An internet advertisement promotion strategy optimization method and system

Abstract

The application discloses an internet advertisement promotion strategy optimization method and system, relates to the technical field of internet advertisement promotion, and aims to solve the problems of real-time data processing delay and information loss in the existing internet advertisement promotion strategy, thereby affecting the stability and improvement of the advertisement investment return rate. The method comprises the following steps: collecting real-time operation data of advertisement audiences on the side of an advertisement investment platform and external public opinion trend information, identifying structural deviation of the behavior mode of the advertisement audiences according to the interactive behavior sequence and the behavior context identifier in the platform, obtaining behavior deviation features, associating and matching the behavior deviation features with the external public opinion trend information, obtaining an association verification result, dynamically correcting the portrait label set of the advertisement audiences according to the association verification result, obtaining an updated audience portrait, and generating and issuing an adjusted advertisement investment strategy according to the updated audience portrait.

Description

Technical Field

[0001] This application relates to the field of internet advertising and promotion technology, and in particular to a method and system for optimizing internet advertising and promotion strategies. Background Technology

[0002] In the field of internet advertising, a significant challenge lies in making advertising strategies more intelligent, reducing reliance on human experience, and rapidly adapting to the ever-changing market environment and competitive landscape. Currently, advertising effectiveness often fluctuates significantly due to untimely or inaccurate strategy adjustments. These problems become more pronounced, especially when there are subtle deviations in data processing, structural changes in user behavior patterns, or conflicts between human intervention and automated systems. These issues can even lead to delays and information loss in real-time data processing, making it difficult to stabilize and improve the overall return on investment (ROI) of advertising campaigns. Summary of the Invention

[0003] This application provides a method and system for optimizing internet advertising promotion strategies, aiming to solve the problems of real-time data processing delays and information loss caused by data processing deviations, structural changes in user behavior patterns, and conflicts between manual intervention and automated systems, which in turn affect the stability and improvement of advertising return on investment.

[0004] Firstly, to address the aforementioned technical problems, this invention provides a method for optimizing internet advertising promotion strategies. This method includes: collecting real-time operational data of advertising audiences on the advertising platform side and external public opinion trend information. The advertising audience refers to end users receiving advertising pushes. The real-time operational data includes a sequence of interactive behaviors within the platform and corresponding behavioral context identifiers. The sequence of interactive behaviors within the platform records browsing, clicking, and searching actions generated by the advertising audience on the advertising platform side. The external public opinion trend information includes trending topics on third-party content publishing platforms and corresponding trend popularity values. Based on the sequence of interactive behaviors within the platform and the behavioral context identifiers, the method identifies the structural patterns of the advertising audience's behavior. The behavioral shift features are derived from the offset analysis. These features characterize the degree of migration in content consumption preferences or conversion path selection among advertising audiences. The behavioral shift features are then correlated and matched with external public opinion trend information to obtain correlation verification results. These results characterize the causal relationship between the structural shift in behavioral patterns and external trending topics. Based on the correlation verification results, the set of profile tags for advertising audiences is dynamically revised to obtain an updated audience profile. This profile tag set includes interest dimension tags and media touchpoint tags. Based on the updated audience profile, an adjusted advertising strategy is generated and deployed. This strategy includes the display format of advertising materials, bid amount, and channel allocation ratio.

[0005] Secondly, this application provides an internet advertising promotion strategy optimization system, which includes: a data collection unit for collecting real-time operation data of advertising audiences on the advertising platform and external public opinion trend information. The advertising audience refers to the terminal users who receive advertising pushes. The real-time operation data includes a sequence of interactive behaviors within the platform and corresponding behavioral context identifiers. The sequence of interactive behaviors within the platform is used to record browsing, clicking, and searching actions generated by the advertising audience on the advertising platform. The external public opinion trend information includes hot topics on third-party content publishing platforms and corresponding trend popularity values; and an identification unit for identifying structural shifts in the behavioral patterns of advertising audiences based on the sequence of interactive behaviors within the platform and behavioral context identifiers, thereby obtaining behavioral shifts. The system comprises several components: a behavioral shift feature to characterize the degree of migration in content consumption preferences or conversion path selection among advertising audiences; a matching unit to correlate behavioral shift features with external public opinion trend information to obtain correlation verification results, which characterize the strength of the causal relationship between structural shifts in behavioral patterns and external hot topics; a correction unit to dynamically correct the set of profile tags for advertising audiences based on the correlation verification results, resulting in an updated audience profile, which includes interest dimension tags and media touchpoint tags; and a generation unit to generate and distribute adjusted advertising strategies based on the updated audience profile, which include material display formats, bid amounts, and channel allocation ratios.

[0006] This application has at least the following beneficial effects: The internet advertising promotion strategy optimization method disclosed in this application can comprehensively and timely obtain user behavior and market hotspot data by collecting real-time operational data of advertising audiences on the advertising platform and external public opinion trend information. Furthermore, this method effectively captures dynamic changes in user interests by identifying structural shifts in user behavior patterns and quantifying the degree of user migration in content consumption preferences and conversion path selection. Based on this, by correlating and matching behavioral shift features with external public opinion trend information, it can reveal the strength of the causal relationship between changes in user behavior and external hot topics, thereby gaining a deeper understanding of the driving factors behind user behavior. Finally, the ad audience profile tag set is dynamically adjusted based on the correlation verification results, and an adjusted advertising strategy is generated and distributed based on the updated audience profile. Attached Figure Description

[0007] Figure 1 This is one of the flowcharts illustrating an internet advertising promotion strategy optimization method provided in this application; Figure 2 This is one of the flowcharts illustrating an internet advertising promotion strategy optimization method provided in this application; Figure 3 This is a schematic diagram of the structure of an Internet advertising promotion strategy optimization system provided in this application. Detailed Implementation

[0008] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0009] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0010] In the field of internet advertising, traditional advertising strategy optimization methods often exhibit lag and inaccuracy when faced with rapid changes in the market environment, structural shifts in user behavior patterns, and subtle biases in data processing. For example, during data integration and migration, inconsistent timestamps disrupt the sequence of user behavior, thus affecting the accuracy of click-through rate (CTR) prediction methods. Simultaneously, the rise of new platforms has caused structural shifts in user behavior patterns, which existing methods struggle to adapt to quickly. Furthermore, the interaction between frequent human intervention and system defects exacerbates the volatility of advertising effectiveness, and delays and data loss in the real-time data processing chain further deteriorate data quality, ultimately leading to a significant decline in advertising ROI.

[0011] In view of the above problems, this application provides a method for optimizing Internet advertising promotion strategies. This application aims to achieve intelligent and real-time optimization of advertising promotion strategies by introducing the identification of structural shifts in user behavior patterns, correlation and matching with external public opinion trend information, and dynamic correction mechanism of audience profiles. This will effectively address the challenges brought about by market changes and user behavior migration, and improve the accuracy and return on investment of advertising.

[0012] The following specific embodiments will provide a detailed introduction and explanation of the Internet advertising promotion strategy optimization method provided in this application.

[0013] Reference Figure 1 This application provides a method for optimizing internet advertising promotion strategies, which may include the following steps: S101. Collect real-time operational data of advertising audiences on the advertising platform and external public opinion trend information.

