Search

CN-121998710-A - Advertisement attribute consistency monitoring method, system, equipment and medium

CN121998710ACN 121998710 ACN121998710 ACN 121998710ACN-121998710-A

Abstract

The invention discloses a method, a system, equipment and a medium for monitoring consistency of advertisement attributes, wherein the method specifically comprises the steps of collecting snapshot and metadata of the advertisement attributes in each processing link of a creation flow, and constructing an advertisement attribute blood-margin map; the method comprises the steps of extracting time sequence data of attribute values based on advertisement attribute blood-edge maps, modeling a normal fluctuation mode of attributes by using an anomaly detection algorithm, sending out risk early warning in advance on abnormal fluctuation deviating from the normal mode, analyzing a conduction path with inconsistent attributes and locating root cause nodes by combining current inconsistent node information and the state of the advertisement attribute blood-edge maps by using a pre-trained causal map model, inputting an alarm event into an alarm strategy model based on reinforcement learning training, and dynamically deciding the sending, priority and informing channels of the alarm. The invention realizes high-efficiency, accurate and real-time monitoring of the consistency of the advertisement attributes, improves the accuracy and efficiency of advertisement delivery, and can meet the continuous development requirement of advertisement business.

Inventors

  • WANG CHUANPENG
  • LI BO

Assignees

  • 安徽三七极域网络科技有限公司

Dates

Publication Date
20260508
Application Date
20251217

Claims (10)

