from the interactive client to clear the content delivery account, and parse the request to determine the target content delivery account.

[0015] The resource execution end is triggered to attempt to clear the target content delivery account and a new clearing work order is created locally;

[0016] When the received clearing operation result indicates that the target content delivery account is in an unclearable state, the target content delivery account is added to the monitoring pool configured by itself, and the clearing work order amount is set in the new clearing work order.

[0017] The status of the target content delivery account in the monitoring pool is scanned periodically. When the target content delivery account changes from a non-clearable status to a clearable status, a clearing request is sent to the resource execution end.

[0018] After receiving the successful clearing operation result from the resource execution terminal, update the clearing work order amount in the clearing work order;

[0019] The resource execution terminal is used to perform the following steps:

[0020] Based on the trigger from the monitoring terminal, perform an attempt to clear the target content delivery account and report the result back to the monitoring terminal.

[0021] Receive the clearing request from the monitoring terminal, perform the clearing operation on the target content delivery account, and feed back the operation result to the monitoring terminal.

[0022] According to a second aspect of the present invention, a content delivery system is provided, comprising a backend system, a media terminal, and an agency terminal, wherein the collaborative workflow of each terminal is as follows:

[0023] The agent generates a request to clear the content delivery account and sends it to the backend system.

[0024] After receiving the request, the backend system parses it to determine the target content delivery account, and then triggers the media side to perform an attempt to clear the target content delivery account, while simultaneously creating a clearing work order locally;

[0025] The media platform, based on the triggering instructions from the backend system, attempts to clear the target content distribution account and then sends the result back to the backend system.

[0026] If the feedback indicates that the target content delivery account is in a non-resetable state, the backend system will add the target content delivery account to its own configured monitoring pool and set the reset work order amount in the new reset work order; then, the backend system will periodically scan the status of the target content delivery account in the monitoring pool.

[0027] When the target content delivery account changes from a non-resettable state to a resettable state, the backend system sends a reset request to the media platform.

[0028] The media platform receives a clearing request from the backend system, performs a clearing operation on the target content distribution account, and feeds back the operation result to the backend system.

[0029] After receiving confirmation from the media that the clearing operation was successful, the backend system updates the clearing order amount in the clearing order.

[0030] The solutions in the embodiments of the present invention have the following technical advantages:

[0031] 1. Precise Target Account Positioning: By receiving and parsing content delivery account clearing requests, the system identifies the target content delivery account, accurately pinpointing the specific account requiring clearing. In the background, due to technical limitations and complex account statuses on media platforms, it's often impossible to accurately locate the account to be processed. This precise positioning method avoids misoperation, ensuring that resource release and allocation are targeted at the correct account, thus improving the accuracy of resource management.

[0032] 2. Attempt to Clear Accounts and Work Order Management: This mechanism performs an attempt to clear accounts for target content and creates a new clearing work order. This provides a clear record and management basis for the entire clearing process. When an account cannot be cleared, the clearing work order records relevant information for easy tracking and processing. Compared to the chaotic resource management and inability to effectively trace account status in existing technologies, work order management makes the entire process more orderly and improves the standardization of resource management.

[0033] 3. Monitoring Pool Mechanism for Handling Non-Resettable Accounts: When an account used for delivering target content is in a non-resettable state, it is added to the monitoring pool and a reset ticket amount is set. This solves the problem of resources not being released in a timely manner when an account cannot be reset in existing technologies. The monitoring pool acts as a "buffer," centrally managing non-resettable accounts and waiting for their status to change. Simultaneously, setting a reset ticket amount helps in accurate subsequent resource accounting and allocation.

[0034] 4. Scheduled Scanning and State Transition Handling: The system periodically scans the accounts in the monitoring pool that are targeted for content delivery. When an account changes from a non-resettable state to a resettable state, a reset request is promptly initiated and executed. This proactive monitoring and timely response mechanism effectively solves the problems of resource idleness and unreasonable allocation in existing technologies. It enables reset operations to be performed as soon as the account status allows, releasing the resources occupied by the account and allowing resources to be quickly allocated to other accounts that need them, thus improving the utilization efficiency of information dissemination resources.

[0035] 5. Work Order Amount Update: After a successful clearing operation, the amount in the cleared work order is updated. This ensures the accuracy and real-time nature of resource management data. Compared to the inefficiency and untimely data updates in existing technologies, timely updates to work order amounts provide accurate data support for subsequent resource allocation and decision-making, further enhancing the overall performance and functionality of the information dissemination system. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0037] Figure 1 This application provides a method for clearing content delivery accounts. Detailed Implementation

[0038] like Figure 1 As shown in the embodiment of this application, a method for clearing content delivery accounts is provided, which includes:

[0039] Receive a request to clear the content delivery account and parse the request to determine the target content delivery account;

[0040] Perform a zeroing operation on the target content delivery account and add a zeroing work order;

[0041] In response to the result of the clearing operation, the target content delivery account is in a non-clearable state, so it is added to the monitoring pool and the clearing work order amount is set in the new clearing work order.

[0042] The status of the target content delivery account in the monitoring pool is scanned periodically. In response to the target content delivery account changing from a non-clearable state to a clearable state, a clearing request is initiated and the target content delivery account is cleared.

[0043] In response to the successful execution of the clearing operation, the amount in the clearing work order is updated.

[0044] Alternatively, methods for clearing content delivery accounts may also include:

[0045] If the result of the clearing operation indicates that the target content delivery account is in a clearing-available state, a new clearing work order is created, a clearing request is initiated, and the clearing operation is performed on the target content delivery account. If the clearing operation is successfully executed, the actual clearing amount is written into the clearing work order.