[0014] Among them, the advertising audience is the end users who receive the advertising push, and the real-time operation data includes the platform interaction behavior sequence and the corresponding behavior context identifier. The platform interaction behavior sequence is used to record the browsing, clicking and search actions generated by the advertising audience on the advertising platform side. The external public opinion trend information includes hot topics on third-party content publishing platforms and the corresponding trend popularity value.

[0015] To better understand the technical solution proposed in this application, some key terms involved will be explained first.

[0016] "Advertising audience" refers to the end users who receive advertising pushes, and is the target group for advertising.

[0017] "Real-time action data" refers to the instantaneous behavioral data generated by advertising audiences on the advertising platform side. Its core is to reflect the user's current activity status and interests.

[0018] "Intra-platform interaction behavior sequence" is an important component of real-time operational data. It records a series of actions taken by the advertising audience within the advertising platform, such as which ads they viewed, which links they clicked, and which keywords they searched. These actions are arranged in chronological order and constitute the user's behavioral trajectory within the platform.

[0019] "Behavioral context identifiers" are auxiliary information associated with the sequence of interactive behaviors within the platform. They provide the environmental context in which the behavior occurs, such as the type of page the user is on, device information, and geographical location. These identifiers help to more accurately understand the intent behind the user's behavior.

[0020] "External public opinion trend information" refers to publicly available discussion data from third-party content publishing platforms (such as social media, news websites, short video platforms, etc.). It reflects current social hotspots, popular trends, and public attention to specific topics.

[0021] "Hot topics" are the core content of external public opinion trend information, referring to events, people or concepts that receive widespread attention and discussion within a specific time period.

[0022] The "Trend Popularity Value" is used to quantify the popularity and spread speed of trending topics. It is usually calculated by statistically analyzing indicators such as the frequency of mentions, interaction volume, and breadth of dissemination of the topic.

[0023] "Structural shift in behavioral patterns" refers to significant changes in advertising audiences' content consumption preferences or conversion path choices. These changes are not simply fluctuations in interest, but rather deep-seated and somewhat persistent shifts in behavioral patterns.

[0024] "Behavioral shift features" are the result of identifying structural shifts in behavioral patterns. They quantitatively characterize the magnitude of the shift in user interest focus or the degree of shift in the conversion path.

[0025] "Association verification results" are the product of matching behavioral deviation features with external public opinion trend information. They are used to assess whether there is a causal relationship between the structural deviation of user behavior patterns and external hot topics, and to quantify the strength of this relationship.

[0026] A “profile tag set” is a set of tags that describe the characteristics of an advertising audience, including “interest dimension tags” (such as preferences for specific products, services or content) and “media touchpoint tags” (such as the types of apps, websites or devices that users frequently use).

[0027] "Dynamic correction" refers to adjusting the set of profile tags in real time or near real time based on the results of correlation verification to ensure the accuracy and timeliness of the audience profile.

[0028] The "updated audience profile" is a dynamically revised set of profile tags that more accurately reflects the current content consumption preferences and conversion paths of the advertising audience.

[0029] "Ad placement strategy" is a specific plan that guides ad placement, including "material display format" (such as images, videos, text links, etc.), "bid amount" (the cost paid for each click, impression, or conversion), and "channel allocation ratio" (the budget allocation of ads across different placement channels).

[0030] Specifically, various methods can be used to collect real-time operational data of advertising audiences on advertising platforms and external public opinion trend information. For example, data collection agents can be deployed on various front-end pages and back-end services of the advertising platform to capture the browsing, clicking, and searching behaviors of advertising audiences in real time. This data, along with contextual identifiers (such as device ID, IP address, page URL, and timestamp), is then encapsulated into data packets and transmitted to the data processing center via message queues. Simultaneously, specialized web crawler systems can be used to periodically or in real-time crawl publicly discussed data from mainstream social media platforms, news aggregation websites, and short video platforms. These crawler systems can be configured with keyword lists to specifically crawl trending topics related to the advertising industry or specific product areas, extracting information such as titles, text, comments, and publication times. Subsequently, natural language processing technology is used to analyze the crawled text data, identify trending topics, and calculate their trend popularity value based on indicators such as the frequency of mentions, number of reposts, and number of comments.

[0031] S102. Based on the platform's interactive behavior sequence and behavioral context identifiers, identify the structural shifts in the behavioral patterns of the advertising audience to obtain behavioral shift features.

[0032] Among them, behavioral shift features are used to characterize the degree of migration of advertising audiences in terms of content consumption preferences or conversion path selection.

[0033] When identifying structural shifts in the behavioral patterns of advertising audiences based on in-platform interaction sequences and behavioral context identifiers, and obtaining behavioral shift features, the following approach can be used. For example, a behavioral pattern analysis module can be constructed, which receives pre-processed in-platform interaction sequences and behavioral context identifiers. First, this module can statistically analyze the distribution of advertising audience behavior types across different time windows, for example, calculating the proportion of users browsing and clicking on different content categories over the past week. Then, the behavioral type distribution of the current monitoring window (e.g., the most recent day) is compared with the behavioral type distribution of a historical baseline window (e.g., the past month), and the difference between the two is calculated. If the difference exceeds a preset threshold, it indicates a possible structural shift in user content consumption preferences. Furthermore, this module can also monitor the number of clicks on external links within the user behavior sequence. If a significant increase in the number of external link clicks is found in a short period, and the growth rate exceeds a preset threshold, it may mean that users are exploring new information sources or conversion paths, resulting in a shift in conversion paths. Combining these analytical results, a quantitative behavioral shift feature can be generated, which characterizes the degree of migration in user content consumption preferences or conversion path selection.

[0034] S103. The behavioral deviation features are correlated and matched with external public opinion trend information to obtain the correlation verification results.

[0035] Among them, the correlation verification results are used to characterize the strength of the causal association between the structural shift in behavioral patterns and external hot topics.

[0036] When correlating and matching behavioral shift features with external public opinion trend information to obtain correlation verification results, the following methods can be used. For example, a correlation matching engine can be designed. This engine first extracts the key behavioral patterns or points of interest that lead to the shift from the behavioral shift features. For instance, if the behavioral shift features indicate a significant increase in user attention to "smart wearable devices," then "smart wearable devices" becomes a key behavioral keyword. Simultaneously, the engine extracts current trending topics and their keyword sets from external public opinion trend information. Then, these key behavioral keywords are compared with the keyword sets in the external public opinion trend information to find overlaps. If it is found that the key keywords of user behavioral shifts highly overlap with the keywords of external trending topics, and the trend popularity value of the external trending topics is high, then it can be determined that there is a strong causal relationship between the structural shift in user behavioral patterns and the external trending topics. The correlation verification result can be a quantitative indicator, such as a correlation strength score, to characterize the strength of this causal relationship.

[0037] S104. Based on the association verification results, dynamically correct the set of profile tags for the advertising audience to obtain the updated audience profile.

[0038] The profile tag set includes interest dimension tags and media touchpoint tags.

[0039] When dynamically revising the audience profile tag set based on association verification results to obtain an updated audience profile, the following approach can be used. For example, when association verification results indicate a strong correlation between a structural shift in user behavior patterns and a certain external trending topic, the revision module extracts core keywords from that trending topic and compares them with the original audience profile tag set. If these core keywords are found to be absent from the original tag set, they are added as new interest tags to the audience profile. If these core keywords have a high semantic similarity to a tag in the original tag set, the weight of that original tag is increased, indicating that the user's preference for that interest is strengthened. For example, if a user behavior shift is strongly correlated with the trending topic "metaverse games," but the original profile does not have the "metaverse" tag, then the "metaverse" tag is added; if the original profile has the "games" tag, then the weight of the "games" tag is increased. In this way, the audience profile can reflect changes in user interests in real time, resulting in an updated audience profile.

