Synchronization method for interactive data of social media account works

Abstract

The invention discloses a synchronization method for interactive data of social media account works, which relates to the technical field of data synchronization and comprises the following steps: receiving account authorization certificates of a plurality of social media platforms and establishing a data connection channel; creating a global unique entity identifier, and constructing a mapping relation table; acquiring work data published by the account on each platform, and extracting work feature information; constructing a cross-platform content entity recognition model, and generating a content entity identifier; establishing an association table of the content entity identifier and each platform work identifier; obtaining interactive data regularly, and normalizing, merging and storing the interactive data of the same content entity; calculating an interactive data aggregation index of the content entity, and generating a content propagation effect evaluation report; according to the method, automatic association of cross-platform content entities and unified management of interactive data are realized, the problems of low efficiency and high error rate of manual maintenance of the mapping table are solved, and the accuracy and efficiency of multi-platform content propagation effect analysis are improved.

Description

Technical Field

[0001] This invention relates to the field of data synchronization technology, and in particular to a method for synchronizing interactive data of works on social media accounts. Background Technology

[0002] In the field of digital marketing, businesses typically need to simultaneously publish content on multiple social media platforms to achieve widespread dissemination. However, the independent storage of interaction data across each platform makes it impossible to comprehensively assess the overall impact of the same content across the entire internet.

[0003] Existing technologies typically employ a central server that periodically polls the application programming interfaces (APIs) of various platforms to obtain interaction data for the same work across different platforms, and then aggregates and summarizes this data. However, because the content identifiers on each platform are independent, and the same work may be published at different times on each platform, the system cannot automatically identify data belonging to the same work across different platforms. Therefore, it is necessary to manually maintain a content identifier mapping table in advance to achieve correlation and matching.

[0004] This method is not only inefficient and difficult to scale, but it is also prone to data mis-association or failure due to manual input errors or omissions, which distorts the subsequent data aggregation results and makes it impossible to provide accurate analysis of the spread across the entire network. Summary of the Invention

[0005] The purpose of this invention is to provide a method for synchronizing interactive data of works on social media accounts, thereby solving the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for synchronizing interactive data of works on social media accounts, comprising the following steps: Receive account authorization credentials from multiple social media platforms and establish data connection channels with each platform; Create a globally unique entity identifier for each account and construct a mapping table between social media accounts and global identifiers; Collect data on works published by accounts on various platforms and extract feature information of the works; Based on the extracted feature information of the works, a cross-platform content entity recognition model is constructed to generate content entity identifiers for each work. Establish a table linking content entity identifiers with work identifiers on various platforms to record the identifier correspondence of the same content entity on different platforms; Regularly collect interaction data for the works from various social media platforms; Based on the association table, query the content entity identifier to which the interactive data belongs, and normalize the interactive data from different platforms but belonging to the same content entity. The normalized interaction data is merged and stored in chronological order; Calculate the interaction data aggregation metrics of content entities and generate a content dissemination effect evaluation report.

[0007] Furthermore, the authorization credentials include, but are not limited to, OAuth tokens, API keys, and application keys.

[0008] Furthermore, the mapping table adopts a key-value pair storage structure, where the key is a combination of the platform identifier and the platform internal account ID, and the value is the corresponding global identifier.

[0009] Furthermore, the collection of work data published by the account on various platforms includes obtaining work data published by the account from various social media platforms through established data connection channels and in accordance with a preset collection strategy; The data collection strategies include two modes: initial full data collection and incremental timed data collection. The first full collection retrieved all works published by the account throughout its history. Incremental acquisition retrieves only newly published works based on timestamps.

[0010] Furthermore, the work's feature information includes: work title, content summary, publication time, and multimedia fingerprint.

[0011] Furthermore, the methods for extracting the work title, content summary, publication time, and multimedia fingerprint are as follows: Title of the work: Extract text content and perform word segmentation, stop word filtering, and semantic standardization; Content Summary: This function automatically summarizes long text content, using the TextRank algorithm to extract key sentences. Release time: uniformly converted to UTC timestamp format; Multimedia fingerprint: Generates a hash fingerprint of multimedia content, including: Image content: Fingerprint values ​​are generated using a perceptual hashing algorithm; Video content: After extracting keyframes, an image fingerprinting algorithm is applied, and combined with audio fingerprints to form a composite fingerprint; Audio content: Spectral analysis is used to extract audio feature vectors.

