Ad text matching correction method and system for ad design

Abstract

This invention discloses a method and system for correcting advertising text collocation errors in advertising design. The method includes the following steps: constructing a time-series corpus and generating high-dimensional semantic vectors for the text; calculating the dynamic semantic correlation between the advertising text to be examined and the corpus to assess potential risks; classifying risk levels based on the risk assessment results and dynamically generating corresponding elastic error correction rule sets; and finally performing hierarchical error correction processing on the advertising text based on the elastic error correction rule sets. This invention, by constructing a time-series corpus, calculating dynamic semantic correlations, and generating elastic error correction rule sets, achieves dynamic perception and accurate error correction of advertising text collocation risks, solving the problem of low accuracy in identifying advertising text collocation errors in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a method and system for correcting the collocation of advertising text in advertising design. Background Technology

[0002] With the rapid development of the digital advertising industry, the creation and arrangement of text content in advertising design has become increasingly frequent and diverse. Advertising text not only needs to be grammatically correct but also needs to precisely match the brand tone, product characteristics, and marketing strategy within a specific context. However, in advertising design practice, there are often contradictions between the compliance of text arrangement, semantic appropriateness, and creative expression.

[0003] In related technologies, the inspection of advertising text mainly relies on keyword matching based on static rule bases or grammatical error correction using general natural language processing models. Static rule base methods scan text by creating a list of prohibited words, but they cannot identify expressions that pose a risk of commercial misrepresentation even if they do not contain prohibited words. While general language models can detect some grammatical errors, they lack the ability to perceive the dynamic changes in specific collocation rules and social consensus within the advertising field, easily misjudging creative rhetorical devices as errors. All these methods suffer from low accuracy in identifying collocation errors in advertising text, making it difficult to meet the high text quality requirements of the advertising design stage. Summary of the Invention

[0004] This application provides a method and system for correcting advertising text pairing errors in advertising design, which solves the problem of low accuracy in identifying advertising text pairing errors in the prior art, and realizes dynamic perception and accurate correction of advertising text pairing risks.

[0005] This application provides a method for correcting advertising text collocations in advertising design. This method is applied to an advertising text collocation correction system for advertising design, and includes: Obtain the text of the advertisement to be inspected, and collect social corpus data and historical regulatory case data in real time to build a time-series corpus containing timestamp information; Based on a time-series corpus, a pre-defined semantic vector model is used to process the text to be inspected and the text data in the corpus to generate corresponding high-dimensional semantic vectors. By performing temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus, dynamic semantic relevance is obtained, and a potential risk assessment value of the advertisement text to be inspected is generated based on the dynamic semantic relevance. Based on the potential risk assessment value, the advertising text to be inspected is classified into risk levels, and a set of flexible error correction rules corresponding to the risk levels is dynamically generated. Based on the set of flexible error correction rules, the advertising text to be inspected is subjected to hierarchical error correction processing.

[0006] Furthermore, the specific steps for constructing a time-series corpus containing timestamp information include: The timestamp information of each data item is extracted from real-time collected social corpus data and historical regulatory case data, and these data items are sorted from early to late according to the timestamp to form a time series corpus with a time series structure. In a time series structure, a corresponding timestamp is appended to each data item as metadata.

[0007] Furthermore, the step of using a preset semantic vector model to process the advertisement text to be inspected and the text data in the corpus to generate corresponding high-dimensional semantic vectors includes the following specific steps: Load the preset semantic vector model, perform word segmentation on the advertisement text to be inspected and extract semantic features to generate a fixed-dimensional high-dimensional semantic vector; The same word segmentation and feature extraction are applied to each text data item in the time series corpus to generate its own high-dimensional semantic vector. Among them, all high-dimensional semantic vectors have the same dimension.

[0008] Furthermore, the dynamic semantic relevance is obtained by performing temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus. Specific steps include: For an advertisement text to be inspected and its corresponding high-dimensional semantic vector Calculate its relationship with the time series corpus within the time window. Subset of the corpus Dynamic semantic relevance The calculation formula is as follows: ; in, For a subset of the corpus The first in corpus; For corpus The corresponding high-dimensional semantic vector; For corpus timestamp; This is the current timestamp; The preset time decay constant; For a subset of the corpus The number of words in the corpus, where e is a natural constant.