[0040] S105. Based on the updated audience profile, generate and distribute the adjusted advertising strategy.

[0041] The advertising strategy includes the format of the advertising materials, the bid amount, and the allocation ratio of channels.

[0042] When generating and deploying adjusted advertising strategies based on updated audience profiles, the following methods can be used. For example, the strategy generation module receives the updated audience profile and combines it with the advertiser's campaign objectives (such as increasing click-through rate, conversion rate, or brand exposure). Based on the updated audience profile, this module can intelligently adjust various parameters of the advertising strategy. For instance, if the updated audience profile shows a significant increase in user interest in "short video content," and media touchpoint tags indicate that users are active on a particular short video platform, the system may suggest increasing the proportion of advertising budget allocated to short video platforms and prioritizing video creative display formats. Simultaneously, based on the intensity of user interest in specific products or services, the system dynamically adjusts ad bids to ensure optimal performance in a highly competitive advertising market. Finally, the generated adjusted advertising strategy is deployed to various advertising channels through the advertising delivery system to guide the actual display of ads.

[0043] Through the steps described above, the method of this application forms an intelligent and adaptive closed loop for optimizing advertising promotion strategies. It effectively solves problems in existing technologies such as data deviations caused by inconsistent timestamps, difficulty in capturing structural changes in user behavior patterns, conflicts between human intervention and system predictions, and delays and loss in real-time data processing. Through deep insights into user behavior and intelligent correlation with the external environment, the method of this application can significantly improve the accuracy, stability, and return on investment of advertising, reduce reliance on human experience, and make adjustments to advertising strategies more intelligent and efficient.

[0044] In some of the embodiments described above in this application, such as Figure 2 As shown, the aforementioned collection of real-time operational data of advertising audiences on the advertising platform and external public opinion trend information includes: S201. Capture the operation logs of the advertising audience within a preset time period through the streaming computing engine, and analyze the sequence of interactive behaviors within the platform.

[0045] The platform's interactive behavior sequence includes behavior type identifiers and corresponding timestamps.

[0046] S202. Use web crawler components to selectively capture publicly available discussion data from mainstream social media platforms and short video platforms, and analyze it to obtain external public opinion trend information.

[0047] Specifically, a streaming computing engine refers to a computing framework capable of processing continuous data streams in real time. It is used to capture the operation logs of advertising audiences within a preset time period, aiming to ensure real-time monitoring and data acquisition of user behavior. Operation logs can be understood as the raw records of user actions such as browsing, clicking, and searching on the advertising platform. The platform-interactive behavior sequence refers to the set of behavioral events arranged chronologically after parsing these operation logs. Behavior type identifiers distinguish different user behaviors, such as "browsing," "clicking," and "searching," while the timestamp precisely records the time each behavior occurred, which is crucial for subsequent analysis of structural shifts in user behavior patterns.

[0048] The web crawler component can be understood as an automated program configured to selectively scrape publicly available discussion data from mainstream social media and short video platforms. This publicly available discussion data is a primary source of external public opinion trend information, such as public posts, comments, and topics on platforms like Weibo, Douyin, and Kuaishou. By analyzing this data, trending topics and their corresponding trend popularity values ​​can be extracted, thereby providing a comprehensive understanding of the impact of the external environment on user interests and behaviors.

[0049] Through the aforementioned technical solutions, this application enables more efficient and comprehensive collection of real-time operational data of advertising audiences and external public opinion trend information. The application of the streaming computing engine significantly improves the real-time nature of interactive behavior sequences within the platform, allowing structural shifts in user behavior patterns to be detected promptly and avoiding untimely strategy adjustments due to data lag. The targeted crawling capabilities of the web crawler component ensure comprehensive coverage and accurate acquisition of external hot topics and trend popularity values, enhancing the ability to perceive the impact of the external environment. Therefore, this application can provide more accurate and real-time input data for subsequent dynamic correction of audience profiles and generation of advertising delivery strategies, thereby significantly improving the optimization effect and response speed of advertising promotion strategies.

[0050] In some preferred embodiments, to collect real-time action data from advertising audiences, a streaming engine based on Apache Flink can be deployed. When advertising audiences browse products, click on ads, or search for keywords on an advertising platform (such as an e-commerce platform), these actions immediately generate log data and are sent to a Kafka message queue. Apache Flink, as the streaming engine, consumes the log data in Kafka in real time and parses it. For example, for a log record "User A clicked on product B on 2023-10-27 10:30:05", the streaming engine will extract the behavior type identifier as "click", the occurrence timestamp as "2023-10-27 10:30:05", and categorize it into User A's platform interaction behavior sequence.

[0051] Meanwhile, in order to obtain external public opinion trend information, a web crawler component based on the Scrapy framework may be configured. The crawler component is set to access mainstream social platforms and short video platforms such as the Weibo hot search list and the Douyin popular video list at regular intervals (for example, every hour). The crawler captures public discussion data on these platforms, including the titles and introductions of hot topics and related comment content. The captured data is then sent to a data processing module for parsing, to extract current hot topics (for example, "a certain celebrity's new drama is released", "a certain technology product is launched") and their corresponding trend popularity values (for example, comprehensively calculated according to the discussion volume, forwarding volume, like volume, etc. of the topics). These real-time and comprehensive data collection results will serve as an important basis for subsequent analysis and strategy optimization.

[0052] Specifically, the step of parsing to obtain the external public opinion trend information described above may be performed in the following manner.

[0053] The step of parsing to obtain the external public opinion trend information described above includes: performing data cleaning on the public discussion data, and extracting unstructured text, wherein the unstructured text is the titles, texts and comment content in the public discussion data; performing semantic word segmentation processing on the unstructured text, removing stop words, and extracting core entity words as the keyword set; counting the occurrence frequency and propagation rate of the core entity words within a preset time window, and performing normalization processing to obtain the popularity index value.

[0054] Herein, the data cleaning aims to remove noise, redundant information or non-standard content in the public discussion data, such as advertisements, repeated comments, special symbols, etc., so as to ensure the accuracy of subsequent analysis. Through data cleaning, unstructured text with actual analysis value can be effectively extracted from the original public discussion data. The unstructured text specifically includes titles, texts and comment content, which are the core carriers for users to express views and emotions.

[0055] Further, the purpose of performing semantic word segmentation processing on the unstructured text is to split the continuous text into words or phrases with independent semantics. In this process, stop words, such as commonly used words without actual semantics like "de", "shi", "zai", will be removed to reduce data dimensions and highlight core information. Through the semantic word segmentation processing, core entity words can be accurately extracted from the text. These core entity words constitute the keyword set, and represent the focus and theme of user discussions.

[0056] Furthermore, to quantify the popularity of external public opinion, statistical analysis of the core entity words in the keyword set is necessary. Specifically, the frequency of each core entity word's appearance within a preset time window will be counted to reflect its level of attention during a specific time period; simultaneously, its dissemination rate will also be counted to measure the speed and breadth of its spread. After normalization, these statistical results can be used to obtain the popularity index value, which can objectively characterize the trend popularity of external hot topics.