[0046] Preferably, in a specific scenario, the preferred or alternative implementation of the above solution is as follows:

[0047] For content delivery accounts, a multidimensional feature array is used to store their detailed information. Each account is represented in the array as a multidimensional vector, with each dimension of the vector representing a feature of the account, such as account creation time, last delivery time, delivery frequency, historical clearing records, and current balance. This multidimensional feature array not only comprehensively describes the account's status but also facilitates subsequent feature engineering and data analysis.

[0048]

[0049] To efficiently manage and query account status information, a quadtree data structure is used to construct a status index. Each node in the quadtree represents a status interval, and the node stores the account information within that status interval. Through the hierarchical structure of the quadtree, accounts in a specific status can be quickly located, greatly improving the efficiency of status queries.

[0050] A knowledge graph is constructed for content delivery accounts, linking accounts with related users, delivery channels, and content. Nodes in the knowledge graph represent entities, and edges represent relationships between entities, such as an account belonging to a particular user or being delivered through a specific channel. Through association analysis of the knowledge graph, the potential relationships between accounts can be deeply explored, providing a more comprehensive basis for decision-making regarding account cleanup operations.

[0051] Based on the above, the technical implementation steps are as follows:

[0052] 1. Receive a request to clear content delivery accounts, and parse the request to determine the target content delivery account.

[0053] The distributed message queue Kafka is used to receive content delivery account reset requests from different clients. Kafka's high throughput, scalability, and fault tolerance ensure efficient request reception and processing. Requests are transmitted using a custom binary protocol that includes key request information such as account ID, request timestamp, and request type.

[0054]

[0055] Upon receiving a request, the account ID is feature extracted and encoded. A hash function is used to map the account ID to a high-dimensional feature space, and then a feature matching algorithm is used to find the corresponding target content delivery account in the multi-dimensional feature array. Simultaneously, by combining a knowledge graph, the relationships between this account and other entities are analyzed to further verify the accuracy of the target account.

[0056] 2. Perform a trial clearing operation on the target content delivery account and add a clearing work order.

[0057] Based on a multidimensional feature array of a target account, a deep learning model is used to predict whether the account can be zeroed out. The deep learning model takes the account's feature vector as input and outputs a zeroing-out score. If the score exceeds a set threshold, an attempt is made to zero out the account.

[0058]

[0059] While attempting to reset the data, a new reset work order is created. Work order information is stored in the distributed file system HBase, using the account ID and timestamp as a composite primary key to ensure uniqueness and efficient querying. The work order record includes information such as request time, target account ID, predicted score, and work order status.

[0060]

[0061] 3. In response to the result of the clearing operation indicating that the target content delivery account is in a non-clearable state, add it to the monitoring pool and set the clearing work order amount in the new clearing work order.

[0062] If the zeroing operation fails, the target account will be added to the monitoring pool. The monitoring pool is implemented using a distributed cache, Redis, to improve data read and write speed. The account's multi-dimensional feature vector and current status information are stored in Redis for easy subsequent status scanning and updates.

[0063]

[0064] Set the clearing amount in the new clearing work order. Based on the account's current balance and historical clearing records, use a machine learning algorithm to predict the final clearing amount for the account. Write the predicted amount into the work order record in HBase.

[0065]

[0066] 4. Periodically scan the status of the target content delivery accounts in the monitoring pool. In response to a target content delivery account changing from a non-clearable state to a clearable state, initiate a clearing request and perform the clearing operation on the target content delivery account.

[0067] The Quartz scheduled task framework is used to periodically scan the accounts in the monitoring pool for target content. At regular intervals, the status information of the accounts is retrieved from Redis, and the accounts whose status has changed are quickly located using a quadtree status index.

[0068]

[0069] When the account status is found to be eligible for reset, a reset request is initiated. The deep learning model is used again to evaluate the account's eligibility for reset. If the score still exceeds the threshold, the reset operation is performed.

[0070] 5. In response to the successful execution of the clearing operation, update the clearing order amount in the clearing work order.

[0071] After the reset operation is successful, the corresponding reset work order record is retrieved from HBase, and the reset work order amount field in the work order is updated according to the actual reset amount. At the same time, the relevant information of the account in the knowledge graph is updated to record the detailed information of this reset operation.

[0072]

[0073] When an account is determined to be in a state where it can be cleared, the process of creating a new clearing work order, initiating a clearing request, and updating the work order amount is executed directly, similar to the steps described above. However, when creating a new work order, it can be more finely categorized and labeled based on the account's feature vector and knowledge graph information for subsequent data analysis and statistics.

[0074] In particular, in another specific application scenario, the implementation process of the above solution is as follows:

[0075] 1. Receiving requests and locating target accounts

[0076] 1. Access Request: Receive content delivery account clearing requests from clients via Kafka distributed message queue. The messages are encapsulated using a custom binary protocol (including account ID, timestamp, etc.).

[0077] 2. ID Feature Extraction and Mapping:

[0078] Using the Locality Sensitive Hash (LSH) function family Execute formula y on account ID x i =h i (x) is calculated and mapped to binary code y. i Construct high-dimensional feature representations.

[0079] Parameter ω i f(x), b i Each account is trained and optimized using historical account data to ensure that similar accounts are distributed in close proximity in the feature space.

[0080] 3. Feature matching and validation:

[0081] Calculate the target account feature vector v t Account v in a multidimensional feature array j cosine similarity Filter out S jt >τ candidate accounts.

[0082] For candidate accounts, a relevance score is calculated using a graph convolutional network (GCN). By combining the entity relationships in the knowledge graph, the target account is finally determined.

