Search

CN-121981545-A - Advertisement putting risk control method based on big data analysis

CN121981545ACN 121981545 ACN121981545 ACN 121981545ACN-121981545-A

Abstract

The invention discloses an advertisement putting risk control method based on big data analysis, which comprises the steps of obtaining advertisement putting multisource behavior data, constructing a putting behavior time sequence, building a multi-level time sequence structure based on the putting behavior time sequence, constructing a multi-level variable point detection structure based on a Bayesian online variable point detection algorithm, generating multi-level running length distribution and multi-level variable point posterior, generating a multi-level running length state based on the multi-level running length distribution and the multi-level variable point posterior, building a risk hypothesis set and generating a risk hypothesis posterior weight set, building a risk state structural representation, generating a risk control action parameter based on the risk state structural representation, acting the risk control action parameter on an advertisement putting process and executing additional update. The invention realizes multi-level accurate identification and closed-loop self-adaptive control of advertisement putting risks.

Inventors

  • YUE DANQING

Assignees

  • 杭州转化引擎互联网科技有限公司

Dates

Publication Date
20260505
Application Date
20260123

Claims (9)

  1. 1. The advertisement putting risk control method based on big data analysis is characterized by comprising the following steps: Acquiring advertisement putting multisource behavior data, executing time alignment processing and structuring processing, and constructing a putting behavior time sequence; Based on the delivery behavior time sequence, performing hierarchical mapping and time aggregation processing, and establishing a multi-hierarchy time sequence structure; Based on a Bayes online variable point detection algorithm, constructing a multi-level variable point detection structure, executing online variable point inference processing by combining a multi-level time sequence structure to generate multi-level running length distribution, and executing joint updating to generate a multi-level variable point posterior; Performing run length state determination processing based on the multi-level run length distribution and the multi-level change point posterior, determining a multi-level run length state; Based on the multi-level running length state and the multi-level variable point posterior, performing hypothesis space division inference processing, constructing a risk hypothesis set, performing posterior update and evidence accumulation on each risk hypothesis, and generating a risk hypothesis posterior weight set; constructing a risk state structured representation based on the multi-level run length state, the risk hypothesis posterior weight set and the abnormal accumulation; based on the structured representation of the risk state, generating risk control action parameters through a mapping rule from the risk state to the control parameters; And (3) acting the risk control action parameters on the advertisement putting process, generating risk control action feedback, and executing additional updating.
  2. 2. The advertisement delivery risk control method based on big data analysis according to claim 1, wherein the generation of the delivery behavior time sequence comprises: collecting advertisement putting multisource behavior data corresponding to an advertisement putting process from an advertisement putting system, wherein the advertisement putting multisource behavior data comprises exposure data, click data, conversion data and control data; performing time alignment processing under a unified time reference on the advertisement putting multisource behavior data, and performing synchronous merging processing by taking a fixed time window as alignment granularity; after the time alignment process is completed, performing a structuring process, converting the exposure data into exposure statistics, converting the click data into click statistics, converting the conversion data into conversion statistics, and converting the control data into control statistics; based on the exposure statistic, click statistic, conversion statistic and control statistic, a time series of throwing actions is constructed.
  3. 3. The advertising risk control method based on big data analysis of claim 1, wherein the generating of the multi-level time series structure comprises: based on each behavior record in the delivery behavior time sequence, performing hierarchical mapping processing according to a predetermined hierarchical identification relationship, and mapping each behavior record to a material hierarchy, a delivery unit hierarchy and an account media hierarchy respectively and uniquely; after the hierarchical mapping processing is completed, performing time aggregation processing on the behavior records mapped to the same hierarchy based on the time window boundaries to which each behavior record belongs, and respectively forming a material hierarchy time sequence, a delivery unit hierarchy time sequence and an account media hierarchy time sequence; and constructing a multi-level time sequence structure based on the material level time sequence, the delivery unit level time sequence and the account media level time sequence.
  4. 4. The advertising risk control method based on big data analysis of claim 1, wherein the generating of the multi-level run length distribution and multi-level variability posterior comprises: Based on a multi-level time sequence structure, performing online variable point inference processing, and independently maintaining corresponding running length distribution under each level to obtain material level running length distribution, unit level running length distribution and account media level running length distribution, thereby forming multi-level running length distribution; Performing a joint update on the multi-level run-length distribution based on the level-dependent update rules; after the level dependence joint updating is completed, on the basis of the online updating result of the running length distribution of each level, a material level variable point posterior, a unit level variable point posterior and an account media level variable point posterior are calculated respectively, and a multi-level variable point posterior is generated through combination.