[0012] Furthermore, the cross-platform content entity recognition model combines text similarity, multimedia fingerprint similarity, and temporal correlation for comprehensive judgment. The calculation process includes: Text similarity calculation: The text is first segmented into words and converted into TF-IDF vectors. Then, the cosine similarity between the vectors is calculated using the following formula: ; in, and These represent the TF-IDF vectors of the two texts, and Let them represent the nth vector in vectors A and B, respectively. The value of each element, Represents the vector dimension. Indicates the text similarity between text A and text B; Multimedia fingerprint similarity calculation: Calculate the Hamming distance for image or video fingerprints, and then convert the distance into a similarity score. ; in, This represents the Hamming distance between two fingerprints. Indicates the total number of digits in the fingerprint. This represents the multimedia fingerprint similarity between text A and text B; The formula for calculating time correlation is: ; in, and These represent the publication timestamps of works A and B, respectively. It is the time decay factor. This indicates the temporal correlation between text A and text B; The formula for calculating the overall similarity is: ; in, , and These are the first, second, and third comprehensive similarity weight coefficients, respectively. This indicates the overall similarity between text A and text B.

[0013] Furthermore, the data in the associated table includes: Content entity identifier is mapped to entity_id; platform identifier is mapped to platform_id; platform work identifier is mapped to content_id; similarity score is mapped to similarity_score; association time is mapped to link_time; association status is mapped to status.

[0014] Furthermore, the normalization process also includes data anomaly detection and correction, implemented as follows: Outliers are identified using the Moving Median Absolute Deviation (MAD) method, calculated as follows: ; Where X is a set of interactive data values. Let i be the i-th data point in set X. The median of the data value X. The absolute deviation of the median is used. When a data point deviates from the median by more than a preset threshold, it is judged as an outlier and corrected by time series interpolation.

[0015] Furthermore, the following technical measures are implemented during the acquisition of the interactive data: Traffic control: Adhering to the API access frequency limits of each platform, request rate limiting is implemented using the token bucket algorithm. The token bucket algorithm is implemented in the following ways: ; in, For the capacity of the bucket, For token generation rate, The time of the last request. This represents the number of tokens remaining from the previous round. This represents the current number of available tokens. Retry on Failure: Implements the exponential backoff algorithm to handle request failures. The retry interval is calculated according to the following formula: ; in, This represents the number of retries. Based on waiting time, Maximum waiting time A random number between 0 and 1, used to prevent synchronization requests. The retry interval is the interval before n retries.

[0016] Beneficial effects of this invention: By building a cross-platform content entity recognition model, works belonging to the same content entity on different platforms can be automatically identified without the need for manual maintenance of the content identifier mapping table, which greatly improves efficiency and reduces the human error rate. A normalization processing mechanism is adopted to unify the interactive data standards of different platforms, and anomaly detection algorithms (such as the moving median absolute deviation method) are combined to ensure data quality, making the cross-platform data aggregation results more accurate and reliable. Interactive data is acquired through a dual-mode approach of scheduled retrieval and event subscription, and combined with a time-series database storage structure to achieve real-time monitoring and analysis of content dissemination effects; Construct a two-tier evaluation system that includes basic and composite indicators. Through indicators such as platform collaboration coefficient, comprehensively quantify the dissemination effect of content across the entire network and provide data support for enterprise decision-making. By employing a connection pool management mechanism and a token bucket traffic control algorithm, the system can effectively comply with the API access restrictions of various platforms while ensuring the real-time acquisition of data, thereby improving system stability and resource utilization.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a method for synchronizing interactive data of works on social media accounts according to the present invention. Detailed Implementation

[0019] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] Please see Figure 1 A method for synchronizing interaction data of works on a social media account, comprising the following steps: Step 100: Receive account authorization credentials from multiple social media platforms and establish data connection channels with each platform; In embodiments of the present invention, the authorization management module first receives account authorization credentials from multiple social media platforms. These authorization credentials include, but are not limited to, OAuth tokens, API keys, and application keys, and are used to verify the system's permission to access user social media account data. The authorization process employs the standard OAuth 2.0 protocol, obtaining access tokens and refresh tokens through an authorization code process. The access token is used for short-term access, and the refresh token is used to obtain a new access token after the previous one expires.