[0083] 2. Zeroing prediction and work order creation

[0084] 1. Feature Preprocessing: Input the multidimensional feature array of the target account (such as creation time, balance, etc.) into the embedding layer and convert it into a low-dimensional vector e. i .

[0085] 2. Deep prediction model computation:

[0086] Attention layer based on formula Calculate the weights α for each feature i and through Generate a weighted feature vector.

[0087] The fully connected layer outputs a zeroing score P through P = σW2·σW1·z + b1) + b2).

[0088] 3. Decision-making and work order creation: If P > 0.8, attempt to clear the value; at the same time, create a clear work order containing the value P in HBase using the account ID and timestamp as the composite primary key, and record the request time, status and other information.

[0089] 3. Handling of Non-Zeroing Status and Amount Prediction

[0090] 1. Monitoring pool storage: If clearing fails, store the account's multidimensional feature vector x and the status "cannot be cleared" in the Redis monitoring pool, with the account ID as the key.

[0091] 2. Estimated amount to be cleared:

[0092] Using an LSTM network to process the historical zeroing amount sequence y = [y1, y2, ..., y T ], through formula f t i t ,o t ,c t ,h t Calculate cell state and hidden state.

[0093] The fully connected layer will eventually hide the state h. T pass Convert to predicted amount And update the "Clear Amount" field in the HBase work order.

[0094] 4. State Scanning and Priority Scheduling

[0095] 1. Quadtree Index Scan: A Quartz scheduled task triggers a scan to retrieve account status from Redis and indexes it using a quadtree status index function. Quickly locate accounts whose status has changed.

[0096] 2. Priority is dynamically adjusted:

[0097] For monitoring pool accounts, priority scores are calculated based on the Multi-Armed Slot Machine (UCB) algorithm. in It is dynamically updated based on historical scans and status change feedback.

[0098] According to U a Adjust the scanning frequency: shorten the scanning interval for high-priority accounts and lengthen the interval for low-priority accounts.

[0099] 3. Status transition handling: If the account becomes "can be cleared", repeat the prediction model calculation in step 2. After confirming that the score meets the standard, perform the clearing operation and update the HBase work order amount and knowledge graph record.

[0100] 5. Information update after successful clearing

[0101] After the clearing operation is successful, the actual cleared amount is obtained, and the "cleared amount" field in the HBase work order is updated; at the same time, the knowledge graph update function is called to record the operation details and improve the account association data.

[0102] The algorithms used in the above steps are explained below:

[0103] 1. Account ID Feature Extraction and Target Localization

[0104] Hash function mapping: The Locality Sensitive Hash (LSH) algorithm is used to map account IDs to a high-dimensional feature space. A family of hash functions is defined. Where d is the number of hash functions, and each h i Map account ID x to binary code y i ∈{0,1}.

[0105] in: The weight vector of the i-th hash function, where n is the dimension of the account feature vector (e.g., the dimension in a multidimensional feature array); f(x): the feature extraction function for account ID x, outputting its feature vector; b i : The bias term of the i-th hash function; <·,·>: Vector inner product operation.

[0106] Feature matching algorithm: Calculates the matching degree between the target account and accounts in the multidimensional feature array based on cosine similarity. Let v be the feature vector of a certain account in the multidimensional feature array. j The target account feature vector is v t Match S jt for:

[0107] Where: ‖·‖: L2 norm of the vector; when S jt When the threshold τ is exceeded, the match is considered successful.

[0108] Knowledge graph verification: Calculate the association degree between accounts using a Graph Convolutional Network (GCN). Let the knowledge graph be... node It includes entities such as accounts, users, and channels, with edges ε representing the relationships between entities. For the target account v t Its correlation score R t for:

[0109]

[0110] in: Target account v t The set of adjacent nodes; h i Node v i λ: Feature vector; W: Learnable weight matrix; b: Bias vector; σ: Activation function (e.g., ReLU).

[0111] 2. Zeroing Prediction Model

[0112] A prediction model is constructed using a deep neural network (DNN) combined with an attention mechanism. The input is the account feature vector x = [x1, x2, ..., x...]. nThe output is the zeroing score P.

[0113] Network structure:

[0114] 1. Embedding layer: Maps discrete features (such as account type) to a low-dimensional vector e. i ;

[0115] 2. Attention layer: Calculate feature weights α i :

[0116] in:

[0117] Query vector and key vector, where m is the embedding dimension;

[0118] 3. Weighted Summation Layer: Generates a weighted feature vector z:

[0119] Fully connected layer: Outputs scores via a multilayer perceptron (MLP):

[0120] P = σ(W2·σ(W1·z+b1)+b2) where:

[0121] W1, W2: Weight matrices;

[0122] b1, b2: Bias vectors.

[0123] 3. Zero-out amount prediction model

[0124] A Long Short-Term Memory (LSTM) network combined with time series forecasting techniques is employed. Let the historical zero-out amount sequence be y = [y1, y2, ..., y]. T The current account feature vector is x, and the predicted amount is... for:

[0125] LSTM structure:

[0126] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0127] i t =σ(W i h t-1 ,x t ]+b i )

[0128] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0129]

[0130] h t =o t ☉tanh(c t )

[0131] -where: f t i t ,o t Forget gate, input gate, output gate vector; c t ,h t : Cell state and hidden state; ⊙: Element-level multiplication; W * ,b * Learnable network parameters.