[0057] Through the aforementioned technical solutions, this application can obtain more accurate and reliable external public opinion trend information. Specifically, data cleaning avoids interference from noisy data in public opinion analysis; semantic word segmentation and core entity word extraction ensure an accurate grasp of the core content of hot topics; and comprehensive statistical analysis of frequency of occurrence and dissemination rate, followed by normalization, enables the popularity index value to more comprehensively and objectively reflect public opinion dynamics. This provides high-quality external data input for subsequent correlation matching of structural shifts in behavioral patterns, significantly improving the accuracy and timeliness of advertising promotion strategy optimization.

[0058] In some embodiments described above, a streaming computing engine captures the operation logs of advertising audiences within a preset time period and parses the interaction behavior sequences within the platform. However, in practical applications, the original operation log data may have data quality issues such as out-of-order or missing timestamps. If these issues are not addressed, the parsed interaction behavior sequences within the platform will not accurately reflect the actual behavioral trajectory and temporal order of the advertising audience, potentially leading to deviations in subsequent behavior pattern recognition, profile correction, and advertising delivery strategy generation, thus affecting the accuracy and effectiveness of advertising promotion. Therefore, this application further proposes optimizations to the above-mentioned method for parsing interaction behavior sequences within the platform to ensure data quality and analytical accuracy.

[0059] In this regard, this application further proposes that the steps for parsing the above-mentioned sequence of interactive behaviors within the platform include: The timestamps are checked for integrity. If there are out-of-order or missing timestamps, the out-of-order data is reordered according to the logical relationship between adjacent behaviors, and the missing data is filled in according to the preset completion rules to obtain the cleaned platform interaction behavior sequence.

[0060] Specifically, integrity verification refers to checking the timestamps associated with each action record in a sequence of interactive behaviors within the platform to ensure they conform to the expected format, range, and logical order. For example, it can check whether the timestamps are in a valid date and time format, whether they fall within a reasonable business timeframe, and whether the timestamps of adjacent actions show an increasing trend. Timestamp disorder refers to a situation in the action sequence where the timestamp of a subsequent action is earlier than or equal to the timestamp of a preceding action. Missing data refers to timestamp fields in some action records being empty or invalid. When timestamp disorder is detected, the disordered data is reordered according to the logical relationship between adjacent actions. For example, if the timestamp of a click action is earlier than its preceding browsing action, but browsing typically occurs before clicking from a business logic perspective, their order can be adjusted based on this logical relationship. The logical relationship between adjacent actions can be predefined; for example, browsing-click-purchase constitutes a typical conversion path. When missing timestamps are detected, the missing data is filled in according to preset completion rules. The completion rules can include, but are not limited to: estimating using the average timestamps of consecutive actions, inferring using the average interval of the user's historical actions, or setting a default timestamp based on the business scenario. Through the above verification, reordering, and filling processes, the final result is a cleaned sequence of platform interaction behaviors with continuous timestamps and correct logic, providing a high-quality foundation for subsequent data analysis.

[0061] The above technical solutions effectively address the issue of out-of-order or missing timestamps in raw operation log data, significantly improving the data quality and accuracy of interactive behavior sequences within the platform. This makes subsequent identification of structural shifts in advertising audience behavior patterns more accurate, avoiding misjudgments caused by data errors. Simultaneously, the cleaned behavior sequences more realistically reflect the content consumption preferences and conversion path choices of advertising audiences, providing a reliable basis for dynamically revising the audience profile tag set. Ultimately, this leads to more accurate and effective advertising strategies, significantly improving advertising efficiency and user experience.

[0062] In some preferred embodiments, a specific example is given below. Suppose that an advertising audience recorded the following sequence of behaviors in the operation log of the advertising platform (only the behavior type and timestamp are shown): 1. Browse product A (timestamp: T1) 2. Click on ad X (timestamp: T3) 3. Search keyword Y (timestamp: T2) 4. Browse product B (timestamp: T4) 5. Click on ad Z (timestamp: missing) The following data quality issues can be observed in this original sequence: The timestamp T3 of behavior 2 (clicking ad X) is later than the timestamp T2 of behavior 3 (searching keyword Y), but logically, the search usually occurs before or after the click, so there is an out-of-order situation here.

[0063] The timestamp for behavior 5 (clicking ad Z) is missing.

[0064] According to the scheme in this application, the integrity of the timestamps is first checked. It is found that the timestamps of behavior 2 and behavior 3 are out of order, and the timestamp of behavior 5 is missing. For the out-of-order data, it is reordered according to the logical relationship between adjacent behaviors. For example, the system recognizes that "searching for keyword Y" is usually a user-initiated exploration behavior, which may occur before "clicking ad X," or that "clicking ad X" is a result based on "searching for keyword Y." Assuming that the preset logical relationship indicates that the "search" behavior should precede the "click" behavior based on the search result, the order of behavior 3 and behavior 2 is adjusted, so that the sequence becomes: 1. Browse product A (timestamp: T1) 2. Search keyword Y (timestamp: T2) 3. Click on ad X (timestamp: T3) 4. Browse product B (timestamp: T4) For missing data, it is filled in according to preset completion rules. For example, the average time interval between consecutive actions can be used for estimation. Assuming that the interval between actions 4 and 5 is usually 5 seconds, the timestamp of action 5 is filled in as T4 + 5 seconds. Alternatively, if the average duration of the user's historical click actions is 10 seconds, the timestamp of action 5 can be inferred from the timestamp T4 of action 4. After the above processing, a cleaned sequence of interaction behaviors within the platform is finally obtained, with continuous timestamps and logical correctness, for example: 1. Browse product A (timestamp: T1) 2. Search keyword Y (timestamp: T2) 3. Click on ad X (timestamp: T3) 4. Browse product B (timestamp: T4) 5. Click on ad Z (timestamp: T4 + 5 seconds) This cleaned sequence can more accurately reflect the actual behavioral path and time sequence of the advertising audience, providing a reliable data foundation for subsequent advertising strategy optimization.

[0065] In some embodiments described above, this application proposes a structural shift in the behavioral patterns of advertising audiences based on in-platform interaction sequences and behavioral context identifiers. However, in practical applications, relying solely on a single dimension or coarse analysis method may fail to accurately capture subtle changes in user behavior patterns and their underlying causes, leading to insufficient accuracy in identifying behavioral shift features and thus affecting the optimization of subsequent advertising strategies. Therefore, this application further proposes a specific method for identifying structural shifts in the behavioral patterns of advertising audiences and obtaining behavioral shift features, in order to more comprehensively and precisely depict the migration of user behavior.

[0066] Based on the cleaned platform interaction behavior sequences and behavior context identifiers, structural shifts in the behavioral patterns of advertising audiences are identified, yielding behavioral shift features, specifically including: Based on the cleaned platform interaction behavior sequence, the distribution ratio of behavior types in the current monitoring window and the historical baseline window are statistically analyzed, and the difference between the two is calculated to obtain the percentage change. The percentage change is used to characterize the magnitude of the shift in user interest focus. The number of clicks pointing to external links in the cleaned platform interaction behavior sequence is detected. When the growth rate of the number of clicks exceeds a preset growth rate threshold, a path change marker is generated. The path change marker is used to characterize the offset of the conversion path. The behavior offset feature is generated by combining the percentage change and the path change marker.