  5. 5. The advertising risk control method based on big data analysis of claim 1, wherein the generating of the multi-level run length status comprises: Based on the multi-level running length distribution and the multi-level variable point posterior, performing comprehensive judgment on the running length values of all levels at the current time point and the probability states of the corresponding points, and determining a multi-level running length state; based on the determination of the multi-level running length state, performing accumulation processing on the multi-level running length state and the change result of the multi-level change point posterior along the time dimension, and calculating the abnormal accumulation amount; the modulation process is performed on a posterior update rate of the delivery unit tier run length distribution and the account media tier run length distribution based on the abnormal cumulative amount.
  6. 6. The advertising risk control method based on big data analysis of claim 1, wherein the generation of the risk hypothesis posterior weight set comprises: under the current time point, trigger judgment is executed based on the multi-level variable point posterior and a preset variable point posterior threshold trigger condition, and when the multi-level variable point posterior meets the preset variable point posterior threshold trigger condition, a hypothesis space division inference state is entered; After entering the hypothesis space division inference state, constructing a risk hypothesis set, wherein the risk hypothesis set is formed by a normal change hypothesis branch, a strategy change hypothesis branch and a risk injection hypothesis branch in parallel; Performing evidence accumulation processing on each risk hypothesis in the risk hypothesis set based on the multi-level running length state, and respectively generating corresponding evidence accumulation results for each risk hypothesis branch by taking the change trend of the multi-level running length state in the time dimension as input; And on the basis of the evidence accumulation processing, executing posterior update processing on each risk hypothesis branch in the risk hypothesis set, mapping the evidence accumulation result into corresponding risk hypothesis posterior weights, and generating a risk hypothesis posterior weight set.
  7. 7. The advertising risk control method based on big data analysis of claim 1, wherein the generation of the risk status structured representation comprises: According to the running length value states of the multi-level running length states under different levels and the evolution relation thereof, performing level priority comparison judgment processing to determine the risk level position; Performing dominant weight judgment processing by the relative weight relation of all risk hypothesis branches in the risk hypothesis posterior weight set, and determining the type of a risk generation mechanism; performing time accumulation trend judgment processing by the accumulation result of the abnormal accumulation amount in the time dimension to determine the evolution intensity; and combining the risk level position, the risk generation mechanism type and the evolution intensity to construct a risk state structured representation.
  8. 8. The advertising risk control method based on big data analysis of claim 1, wherein the generation of the risk control action parameters comprises: performing a risk level location and amplitude selection decision process based on the risk level location based on the risk state structured representation, determining an adjustment amplitude of the delivery rate parameter; based on the risk generation mechanism type, executing dominant mechanism judgment and audit level selection processing based on the risk generation mechanism type, and determining the value level of the audit intensity parameter; Based on the evolution intensity in the risk state structured representation, executing evolution stage judgment and execution process determination processing based on the evolution intensity, and determining an execution process of the freezing path parameter; and combining the release rate parameter, the auditing strength parameter and the freezing path parameter to generate a risk control action parameter.
  9. 9. The advertising risk control method based on big data analysis of claim 1, wherein the additional updating includes: The risk control action parameters are acted on the advertisement putting process, corresponding putting rate modulation, auditing intervention configuration and freezing path pushing operation are executed on the current advertisement putting behavior, and risk control action feedback is generated; based on risk control action feedback, corresponding advertisement putting multisource behavior data is obtained, the advertisement putting multisource behavior data is added to a putting behavior time sequence according to time alignment processing and structuring processing rules, and additional updating is executed; Inputting the added updated delivery behavior time sequence into a multi-level variable point detection structure, and executing online updating processing on multi-level running length distribution and multi-level variable point posterior to update the multi-level running length state; On the basis of the multi-level running length state update, performing posterior update processing on the risk hypothesis posterior weight set based on the updated multi-level variable point posterior; Based on the updated multi-level run length state, risk hypothesis posterior weight set and abnormal accumulation, reconstructing a risk state structured representation, and updating risk control action parameters.