[0132] Predicted output: The LSTM output h is passed through a fully connected layer. T Mapped to predicted amount:

[0133] 4. State Scanning and Priority Scheduling

[0134] Quadtree State Index: Let N be the quadtree node that stores the set of accounts. Its state interval is [s min ,s max Define the status query function. Return the subset of accounts with status 's':

[0135]

[0136] in: The set of accounts for the i-th child node of node N. Priority scheduling: Dynamically adjust the scan frequency based on a multi-armed slot machine algorithm (such as UCB). Let n be the historical scan count of account a. a The reward value is r a Its priority score U a for:

[0137]

[0138] in: Average reward value for account a; c: Exploration - utilizing balance parameters; A collection of all accounts in the monitoring pool.

[0139] In this application, for Locality Sensitive Hash (LSH) and cosine similarity, LSH designs a family of hash functions such that points that are close in distance in the original space also have the same hash code with a high probability in the hash space. In account ID feature extraction, the formula y is used... i =h i(x) Maps the account ID to a high-dimensional feature space using the weight vector ω of the hash function. i and bias term b i A linear transformation and thresholding are applied to the account feature vector f(x) to generate a binary code. Cosine similarity is then used. This is used to measure the similarity between feature vectors; the closer the value is to 1, the more consistent the vector directions. In the process of clearing content delivery accounts, there are numerous account IDs from different clients with varying formats. LSH can quickly map account IDs to a feature space that facilitates comparison, and cosine similarity can efficiently filter out target accounts. For example, when multiple accounts share similar delivery channels, user groups, or other characteristics, LSH can map these accounts to similar hash encoding regions, and cosine similarity can further pinpoint the target account.

[0140] In this application, for Graph Convolutional Networks (GCNs), GCNs update the features of nodes themselves by aggregating the feature information of neighboring nodes based on graph structure data. (Formula) In the middle, by targeting account v t The adjacent node v i eigenvector h i The weighted summation is then applied, followed by an activation function σ to obtain the account's relevance score. The learnable weight matrix W and bias vector b are continuously optimized during training to capture the complex relationships between accounts. In content delivery scenarios, accounts have complex relationships with entities such as users, delivery channels, and delivery content. GCN can leverage knowledge graphs to mine these relationships, providing a more comprehensive basis for decision-making regarding account termination. For example, if an account is closely associated with high-value users or high-quality delivery channels, the relevance score calculated by GCN will be higher. The system can use this to assess the impact of terminating the account and avoid blind actions.

[0141] In this application, a degree neural network (DNN) combined with an attention mechanism is used. The DNN extracts features and recognizes patterns from the input data through nonlinear transformations of multiple layers of neurons. The attention mechanism is introduced using the following formula: Calculate feature weights so that the network can focus on account features that are more important for predicting zero-out rates. Weighted summation layer. A weighted feature vector is generated, and then a fully connected layer is used to output the predicted score P. Content delivery accounts have numerous features, such as creation time, delivery frequency, and account balance. Different features have varying degrees of impact on whether an account can be cleared. An attention mechanism allows the model to automatically learn the importance of each feature. For example, in some scenarios, account balance and recent delivery anomalies are key factors in determining whether an account can be cleared. The model will assign higher weights to these features, thus more accurately predicting whether an account can be cleared.

[0142] In this application, for Long Short-Term Memory (LSTM) networks, LSTM uses a forgetting gate ft Input gate i t and output gate o t Controlling cell state c t This information flow solves the gradient vanishing and gradient exploding problems of traditional recurrent neural networks (RNNs), making it suitable for processing time series data. Formula f t i t ,o t ,c t ,h t The process of LSTM operation at each time step is described. By learning from the historical zeroing amount sequence, it captures long-term dependencies in the time series and predicts the future zeroing amount.

[0143] The amount of money cleared from account balances for content delivery shows certain patterns over time and is influenced by historical data. LSTM can learn the trends and periodicity of historical cleared amounts for accounts. For example, there are significant fluctuations in the amount of money cleared from account balances before and after promotional seasons. LSTM can predict the current amount of money cleared from account balances based on data from similar historical periods, providing an accurate reference for setting ticket amounts.

[0144] In this application, the Multi-Armed Slot Machine (UCB) algorithm is used to balance exploration (trying new strategies) and exploitation (selecting a known optimal strategy). Formula middle, This represents the average reward value for account a, reflecting the known changes in the account's status. For the exploration term, as the number of scans n increases... a As the value increases, the number of exploration terms gradually decreases. The algorithm maximizes U. a The algorithm dynamically adjusts the scanning priority of accounts. Since the status changes of accounts in the monitoring pool are uncertain, the UCB algorithm can dynamically adjust the scanning frequency based on historical scanning results and status change feedback. For example, for accounts with frequent status changes but not yet at a state where they can be cleared, the algorithm will increase their scanning priority and intensify the exploration; for accounts with stable status that cannot be cleared for a long time, the algorithm will reduce the scanning frequency to minimize resource consumption.

[0145] Therefore, the above-mentioned solution of this application has the following technical advantages compared with the traditional technology:

[0146] 1. For target account positioning, simple string matching or single-attribute-based retrieval are typically used, such as precise matching based solely on account ID. This approach cannot handle complex feature associations and similar account filtering. This solution combines LSH, cosine similarity, and GCN to quickly filter target accounts from a high-dimensional feature space and uses knowledge graphs to mine potential relationships between accounts for verification. Compared to traditional techniques, this significantly improves the accuracy and efficiency of target account positioning, reduces false positives caused by similar account features, and is particularly suitable for large-scale content delivery account scenarios with diverse features.

[0147] 2. For predicting account balance reset, traditional techniques often rely on rule engines or simple statistical models, such as setting fixed thresholds to determine if an account balance meets the reset criteria. These methods fail to comprehensively consider the various dynamic characteristics of the account. This application employs a DNN combined with an attention mechanism, which automatically learns the importance of each account feature and extracts key information from massive amounts of data for prediction. Compared to traditional techniques, the prediction model has stronger non-linear fitting and generalization capabilities, enabling more accurate judgment of account reset potential under complex conditions, reducing false positive rates, and improving the success rate and security of the reset operation.