[0067] Specifically, the cleaned platform-wide interaction behavior sequence refers to an ordered record of user browsing, clicking, and searching actions on the advertising platform after integrity verification, reordering, and data population. The current monitoring window can be understood as the set of user behavior data within a recent period (e.g., the past 24 hours, 7 days, or 30 days), while the historical baseline window refers to the set of user behavior data from an earlier period (e.g., the past 30 to 60 days), used to establish a normal baseline for user behavior. The behavior type distribution ratio refers to the proportion of different behavior types (such as browsing, clicking, and searching) among all behaviors within a specific time window. By calculating the difference in behavior type distribution ratios between the current monitoring window and the historical baseline window, the magnitude of the shift in user interest focus can be quantified. For example, if the proportion of a user's browsing behavior for a certain type of product increases significantly, it indicates that their interest may have shifted to that type of product.

[0068] The number of clicks on external links refers to the number of times a user clicks a link within the advertising platform that redirects to an external website or application. The preset growth rate threshold is a configurable parameter used to determine whether the increase in external clicks is abnormal or significant. When the growth rate of external clicks exceeds this threshold, it indicates that users may be seeking information or services outside the platform. This usually foreshadows a shift in their conversion path, such as moving from direct purchase within the platform to comparing prices or seeking more reviews externally. The generation of path change markers aims to clearly indicate this potential change in the user's conversion path.

[0069] In practical applications, behavioral shift features are generated by comprehensively considering both percentage changes and path change markers. For example, weighted summation, rule engines, or machine learning models can be used to fuse these two indicators. Percentage changes reflect shifts in user content consumption preferences, while path change markers reflect changes in the paths users choose to achieve their goals (such as purchases or registrations). By integrating information from both dimensions, a more comprehensive and accurate structural shift in advertising audience behavior patterns can be characterized.

[0070] The aforementioned technical solution significantly improves the accuracy and comprehensiveness of identifying structural shifts in advertising audience behavior patterns. Compared to identification methods relying solely on a single behavioral indicator, this solution combines the magnitude of shifts in user interest focus (changes in percentage) and shifts in conversion paths (path change markers) to gain deeper insights into the underlying motivations and trends behind user behavior. Consequently, the obtained behavioral shift features more accurately reflect the true migration in user content consumption preferences and conversion path choices, providing a more reliable basis for the subsequent dynamic correction of the advertising audience profile tag set. This allows the generated advertising delivery strategy to more precisely match users' current needs and behavioral habits, effectively improving advertising reach efficiency and conversion rates.

[0071] In some preferred embodiments, a specific example is given below. Suppose that an advertising audience primarily browsed and clicked on fashion apparel products on an e-commerce platform over the past month (historical baseline window), with a high proportion of "browsing fashion apparel" and "clicking on fashion apparel" in their behavior type distribution. However, in the most recent week (current monitoring window), this audience's behavior type distribution has changed, with a significant increase in the proportion of "browsing smart home" and "clicking on smart home" while their interest in fashion apparel has decreased. At this point, by calculating the difference in behavior type distribution proportions between the current monitoring window and the historical baseline window, a large change in proportion can be obtained, indicating that the audience's focus of interest has shifted from fashion apparel to smart home products.

[0072] Meanwhile, the system detected that this audience member had clicked on a large number of smart home product links to external review websites or brand websites within the past week, and the growth rate of these external clicks exceeded a preset growth rate threshold. This indicates that the audience's conversion path may be shifting from direct purchase within the platform to external information gathering and price comparison. In this case, the system will generate a path change marker.

[0073] Ultimately, by comprehensively considering this significant change in proportion and the generated path change markers, the system can generate a precise behavioral shift feature. This feature clearly indicates that the audience's interest has shifted towards smart home technology, and their conversion path tends to involve external decision-making. Based on this behavioral shift feature, the advertising platform can promptly adjust the audience's profile tags, such as adding a "smart home enthusiast" tag, and accordingly adjust its advertising strategies, such as pushing smart home-related ad creatives and optimizing delivery channels to adapt to their new conversion path preferences.

[0074] In some embodiments described above, the change in proportion is obtained by calculating the difference between the distribution ratios of behavior types within the current monitoring window and the historical baseline window, thus characterizing the magnitude of the shift in user interest focus. However, in practical applications, simple difference calculations may not fully reflect the dynamics and complexity of behavioral patterns, especially in internet advertising scenarios with massive and frequently changing data, potentially leading to inaccurate or insensitive judgments of user interest shifts.

[0075] In this regard, this application further proposes the following steps for calculating the difference between the two to obtain the change in proportion: The weighted sliding window algorithm is used to calculate the distribution ratio of behavior types in the current monitoring window and the historical baseline window, so as to obtain the first distribution ratio vector and the second distribution ratio vector. Calculate the preset distance metric between the first distribution ratio vector and the second distribution ratio vector, and determine the preset distance metric as the percentage change.

[0076] Specifically, the weighted sliding window algorithm is a commonly used method in time series data analysis. Its core principle is to assign different weights to data at different time points, typically with data closer to the current time point having higher weights. This approach allows for more sensitive capture of the latest trends in behavioral patterns while also considering the influence of historical data, avoiding misjudgments due to short-term fluctuations. When calculating the distribution proportion of behavioral types, this algorithm can dynamically adjust the contribution of each behavioral type within the window, thus obtaining a more representative distribution proportion.

[0077] The first distribution ratio vector is a vector representation of the distribution ratio of the advertising audience's behavior types within the current monitoring window, calculated using a weighted sliding window algorithm. For example, if the behavior types include browsing, clicking, and searching, the vector may contain the weighted ratio values ​​of these three behavior types within the current window. The second distribution ratio vector is a vector representation of the distribution ratio of the advertising audience's behavior types within the historical baseline window, calculated using the same algorithm. These two vectors represent the behavioral preference structure of the advertising audience over different time periods.

[0078] In practical applications, a preset distance metric can be understood as an indicator used to quantify the degree of difference between two distribution ratio vectors. For example, various mathematical methods such as Euclidean distance, cosine similarity (taking 1 minus the cosine similarity as the distance), Kullback-Leibler divergence, or Jensen-Shannon divergence can be used to calculate the distance between vectors. Choosing an appropriate distance metric can more accurately reflect the degree of migration of advertising audiences in content consumption preferences or conversion path choices, i.e., the change in proportion.

[0079] Through the aforementioned technical solutions, this application can more accurately and sensitively capture structural shifts in advertising audience behavior patterns. The application of the weighted sliding window algorithm enables the system to prioritize responses to users' latest behavioral preferences, ensuring timely judgment of the magnitude of shifts in interest focus. Simultaneously, using a preset distance metric to quantify the differences between the distribution ratios of behavioral types not only improves the accuracy of calculating percentage changes but also enhances the ability to identify complex behavioral pattern changes, effectively reducing interference from data noise or short-term fluctuations. Therefore, it can provide more reliable and refined behavioral shift features for subsequent dynamic correction of audience profiles, thereby generating more targeted and effective advertising strategies and improving the overall effectiveness of advertising promotion and user experience.

