Method and system for PR full-process arrangement and data closed-loop engine

Abstract

The invention discloses a method and a system for PR full-flow arrangement and a data closed-loop engine, and relates to the field of intelligent data processing. According to the PR full-process arrangement and data closed-loop engine method, a PR process directed graph is constructed by obtaining a delivery task data set of a plurality of delivery objects; collecting multi-source event data of a putting process, and driving a state machine to perform state transition; in the state transition process, collecting an evidence set of each delivery object, and extracting the closed-loop confidence of the corresponding delivery object; the method comprises the following steps: when each delivery object arrives at a preset task node, identifying a to-be-executed task node set which comprises a plurality of to-be-executed task nodes, analyzing a corresponding task priority score value, and performing dynamic scheduling based on the task priority score value. Therefore, full-link standardized management of PR delivery from process arrangement to acceptance and archiving is realized, the delivery effect is effectively guaranteed, and the PR cooperation efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent data processing, specifically to a method and system for a full-process orchestration and data closed-loop engine for public relations. Background Technology

[0002] In the field of public relations and content marketing, in order to achieve goals such as brand exposure, product promotion or reputation building, executing a PR campaign is usually a complex collaborative process involving multiple steps, multiple roles and multiple platforms. A typical PR campaign usually begins with event planning and influencer selection, and then goes through multiple stages such as establishing connections and communication, sending samples for shooting, content creation and review, multi-platform release, data collection, and finally ends with effect acceptance and review.

[0003] Currently, the mainstream solution for managing such processes in the industry is essentially a loose combination model with human intervention as the core and general tools as an aid. Specifically, the process is usually statically defined and recorded in the form of documents or tables. For example, the project leader lists the task list, responsible persons and deadlines in Excel or online collaborative documents, and makes reminders and progress through instant messaging tools or emails. The dependencies between the task nodes rely entirely on the personal memory or verbal communication of the participants, lacking the ability to automatically arrange and enforce the process at the system level.

[0004] The limitations of existing technologies include at least the following issues: disorganized process progression, unclear task node dependencies, and a tendency for execution to skip steps or become blocked; ineffective integration of multi-source event data, resulting in fragmented and broken evidence chains, making it difficult to quantify the completeness of evidence throughout the entire deployment process; lack of a priority evaluation system for pending tasks, leading to blind resource allocation and neglect of high-value, high-risk tasks; lack of anomaly detection for key deployment indicators, resulting in delayed risk warnings; and reliance on subjective judgment for deployment effectiveness acceptance, failing to coordinate KPI achievement with closed-loop confidence verification, leading to low credibility of acceptance results and potentially causing inefficient PR deployment and resource waste. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for a full-process orchestration and data closed-loop engine for public relations (PR), which solves the problems of lack of end-to-end collaboration in existing technologies, which can easily lead to process chaos, data fragmentation, and resource inefficiency.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for a PR full-process orchestration and data closed-loop engine, comprising the following steps: acquiring a dataset of delivery tasks for several delivery targets, and constructing a directed graph of the PR process for the corresponding delivery targets, which includes several task nodes and corresponding state machines; collecting multi-source event data of the delivery process to drive the state machine to perform state transitions; during the state transition process, collecting the evidence set of each delivery target and extracting the closed-loop confidence of the corresponding delivery target; simultaneously identifying the set of task nodes to be executed for each delivery target, including several task nodes to be executed, and analyzing the corresponding task priority score, and performing dynamic scheduling based on the task priority score; and automatically accepting and archiving each delivery target when it reaches a preset task node.

[0007] Furthermore, the deployment task dataset includes several deployment tasks and corresponding task types. The specific steps for constructing the directed graph of the PR process are as follows: Based on the task types of the several deployment tasks, a task node set is generated, which includes several task nodes and corresponding state machines; Based on preset task constraint rules, the task node set is identified and processed to generate a set of directed dependency edges between the task nodes; The task node set and the set of directed dependency edges constitute the directed graph of the PR process.

[0008] Furthermore, the specific steps of the state transition are as follows: the multi-source event data of the deployment process is normalized and stored in the event bus; the event bus is distributed to each task node according to preset rules to drive the corresponding state machine to perform state transition.

[0009] Furthermore, the evidence set includes several types of evidence. The specific steps for extracting the closed-loop confidence of the target are as follows: evaluate and process several types of data for each target to obtain the confidence of the corresponding type of evidence; and fuse the confidence of several types of evidence to obtain the closed-loop confidence of each target.