[0148] 3. For predicting zero-out amounts, simple models such as linear regression are generally used, assuming a linear relationship in the data, which makes it difficult to capture complex patterns and long-term dependencies in time series. This application uses LSTM to process historical zero-out amount sequences, effectively learning features such as trends, periodicity, and abnormal fluctuations in the time series. Compared with traditional techniques, it is more accurate in predicting account zero-out amounts, providing a reliable basis for setting work order amounts and avoiding financial disputes or unreasonable resource allocation problems caused by inaccurate amount predictions.

[0149] 4. Current methods for state scanning and priority scheduling often employ fixed scanning cycles or manually set priority rules, lacking dynamic adaptability to account state changes. This application achieves efficient state scanning and dynamic priority scheduling based on quadtree state indexing and the UCB algorithm. The quadtree can quickly locate accounts with changing states, and the UCB algorithm automatically adjusts the scanning frequency based on historical account feedback. Compared to traditional techniques, this significantly improves resource utilization efficiency, reduces invalid scans, ensures the system can respond promptly to account state changes, and accelerates the processing speed of accounts that can be cleared.

[0150] Preferably, the method for clearing content delivery accounts further includes: assigning a priority order to all target content delivery accounts added to the monitoring pool, so that the accounts are scanned according to the priority order when the status of the target content delivery accounts in the monitoring pool is scanned periodically.

[0151] Preferably, it also includes: configuring scanning frequencies from high to low priority, so that when periodically scanning the status of target content delivery accounts in the monitoring pool, the higher the priority of the target content delivery account, the higher the scanning frequency for it.

[0152] Preferably, the priority order for all target content delivery accounts added to the monitoring pool is assigned, including: sorting the target content delivery accounts according to the order in which they were added to the monitoring pool, and assigning priority order based on the sorted queue.

[0153] Preferably, in a specific application scenario, the following description is provided in a preferred or alternative manner.

[0154] 1. Priority dynamic allocation model based on time entropy

[0155] 1.1 Calculation of Time Entropy

[0156] In this application, time entropy is used to quantify the temporal relationship between accounts joining the monitoring pool, and a time entropy function H(T) is defined to measure the account set T = {t1, t2, ..., t}. n Uniformity of time distribution:

[0157]

[0158] Where: t i : Timestamp of account i joining the monitoring pool; λ: Time decay coefficient, optimized through historical data fitting, used to adjust time weight (the larger λ is, the higher the priority of recently joined accounts); exponential term exp(-λt) i The timestamp is decayed to give higher weight to accounts that joined earlier.

[0159] 1.2 Priority Calculation

[0160] In this application, account a is defined based on time entropy. i Priority score

[0161] in: Account A i The cumulative number of status changes since joining the monitoring pool (based on knowledge graph analysis of related events); max(ΔS): the maximum number of status changes for all accounts in the monitoring pool; the second term. This is used to dynamically adjust priorities, and accounts with frequent status changes receive additional weight.

[0162] 2. Adaptive Scan Frequency Dynamic Adjustment Model

[0163] 2.1 Scanning Value Assessment Function

[0164] In this application, the scanning value function V(a) is defined. i ) Quantify each scan of account a i Expected returns:

[0165] Where: α, β: weight coefficients, optimized through reinforcement learning to balance the influence of priority, time interval, and prediction score; TimeDiff(t last ,t now Account a i Last scan time t last With current time t now Time difference; MaxTimeDiff: the maximum time difference among all accounts in the monitoring pool; PredictionScore(a i Account a i The probability prediction score for zeroing out (output by the DNN model).

[0166] 2.2 Calculation of scanning frequency

[0167] In this application, account a is calculated using a piecewise function based on scan value. i Scan frequency

[0168] Where: τ1, τ2, τ3: frequency adjustment coefficients, satisfying τ1<τ2<τ3; θ1, λ2: value thresholds, determined through cluster analysis of historical scan data; the scanning frequency is inversely proportional to the scanning value, and high-value accounts receive a higher scanning frequency.

[0169] 3. Quadtree State Indexing and Scan Scheduling Optimization

[0170] 3.1 Quadtree Node State Entropy

[0171] In this application, the state entropy H of a quadtree node N is defined. N To measure the uncertainty of the distribution of account status within a node: Where: S: Account status set (e.g., {'cannot be cleared', 'can be cleared'}); |A N,s |: The number of accounts in node N with state s; |A N |: The total number of accounts in node N.

[0172] 3.2 Scan Path Planning Algorithm

[0173] In this application, Monte Carlo Tree Search (MCTS) combined with quadtree state entropy is used to plan the scan path. During each scan, the branch with the highest state entropy is visited first, as shown in the formula:

[0174]

[0175] Where: N * : Optimal child node; c: Explore - utilize the balancing parameter; N parent The number of times the parent node has been visited; Child node N i The number of visits.

[0176] In summary, the technical principles of the above algorithm in this application are as follows:

[0177] 1. Priority dynamic allocation model based on time entropy

[0178] This model utilizes the concept of entropy from information theory, quantifying the uniformity of the time distribution of accounts joining the monitoring pool through the time entropy function H(T). The exponential term exp(<λt) i The timestamp is decayed to give higher weight to accounts that joined earlier in the priority calculation. λ, as the time decay coefficient, can adjust the rate of weight decay. Priority scoring formula. Based on this, combined with the number of account status changes The system dynamically adjusts priorities, giving accounts with frequent status changes an extra boost. During the clearing of content delivery accounts, the monitoring pool contains accounts added at different times, and some accounts experience frequent status changes. For example, a newly launched campaign may cause frequent status changes for related accounts, and traditional priority allocation based on a fixed time sequence cannot respond promptly to such changes. This model, however, dynamically adjusts priorities based on account addition time and status changes, ensuring that accounts requiring urgent processing are scanned and processed first.