[0021] After authorization, the system establishes data connection channels with each platform. These channels employ Secure Sockets Layer (SSL) encryption to ensure data transmission security. Connection configurations for each platform are stored separately, including parameters such as connection address, authentication method, request frequency limits, and retry policies. A connection pool management mechanism dynamically adjusts the number of connections to ensure the stability and efficiency of the data channels.

[0022] It should be noted that the system structure for synchronizing interactive data of social media account works (hereinafter referred to as the system) of the present invention includes an authorization management module, an account mapping module, a content acquisition module, an entity recognition module, an association management module, a data synchronization module, a normalization processing module, a data storage module, and an analysis and evaluation module.

[0023] Step 200: Create a globally unique entity identifier for each account and construct a mapping table between social media accounts and global identifiers; To achieve unified management of cross-platform data, this invention creates a globally unique entity identifier (GUID) for each social media account in the account mapping module. The GUID adopts a 128-bit UUID (Universally Unique Identifier) ​​format and is generated by combining a timestamp, a random number, and a system identifier to ensure global uniqueness.

[0024] The account mapping module constructs a mapping table between social media accounts and global identifiers. This table uses a key-value pair storage structure, where the key is a combination of the platform identifier and the platform's internal account ID, and the value is the corresponding global identifier. The mapping relationship is as follows: The platform identifier is mapped to platform_id; the platform account ID is mapped to account_id; the global identifier is mapped to guid; the account information is mapped to account_info; the creation time is mapped to create_time; and the update time is mapped to update_time. The `account_info` table contains a JSON structure with basic information such as account name, type, and number of followers. This mapping table is stored in a distributed database, supporting high-concurrency read / write operations and fast queries.

[0025] Step 300: Collect data on works published by the account on various platforms and extract feature information of the works, including title, content summary, publication time and multimedia fingerprint; The content acquisition module retrieves content data from various social media platforms through established data connection channels, following a preset acquisition strategy. The acquisition strategy includes two modes: initial full-volume acquisition and incremental timed acquisition. Initial full-volume acquisition retrieves all historically published content from the account, while incremental acquisition retrieves only newly published content based on timestamps.

[0026] The original work data structure varies across platforms. An adapter pattern is used to design a unified data conversion interface, converting data formats from different platforms into a standardized internal structure. The standardized work data includes three parts: basic information, content information, and platform metadata.

[0027] The following features were extracted from the standardized work data: Title of the work: Extracting text content and performing word segmentation, stop word filtering, and semantic standardization.

[0028] Content Summary: This section describes an automatic summary of long text content, using the TextRank algorithm to extract key sentences. The TextRank algorithm calculates sentence importance scores as follows: ; in, Sentence Importance score This is the damping coefficient (default value is 0.85). Indicates pointing to a sentence A collection of sentences, Sentence The set of sentences it points to. Sentence to sentence The transfer weights.

[0029] Release time: Converted to UTC timestamp format.

[0030] Multimedia fingerprinting: Generating hash fingerprints for multimedia content such as images and videos, including: Image content: A 64-bit fingerprint value is generated using the Perceptual Hash (pHash) algorithm; Video content: After extracting keyframes, an image fingerprinting algorithm is applied, and combined with audio fingerprints to form a composite fingerprint; Audio content: Spectral analysis is applied to extract audio feature vectors; The image-aware hashing algorithm process is as follows: scale the image to 32×32, convert it to grayscale, apply discrete cosine transform (DCT), extract low-frequency coefficients, calculate the mean, and compare each coefficient with the mean to generate a bit string.

[0031] Step 400: Based on the extracted work feature information, construct a cross-platform content entity recognition model and generate a content entity identifier for each work; Based on the extracted feature information of the works, this invention constructs a cross-platform content entity recognition model to determine whether works published on different platforms belong to the same content entity. This model employs a multi-feature fusion method, combining text similarity, multimedia fingerprint similarity, and temporal correlation for a comprehensive judgment.