[0010] Further, the specific steps for identifying the set of task nodes to be executed for each of the aforementioned deployment objects are as follows: read the state machine of each task node and determine whether it meets the preset executable state conditions; mark the task nodes that meet the preset executable state conditions as task nodes to be executed, and perform statistical processing to obtain the set of task nodes to be executed for each of the aforementioned deployment objects.

[0011] Further, the specific steps for analyzing the task priority score of each of the pending task nodes are as follows: Based on the multi-source event data of the deployment process and the evidence set of each of the deployment objects, analyze the score dataset of each of the pending task nodes, including urgency score, impact score, risk score, blockage score, and cost score; based on the score dataset, analyze the task priority score of each of the pending task nodes.

[0012] Furthermore, the dynamic scheduling specifically involves: reading the task priority score values ​​of each of the task nodes to be executed, arranging them in descending order, and generating a task execution sequence; and performing task scheduling processing on each of the deployment objects based on the task execution sequence.

[0013] Furthermore, based on the multi-source event data of the delivery process, key indicator time-series data of each delivery target are extracted; anomaly analysis is performed on the key indicator time-series data based on a preset anomaly detection algorithm, and anomaly scores are calculated; when the anomaly score exceeds a preset warning threshold, warning processing is triggered.

[0014] Furthermore, the specific steps of the automatic acceptance process are as follows: based on the multi-source event data of the delivery process, extract the delivery indicator vectors for each delivery object; based on the delivery indicator vectors, analyze the KPI achievement degree of each delivery object; read the closed-loop confidence of each delivery object, and perform collaborative verification with the KPI achievement degree.

[0015] A PR (Public Relations) end-to-end orchestration and data closure engine system includes: a process orchestration module for acquiring a dataset of campaign tasks for several campaign targets and constructing a directed graph of the PR process for each campaign target, which includes several task nodes and corresponding state machines; an event bus module for collecting multi-source event data of the campaign process and performing normalization processing for storage on the event bus; a state machine advancement module for driving the state machine to perform state transitions based on the event bus; an evidence alignment and closure calculation module for collecting evidence sets for each campaign target during state transitions and extracting the closure confidence of the corresponding campaign target; a task scheduling module for identifying the set of task nodes to be executed for each campaign target, including several task nodes to be executed, and analyzing the corresponding task priority score value, and performing dynamic scheduling based on the task priority score value; an anomaly detection and early warning module for performing detection and early warning processing based on the multi-source event data of the campaign process and combined with a preset anomaly detection algorithm; and a KPI acceptance and archiving module for automatically accepting and archiving each campaign target when it reaches a preset task node.

[0016] The present invention has the following beneficial effects:

[0017] (1) The method of PR full-process orchestration and data closed-loop engine is to construct a PR process directed graph containing task nodes and state machines, clarify the node dependency relationship, combine multi-source event data regularization and event bus distribution, drive the orderly transition of state machines, avoid task skipping or blocking from the root, and ensure the smooth progress of the process. At the same time, evidence sets are collected and closed-loop confidence is extracted in the state transition, realizing data integration and full-process traceability. On this basis, the task priority score value is analyzed and dynamically scheduled through the scoring dataset to ensure that high-value and high-risk tasks are allocated resources first, improve resource utilization efficiency, and automatically accept the results in conjunction with anomaly detection and early warning and KPI achievement and closed-loop confidence. This not only avoids risks in advance, but also replaces subjective judgment to improve the credibility of acceptance, and realizes the standardization of PR delivery process and the refinement of management in all aspects.

[0018] (2) The PR full-process orchestration and data closed-loop engine system, through the directed graph of the PR process constructed by the process orchestration module, clearly sorts out the task nodes and state machine associations, laying the foundation for the orderly advancement of the process; the event bus module realizes the centralized organization and storage of multi-source event data, solving the problem of scattered and difficult-to-reuse data in the deployment; the state machine advancement module relies on the event bus to accurately drive state transitions, ensuring that the process advances according to the established logic and avoiding disorder and chaos; the evidence alignment and closed-loop calculation module synchronously collects evidence and extracts closed-loop confidence, making the entire deployment process traceable and verifiable; the task scheduling module achieves accurate resource matching through scientific analysis of priorities, avoiding resource mismatch and waste; the anomaly detection and early warning module investigates risks in advance; the KPI acceptance and archiving module completes acceptance and archiving through automatic collaborative verification, reducing human intervention errors; all modules work together to achieve standardized management of the entire PR deployment chain from process orchestration to acceptance and archiving, effectively ensuring the deployment effect.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for a PR full-process orchestration and data closed-loop engine according to the present invention.