  1. 1. The advertisement attribute consistency monitoring method is characterized by comprising the following steps: collecting snapshot and metadata of advertisement attributes in each processing link of the creation flow, and constructing an advertisement attribute blood-margin map for recording the change history of all-link data; Extracting time sequence data of attribute values based on the advertisement attribute blood-margin map, modeling a normal fluctuation mode of the attribute by using an anomaly detection algorithm, and sending out risk early warning on anomaly fluctuation deviating from the normal mode in advance; When detecting that the attributes are actually inconsistent or risk early warning is received, analyzing a conduction path with inconsistent attributes and positioning root cause nodes by utilizing a pre-trained causal graph model and combining current inconsistent node information and the state of advertisement attribute blood-margin maps; Dynamically highlighting root cause nodes and affected attribute nodes and conduction paths in a visual interface; inputting an alarm event into an alarm strategy model based on reinforcement learning training, and dynamically deciding the sending, priority and notification channel of an alarm according to historical alarm response data and the current event context; and combining a plurality of alarm events with the same or similar root cause nodes into a single aggregated alarm event by utilizing an intelligent alarm aggregation algorithm.
  2. 2. The method of claim 1, wherein the step of collecting the snapshot and the metadata of the advertisement attribute in each processing link of the creation flow to construct an advertisement attribute blood-margin map for recording the change history of the full-link data specifically comprises: Deploying data acquisition probes in a plurality of processing links of the advertisement creation flow, and synchronously capturing complete snapshots of advertisement attribute of different processing links, wherein the process metadata comprises service identifiers, link identifiers and request chain contexts; packaging the advertisement attribute complete snapshot and the process metadata into standardized events, and asynchronously sending the standardized events to a message middleware; Creating a corresponding attribute version node for the current link in a graph database based on a service identifier and a request chain context in an event by stream processing of the event in the service consumption message middleware; Searching a father version node generated by an upstream ring node in a graph database according to the request chain context, and creating a derivative relation edge pointing to the father version node from the current attribute version node to form a graph structure for describing data flow and version evolution to form an advertisement attribute blood-edge graph.
  3. 3. The method according to claim 1, wherein the extracting the time series data of the attribute value based on the advertisement attribute blood-margin map uses an anomaly detection algorithm to model a normal fluctuation mode of the attribute, and the early warning of risk is sent out in advance for the anomaly fluctuation deviating from the normal mode, specifically comprising: based on the advertisement attribute blood-margin map, dividing according to the combination dimension of the attribute and the processing link, and extracting multi-dimensional time sequence characteristic data for reflecting the evolution of the attribute content and the processing behavior characteristic from the historical sequence of the map node; Aiming at each combination of the attribute and the processing link, training the multi-dimensional time sequence characteristic data by using an unsupervised machine learning algorithm, and respectively establishing a baseline model for describing a normal fluctuation mode of each combination; when the advertisement attribute data flows through any processing link and the blood-margin map is updated, calculating a corresponding current time sequence feature vector according to the newly generated map node and the context information thereof in real time; the current time sequence feature vector is input into a corresponding baseline model to be inferred, and an abnormal quantification score used for representing the degree of deviation from a normal mode is obtained; and when the abnormal quantitative score exceeds a preset early warning threshold, generating a risk early warning prompt.
  4. 4. The method according to claim 1, wherein when detecting that the attribute is actually inconsistent or receiving the risk pre-warning, analyzing the conduction path of the attribute inconsistency and locating the root cause node by using a pre-trained causal graph model and combining the current inconsistent node information and the state of the advertisement attribute blood edge map, and further comprising: abstracting a system topology graph containing a logic processing unit and a dependency relationship between units based on a data flow path reflected by an advertisement attribute blood-edge map; constructing a causal graph model in a Bayesian network form by taking a system topological graph as a structural framework, wherein network nodes of the causal graph model are used for representing state variables of processing units or attribute consistency; and training and learning the conditional probability parameters of each node in the causal graph model by using the historical fault event and the corresponding component abnormal evidence data set.
  5. 5. The method according to claim 4, wherein when detecting that the attribute is actually inconsistent or receiving the risk pre-warning, analyzing the conduction path of the inconsistent attribute and locating the root cause node by using a pre-trained causal graph model and combining the current inconsistent node information and the state of the advertisement attribute blood-edge map, specifically comprising: When an attribute inconsistent alarm or risk early warning is received, extracting currently observed node state evidence from alarm and real-time monitoring data; Injecting the current observation evidence into the trained causal graph model, and executing probabilistic reasoning calculation to obtain posterior probabilities that all undefined state nodes are identified as abnormal under the current evidence; And sorting the nodes according to the posterior probability, performing causal analysis, screening out at least one node with the highest posterior probability and without an abnormal father node, judging as a root node causing inconsistent attribute at this time, and determining a conduction path from the root node to the alarm node.
  6. 6. The method according to claim 1, wherein the inputting the alarm event into the reinforcement learning training based alarm policy model, and dynamically deciding the sending, priority and notification channel of the alarm according to the historical alarm response data and the current event context, specifically comprises: Based on the historical alarm response data, forming an alarm decision process into a Markov decision process, setting a state space as a comprehensive vector for describing alarm event characteristics, system context and receiver state, and setting an action space as a combined action set comprising whether to suppress alarms, set priorities and select notification channels; Constructing a multi-objective rewarding function, wherein the multi-objective rewarding function is used for calculating numeric rewards or penalties obtained by each decision according to whether the follow-up alarming is confirmed to be valid, the response efficiency is high or low and the channel cost benefit is notified; Training the reinforcement learning model based on the state space, the action space and the multi-objective rewarding function to obtain an alarm strategy model; when a new alarm event is generated, constructing a corresponding current state vector in real time, inputting the current state vector into an alarm strategy model for strategy reasoning, and obtaining a recommended decision action aiming at the alarm event; And executing a recommended decision action, controlling the sending of the alarm, the priority allocation and the notification channel routing, and completely recording the state, action and time information of the decision.
  7. 7. The method according to any one of claims 1 to 6, wherein the merging of multiple alarm events with the same or similar root cause node into a single aggregated alarm event using an intelligent alarm aggregation algorithm, in particular comprises: extracting an alarm feature vector containing root node identification, an affected attribute path diagram structure, alarm semantics and space-time information for each input alarm event, and representing the alarm feature vector as a small directed graph centering on the root node; Based on the small directed graph, performing incremental clustering processing on a plurality of alarm events in a dynamic sliding time window, and determining to merge or create a new cluster the new alarm event by calculating the composite similarity between the new alarm event and the existing cluster, wherein the composite similarity comprises the accurate matching degree of root node identification, the similarity of affected attribute path graph, the semantic similarity of alarm text and the immediate proximity; When one aggregation cluster meets a preset output triggering condition, generating one aggregation alarm event based on all alarm events in the aggregation cluster.
  8. 8. An advertisement attribute consistency monitoring system, the system is characterized by comprising the following specific components: the data recording module is used for collecting snapshot and metadata of advertisement attributes in each processing link of the creation flow and constructing an advertisement attribute blood-margin map for recording the change history of the full-link data; the wind direction early warning module is used for extracting time sequence data of attribute values based on the advertisement attribute blood-edge map, modeling a normal fluctuation mode of the attribute by utilizing an anomaly detection algorithm, and sending out risk early warning on anomaly fluctuation deviating from the normal mode in advance; The risk attribution module is used for analyzing a conduction path with inconsistent attributes and positioning root cause nodes by utilizing a pre-trained causal graph model and combining current inconsistent node information and the state of an advertisement attribute blood margin map when detecting that the attributes are actually inconsistent or receiving risk early warning; The dynamic display module is used for dynamically highlighting the root cause node, the affected attribute nodes and the conduction path in the visual interface; The alarm decision module is used for inputting an alarm event into an alarm strategy model based on reinforcement learning training, and dynamically deciding the sending, priority and notification channel of the alarm according to historical alarm response data and the current event context; And the alarm aggregation module is used for combining a plurality of alarm events with the same or similar root cause nodes into a single aggregated alarm event by utilizing an intelligent alarm aggregation algorithm.
  9. 9. A computer device comprising a memory and a processor and a computer program stored on the memory, which when executed on the processor implements the advertisement attribute consistency monitoring method according to any of claims 1 to 7.
  10. 10. A computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the advertisement attribute consistency monitoring method according to any of claims 1 to 7.