[0009] Furthermore, the specific steps for generating the potential risk assessment value of the advertisement text to be inspected based on dynamic semantic relevance include: A series of dynamic semantic relevances obtained based on periodic calculations Construct its time series data; The potential risk assessment value is calculated by statistically analyzing the moving average value of the time series data. The calculation formula is as follows: ; in, For dynamic semantic relevance In the short time window Moving average within; For dynamic semantic relevance In the long time window Moving average within; for In the long time window Within the standard deviation; It is a very small normal quantity.

[0010] Furthermore, based on the potential risk assessment value, the risk level of the advertisement text to be inspected is classified, and the specific steps include: Based on the potential risk assessment values, the advertising texts to be inspected are divided into four risk levels: When the potential risk assessment value is lower than the threshold, it is classified as a safety level; When the potential risk assessment value is greater than threshold one and less than or equal to threshold two, and the text content conforms to the existing rule base, it is classified as a warning level. If the potential risk assessment value is greater than threshold two, or if the text content matches the observation period rules, it will be classified as a warning level. If the text content violates a definitive rule, it is classified as prohibited.

[0011] Furthermore, the specific steps for dynamically generating the elastic error correction rule set corresponding to the risk level include: When the risk level is a warning level or a prohibition level, candidate error correction rules are automatically generated based on the historical processing case library; The steps for establishing a historical processing case library include: storing the text content, semantic vectors, application rules, and processing results of each processing record; Calculate the applicability score of the candidate error correction rules, and rank the candidate rules based on the applicability score. The higher the applicability score, the higher the priority of the rule. A matching threshold is set for each rule, and the matching threshold is dynamically adjusted based on the rule's weight coefficient. The sorted rule set is stored as an elastic error correction rule set, which includes rule content, priority order, matching threshold and weight coefficient. Version management is applied to candidate error correction rules, recording the creation timestamp and modification history of each rule; A canary release strategy was adopted to initially push the newly generated elastic error correction rule set to a portion of server traffic for testing; Monitor the error correction accuracy and recall rate in the test, and adjust the rule parameters based on the feedback of the indicators. Once the accuracy and recall metrics reach preset thresholds, the elastic error correction rule set will be expanded to all servers.

[0012] Furthermore, the hierarchical error correction processing performed on the advertisement text to be inspected based on the elastic error correction rule set includes the following specific steps: For texts related to security levels, approval is granted directly without any intervention; For text indicating a warning level, return potential risk alerts to the user interface without blocking them; For text at the warning level, issue a warning notification and suggest text modifications; For text at the prohibited level, enforce blocking and provide specific explanations of the violation.

[0013] This application provides an advertising text collocation error correction system for advertising design, which implements an advertising text collocation error correction method for advertising design, including: a time-series corpus construction module, a semantic vectorization processing module, a dynamic risk assessment module, and a hierarchical error correction processing module; The time-series corpus construction module is used to acquire the advertising text to be inspected and to collect social corpus data and historical regulatory case data in real time to construct a time-series corpus containing timestamp information. The semantic vectorization processing module is used to process the advertisement text to be inspected and the text data in the corpus based on the time series corpus and to generate corresponding high-dimensional semantic vectors using a preset semantic vector model. The dynamic risk assessment module is used to perform temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vector of each text data in the corpus to obtain dynamic semantic correlation, and generate a potential risk assessment value of the advertisement text to be inspected based on the dynamic semantic correlation. The hierarchical error correction processing module is used to classify the risk level of the advertisement text to be inspected based on the potential risk assessment value, and dynamically generate a set of flexible error correction rules corresponding to the risk level, and perform hierarchical error correction processing on the advertisement text to be inspected based on the set of flexible error correction rules.