[0021] Figure 2 This is a schematic diagram of the directed graph of the PR process in the method of PR full-process orchestration and data closed-loop engine of the present invention.

[0022] Figure 3 This is a system block diagram of a PR full-process orchestration and data closed-loop engine of the present invention. Detailed Implementation

[0023] Please see Figure 1This invention provides a technical solution: a method for a PR full-process orchestration and data closed-loop engine, comprising the following steps: acquiring a dataset of deployment tasks for several deployment objects (in the PR deployment process), and constructing a directed graph of the PR process for the corresponding deployment objects, which includes several task nodes and corresponding state machines; collecting multi-source event data of the deployment process, driving the state machine to perform state transitions (and updating the state machine's state labels); during the state transition process, collecting the evidence set of each deployment object, and extracting the closed-loop confidence of the corresponding deployment object; simultaneously (during the state transition process) identifying the set of task nodes to be executed for each deployment object, including several task nodes to be executed, and analyzing the corresponding task priority score value, and performing dynamic scheduling based on the task priority score value; when each deployment object (deployment process) reaches a preset task node (such as a KPI acceptance task node), performing automatic acceptance and archiving processing (after acceptance, archiving and accumulating the process, evidence package, indicators, and anomaly handling results to form reusable deployment assets and review samples).

[0024] It should be noted that, in this implementation example, the target audience refers to the smallest business unit in a PR campaign that can be independently orchestrated, executed, and accepted, such as a combination of <campaign ID, platform, influencer ID, content / note ID>; multi-source event data refers to events generated from emails / IM private messages / form submissions / review results / posting logistics / publishing links / platform data feedback, etc.; the event bus is used for unified access, deduplication, and distribution of events; and the evidence set refers to communication evidence, link evidence, publishing evidence, indicator evidence, and review evidence formed around the target audience.

[0025] Based on multi-source event data from the campaign delivery process, key performance indicator (KPI) time-series data for each campaign target is extracted. Specifically, when an event type belongs to platform data feedback (such as data interface callbacks from social media platforms or e-commerce platforms), the event is parsed. , It typically contains actual values ​​for one or more key business metrics, such as It can represent a specific timestamp The amount of exposure, interactions (likes, favorites, comments, and shares), number of visitors to the store, or transaction amount, etc., based on the information carried by the event. The parsed index values and its corresponding timestamp Add to The time series data sequence of key indicators under the identified target audience, i.e., the time series data of key indicators;

[0026] Anomaly analysis is performed on the time series data of key indicators based on a pre-defined anomaly detection algorithm, and anomaly scores are calculated. Specifically, for the current time... data points The algorithm estimates the expected normal value at a given moment based on historical data within a previous time window (or through exponential weighting), which is usually expressed as a moving average (or exponentially weighted average). Simultaneously calculate its sliding standard deviation. ;

[0027] The algorithm calculates the absolute deviation between the current observation and the historical baseline, and normalizes this deviation using the moving standard deviation to obtain the outlier score. The calculation formula is: A_t=|y_t-μ_t| / (σ_t+ε); where y_t is the current observation, μ_t is the moving mean or exponentially weighted mean, σ_t is the moving standard deviation, and ε is a very small positive number to prevent division by zero.

[0028] When the abnormal score exceeds the preset warning threshold When a warning is triggered, the specific steps are as follows: When ≥ When an abnormal event e_exception is triggered, it will be pushed to the corresponding responsible person or manager according to the alert, escalation, freeze or review strategy, so as to realize the timely handling of risks such as viral articles, deletion of articles, interruption of traffic, and delay.

[0029] Specifically, such as Figure 2 As shown, the campaign dataset includes several campaign tasks and their corresponding task types (including but not limited to link building, photo submission, review, publication, data collection, etc.). The specific steps for constructing the directed graph of the PR process are as follows:

[0030] Based on several task types for task deployment, a set of task nodes is generated, which includes several task nodes and their corresponding state machines. Specifically, for each task type, a corresponding task node is created; at the same time, a state machine is initialized for this node, namely: S={s_created,s_assigned,s_in_progress,s_blocked,s_done,s_failed}; where S is the set of node states, s_created=created, s_assigned=assigned, s_in_progress=in progress, s_blocked=blocked, s_done=completed, s_failed=failed / terminated, and the above are all state labels;