[0080] In some preferred embodiments, it is assumed that the behavioral types of the advertising audience include "browsing product A", "clicking ad B", and "searching keyword C". The system can set a current monitoring window (e.g., the last 24 hours) and a historical baseline window (e.g., the last 7 days). Using a weighted sliding window algorithm, the behavioral types within these two windows are statistically analyzed, and newer behaviors are assigned higher weights (e.g., behaviors from the last hour have a weight of 0.5, and behaviors from the previous 23 hours have decreasing weights). This yields the behavioral type distribution vectors for the current monitoring window (e.g., [browsing A: 0.4, clicking B: 0.3, searching C: 0.3]) and the historical baseline window (e.g., [browsing A: 0.6, clicking B: 0.2, searching C: 0.2]). Subsequently, cosine similarity can be used as a preset distance metric to calculate the similarity between these two vectors. For example, if the cosine similarity is 0.8, the change in proportion can be defined as 1 - 0.8 = 0.2. This value of 0.2 quantifies the degree of structural shift in the advertising audience's behavioral patterns. When this percentage change exceeds a preset threshold, the system can identify a significant shift in the user's focus of interest, such as a shift from interest in product A to interest in product B, thereby triggering subsequent profile correction and strategy adjustment.

[0081] Traditional internet advertising optimization methods often employ simple matching mechanisms, such as direct keyword comparison, when correlating structural shifts in audience behavior patterns with external public opinion trends. However, this approach may struggle to accurately capture the deep causal relationship between these shifts, especially when dealing with complex and ever-changing user behavior and massive amounts of public opinion data. This can lead to insufficient precision in correlation verification results, impacting the effectiveness of subsequent audience profiling and advertising strategy adjustments. Without addressing these issues, advertising strategy optimization will lack the depth and accuracy supported by data, potentially resulting in inefficient advertising and an inability to respond promptly to market changes. Therefore, this application proposes a more refined correlation matching method, aiming to more accurately assess the causal relationship between behavioral shifts and external public opinion through quantitative analysis.

[0082] In some embodiments of this application, the step of associating and matching behavioral deviation features with external public opinion trend information to obtain association verification results includes: Extract high-frequency behavioral keywords from the cleaned platform interaction sequence; The intersection of high-frequency behavioral keywords and keyword sets from external public opinion trend information is performed to obtain matching keyword groups. Calculate the frequency of occurrence of matching keyword phrases in the cleaned platform interaction behavior sequence, and the weighted value of the popularity index in external public opinion trend information; If the product of the frequency of occurrence and the weighted value of the popularity index is greater than the preset association threshold, then an association verification result indicating a strong association is generated. The association verification result is also used to quantify the influence weight of external hot topics on the structural shift of behavioral patterns.

[0083] Specifically, extracting high-frequency behavioral keywords from the cleaned platform interaction sequences refers to identifying core words that advertisers have recently frequently paid attention to or operated on from browsing, clicking, and searching behavior data of advertising audiences on the advertising platform after integrity verification, reordering, and filling processing. These keywords can directly reflect the current content consumption preferences and focus of interest of advertising audiences. For example, TF-IDF (Term Frequency-Inverse Document Frequency) algorithms or TextRank algorithms can be used to identify and extract topic words that appear frequently and have distinctiveness within a specific time window.

[0084] This process involves performing an intersection operation between high-frequency behavioral keywords and keyword sets from external public opinion trend information to obtain matching keyword groups. This can be understood as comparing high-frequency keywords extracted from behavioral data within the advertising audience platform with external trending topic keywords crawled and parsed from mainstream social media platforms and short video platforms to identify keywords commonly found in both. The aim is to discover whether there are shared concerns between advertising audience behavioral shifts and external trending topics, thereby establishing a preliminary potential connection between the two.

[0085] In practical applications, calculating the frequency of occurrence of matching keyword groups in the cleaned platform-wide interaction behavior sequence, and the weighted value of their popularity index in external public opinion trend information, refers to statistically analyzing the number of times the aforementioned matching keyword groups appear in the advertising audience's platform-wide behavior sequence, as well as their level of activity in external public opinion data discussion and dissemination. The weighted value of the popularity index comprehensively considers factors such as keyword frequency, dissemination speed, and discussion breadth, calculating a comprehensive indicator to more fully reflect the keyword's influence in external public opinion.

[0086] Furthermore, it is determined whether the product of the frequency of occurrence and the weighted value of the popularity index exceeds a preset association threshold. If so, an association verification result indicating a strong association is generated. This result is also used to quantify the influence weight of external trending topics on the structural shift in behavioral patterns. The aim is to assess the strength of the association between advertising audience behavioral shifts and external trending topics using a quantitative indicator. When the product exceeds the preset threshold, it indicates a significant, non-accidental association between the structural shift in advertising audience behavior and external trending topics; this association is considered a strong association. This product value, either directly or after further processing, can serve as a basis for quantifying the influence weight of external trending topics on the structural shift in advertising audience behavioral patterns, providing more precise guidance for subsequent profile adjustments.

[0087] Through the above technical solution, this application overcomes the shortcomings of traditional association matching methods in capturing deep causal relationships. By introducing the extraction of high-frequency behavioral keywords, the intersection operation of keyword groups, and the product judgment mechanism of occurrence frequency and popularity index weighted value, this solution can more accurately and quantitatively assess the strength of the causal relationship between the structural shift in advertising audience behavior patterns and external hot topics. Therefore, the generated association verification results not only indicate the existence of strong associations but also provide the influence weight of external hot topics on the structural shift in behavior patterns, providing a more refined and reliable basis for the subsequent dynamic correction of the advertising audience profile tag set, significantly improving the accuracy of audience profiling and the effectiveness of advertising placement strategies.

[0088] In some preferred embodiments, it is assumed that an advertising audience, in the cleaned platform interaction behavior sequence, recently frequently searched and browsed a large amount of content about "AI painting tools," thus "AI painting tools" is extracted as a high-frequency behavioral keyword. Simultaneously, external public opinion trend information crawled by a web crawler component shows that "AI painting" has also become a highly discussed topic on mainstream social media platforms and short video platforms, with its keyword set containing "AI painting" and a high weighted popularity index. At this point, the intersection operation of the high-frequency keyword "AI painting tools" in the advertising audience's behavior and the keyword "AI painting" in external public opinion is performed to obtain the matching keyword group "AI painting." Further, the system calculates the frequency of "AI painting" in the advertising audience's behavior sequence (e.g., appearing 50 times in the past week) and its weighted popularity index in external public opinion (e.g., 0.8). If the product of the two (50 * 0.8 = 40) is greater than the preset association threshold (e.g., set to 30), the system will generate an association verification result indicating a strong association and quantify the influence weight of the external hot topic "AI painting" on the structural shift in the behavioral pattern of the advertising audience as 40. Based on this strong association and quantified weight, subsequent audience profile correction will be able to more accurately identify the strong interest of the advertising audience in "AI painting" and adjust the advertising strategy accordingly, such as prioritizing the push of advertising materials related to AI painting tools or increasing the proportion of advertising on related channels.

[0089] In some embodiments described above, by associating behavioral shift features with external public opinion trend information, a correlation verification result can be obtained. This result is used to characterize the strength of the causal relationship between structural shifts in behavioral patterns and external trending topics. However, when dynamically modifying the set of tags for the advertising audience profile based on this correlation verification result, simply adding or deleting tags or adjusting weights may not accurately distinguish whether the audience has developed entirely new interest dimensions or whether their existing interests have been reinforced. This coarse-grained modification method may result in inaccurate audience profile updates, thereby affecting the optimization effect of advertising placement strategies.