[0179] 2. Adaptive Scan Frequency Dynamic Adjustment Model

[0180] Scanning value function V(a) i Taking into account account priority Time interval since the last scan (TimeDiff(t)) last ,t now ) and the probability of being reset to zero PredictionScore(a i The influence of various factors is balanced through weighting coefficients α and β. The formula for calculating scan frequency is as follows: A piecewise function is used to allocate scanning resources based on scanning value, with accounts of higher scanning value receiving more frequent scans, thus achieving dynamic allocation of scanning resources. The value and status of content delivery accounts vary; some accounts are more likely to be cleared and have higher priority, requiring increased scanning frequency; while some low-value, stable accounts do not need frequent scanning. For example, accounts with high budgets and approaching expiration are more likely to be cleared and have higher priority; this model will allocate a higher scanning frequency to them to ensure timely clearing and avoid resource waste.

[0181] 3. Quadtree State Indexing and Scan Scheduling Optimization

[0182] The state entropy H of a quadtree node N Entropy is used to measure the uncertainty of account state distribution within a node; the more dispersed the state distribution, the higher the entropy value. Monte Carlo Tree Search (MCTS) combines state entropy with the formula... Prioritizing nodes with high state entropy and relatively few visits optimizes the scanning path. The monitoring pool contains numerous accounts with complex states, making traditional sequential or random scanning inefficient. A quadtree divides accounts into state ranges, and MCTS prioritizes exploring regions with uncertain states based on state entropy. For example, when many accounts are in a "cannot be cleared" state, the system can prioritize exploring node branches where state transitions are possible, quickly locating accounts that can be cleared and improving scanning efficiency.

[0183] Therefore, the above-mentioned technical implementation of this application has the following technical advantages compared with the traditional technology:

[0184] 1. Priority allocation

[0185] Traditional methods using fixed priority rules, such as simply arranging accounts according to the order in which they were added to the monitoring pool, or relying on manually set static priorities, cannot adjust to dynamic changes in account status. This solution dynamically adjusts priorities based on time entropy and status changes, considering not only time factors but also changes in account status. Compared to traditional technologies, it can more flexibly respond to dynamic changes in account status, making priority allocation more aligned with actual needs, ensuring that urgent or critical accounts are processed first, reducing overall processing time, and improving system response speed.

[0186] 2. Scan frequency configuration

[0187] Scanning frequency is typically statically configured, with all accounts using the same or a few fixed scanning cycles, lacking specificity based on account value and status changes. This solution employs an adaptive scanning frequency dynamic adjustment model, comprehensively considering account priority, time intervals, and zeroing probability prediction scores to dynamically calculate the scanning frequency. Compared to traditional technologies, scanning resources can be precisely allocated to high-value, high-probability-of-zeroing accounts, avoiding ineffective scanning of low-value accounts. This significantly improves the success rate and processing efficiency of zeroing operations while saving system resources.

[0188] 3. Scan scheduling

[0189] Traditional methods, such as sequential or random scanning without optimized scanning paths, are inefficient when handling large numbers of accounts and struggle to quickly locate accounts that can be cleared. This solution utilizes a quadtree state index and MCTS for optimized scan scheduling, prioritizing the exploration of regions with uncertain states. Compared to traditional techniques, this significantly reduces invalid traversals during the scanning process, enabling rapid location of accounts that can be cleared. Especially in scenarios with a large number of accounts and complex states, this significantly improves scanning efficiency and accelerates the overall progress of account clearing.

[0190] Optionally, it also includes: in response to the clearing request, releasing the available resource quota for the target content delivery account, so that the right to use the available resource quota is transferred to the resource management module of the target content delivery account.

[0191] Optionally, it also includes: generating and sending an alarm notification in response to a failure of the zeroing operation.

[0192] The following provides a detailed explanation of how to release resource quotas in response to a zeroing request and how to generate alarm notifications when a zeroing operation fails.

[0193] 1. Respond to the clear request to release resource quota.

[0194] Use a ResourceManager object to manage the resource quotas for all content delivery accounts. This object can be implemented as a class, containing a hash table (dictionary) where the key is the account ID and the value is the corresponding resource quota object (ResourceQuota).

[0195] Each resource quota object contains information such as the currently available resource quota, the total resource quota, and resource usage permissions.

[0196] Use a message queue (such as Kafka) to receive zeroing request messages to ensure asynchronous processing of requests and high concurrency performance.

[0197] 1.1 Receive reset request

[0198] When the system receives a reset request, the request is transmitted via a message queue (such as Kafka). The request message contains the ID of the target content delivery account and the request type (in this case, a reset request). The processor reads the request message from the message queue and parses it.

[0199] The following is a Python code example simulating receiving messages from Kafka:

[0200] from kafka import KafkaConsumer

[0201] # Connect to Kafka

[0202] consumer=KafkaConsumer('clearance_request_topic',bootstrap_servers='localhost:9092')

[0203] For message in consumer:

[0204] # Message parsing

[0205] request=eval(message.value.decode('utf-8'))

[0206] account_id=request.get('account_id')

[0207] ifrequest.get('request_type')=='clearance':

[0208] # Handling zeroing requests

[0209] release_resources(account_id)

[0210] 1.2 Release of resource quota

[0211] In the resource management module, locate the resource quota object corresponding to the target content delivery account. Release all available resource quotas for that account and transfer resource usage permissions to the resource management module of that account.