[0031] Based on preset task constraint rules, the set of task nodes is identified and processed to generate a set of directed dependency edges between task nodes. Specifically, the preset task constraint rules clarify the execution order of PR tasks (e.g., compliance verification / auction submission after connection establishment, content review after auction submission, publication after approval, data retrieval after publication, KPI acceptance after data retrieval, etc.). Each task node's preceding dependent nodes are selected according to these rules, and the dependency relationship is represented by directed edges (from preceding node to successor node). All directed edges are then aggregated to form a set of directed dependency edges. ;

[0032] From the set of task nodes and the set of directed dependent edges This constitutes a directed graph of the PR process. Specifically, it refers to: a set of task nodes Using the core nodes (including connection establishment, bidding, compliance verification, content review, release, data collection, KPI acceptance / archiving, etc.) as vertices, and the set of directed dependent edges E (reflecting the execution order constraints between nodes) as directed edges, construct a directed graph of the PR process without circular dependencies. The graph associates the state sets and transition rules of each node, supporting the automated advancement of subsequent nodes when the preceding node is completed.

[0033] The specific steps of state transition are as follows: The multi-source event data of the deployment process is normalized and stored in the event bus, specifically:

[0034] The multi-source event data for the campaign process aims to comprehensively capture various key event data throughout the entire PR campaign process. The sources of event data include, but are not limited to, email receipts, influencer private message communication records, compliance verification result feedback, package delivery / return notifications, material submission records, content review comments, posting link uploads, platform data feedback (such as core metrics such as exposure and likes), and anomaly detection trigger signals (such as post deletion reminders and data interruption alarms).

[0035] The collected multi-source event data is uniformly organized according to a pre-defined standardized event model, which is defined as follows: ,in: For event types (such as private message communication completed, material review approved, and publication successful). The event triggering entity (such as PR staff, influencers, or system-triggered modules). This is the identifier for the associated business object (i.e., the unique ID of the PR delivery object, used to bind it to a specific delivery task). For the precise timestamp of the event, This includes the core business data carried by the event (such as the original communication records, details of review comments, publication links, and specific values ​​of platform metrics). This is a unique hash digest calculated based on the core field of the event, used to achieve idempotent deduplication of events (avoiding repeated processing of the same event). During the normalization process, unstructured data (such as screenshots of private chats and handwritten review comments) is transformed into structured data (such as OCR recognition of screenshots to extract text and digitization of handwritten comments). Data with inconsistent formats (such as indicator units and time formats returned from different platforms) are standardized to ensure that all event data meets the reception and distribution requirements of the event bus.

[0036] The standardized event data after normalization is written into the event bus. The event bus persists the event data according to the preset storage strategy, and records metadata such as the event reception time and storage path to facilitate subsequent traceability and verification.

[0037] The event bus is distributed to each task node according to preset rules to drive the corresponding state machine to perform state transitions. Specifically:

[0038] The event bus adopts a publish-subscribe model. Each task node's state machine has pre-subscribed to its associated event types (e.g., the content review node's state machine subscribes to material submission and review feedback events). Upon receiving event data, the event bus determines the appropriate event type based on the published event data. The association relationship accurately distributes events to the state machines of the corresponding task nodes, achieving precise matching between events and state machines;

[0039] After receiving the event data, the state machine of the dispatched event calls the preset state transition function. ,in, This is the state transition function. This is the current state. To trigger the event, The new state after the transition is based on the event type and The business data in the data drives the state of the corresponding task node to transfer between preset state set node state sets;

[0040] For example, after the state machine of the sampling node receives the event of the expert signing for the sample, from... State transition to Status; After the content review node's state machine receives a review failure event, it then... State transition to Simultaneously, the state machine updates core related information, including but not limited to: the reason for the state change (i.e., the details of the event that triggered this state transition, such as the event type and the triggering entity), the state change timestamp, the business progress corresponding to the current state, and the responsible person's related information (such as the PR staff or experts who need to be contacted after the state transition). The above updated information is synchronized to the corresponding nodes of the PR process directed graph in real time.

[0041] Collection of evidence This includes several types of evidence, such as communication evidence (emails / private messages). Publish link evidence Release status evidence Platform metrics evidence Audit / compliance evidence ,Right now The specific steps for extracting the closed-loop confidence of the target audience are as follows:

[0042] Several categories of data from each target audience are evaluated to obtain the confidence level of the corresponding category of evidence. Specifically, each category of data has a corresponding evaluation rule. The following rules are used to evaluate each category of evidence to obtain the confidence level of each type of evidence. , ( (∈[0, 1], the closer the value is to 1, the more sufficient and credible the evidence is).