[0014] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: The advertising text collocation correction method and system provided in this application constructs a time-series corpus and generates high-dimensional semantic vectors for the text. It calculates the dynamic semantic relevance between the advertising text to be examined and the corpus to assess potential risks. Based on the risk assessment results, it classifies risk levels and dynamically generates corresponding flexible correction rule sets. Finally, it performs hierarchical correction processing on the advertising text based on these rule sets. In this process, considering the dynamic changes in the compliance boundaries and semantic appropriateness of advertising text collocations with social consensus, it achieves accurate risk assessment of advertising texts by constructing a time-series corpus and calculating dynamic semantic relevance. Furthermore, by establishing a flexible correction rule set corresponding to risk levels, it achieves hierarchical processing of advertising texts, effectively identifying text collocations with commercial misleading risks while protecting reasonable creative expression, thus improving the accuracy and applicability of advertising text collocation correction. Attached Figure Description

[0015] Figure 1 A flowchart of an advertising text matching error correction method for advertising design provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of an advertising text matching error correction system for advertising design provided in an embodiment of this application. Detailed Implementation

[0016] This application provides a method and system for correcting advertising text pairing errors in advertising design, which solves the problem of low accuracy in identifying advertising text pairing errors in the prior art. By constructing a time-series corpus, calculating dynamic semantic relevance, and generating an elastic error correction rule set, it achieves dynamic perception and accurate error correction of advertising text pairing risks.

[0017] In related technologies, text inspection methods based on static rule bases fail to consider the dynamic nature of advertising compliance boundaries and semantic appropriateness in practical applications. The evolution of social consensus and adjustments in regulatory policies can lead to significant differences in the evaluation standards for text collocation. For example, certain expressions may be considered creative at one time but deemed inappropriate at another. Static rule bases cannot adapt to such dynamic changes. Inspection methods based on general language models lack an understanding of the specific context and social consensus in the advertising field, making them prone to misjudgments. Whether static rules or general models are used for advertising text collocation inspection, both will struggle to accurately identify collocation errors, leading to false positives or false negatives.

[0018] To address the aforementioned technical issues, this application provides a method for correcting advertising text collocations in advertising design. This method collects social corpora and regulatory cases in real time to construct a temporal corpus. Through semantic vectorization and temporal correlation calculations, it dynamically perceives the potential risks of advertising texts. Based on the risk assessment results, it generates a flexible set of correction rules, enabling differentiated processing of texts with different risk levels, thereby improving the accuracy and adaptability of advertising text collocation correction. To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] like Figure 1 As shown, this application provides a method for correcting advertising text collocation errors in advertising design. This method is applied to an advertising text collocation error correction system for advertising design, and includes: Obtain the text of the advertisement to be inspected, and collect social corpus data and historical regulatory case data in real time to build a time-series corpus containing timestamp information; Based on a time-series corpus, a pre-defined semantic vector model is used to process the text to be inspected and the text data in the corpus to generate corresponding high-dimensional semantic vectors. By performing temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus, dynamic semantic relevance is obtained, and a potential risk assessment value of the advertisement text to be inspected is generated based on the dynamic semantic relevance. Based on the potential risk assessment value, the advertising text to be inspected is classified into risk levels, and a set of flexible error correction rules corresponding to the risk levels is dynamically generated. Based on the set of flexible error correction rules, the advertising text to be inspected is subjected to hierarchical error correction processing.

[0020] The high-dimensional semantic vector, dynamic semantic correlation, and potential risk assessment value are all stored in the database along with their corresponding timestamps for continuous model iteration and retrospective analysis of historical risk events.

[0021] Furthermore, the specific steps for constructing a time-series corpus containing timestamp information include: The timestamp information of each data item is extracted from real-time collected social corpus data and historical regulatory case data, and these data items are sorted from early to late according to the timestamp to form a time series corpus with a time series structure. In the time series structure, a corresponding timestamp is attached to each data item as metadata. The time series corpus stores all data items and their timestamps.

[0022] The time series corpus is updated regularly by adding newly collected data items and maintaining the timestamp order. During the construction process, older data items are given lower weights to reflect dynamic changes in consensus; Verify the integrity of the time series corpus, ensuring that the timestamp of each data item corresponds one-to-one with the data content; The completed time-series corpus is stored in a database. The update frequency of the time-series corpus can be set to once per hour, the data sampling frequency to collect social corpus data once per minute, and the sampling frequency of historical regulatory case data to once per day.