[0212] Here is a Python code example:

[0213]

[0214] 2. Generate an alarm notification if the zeroing operation fails.

[0215] Use a logging system (such as the ELK stack: Elasticsearch, Logstash, Kibana) to record detailed information about the zeroing operation, including the operation time, account ID, and operation result.

[0216] Use a message queue (such as RabbitMQ) to send alert notifications. Alert notifications can be sent via email, SMS, or internal system messages.

[0217] Use a rule engine (such as Drools) to define alert rules and generate different levels of alert notifications based on different failure reasons and account attributes.

[0218] 2.1 Zeroing operation failed.

[0219] During the zeroing operation, any exceptions or errors that may occur are captured. Examples include database operation failures and network connection interruptions. When a zeroing operation failure is detected, detailed error information is logged to the logging system.

[0220] The following is a code example that implements a zeroing operation and detects failure:

[0221]

[0222]

[0223] 2.2 Generate alarm notifications

[0224] Based on the recorded error information and predefined alarm rules, a corresponding alarm notification is generated. The alarm notification includes information such as the account ID, error reason, and operation time. The alarm notification is then sent to a message queue for processing by the alarm notification system.

[0225] The following is a Python code example that generates an alert notification and sends it to RabbitMQ:

[0226]

[0227] Based on the above embodiments, this application also provides a content delivery account management system, which includes: a monitoring terminal, a resource execution terminal, and an interaction terminal, wherein:

[0228] The interactive end is used to generate a request to clear the content delivery account and send it to the monitoring end;

[0229] The monitoring terminal is used to perform the following steps:

[0230] Receive a request from the interactive client to clear the content delivery account, and parse the request to determine the target content delivery account.

[0231] The resource execution end is triggered to attempt to clear the target content delivery account and a new clearing work order is created locally;

[0232] When the received clearing operation result indicates that the target content delivery account is in an unclearable state, the target content delivery account is added to the monitoring pool configured by itself, and the clearing work order amount is set in the new clearing work order.

[0233] The status of the target content delivery account in the monitoring pool is scanned periodically. When the target content delivery account changes from a non-clearable status to a clearable status, a clearing request is sent to the resource execution end.

[0234] After receiving the successful clearing operation result from the resource execution terminal, update the clearing work order amount in the clearing work order;

[0235] The resource execution terminal is used to perform the following steps:

[0236] Based on the trigger from the monitoring terminal, perform an attempt to clear the target content delivery account and report the result back to the monitoring terminal.

[0237] Receive the clearing request from the monitoring terminal, perform the clearing operation on the target content delivery account, and feed back the operation result to the monitoring terminal.

[0238] Based on the above embodiments, this application also provides a content delivery system, which includes a backend system, a media terminal, and an agency terminal, and the collaborative workflow of each terminal is as follows:

[0239] The agent generates a request to clear the content delivery account and sends it to the backend system.

[0240] After receiving the request, the backend system parses it to determine the target content delivery account, and then triggers the media side to perform an attempt to clear the target content delivery account, while simultaneously creating a clearing work order locally;

[0241] The media platform, based on the triggering instructions from the backend system, attempts to clear the target content distribution account and then sends the result back to the backend system.

[0242] If the feedback indicates that the target content delivery account is in a non-resetable state, the backend system will add the target content delivery account to its own configured monitoring pool and set the reset work order amount in the new reset work order; then, the backend system will periodically scan the status of the target content delivery account in the monitoring pool.

[0243] When the target content delivery account changes from a non-resettable state to a resettable state, the backend system sends a reset request to the media platform.

[0244] The media platform receives a clearing request from the backend system, performs a clearing operation on the target content distribution account, and feeds back the operation result to the backend system.

[0245] After receiving feedback from the media that the clearing operation was successful, the backend system updates the clearing work order amount in the clearing work order.

[0246] The above-mentioned method for clearing content delivery accounts can be applied to various scenarios. The following examples illustrate this in three typical areas: internet advertising, online education resource distribution, and streaming media content delivery. For instance, in the online education resource distribution scenario, online education platforms have a large number of course delivery accounts used to push courses to different user groups. For example, the platform sets up course delivery accounts for K-12 students, university students, and working professionals, allocating corresponding resource usage quotas. For instance, a K-12 account can push 1000 course advertisements, a university student account can push 800, and a working professional account can push 600. When the semester ends or courses are updated, the relevant accounts need to be cleared to reallocate resources. However, some accounts may not be cleared due to reasons such as incomplete user learning progress tracking or failed course copyright review. After this solution is implemented, the agent sends a clearing request, and the backend system coordinates the media side's operation. If a K-12 account cannot be cleared due to individual students' learning progress not being synchronized, it is added to the monitoring pool and scanned with high priority and high frequency, considering the urgent need for resources for that account as the new semester approaches. Once the progress is synchronized, the account will be immediately cleared, releasing 1000 push notifications for the promotion of courses in the new semester. If the clearing fails, and it is due to a malfunction in the copyright review system, an alert will be sent to the platform's technical staff, including account information and fault details, to help quickly resolve the issue.