[0043] Confidence assessment: The core assessment dimension is whether the key information of the cooperation is fully covered. Key information includes, but is not limited to, confirmation of cooperation intention, requirements for promotional content, agreed release time, and fee settlement instructions. If the event data contains complete two-way communication records (such as email exchanges and private chat logs) and no key information is missing, it is considered... Confidence level = 1; if only one-way communication records are included or key information is partially missing, assign a value according to the proportion of missing information (e.g., if one key piece of information is missing, then...). Confidence level = 0.8, if 2 terms are missing... (Confidence level = 0.6, and so on); if no communication data is collected, then... Confidence level = 0;

[0044] Confidence assessment: The core assessment dimensions are link validity and binding consistency; if the collected publishing links are accessible and the content corresponding to the links matches the promotion theme and influencer identity of the target audience (verified through link content parsing and influencer account matching), then the confidence level is determined to be high. Confidence level = 1; if the link is accessible but the content relevance is disputed (e.g., the topic is relevant but the brand is not mentioned), then the following judgment is made. Confidence level = 0.7; if the link is inaccessible or unrelated to the target audience, it is considered invalid. Confidence level = 0;

[0045] Confidence assessment: The core assessment dimensions are the authenticity and completeness of the posting status; if the event data includes a platform posting success receipt and a posting timestamp, and is consistent with the posting information provided by the influencer, then it is considered credible. Confidence level = 1; if only a single source of posting status information is obtained (e.g., only influencer feedback confirms successful posting, without platform confirmation), then the following is determined: Confidence level = 0.6; if the publication status is not confirmed or confirmed as not published, then... Confidence level = 0;

[0046] Confidence assessment: The core assessment dimensions are data completeness and timeliness; if complete data on preset core indicators such as exposure, likes, comments, reposts, followers, store visits, and sales are collected, and the data is real-time data transmitted within a preset period after publication (e.g., within 72 hours), then the confidence level is determined to be high. Confidence level = 1; if some core indicators are missing (e.g., lack of trading volume data), assign a value according to the proportion of missing indicators (e.g., if one core indicator is missing, then...). Confidence level = 0.85, if 2 terms are missing... (Confidence level = 0.7, and so on); if no platform indicator data is collected, then... Confidence level = 0;

[0047] Confidence assessment: The core assessment dimensions are the completeness of the audit process and the clarity of the compliance conclusion; if it includes complete audit opinion records and compliance verification reports, and clearly indicates that the audit has passed and compliance has been achieved, then it is considered credible. Confidence level = 1; if the review process is complete but the conclusion is that the document passes after rectification (rectification record attached), then the judgment is... Confidence level = 0.9; if only part of the review process is completed or the compliance conclusion is unclear, the judgment is... Confidence level = 0.3; if no audit / compliance verification has been conducted, the judgment is... Confidence level = 0;

[0048] The confidence levels of several types of evidence are then fused to obtain the closed-loop confidence level for each target audience. Specifically, this involves calculating the closed-loop confidence level for a given target audience. The specific formula is as follows: ,in, As a product (integration over all types of evidence), this fusion form can be understood as meaning that sufficient evidence from any one of the multiple evidences can significantly improve the overall confidence level.

[0049] In this implementation plan, a standardized directed graph of the public relations process is constructed to clarify the dependencies between task nodes, avoid execution skips or blockages, and ensure that the process progresses in a systematic manner. Multi-source event data is processed and stored uniformly in the event bus, which not only achieves data structure transformation and format unification, but also avoids duplicate processing through hash deduplication. The state machine drives state transitions precisely based on events and synchronizes node progress information in real time to ensure the automated progress of the process. At the same time, the confidence of various types of evidence is evaluated according to clear rules and the closed-loop confidence is calculated to achieve quantitative verification of the entire evidence chain for deployment, so as to ensure the orderly progress of the process.

[0050] Specifically, the steps for identifying the set of task nodes to be executed for each deployment object are as follows: Read the state machine of each task node and determine whether it meets the preset executable state conditions. Specifically, this involves: traversing the current state of the state machine of each task node. The preset executable state conditions for a task node are that two conditions are met simultaneously:

[0051] Condition 1: The current state of the task node is or ,in, The status indicates that the node has been created but has not yet started execution. The status indicates that the node was previously blocked due to unmet or abnormal dependencies, and it is now necessary to re-verify whether it meets the execution conditions.