[0023] Furthermore, the step of using a preset semantic vector model to process the advertisement text to be inspected and the text data in the corpus to generate corresponding high-dimensional semantic vectors includes the following specific steps: Load the preset semantic vector model, perform word segmentation on the advertisement text to be inspected and extract semantic features to generate a fixed-dimensional high-dimensional semantic vector; The same word segmentation and feature extraction are applied to each text data item in the time series corpus to generate its own high-dimensional semantic vector. Among them, all high-dimensional semantic vectors have the same dimension.

[0024] In one specific embodiment, the advertising text to be inspected, "This skincare product has excellent effects," is segmented into words to obtain the segmented word sequence ["this", "skincare product", "effects", "excellent"]. The semantic features of each word are extracted using a semantic vector model, and these features are weighted and combined to generate a 768-dimensional floating-point vector, for example, [0.12, -0.45, 0.83, ..., 0.67].

[0025] The steps involved in constructing a semantic vector model include: Prepare a training dataset containing text content and corresponding timestamps from social corpus data and historical regulatory case data; initialize the parameters of the pre-trained language model; input the training data into the model for forward propagation and calculate the semantic similarity loss; update the model parameters through the backpropagation algorithm and minimize the loss function; repeat the training process until the model converges; use the validation set to evaluate the model performance until the generated semantic vectors can accurately reflect the semantic relationships of the text.

[0026] The text data in the corpus refers to a combination of social corpus data and historical regulatory case data, with each data item containing text content and a timestamp.

[0027] Furthermore, the dynamic semantic relevance is obtained by performing temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus. Specific steps include: For an advertisement text to be inspected and its corresponding high-dimensional semantic vector Calculate its relationship with the time series corpus within the time window. Subset of the corpus Dynamic semantic relevance The calculation formula is as follows: ; in, For a subset of the corpus The first in corpus; For corpus The corresponding high-dimensional semantic vector; For corpus timestamp; This is the current timestamp; This is a preset time decay constant used to adjust the weight of historical data; its value ranges from [0.01, 0.5]. A larger value indicates a higher weighting of recent data, reflecting the dynamic temporal changes in semantic relevance; For a subset of the corpus The number of words in the corpus, where e is a natural constant.

[0028] The time decay constant is obtained as follows: Select corpus data from historical time periods and calculate the semantic similarity between corpora at different time points and the current corpus; Plot a decay scatter plot with time interval on the x-axis and semantic similarity on the y-axis. The exponential decay curve was fitted using the least squares method. The one with the highest goodness of fit was obtained. value; This The value is used as a preset time decay constant.

[0029] Repeat this process to cover multiple time windows to achieve precise quantification of the evolving trend of social consensus; The system stores the calculated dynamic semantic relevance and its corresponding timestamp. The time window length is set to range from seven to ninety days, and the data sampling frequency is to calculate the dynamic semantic relevance once per hour.

[0030] Furthermore, the specific steps for generating the potential risk assessment value of the advertisement text to be inspected based on dynamic semantic relevance include: A series of dynamic semantic relevances obtained based on periodic calculations Values, constructing their time series data; The potential risk assessment value is calculated by statistically analyzing the moving average value of the time series data. The calculation formula is as follows: ; in, For dynamic semantic relevance In the short time window Moving average within; For dynamic semantic relevance In the long time window Moving average within; for In the long time window The standard deviation within the range is used to measure data volatility; For a very small normal quantity, for example This is used to prevent the denominator from being zero.

[0031] The distinction between the short-term and long-term time windows is determined through the following steps: Obtain historical dynamic semantic relevance time series data, which includes daily calculated semantic relevance values; Calculate the autocorrelation function of dynamic semantic correlation time series; Identify the time span corresponding to the first local minimum in the autocorrelation function and use that time span as the length of the short-term time window; Calculate the time span corresponding to when the autocorrelation function decays to one-tenth of its initial value, and use this time span as the length of the long-term time window.

[0032] In one specific embodiment, a short-term window of seven days and a long-term window of ninety days are set to compare changes in short-term and long-term semantic consensus; potential risk assessment values ​​and their timestamps are stored for trend prediction of dynamic risks.