[0032] The core computational process of the content entity recognition model includes: Text similarity calculation: Cosine similarity is calculated for both the title and content summary. The text is first segmented and converted into TF-IDF (Term Frequency-Inverse Document Frequency) vectors, and then the cosine similarity between the vectors is calculated using the following formula: ; in, and These represent the TF-IDF vectors of the two texts, and Let them represent the nth vector in vectors A and B, respectively. The value of each element, Represents the vector dimension. This indicates the text similarity between text A and text B.

[0033] Multimedia fingerprint similarity calculation: Calculate the Hamming distance (number of distinct bits) for image or video fingerprints, and then convert the distance into a similarity score. ; in, This represents the Hamming distance between two fingerprints. Indicates the total number of digits in the fingerprint. This represents the multimedia fingerprint similarity between text A and text B.

[0034] Time relevance calculation: Taking into account the proximity of the release times of the works, the calculation formula is as follows: ; in, and These represent the publication timestamps of works A and B, respectively. This is the time decay factor, with a default value set to... , This indicates the temporal correlation between text A and text B.

[0035] Multi-feature fusion calculation of comprehensive similarity: ; in, , and These are the first, second, and third comprehensive similarity weight coefficients, which satisfy... , This indicates the overall similarity between text A and text B.

[0036] When the overall similarity When the similarity exceeds a preset threshold (default value is 0.85), the two works are determined to belong to the same content entity. Through iterative clustering, all related works are grouped into the same content entity, and a unique content entity identifier is generated for each content entity.

[0037] Step 500: Establish a table linking content entity identifiers with work identifiers on various platforms, recording the identifier correspondence of the same content entity on different platforms; After identifying the content entity, the association management module establishes an association table between the content entity identifier and the work identifiers of each platform, recording the correspondence between the identifiers of the same content entity on different platforms. The association table adopts a many-to-one data structure, where work identifiers from multiple platforms are mapped to the same content entity identifier.

[0038] The data in the related table includes: Content entity identifier is mapped to entity_id; platform identifier is mapped to platform_id; platform work identifier is mapped to content_id; similarity score is mapped to similarity_score; association time is mapped to link_time; status is mapped to status; Among them, entity_id adopts UUID format to ensure global uniqueness; platform_id identifies the social media platform; content_id is a unique identifier for works within the platform; similarity_score records the similarity score between the work and the entity; link_time records the time when the association was established; status indicates the association status (valid association, invalid association).

[0039] The join tables employ a distributed storage structure and use composite indexes to improve query efficiency. A version control mechanism is implemented for the join tables to record the change history of relationships, supporting backtracking and auditing of relationships.

[0040] Step 600: Regularly obtain interaction data for the work from various social media platforms; The data synchronization module periodically retrieves interaction data from various social media platforms, including views, likes, comments, shares, and their temporal changes. Two strategies are employed to acquire this interaction data: Active fetch: Actively call the APIs of various platforms to retrieve the latest interactive data at preset time intervals (default value is 15 minutes).

[0041] Event Subscription: For platforms that support webhook or real-time push functionality, the system registers a data change event listener to receive real-time updates pushed by the platform when interactive data changes.

[0042] The following technical measures are implemented during the acquisition of interactive data: Traffic control: Adhering to the API access frequency limits of each platform, request rate limiting is implemented using the token bucket algorithm. The token bucket algorithm is implemented in the following way: in, For the capacity of the bucket, For token generation rate, The time of the last request. This represents the number of tokens remaining from the previous round. This represents the number of tokens currently available.

[0043] Retry on Failure: Implements the exponential backoff algorithm to handle request failures. The retry interval is calculated according to the following formula: in, This represents the number of retries. Based on waiting time, Maximum waiting time A random number between 0 and 1, used to prevent synchronization requests. The retry interval is the interval before n retries.

[0044] Data consistency verification: Perform consistency verification on the acquired interactive data, detect outliers and mutation points, and ensure data quality.

[0045] Step 700: Query the content entity identifier of the interactive data according to the association table, and normalize the interactive data from different platforms but belonging to the same content entity. Because different social media platforms use different interaction data formats, measurement standards, and statistical methods, the normalization module is responsible for standardizing interaction data from different platforms that belong to the same content entity. Normalization includes the following key steps: Interaction Type Mapping: Establishing a standard mapping relationship between interaction types across different platforms, mapping the platform-specific interaction types to a unified interaction classification system. Standard interaction types include: view, like, comment, share, favorite, download, etc.