[0052] Condition 2: Task Node All preceding dependent nodes All meet This condition is unlocked through the formula. Verification (wherein) This represents a logical AND operation, meaning that all preceding nodes must simultaneously achieve the AND condition. Both conditions are indispensable. Represents a node (Whether the unlocking conditions are met), query each preceding node. The state machine data is checked one by one to ensure that there are no incomplete predecessor dependencies;

[0053] First, determine if the current state meets condition one, then verify condition two by traversing the list of preceding dependent nodes; only when both conditions are met simultaneously, determine the task node. The preset executable state conditions are met;

[0054] Task nodes that meet the preset executable status conditions are marked as pending task nodes, and statistical processing is performed to obtain a set of pending task nodes for each delivery object. Specifically, the judgment results of all task nodes are filtered, and task nodes that meet the executable status conditions are uniformly marked as pending task nodes. Each marked node is attached with unique tag information, such as: tag generation timestamp, node responsible person information, preset deadline, and unique identifier of the associated delivery object. ;

[0055] Based on node association The task nodes to be executed are bound to the corresponding delivery objects to ensure that each node belongs only to its corresponding delivery object and avoid confusion across delivery objects;

[0056] Investigate and remove nodes with duplicate markings (such as duplicate verification markings caused by system retries) to ensure that each node to be executed appears only once in the set;

[0057] The task nodes are initially sorted according to their logical order in the directed graph of the PR process (i.e., the order of nodes from upstream to downstream of the process). The processed task nodes are then aggregated to form a set of task nodes to be executed, with each target object as the unit.

[0058] The specific steps for analyzing the task priority score of each pending task node are as follows: Based on the multi-source event data of the deployment process and the evidence set of each deployment object, analyze the score dataset of each pending task node, including urgency score, impact score, risk score, congestion score, and cost score, specifically as follows:

[0059] Urgency score: Extract the preset deadline of the task node to be executed from the multi-source event data of the deployment process and the current time, calculate the time difference, and obtain the urgency score.

[0060] Impact on score: Extract historical collaboration data (number of followers, past exposure, conversion contribution rate) of the target audience (such as influencer-content combination) corresponding to this task node from multi-source event data of the campaign process, and combine it with platform indicator evidence in the evidence set. The historical average (such as the average exposure / conversion data of similar content) is used to calculate the expected contribution value through weighted summation, and this value is used as the impact score.

[0061] Risk score: Extract the deletion / delay / compliance risk corresponding to this task node from multi-source event data in the delivery process;

[0062] Blocking score: Extract the list of successor nodes of the task node to be executed in the directed graph of the PR process from the multi-source event data of the deployment process, count the number of successor task nodes that directly depend on the node, and use it as the blocking score.

[0063] Cost score: Extract the time cost (such as review time), human cost (such as the number of staff required), and communication cost (such as the number of times to communicate with influencers) required to complete the task from multi-source event data of the campaign process, and perform standardized processing. The scores are then weighted according to preset weights to obtain the cost score. It should be noted that the above scores have all been standardized and mapped to values ​​between 0 and 1.

[0064] Based on the scoring dataset, the task priority score of each task node to be executed is analyzed. The formula for calculating the task priority score of a task node to be executed is as follows: P_i=w1·U_i+w2·I_i+w3·R_i+w4·B_i-w5·C_i;

[0065] Wherein, P_i is the task priority score of a task node to be executed, U_i is the urgency score of a task node to be executed, I_i is the impact score of a task node to be executed, R_i is the risk score of a task node to be executed, B_i is the blocking score of a task node to be executed, C_i is the cost score of a task node to be executed, and w1, w2, w3, w4, and w5 are configurable weights (in this implementation example, they can be 0.4, 0.3, 0.2, 0.3, and 0.2, respectively). Preferably, the dynamic scheduling is triggered after a batch of events is consumed by the event bus or when the set of task nodes to be executed changes, so as to ensure that the task queue is dynamically updated with real-time data and dependency unlock status.

[0066] Dynamic scheduling specifically involves: reading the task priority score of each task node to be executed, sorting them in descending order, and generating a task execution sequence (for each deployment object); based on the task execution sequence, performing task scheduling processing on each deployment object, that is, executing them sequentially according to the order of the task nodes to be executed in the task execution sequence.

[0067] This implementation plan accurately identifies task nodes to be executed through dual-condition verification, ensuring that nodes are in an updatable state and verifying that all prerequisites have been completed, thus avoiding confusion in the process. At the same time, it attaches unique information to nodes and categorizes them according to the target audience, effectively preventing cross-object confusion and duplicate marking issues, making the list of tasks to be executed clear and concise. Based on this, it combines multi-source event data and evidence sets to calculate task priorities based on urgency, impact, risk, blockage, and cost, and then generates an execution sequence and schedules it according to the scores, ensuring that high-urgency, high-value, and high-risk tasks are prioritized, optimizing resource allocation, reducing process blockage, and significantly improving the efficiency and resource utilization of the entire PR deployment process.