Description

Advertisement attribute consistency monitoring method, system, equipment and medium Technical Field The present invention relates to the field of advertisement operation technologies, and in particular, to a method, a system, an apparatus, and a medium for monitoring consistency of advertisement attributes. Background In the field of advertisement delivery, the advertisement creation process is a complex and critical process, wherein the correctness and consistency of advertisement attributes at different stages (e.g. panel input, asynchronous processing, media API output, etc.) play a decisive role in the accurate delivery and efficient operation of advertisements. However, in the existing advertisement creation process, the advertisement attribute is inconsistent, which directly causes advertisement delivery errors or inefficiency, and brings a plurality of adverse effects to advertisers and advertisement platforms. Traditional advertisement attribute consistency monitoring methods rely mainly on manual inspection logs. There are many limitations to this monitoring approach, specifically as follows: 1. The manual checking efficiency is low, in the advertisement creation flow, the advertisement attribute relates to a plurality of links and a large amount of data, and the manual checking needs to compare advertisement attribute values of different stages one by one. In the face of massive data and complex processes, manual operation is time-consuming and labor-consuming, and comprehensive inspection is difficult to complete in a short time, so that monitoring efficiency is extremely low. For example, in a large advertising campaign, tens of thousands of advertising attribute data may be generated each day, with manual one-by-one inspection of tasks that are nearly impossible to accomplish. 2. The error is easy to occur, the manual checking process is easy to be influenced by subjective factors, different checking staff can have difference in understanding and judging standards of data, repeated checking work is easy to cause fatigue of the checking staff for a long time, and misjudgment is caused. For example, an inspector may inadvertently ignore certain subtle but critical attribute differences, or make erroneous determinations of similar but not exactly the same attribute values, which may directly affect the accuracy of the advertisement placement. 3. The real-time monitoring is lacking, and the traditional monitoring method cannot always find the condition of inconsistent advertisement attributes in real time. Manual inspection is typically performed at a specific point in time or stage, meaning that if a problem arises in which the advertisement attributes are inconsistent during the period of time between inspection, it may not be found in time, resulting in delayed discovery of the problem. For example, in a critical period of advertisement delivery, if advertisement attributes are inconsistent and not timely perceived, the best delivery opportunity may be missed, resulting in economic loss for advertisers. 4. The expansibility is insufficient, and the number and complexity of advertisement attributes are increased along with the continuous development of advertisement services. New advertisement types, delivery channels and business rules are continuously emerging, so that advertisement attribute systems are increasingly huge and complex. The traditional manual monitoring method is difficult to adapt to the rapidly-changing business requirement, and cannot be flexibly expanded to cover the newly-added advertisement attribute and monitoring dimension. For example, conventional approaches require reformulation of inspection rules and procedures when introducing new ad delivery platforms or pushing new ad creative forms, which can be labor intensive and time consuming. 5. The warning information provided by the existing monitoring method is quite simple, and usually only informs that the problem of inconsistent advertising attributes exists, but lacks finer information, such as inconsistent specific attribute values, specific time when the problem occurs, related log information and the like. Therefore, after the operator receives the alarm, the operator is difficult to quickly locate the root and specific position of the problem, and the difficulty of problem investigation and solution is increased. For example, an operator may need to manually search a large amount of log data for relevant information based on simple alarm information to gradually determine where the problem is, which clearly greatly prolongs the time for the problem to be solved. Disclosure of Invention The invention aims to provide a method, a system, equipment and a medium for monitoring consistency of advertisement attributes, which realize high-efficiency, accurate and real-time monitoring of consistency of advertisement attributes, improve the accuracy and efficiency of advertisement delivery, and adapt to the continuous devel