[0090] In response, this application further proposes a method for dynamically revising the set of profile tags for the advertising audience based on the aforementioned correlation verification results to obtain an updated audience profile, specifically including: When the above association verification results indicate that there is a strong association, the keywords in the above matching keyword group that are not included in the original profile tag set will be identified as new interest tags. Identify the semantic similarity between the newly added interest tags and each tag in the existing profile tag set; When the semantic similarity is less than the preset similarity threshold, it is determined to be a new interest dimension, and the new interest tag is added to the above-mentioned advertising audience; When the semantic similarity is greater than or equal to the preset similarity threshold, it is determined to be interest enhancement, and the weight value of the tag corresponding to the semantic in the original portrait tag set is increased. The adjusted tag set is determined as the updated audience profile described above.

[0091] Specifically, when the correlation verification results indicate a strong correlation between a structural shift in behavioral patterns and external trending topics, the system first filters out keywords from the matching keyword group that do not yet appear in the current audience profile tag set, and identifies these keywords as potential new interest tags. For example, if the matching keyword group contains "metaverse" but it is not in the audience profile, then "metaverse" is identified as a new interest tag.

[0092] Subsequently, to determine whether these newly added interest tags represent entirely new interest dimensions or a deepening of existing interests, it is necessary to identify the semantic similarity between the new interest tags and each tag in the original profile tag set. This can be achieved by mapping the tags to a predefined semantic vector space and calculating their cosine similarity.

[0093] The preset similarity threshold is a configurable parameter used to define semantic "new" and "old". When the calculated semantic similarity is less than this threshold, it indicates that the new interest tag differs significantly from existing tags semantically, and is therefore judged as a new interest dimension. In this case, the new interest tag will be directly added to the user profile tag set. For example, if the user profile only has the "sports" tag, and the new interest tag is "blockchain", the semantic similarity between the two is low, so "blockchain" is considered a new interest dimension.

[0094] Conversely, when the semantic similarity is greater than or equal to a preset similarity threshold, it indicates that the newly added interest tag has a strong semantic correlation with the existing tags, and this is considered interest enhancement. In this case, the system will not simply add a new tag, but will increase the weight of tags in the existing profile tag set that semantically correspond to the newly added interest tag, to reflect the increased activity and importance of the interest. For example, if the user profile has a "science fiction movie" tag, and the newly added interest tag is "cyberpunk," the two have a high semantic similarity, so the weight of the "science fiction movie" tag will be increased.

[0095] Ultimately, the set of tags formed after the above judgments and adjustments is determined as the updated audience profile.

[0096] Through the aforementioned technical solution, this application enables more refined and accurate dynamic correction of the advertising audience profile tag set. Compared to coarse corrections based solely on correlation verification results, this solution effectively distinguishes whether the audience has developed entirely new content consumption preferences or whether their existing preferences have been reinforced by external trends. Therefore, the updated audience profile more realistically and comprehensively reflects the real-time changes in the advertising audience's interests and conversion path migrations, avoiding profile distortion or redundancy caused by improper tag correction. This precise profile update allows subsequent advertising strategies to more accurately match the audience's actual needs and interests, thereby significantly improving click-through rates, conversion rates, and overall promotional effectiveness, reducing the cost of ineffective ad placements, and optimizing the advertiser's return on investment.

[0097] In some preferred embodiments, it is assumed that the profile tag set of an advertising audience includes interest tags such as "digital products" and "photography". Recently, by associating and matching the audience's real-time operational data on the advertising platform (e.g., frequent browsing of tutorials and works about "AI painting") with external public opinion trend information (e.g., the surge in popularity of the topic "AI painting" on mainstream social media platforms), the system has identified a strong correlation between the audience's structural shift in behavioral patterns and the external hot topic of "AI painting".

[0098] At this point, the system will dynamically modify the audience profile according to the scheme proposed in this application. First, "AI painting" will be identified as a new interest tag because it is not yet included in the original profile tag set. Next, the system will identify the semantic similarity between "AI painting" and the existing tags "digital products" and "photography".

[0099] If the semantic similarity calculation result (e.g., cosine similarity) between "AI painting" and "digital products" or "photography" is less than a preset similarity threshold (e.g., 0.4), then "AI painting" is determined to be a new interest dimension, and the "AI painting" tag is directly added to the audience's profile tag set. This reflects the audience's interest in a completely new field.

[0100] Conversely, if the semantic similarity between "AI painting" and "photography" is greater than or equal to a preset similarity threshold (e.g., 0.6), it is considered interest enhancement, and the system will increase the weight of the "photography" tag in the original image tag set. This indicates that the audience's original interest in "photography" has been deepened and expanded in the direction of "AI painting".

[0101] In this way, the audience's profile tag set can be accurately updated, and whether it is the addition of new interests or the reinforcement of existing interests, it can be appropriately reflected, thus providing a solid foundation for generating more targeted advertising strategies in the future.

[0102] Specifically, in the process of dynamically correcting the set of profile tags for the advertising audience based on the association verification results, identifying the semantic similarity between the newly added interest tags and each tag in the original set of profile tags is one of the key steps.

[0103] The step of identifying the semantic similarity between the newly added interest tags and each tag in the existing profile tag set includes: The newly added interest tags and each tag in the original profile tag set are mapped to a preset semantic vector space to obtain a first semantic vector and a second semantic vector set; Calculate the cosine similarity between the first semantic vector and each vector in the second semantic vector set, and determine the calculation result as the semantic similarity.

[0104] Specifically, newly added interest tags refer to keywords identified during the association matching process that are not yet included in the existing set of tags for the advertising audience's profile. The existing set of tags for the profile includes the advertising audience's existing interest dimension tags and media touchpoint tags. To accurately measure the semantic relevance between these tags, this application adopts a method of mapping tags to a semantic vector space. The semantic vector space is a high-dimensional mathematical space in which each word or phrase is represented as a vector, and the distance or angle between vectors can reflect their semantic similarity. For example, textual tags can be converted into numerical vectors using pre-trained word vector models (such as Word2Vec, GloVe, or BERT).

[0105] The first semantic vector refers to the representation of the newly added interest tag in the semantic vector space, while the second semantic vector set consists of the representations of each tag in the original profile tag set in the semantic vector space. After obtaining these semantic vectors, the strength of their semantic association can be quantified by calculating the cosine similarity between the first semantic vector and each vector in the second semantic vector set. Cosine similarity is a commonly used metric, with values ​​ranging from -1 to 1. Values ​​closer to 1 indicate greater semantic similarity, values ​​closer to -1 indicate less semantic similarity, and values ​​close to 0 indicate independence. In this way, the semantic similarity value between the newly added interest tag and each existing tag in the original profile tag set can be obtained.

[0106] The aforementioned technical solution enables precise quantification of the semantic similarity between newly added interest tags and tags in the existing profile tag set. This accurate semantic similarity calculation allows for a more accurate determination, during subsequent profile tag revision, of whether a newly added interest tag represents a completely new interest dimension or a further reinforcement of existing interests. This helps to build more refined and dynamic advertising audience profiles, thus providing a solid data foundation for generating more targeted and effective advertising strategies, significantly improving the accuracy and conversion efficiency of advertising promotion.