[0068] Specifically, the automatic acceptance processing steps are as follows: Based on multi-source event data of the delivery process, extract the delivery indicator vectors for each delivery object. Specifically, this involves: filtering the multi-source event data of the delivery process written to the event bus, identifying events with the event type of platform data feedback or similar identifiers, and parsing the event when such an event is consumed. ,Should It includes raw metric data directly related to the campaign's effectiveness, such as impressions, likes, favorites, comments, shares, new followers, store visitors, and sales within a specific time window. The system uses this data to inform the campaign. (Business Object Identifier) ​​The parsed metric values ​​and corresponding timestamps are added and aggregated under the name of the target object corresponding to the identifier, forming a structured target object-specific targeting metric vector. =(Exposure, likes, favorites, comments, reposts, follows, store visits, transactions, etc.);

[0069] Based on the delivery indicator vector, the KPI achievement degree of each delivery object is analyzed. Specifically, the delivery target vector is defined, and each indicator in the delivery target vector is the target value corresponding to the delivery indicator vector. The formula for calculating the KPI achievement degree K of a certain delivery object is as follows: K=Σ_{j=1..d}α_j·min(m_j / t_j,cap);

[0070] Where m_j is the value of the j-th indicator in the delivery indicator vector, t_j is the target value of the j-th indicator in the delivery target vector, α_j is the weight of the j-th indicator, and cap is the upper limit coefficient (to prevent distortion caused by excessively inflating a single indicator).

[0071] If an anomaly exists, a penalty is introduced: K'=K·(1-β·A); where K' is the achievement degree after introducing the anomaly penalty, A is the anomaly score (or the normalized value of the anomaly intensity), and β is the penalty weight;

[0072] Read the closed-loop confidence level of each target audience and perform collaborative verification with the KPI achievement level. Specifically, this involves comparing the closed-loop confidence level and the KPI achievement level with the closed-loop confidence threshold, respectively. KPI threshold Perform comparison processing and meet the following requirements. At that time, the PR campaign for the target audience is automatically approved.

[0073] In this implementation plan, relying on multi-source data from the event bus, core indicators such as exposure and sales of the target audience are accurately extracted and structured into indicator vectors. This provides objective and complete data support for KPI calculation. Through weighted calculation of target vectors and indicator vectors, coupled with upper limit coefficients and anomaly penalty mechanisms, the plan effectively avoids the excessive inflation of overall achievement by a single indicator, making the KPI evaluation results more in line with the actual campaign performance. At the same time, the KPI achievement and closed-loop confidence are verified collaboratively. Only when both indicators meet the standards can the campaign be deemed accepted, ensuring that the campaign performance meets expectations and thus achieving accurate PR campaign acceptance.

[0074] Please see Figure 3 This invention provides a technical solution: a PR full-process orchestration and data closed-loop engine system, comprising: a process orchestration module for acquiring a dataset of delivery tasks for several delivery targets and constructing a directed graph of the PR process for each delivery target, including several task nodes and corresponding state machines; an event bus module for collecting multi-source event data of the delivery process and performing normalization processing for storage in the event bus; a state machine advancement module for driving the state machine to perform state transitions based on the event bus; an evidence alignment and closed-loop calculation module for collecting the evidence set of each delivery target during the state transition process and extracting the closed-loop confidence of the corresponding delivery target; a task scheduling module for identifying the set of task nodes to be executed for each delivery target, including several task nodes to be executed, and analyzing the corresponding task priority score value, and performing dynamic scheduling based on the task priority score value; an anomaly detection and early warning module for performing detection and early warning processing based on multi-source event data of the delivery process and combined with a preset anomaly detection algorithm; and a KPI acceptance and archiving module for automatically accepting and archiving each delivery target when it reaches a preset task node.

[0075] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for a PR full-process orchestration and data closed-loop engine, characterized in that, Includes the following steps: Obtain a dataset of campaign tasks for several campaign targets, and construct a directed graph of the PR process for the corresponding campaign targets, which includes several task nodes and their corresponding state machines. It also collects multi-source event data from the delivery process and drives the state machine to perform state transitions. During the state transition process, evidence sets for each of the aforementioned delivery objects are collected, and the closed-loop confidence of the corresponding delivery object is extracted; Simultaneously, the set of task nodes to be executed for each of the aforementioned deployment objects is identified, including several task nodes to be executed, and the corresponding task priority score is analyzed, and dynamic scheduling is performed based on the task priority score. When each of the aforementioned delivery objects reaches the preset task node, automatic acceptance and archiving processes are performed.