[0033] Furthermore, based on the potential risk assessment value, the risk level of the advertisement text to be inspected is classified, and the specific steps include: Based on the potential risk assessment values, the advertising texts to be inspected are divided into four risk levels: When the potential risk assessment value is lower than the threshold, it is classified as a safety level; When the potential risk assessment value is greater than threshold one and less than or equal to threshold two, and the text content conforms to the existing rule base, it is classified as a warning level; the existing rule base is a set of confirmed compliance rules extracted from historical regulatory cases and social corpora; the steps to establish the existing rule base include: extracting text data from historical regulatory cases and social corpora, calculating the semantic features of the text through a semantic vector model, and storing the compliance rules that have been manually reviewed and confirmed as the existing rule base; When the potential risk assessment value is greater than the threshold two, or the text content hits the observation period rules, it is classified as a warning level; the observation period rules are a set of rules that are identified by dynamic semantic relevance calculation, indicating that the risk has increased but the violation has not been confirmed; the steps to establish the observation period rules include: identifying text expressions with increased risk through dynamic semantic relevance calculation, and temporarily storing rules whose risk trend coefficient exceeds the set standard but has not reached the level of violation as observation period rules. When the text content matches a definitive violation rule, it is classified as prohibited. A definitive violation rule is a set of clear violation rules extracted from publicly disclosed penalty cases by regulatory authorities. The steps to establish a definitive violation rule include: extracting violation text expressions from publicly disclosed penalty cases by regulatory authorities, establishing a violation expression feature library through a semantic vector model, and storing violation rules confirmed by legal experts as definitive violation rules.

[0034] Security level: Warning Level: The text content is compliant with existing rules; Warning level: Or, the text content hits the observation period rule; Prohibition level: the text content hits the certainty violation rule.

[0035] Furthermore, the specific steps for dynamically generating the elastic error correction rule set corresponding to the risk level include: When the risk level is a warning level or a prohibition level, candidate error correction rules are automatically generated based on the historical processing case library; The steps for establishing a historical processing case library include: storing the text content, semantic vectors, application rules, and processing results of each processing record; The applicability score of candidate error correction rules is calculated through the following steps: calculating the cosine similarity of the semantic vectors between the current text and historical cases, counting the number of successful rule applications, and then weighting the results to obtain the applicability score. Specifically, the applicability score of the candidate error correction rules is calculated. Use the formula: ; in, Representation rules The applicability score; Indicates the current text and rules Average semantic vector cosine similarity of relevant historical cases; Representation rules Number of successful applications; and These are preset weighting coefficients that satisfy... .

[0036] Candidate rules are ranked based on their applicability scores; the higher the applicability score, the higher the rule priority. A matching threshold is set for each rule, and the matching threshold is dynamically adjusted based on the rule weight coefficient. The calculation steps for the rule weight coefficient include: counting the number of rule triggers and the number of correct judgments, calculating the weight coefficient value based on accuracy and recall metrics, and specifically setting a matching threshold for each rule based on the rule weight coefficient. Dynamically adjusted, including rule weight coefficients The calculation uses the following formula: ; in, Representation rules Weighting coefficients; Representation rules accuracy, ; Representation rules Recall rate, ; Representation rules The number of correct judgments; Representation rules The number of error triggers; Representation rules The number of times that should have been triggered but were not.

[0037] The sorted rule set is stored as an elastic error correction rule set, which includes rule content, priority order, matching threshold and weight coefficient. Version management is applied to candidate error correction rules, recording the creation timestamp and modification history of each rule; A canary release strategy was adopted to initially push the newly generated elastic error correction rule set to a portion of server traffic for testing; Monitor the error correction accuracy and recall rate in the test, and adjust the rule parameters based on the feedback of the indicators. Once the accuracy and recall metrics reach preset thresholds, the elastic error correction rule set will be expanded to all servers.

[0038] Furthermore, the hierarchical error correction processing performed on the advertisement text to be inspected based on the elastic error correction rule set includes the following specific steps: For texts related to security levels, approval is granted directly without any intervention; For text indicating a warning level, return potential risk alerts to the user interface without blocking them; For text at the warning level, issue a warning notification and suggest text modifications; For text at the prohibited level, enforce blocking and provide specific explanations of the violation.