[0046] Unified Data Units: To address the differences in counting rules across different platforms, a unit conversion function is applied to convert the data into a unified metric. In one embodiment of this invention, some platforms count a video view as valid only if it lasts more than 30 seconds, while others may only require 3 seconds. Conversion rules are defined by configuring a matrix to ensure data comparability.

[0047] Data normalization calculation: The raw interaction data is standardized using the following normalization formula: ; in, These are the original interaction data values. and These represent the historical minimum and maximum values ​​for this type of interaction data. This method maps interaction data of different magnitudes uniformly to the [0,1] interval. These are the normalized interaction data values.

[0048] Data anomaly detection and correction: Outliers are identified using the Moving Median Absolute Deviation (MAD) method, calculated as follows: ; Where X is a set of interactive data values. Let i be the i-th data point in set X. The median of the data value X. The absolute deviation of the median is used. When a data point deviates from the median by more than a preset threshold, it is judged as an outlier and corrected by time series interpolation.

[0049] Step 800: Merge and store the normalized interaction data in chronological order; The normalized interactive data is merged and stored through the data storage module. The merged storage employs the following technical solution: Data storage structure: A time series database (TSDB) is used to store the time series of interaction data. The primary key is designed as (entity_id, interaction_type, timestamp), which supports efficient time range query and aggregation statistics.

[0050] Data merging strategy: Interaction data of the same content entity across different platforms is merged chronologically. Three merging strategies are implemented for data with the same timestamp: Weighted summation: Considering a linear combination of platform weights; Maximum value selection: Select the maximum value from each platform; Adaptive fusion: Dynamically selects merging strategies based on the distribution characteristics of historical data.

[0051] Multi-tiered storage: Implements a storage architecture that separates hot and cold data, storing the most recent 30 days of hot data in an in-memory database and archiving historical data to a distributed file system.

[0052] Step 900: Calculate the interaction data aggregation index of the content entity and generate a content dissemination effect evaluation report.

[0053] The analysis and evaluation module calculates aggregated interaction data metrics for content entities based on synchronized interaction data, and generates a content dissemination effectiveness evaluation report. The specific implementation is as follows: Interactive data aggregation index calculation: A two-level index system is defined, including basic indicators and composite indicators.

[0054] Basic metrics: These are derived directly from the raw interaction data and mainly include: Total Pageviews: The sum of pageviews across all platforms; Total Interactions: The total number of all types of interactions; Platform Coverage: The percentage of platforms where this content was published out of the total number of accessed platforms; Peak Interactions: The maximum number of interactions per unit of time; First Interaction Time: The time interval between the first interaction received after the content was published.

[0055] Composite indicators: Calculated by combining basic indicators, including: Interaction Rate: Calculated by dividing total interactions by total views; Average Interaction Response Time: Average response time for comments or replies; Platform Collaboration Coefficient: Measures the effectiveness of content dissemination across multiple platforms, calculated as follows: ; in, Indicates platform and platform The relevance of interactive data on the surface The total number of platforms, This represents the platform collaboration coefficient.

[0056] The analysis and evaluation module generates a content dissemination effectiveness evaluation report based on the calculated aggregate indicators.

[0057] Through the aforementioned technical means, the automated collection, association, and synchronous analysis of interactive data from social media account works have been achieved, solving the technical problems of cross-platform content entity association and interactive data integration, and improving the accuracy and efficiency of multi-platform content dissemination effect analysis.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for synchronizing social media account work interaction data, the method comprising: The method comprises the following steps: Receiving account authorization credentials of multiple social media platforms and establishing data connection channels with each platform; Creating a globally unique entity identifier for each account and building a mapping table of social media accounts and global identifiers; Collecting work data published by the account on each platform and extracting work feature information; Based on the extracted work feature information, a cross-platform content entity identification model is built to generate a content entity identifier for each work; An association table of content entity identifiers and work identifiers on each platform is established to record the identifier correspondence of the same content entity on different platforms; Obtaining interactive data of works from each social media platform at regular intervals; According to the association table, query the content entity identifier to which the interactive data belongs, and normalize the interactive data from different platforms but belonging to the same content entity; Merging and storing the normalized interactive data in chronological order; Calculate the interactive data aggregation index of the content entity, and generate a content propagation effect evaluation report. 2.The method of claim 1, wherein: The authorization credentials include OAuth tokens, API keys, and application keys. 3.The method of claim 2, wherein: The mapping table uses a key-value storage structure, with the key being a combination of platform identifier and platform internal account ID, and the value being the corresponding global identifier.