2. The method for PR full-process orchestration and data closed-loop engine according to claim 1, characterized in that, The deployment task dataset includes several deployment tasks and their corresponding task types. The specific steps for constructing the directed graph of the PR process are as follows: Based on the task types of the aforementioned deployment tasks, a task node set is generated, which includes several task nodes and corresponding state machines. Based on preset task constraint rules, the task node set is identified and processed to generate a set of directed dependency edges between the task nodes. The set of task nodes and the set of directed dependency edges constitute the directed graph of the PR process.

3. The method for PR end-to-end orchestration and data closed-loop engine according to claim 1, characterized in that, The specific steps of the state transition are as follows: The multi-source event data of the delivery process is normalized and stored in the event bus; The event bus is distributed to each task node according to a preset rule to drive the corresponding state machine to perform state transitions.

4. The method for PR full-process orchestration and data closed-loop engine according to claim 1, characterized in that, The evidence set includes several types of evidence. The specific steps for extracting the closed-loop confidence of the target are as follows: Several types of data from each of the aforementioned targets are evaluated and processed to obtain the confidence level of the corresponding type of evidence; The confidence levels of several types of evidence are then fused to obtain the closed-loop confidence level of each of the aforementioned targets.

5. The method for PR full-process orchestration and data closed-loop engine according to claim 1, characterized in that, The specific steps for identifying the set of task nodes to be executed for each of the aforementioned delivery objects are as follows: Read the state machine of each task node and determine whether the preset executable state conditions are met; Task nodes that meet the preset executable state conditions are marked as task nodes to be executed, and statistical processing is performed to obtain the set of task nodes to be executed for each of the deployed objects.

6. The method for PR end-to-end orchestration and data closed-loop engine according to claim 5, characterized in that, The specific steps for analyzing the task priority score of each of the aforementioned task nodes to be executed are as follows: Based on the multi-source event data of the delivery process and the evidence set of each delivery object, the scoring dataset of each task node to be executed is analyzed, including urgency score, impact score, risk score, blockage score, and cost score. Based on the scoring dataset, the task priority score value of each of the task nodes to be executed is analyzed.

7. The method for PR full-process orchestration and data closed-loop engine according to claim 6, characterized in that, The dynamic scheduling specifically refers to: Read the task priority score of each of the task nodes to be executed, sort them in descending order, and generate a task execution sequence; Based on the task execution sequence, task scheduling is performed on each of the deployment objects.

8. The method for PR full-process orchestration and data closed-loop engine according to claim 1, characterized in that, It also includes the following steps: Based on the multi-source event data of the aforementioned delivery process, extract the time-series data of key indicators for each of the delivery targets. Anomaly analysis is performed on the time-series data of the key indicators based on a preset anomaly detection algorithm, and anomaly scores are calculated. When the abnormal score exceeds a preset warning threshold, a warning process is triggered.

9. The method for PR end-to-end orchestration and data closed-loop engine according to claim 1, characterized in that, The specific steps of the automatic acceptance process are as follows: Based on multi-source event data of the delivery process, extract the delivery indicator vectors for each of the delivery targets; Based on the aforementioned targeting metric vector, analyze the KPI achievement rate of each of the aforementioned targeting objects; Read the closed-loop confidence level of each of the aforementioned targets and perform collaborative verification with the KPI achievement level.

10. A system for a PR full-process orchestration and data closed-loop engine, employing the method of the PR full-process orchestration and data closed-loop engine as described in any one of claims 1-9, characterized in that, include: The process orchestration module is used to obtain the deployment task dataset for several deployment objects and construct the PR process directed graph for the corresponding deployment objects, which includes several task nodes and corresponding state machines. The event bus module is used to collect multi-source event data from the deployment process, and to perform normalization processing for storage in the event bus; A state machine advancement module is used to drive the state machine to perform state transitions based on an event bus. The evidence alignment and closed-loop calculation module is used to collect the evidence set of each of the deployed objects during the state transition process and extract the closed-loop confidence of the corresponding deployed objects. The task scheduling module is used to identify the set of task nodes to be executed for each of the deployment objects, including several task nodes to be executed, and to analyze the corresponding task priority score value, and to perform dynamic scheduling based on the task priority score value. An anomaly detection and early warning module is used to perform detection and early warning processing based on multi-source event data of the delivery process and in combination with a preset anomaly detection algorithm; The KPI acceptance and archiving module is used to automatically accept and archive each of the aforementioned targets when they reach the preset task nodes.

Images