Description

Advertisement putting risk control method based on big data analysis Technical Field The invention relates to the field of digital advertisement delivery and programmed advertisement control, in particular to an advertisement delivery risk control method based on big data analysis. Background With the rapid development of digital advertising and programmed delivery technologies, the advertising delivery system has the characteristics of large data scale, multiple structure levels and frequent dynamic change of delivery strategies. Advertisers manage and control advertisements through multi-level structures such as accounts, delivery units, materials and the like, the delivery behaviors form complex business index sequences in the time dimension, the problems of abnormal fluctuation, strategy failure and risk injection in the delivery process are increasingly prominent, and the stability and the fund safety of advertisement delivery are directly influenced. Existing advertising risk control techniques rely mostly on single-index threshold decisions, rule triggers, or static model analysis, often to identify anomalies at a single level or on a single time scale. The method is difficult to describe the linkage relation among different service levels, can not distinguish the release abnormality caused by normal service fluctuation, strategy adjustment or external risk factors, and is easy to cause misjudgment or response lag. Meanwhile, part of technologies only pay attention to whether an abnormality occurs or not, but lack descriptions of an abnormal evolution process and an accumulated effect thereof, and are difficult to support risk state judgment under a continuous operation scene. In addition, in the risk treatment link in the prior art, a fixed rule or a simple feedback control mode is often adopted, and a structural decision basis driven by a risk state inference result is lacked, so that a definite corresponding relation between a control action and an actual risk source is lacked, and refined risk control is difficult to realize while the delivery efficiency is ensured, and the effectiveness and the reliability of the advertisement delivery risk control technology are restricted due to the fact that the method is particularly prominent in a multi-level, large-scale and online advertisement delivery scene. Therefore, how to provide an advertisement delivery risk control method based on big data analysis is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide an advertisement putting risk control method based on big data analysis, which is characterized in that a putting behavior time sequence is constructed around advertisement putting multisource behavior data, a multi-level time sequence structure is formed through level mapping and time aggregation processing, a Bayesian online variable point detection algorithm is introduced to construct a multi-level variable point detection structure, the abnormal evolution process of the putting behavior is continuously inferred by combining multi-level running length distribution and multi-level variable point posterior, a risk state structured representation is further constructed based on a multi-level running length state, a risk hypothesis posterior weight set and an abnormal accumulation amount, risk control action parameters are generated according to the risk state structured representation, deterministic modulation of putting rate parameters, audit intensity parameters and freezing path parameters is realized, and an online closed-loop risk control flow is formed through recharging updating of risk control action feedback. The invention can distinguish abnormal evolution states under different service levels and different risk generation mechanisms, and has the advantages of fine risk identification, continuous response and clear control logic. According to the embodiment of the invention, the advertisement putting risk control method based on big data analysis comprises the following steps: Acquiring advertisement putting multisource behavior data, executing time alignment processing and structuring processing, and constructing a putting behavior time sequence; Based on the delivery behavior time sequence, performing hierarchical mapping and time aggregation processing, and establishing a multi-hierarchy time sequence structure; Based on a Bayes online variable point detection algorithm, constructing a multi-level variable point detection structure, executing online variable point inference processing by combining a multi-level time sequence structure to generate multi-level running length distribution, and executing joint updating to generate a multi-level variable point posterior; Performing run length state determination processing based on the multi-level run length distribution and the multi-level change point posterior, determining a multi-level run length state; Based on the multi-