[0247] In addition, in streaming content delivery scenarios, streaming platforms set up content delivery accounts for different types of content (movies, TV series, variety shows, etc.) and allocate traffic resource quotas. For example, a popular movie delivery account is allocated 500GB of traffic, and a TV series delivery account is allocated 300GB of traffic. When a movie ends its theatrical run or a TV series finishes airing, the account needs to be cleared to release traffic resources. However, there may be situations where clearing is not possible due to incomplete copyright renewal or delays in playback data statistics. After the agent initiates a clearing request, the backend system and the media end work together. If a movie delivery account cannot be cleared due to copyright renewal issues, it is added to the monitoring pool. Based on priority and scanning frequency rules, considering the high traffic demand from subsequent new movie releases, this account is given high priority and scanned frequently. When the copyright renewal is completed, the account is cleared promptly, releasing 500GB of traffic for new movie promotion. If clearing fails, such as due to abnormal data statistics interface, an alarm notification sends detailed information to the platform operations team for targeted handling, ensuring the normal operation of the content delivery business.

[0248] The above embodiments are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the patent protection scope of the embodiments of the present invention should be defined by the claims. The systems, devices, modules, or units described in the above embodiments are specifically implemented by computer chips or entities, or by products with certain functions.

Claims

1. A content delivery account zeroing processing method, characterized by, The method comprises the following steps: receiving a content delivery account zeroing request and analyzing the content delivery account zeroing request to determine a target content delivery account; performing a zeroing operation on the target content delivery account and adding a zeroing work order; in response to the result of the zeroing operation indicating that the target content delivery account is in a non-zeroable state, adding it to a monitoring pool and setting a zeroing work order amount in the added zeroing work order, wherein the zeroing work order amount of the account is predicted using a machine learning algorithm according to the current balance and historical zeroing records of the account; performing a periodic scan on the state of the target content delivery account in the monitoring pool, and in response to the target content delivery account changing from a non-zeroable state to a zeroable state, initiating a zeroing request and performing a zeroing operation on the target content delivery account; in response to the successful execution of the zeroing operation, updating the zeroing work order amount in the zeroing work order, wherein the zeroing work order amount field in the zeroing work order is updated according to the actual zeroing amount; wherein, further comprising: assigning a priority order to all target content delivery accounts added to the monitoring pool, so as to scan in the priority order when performing a periodic scan on the state of the target content delivery account in the monitoring pool; wherein, further comprising: configuring a high-to-low scanning frequency according to the priority order from high to low, so that the higher the priority order of the target content delivery account, the higher the scanning frequency for it when performing a periodic scan on the state of the target content delivery account in the monitoring pool; wherein, further comprising: in response to the zeroing request, releasing the available resource quota of the target content delivery account, so that the use permission of the available resource quota is transferred to the resource management module of the target content delivery account.

2. The content delivery account zeroing method of claim 1, wherein, Further comprising: in response to the result of the zeroing operation indicating that the target content delivery account is in a zeroable state, adding a zeroing work order, initiating a zeroing request and performing a zeroing operation on the target content delivery account, and in response to the successful execution of the zeroing operation, writing the actual zeroing amount in the zeroing work order.

3. The content delivery account zeroing method of claim 1, wherein, Assigning a priority order to all target content delivery accounts added to the monitoring pool, comprising: sorting according to the time sequence of the target content delivery accounts added to the monitoring pool, and assigning a priority order based on the sorted queue.

4. The content delivery account zeroing method of claim 1, wherein, Further comprising: in response to the failure of the zeroing operation, generating an alarm notification and sending it.

5. A content delivery account management system, applying the content delivery account zeroing processing method of any one of claims 1-4, characterized in that, Comprising: a monitoring end, a resource execution end and an interaction end, wherein: the interaction end is used to generate a content delivery account zeroing request and send it to the monitoring end; the monitoring end is used to perform the following steps: receiving the content delivery account zeroing request from the interaction end, analyzing the content delivery account zeroing request to determine a target content delivery account; triggering the resource execution end to perform a zeroing operation on the target content delivery account and adding a zeroing work order locally; when the received zeroing operation result indicates that the target content delivery account is in a non-zeroable state, adding the target content delivery account to the monitoring pool configured by itself and setting a zeroing work order amount in the added zeroing work order; periodically scanning the state of the target content delivery account in the monitoring pool, and when the target content delivery account changes from a non-zeroable state to a zeroable state, initiating a zeroing request to the resource execution end; After receiving the zero-clearing operation success result fed back by the resource execution end, the zero-clearing work order amount in the zero-clearing work order is updated; The resource execution end is used to execute the following steps: According to the trigger of the monitoring end, the target content delivery account is executed to perform a zero-clearing operation, and the operation result is fed back to the monitoring end; The monitoring end receives the zero-clearing request, and the target content delivery account is executed to perform a zero-clearing operation, and the operation result is fed back to the monitoring end.

6. A content delivery system, applying the content delivery account zeroing processing method of any one of claims 1-4, characterized in that, The back-end system, the media end and the agent end are included, and the collaborative work flow is as follows: The agent end generates a content delivery account zero-clearing request and sends it to the back-end system; After receiving the request, the back-end system analyzes to determine the target content delivery account, and then triggers the media end to perform a zero-clearing operation on the target content delivery account, and adds a zero-clearing work order locally; The media end performs a zero-clearing operation on the target content delivery account according to the trigger instruction of the back-end system, and feeds back the operation result to the back-end system; If the feedback result shows that the target content delivery account is in a non-zero-clearing state, the back-end system adds the target content delivery account to the monitoring pool configured by itself, and sets the zero-clearing work order amount in the added zero-clearing work order; then, the back-end system scans the state of the target content delivery account in the monitoring pool regularly; When the target content delivery account changes from a non-zero-clearing state to a zero-clearing state, the back-end system initiates a zero-clearing request to the media end; The media end receives the zero-clearing request from the back-end system, performs a zero-clearing operation on the target content delivery account, and feeds back the operation result to the back-end system; After receiving the zero-clearing operation success result fed back by the media end, the back-end system updates the zero-clearing work order amount in the zero-clearing work order.

Images