4. The method of claim 3, wherein: The collection of work data published by the account on each platform includes obtaining work data published by the account from each social media platform according to the preset collection strategy through the established data connection channel; The collection strategy includes two modes: initial full collection and incremental regular collection: Initial full collection obtains all works published by the account in history; Incremental collection only obtains newly published works according to the timestamp.

5. The method of claim 4, wherein: The work feature information includes: work title, content abstract, publication time and multimedia fingerprint.

6. The method of claim 5, wherein: The extraction methods of work title, content abstract, publication time and multimedia fingerprint are as follows: Work title: extract text content and perform word segmentation, stop word filtering and semantic standardization processing; Content abstract: automatically abstract long text content and extract key sentences using TextRank algorithm; Publication time: convert to UTC timestamp format; Multimedia fingerprint: generate hash fingerprint for multimedia content, including: Image content: generate fingerprint value using perceptual hashing algorithm; Video content: extract key frames and apply image fingerprint algorithm, and combine audio fingerprint to form composite fingerprint; Audio content: apply frequency spectrum analysis to extract audio feature vector.

7. The method of Claim 6, wherein: The cross-platform content entity identification model combines text similarity, multimedia fingerprint similarity and time correlation for comprehensive judgment, and the calculation process includes: Text similarity calculation: text is first processed by word segmentation and converted into TF-IDF vector, then the cosine similarity between vectors is calculated, and the calculation formula is: ； where, and denote the TF-IDF vectors of two texts, and denote the values of the th element in the A vector and the B vector, respectively, denotes the vector dimension, denotes the text similarity of the A text and the B text; Multimedia fingerprint similarity calculation: calculate the Hamming distance of image or video fingerprint, then convert the distance to similarity: ； wherein, denotes the Hamming distance between two fingerprints, denotes the total number of bits of a fingerprint, denotes the multimedia fingerprint similarity between the A-text and the B-text; The calculation formula of time correlation is: ； wherein, and denote the publication timestamps of the A and B works, respectively, is a time decay factor, denotes the temporal correlation of the A and B texts; The calculation formula of comprehensive similarity is: ； wherein, , and are the first, second, third comprehensive similarity weight coefficients, respectively, denotes the comprehensive similarity of the A text and the B text.

8. The method of Claim 7, wherein, The data of the association table includes: The content entity identifier is mapped to entity_id; the platform identifier is mapped to platform_id; the platform work identifier is mapped to content_id; the similarity score is mapped to similarity_score; the association time is mapped to link_time; and the association state is mapped to status. 9.The method of claim 8, wherein: The normalization processing further includes data anomaly detection and correction, and the implementation method is as follows: An abnormal value is identified using a moving median absolute deviation (MAD) method, and the calculation formula is as follows: ； wherein X is a set of interaction data values, is the i-th data point in the set X, is the median of the data values X, is the median absolute deviation, and when a data point deviates from the median by more than a pre-set threshold, it is determined to be an outlier and is corrected by a time series interpolation method.

10. The method of Claim 9, wherein, In the process of obtaining the interaction data, the following contents are included: Flow control: follow the API access frequency limit of each platform, implement token bucket algorithm for request speed limit, and the expression of the token bucket algorithm is as follows: ； wherein, is the bucket capacity, is the token generation rate, is the last request time, is the last remaining tokens, is the current available tokens; Failure retry: implement an exponential backoff algorithm to handle request failure, and the retry interval is calculated according to the following formula: ； wherein, is the number of retries, is the base wait time, is the maximum wait time, is a random number between 0 and 1 to prevent request synchronization, is the retry interval before n retries.

Images