[0039] Provide detailed explanations of violations, including: identifying the specific location of the violation in the text; explaining the specific content of the violation rule; citing relevant regulatory basis; and providing suggested directions for modification. Record a log for each action, including the risk level and applied rules.

[0040] In one specific embodiment, a flexible error correction rule set is applied during processing to ensure adaptation to dynamic consensus changes. The testing process employs an A / B testing method, allocating traffic at a 50 / 50 ratio; the processing frequency is real-time execution, triggered each time a text is submitted.

[0041] like Figure 2 As shown, this application embodiment provides an advertising text collocation error correction system for advertising design, which is used to implement the advertising text collocation error correction method for advertising design, including: a time-series corpus construction module, a semantic vectorization processing module, a dynamic risk assessment module, and a hierarchical error correction processing module; The time-series corpus construction module is used to acquire the advertising text to be inspected and to collect social corpus data and historical regulatory case data in real time to construct a time-series corpus containing timestamp information. The semantic vectorization processing module is used to process the advertisement text to be inspected and the text data in the corpus based on the time series corpus and to generate corresponding high-dimensional semantic vectors using a preset semantic vector model. The dynamic risk assessment module is used to perform temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vector of each text data in the corpus to obtain dynamic semantic correlation, and generate a potential risk assessment value of the advertisement text to be inspected based on the dynamic semantic correlation. The hierarchical error correction processing module is used to classify the risk level of the advertisement text to be inspected based on the potential risk assessment value, and dynamically generate a set of flexible error correction rules corresponding to the risk level, and perform hierarchical error correction processing on the advertisement text to be inspected based on the set of flexible error correction rules.

[0042] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0043] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a program product.

[0044] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0045] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0046] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0047] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for correcting errors in advertising text collocation used in advertising design, characterized in that, Includes the following steps: Obtain the text of the advertisement to be inspected, and collect social corpus data and historical regulatory case data in real time to build a time-series corpus containing timestamp information; Based on a time-series corpus, a pre-defined semantic vector model is used to process the text to be inspected and the text data in the corpus to generate corresponding high-dimensional semantic vectors. By performing temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus, dynamic semantic relevance is obtained, and a potential risk assessment value of the advertisement text to be inspected is generated based on the dynamic semantic relevance. Based on the potential risk assessment value, the advertising text to be inspected is classified into risk levels, and a set of flexible error correction rules corresponding to the risk levels is dynamically generated. Based on the set of flexible error correction rules, the advertising text to be inspected is subjected to hierarchical error correction processing.

2. The advertising text matching error correction method for advertising design as described in claim 1, characterized in that, The specific steps for constructing a time-series corpus containing timestamp information include: The timestamp information of each data item is extracted from real-time collected social corpus data and historical regulatory case data, and these data items are sorted from early to late according to the timestamp to form a time series corpus with a time series structure. In a time series structure, a corresponding timestamp is appended to each data item as metadata.

3. The advertising text matching error correction method for advertising design as described in claim 1, characterized in that, The process of generating corresponding high-dimensional semantic vectors by using a preset semantic vector model to process the advertisement text to be inspected and the text data in the corpus includes the following steps: Load the preset semantic vector model, perform word segmentation on the advertisement text to be inspected and extract semantic features to generate a fixed-dimensional high-dimensional semantic vector; The same word segmentation and feature extraction are applied to each text data item in the time series corpus to generate its own high-dimensional semantic vector. Among them, all high-dimensional semantic vectors have the same dimension.

4. The method for correcting advertising text collocation in advertising design as described in claim 1, characterized in that, The process involves performing a temporal correlation calculation between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vectors of various text data in the corpus to obtain dynamic semantic relevance. Specific steps include: For an advertisement text to be inspected and its corresponding high-dimensional semantic vector Calculate its relationship with the time series corpus within the time window. Subset of the corpus Dynamic semantic relevance The calculation formula is as follows: ； in, For a subset of the corpus The first in corpus; For corpus The corresponding high-dimensional semantic vector; For corpus timestamp; This is the current timestamp; The preset time decay constant; For a subset of the corpus The number of words in the corpus, where e is a natural constant.

5. The method for correcting advertising text collocation in advertising design as described in claim 1, characterized in that, The specific steps for generating the potential risk assessment value of the advertisement text to be inspected based on dynamic semantic relevance include: A series of dynamic semantic relevances obtained based on periodic calculations Construct its time series data; The potential risk assessment value is calculated by statistically analyzing the moving average value of the time series data. The calculation formula is as follows: ； in, For dynamic semantic relevance In the short time window Moving average within; For dynamic semantic relevance In the long time window Moving average within; for In the long time window Within the standard deviation; It is a very small normal quantity.

6. The method for correcting advertising text collocation in advertising design as described in claim 1, characterized in that, The process of classifying the risk level of the advertising text to be inspected based on the potential risk assessment value includes the following steps: Based on the potential risk assessment values, the advertising texts to be inspected are divided into four risk levels: When the potential risk assessment value is lower than the threshold, it is classified as a safety level; When the potential risk assessment value is greater than threshold one and less than or equal to threshold two, and the text content conforms to the existing rule base, it is classified as a warning level. If the potential risk assessment value is greater than threshold two, or if the text content matches the observation period rules, it will be classified as a warning level. If the text content violates a definitive rule, it is classified as prohibited.

7. The method for correcting advertising text collocation in advertising design as described in claim 1, characterized in that, The specific steps for dynamically generating a set of elastic error correction rules corresponding to the risk level include: When the risk level is a warning level or a prohibition level, candidate error correction rules are automatically generated based on the historical processing case library; The steps for establishing a historical processing case library include: storing the text content, semantic vectors, application rules, and processing results of each processing record; Calculate the applicability score of the candidate error correction rules, and rank the candidate rules based on the applicability score. The higher the applicability score, the higher the priority of the rule. A matching threshold is set for each rule, and the matching threshold is dynamically adjusted based on the rule's weight coefficient. The sorted rule set is stored as an elastic error correction rule set, which includes rule content, priority order, matching threshold and weight coefficient. Version management is applied to candidate error correction rules, recording the creation timestamp and modification history of each rule; A canary release strategy was adopted to initially push the newly generated elastic error correction rule set to a portion of server traffic for testing; Monitor the error correction accuracy and recall rate in the test, and adjust the rule parameters based on the feedback of the indicators. Once the accuracy and recall metrics reach preset thresholds, the elastic error correction rule set will be expanded to all servers.

8. The method for correcting advertising text collocation in advertising design as described in claim 1, characterized in that, The hierarchical error correction processing based on the elastic error correction rule set for the advertisement text to be inspected includes the following steps: For texts related to security levels, approval is granted directly without any intervention; For text indicating a warning level, return potential risk alerts to the user interface without blocking them; For text at the warning level, issue a warning notification and suggest text modifications; For text at the prohibited level, enforce blocking and provide specific explanations of the violation.

9. An advertising text matching error correction system for advertising design, used to implement the advertising text matching error correction method for advertising design as described in any one of claims 1-8, characterized in that, include: The module includes a time-series corpus construction module, a semantic vectorization processing module, a dynamic risk assessment module, and a hierarchical error correction processing module. The time-series corpus construction module is used to acquire the advertising text to be inspected and to collect social corpus data and historical regulatory case data in real time to construct a time-series corpus containing timestamp information. The semantic vectorization processing module is used to process the advertisement text to be inspected and the text data in the corpus based on the time series corpus and to generate corresponding high-dimensional semantic vectors using a preset semantic vector model. The dynamic risk assessment module is used to perform temporal correlation calculations between the high-dimensional semantic vector of the advertisement text to be inspected and the high-dimensional semantic vector of each text data in the corpus to obtain dynamic semantic correlation, and generate a potential risk assessment value of the advertisement text to be inspected based on the dynamic semantic correlation. The hierarchical error correction processing module is used to classify the risk level of the advertisement text to be inspected based on the potential risk assessment value, and dynamically generate a set of flexible error correction rules corresponding to the risk level, and perform hierarchical error correction processing on the advertisement text to be inspected based on the set of flexible error correction rules.

Images