[0107] In some embodiments, such as Figure 3 As shown, this application proposes an internet advertising promotion strategy optimization system, including: a data collection unit 301, used to collect real-time operation data of advertising audiences on the advertising platform and external public opinion trend information. The advertising audience refers to the terminal users who receive advertising pushes. The real-time operation data includes the platform's interactive behavior sequence and corresponding behavioral context identifiers. The platform's interactive behavior sequence is used to record the browsing, clicking, and searching actions generated by the advertising audience on the advertising platform. The external public opinion trend information includes hot topics on third-party content publishing platforms and corresponding trend popularity values; and an identification unit 302, used to identify the structural shifts in the advertising audience's behavioral patterns based on the platform's interactive behavior sequence and behavioral context identifiers, and obtain behavioral shift features. The offset feature is used to characterize the degree of migration of advertising audiences in content consumption preferences or conversion path selection; the matching unit 303 is used to match the behavioral offset feature with external public opinion trend information to obtain the association verification result, which is used to characterize the causal relationship between the structural offset of the behavioral pattern and external hot topics; the correction unit 304 is used to dynamically correct the profile tag set of advertising audiences according to the association verification result to obtain the updated audience profile, which includes interest dimension tags and media touchpoint tags; the generation unit 305 is used to generate and distribute the adjusted advertising placement strategy according to the updated audience profile, which includes the display format of the creative materials, the bid amount, and the channel allocation ratio.

[0108] This system aims to effectively address issues in existing internet advertising promotion, such as lagging and inaccurate strategy adjustments, excessive reliance on human experience, and data processing delays and loss, by constructing an intelligent closed-loop optimization process. Specifically, the data collection unit is responsible for comprehensively and in real-time acquiring user behavior data within the platform and external public opinion dynamics, laying the data foundation for subsequent analysis. The identification unit deeply analyzes user behavior sequences, accurately capturing structural shifts in user content consumption preferences or conversion paths. The matching unit further correlates the identified behavioral shifts with external trending topics, revealing their underlying causal relationships. The correction unit dynamically updates user profiles based on the correlation results, ensuring the accuracy and timeliness of the profiles. Finally, the generation unit intelligently generates and distributes adjusted advertising strategies based on the updated and accurate profiles, thereby significantly improving the accuracy, stability, and overall return on investment of advertising, reducing reliance on human experience, and making advertising strategy adjustments more intelligent and efficient.

[0109] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for optimizing internet advertising promotion strategies, characterized in that, include: The system collects real-time operation data of the advertising audience on the advertising platform and external public opinion trend information. The advertising audience refers to the terminal users who receive the advertising push. The real-time operation data includes the platform interaction behavior sequence and the corresponding behavior context identifier. The platform interaction behavior sequence is used to record the browsing, clicking and search actions of the advertising audience on the advertising platform. The external public opinion trend information includes hot topics on third-party content publishing platforms and the corresponding trend popularity value. Based on the platform's interactive behavior sequence and the behavior context identifier, the structural shift in the advertising audience's behavior pattern is identified to obtain behavior shift features. These behavior shift features are used to characterize the degree of migration of the advertising audience in content consumption preferences or conversion path selection. The behavioral shift features are correlated and matched with the external public opinion trend information to obtain the correlation verification results. The correlation verification results are used to characterize the causal correlation strength between the structural shift of the behavioral pattern and external hot topics. Based on the correlation verification results, the set of profile tags for the advertising audience is dynamically modified to obtain an updated audience profile. The set of profile tags includes interest dimension tags and media touchpoint tags. Based on the updated audience profile, an adjusted advertising strategy is generated and distributed. The advertising strategy includes the display format of the advertising materials, the bid amount, and the channel allocation ratio.

2. The method according to claim 1, characterized in that, The collection of real-time operational data of advertising audiences on the advertising platform and external public opinion trend information includes: The operation logs of the advertising audience within a preset time period are captured by the streaming computing engine, and the interaction behavior sequence within the platform is parsed. The interaction behavior sequence within the platform includes behavior type identifiers and corresponding timestamps. By using web crawler components to selectively capture publicly available discussion data from mainstream social media platforms and short video platforms, the external public opinion trend information is obtained through analysis.

3. The method according to claim 2, characterized in that, The analysis yields the external public opinion trend information, including: The publicly discussed data is cleaned to extract unstructured text, which includes the title, body text, and comments from the publicly discussed data. The unstructured text is semantically segmented to remove stop words and extract core entity words as a keyword set. The frequency of occurrence and propagation rate of the core entity words within a preset time window are statistically analyzed, and the popularity index value is obtained after normalization.

4. The method according to claim 2, characterized in that, The parsing of the platform's interactive behavior sequence includes: The timestamps are checked for integrity. If there are out-of-order or missing timestamps, the out-of-order data is reordered according to the logical relationship between adjacent behaviors, and the missing data is filled in according to the preset completion rules to obtain the cleaned platform interaction behavior sequence.

5. The method according to claim 4, characterized in that, The step of identifying structural shifts in the behavioral patterns of the advertising audience based on the platform's in-platform interaction sequence and the behavioral context identifier, to obtain behavioral shift features, includes: Based on the cleaned platform interaction behavior sequence, the distribution ratio of behavior types in the current monitoring window and the historical baseline window are statistically analyzed, and the difference between the two is calculated to obtain the percentage change. The percentage change is used to characterize the magnitude of the shift in user interest focus. The number of clicks pointing to external links in the cleaned platform interaction behavior sequence is detected. When the growth rate of the number of clicks exceeds a preset growth rate threshold, a path change marker is generated. The path change marker is used to characterize the offset of the conversion path. The behavior offset feature is generated by combining the percentage change and the path change marker.

6. The method according to claim 5, characterized in that, The calculation of the difference between the two to obtain the change in proportion includes: The weighted sliding window algorithm is used to calculate the distribution ratio of behavior types in the current monitoring window and the historical baseline window to obtain a first distribution ratio vector and a second distribution ratio vector. Calculate a preset distance metric between the first distribution ratio vector and the second distribution ratio vector, and determine the preset distance metric as the percentage change.

7. The method according to claim 4, characterized in that, The step of associating and matching the behavioral offset features with the external public opinion trend information to obtain the association verification result includes: Extract high-frequency behavioral keywords from the cleaned platform interaction sequence; The intersection of the high-frequency behavioral keywords and the keyword set in the external public opinion trend information is performed to obtain the matching keyword group; Calculate the frequency of occurrence of the matching keyword group in the cleaned platform interaction behavior sequence, and the weighted value of the popularity index in the external public opinion trend information; If the product of the occurrence frequency and the weighted value of the popularity index is greater than a preset association threshold, then an association verification result indicating a strong association is generated. The association verification result is also used to quantify the influence weight of external hot topics on the structural shift of the behavior pattern.

8. The method according to claim 7, characterized in that, The step of dynamically correcting the set of profile tags for the advertising audience based on the association verification results to obtain an updated audience profile includes: When the association verification result indicates that there is a strong association, the keywords in the matching keyword group that are not included in the original profile tag set are identified as new interest tags; Identify the semantic similarity between the newly added interest tags and each tag in the existing profile tag set; When the semantic similarity is less than a preset similarity threshold, it is determined to be a new interest dimension, and the new interest tag is added to the advertising audience; When the semantic similarity is greater than or equal to the preset similarity threshold, it is determined to be interest enhancement, and the weight value of the corresponding semantic tag in the original portrait tag set is increased; The adjusted set of tags is determined as the updated audience profile.

9. The method according to claim 8, characterized in that, The step of identifying the semantic similarity between the newly added interest tags and each tag in the existing profile tag set includes: The newly added interest tags and each tag in the original profile tag set are mapped to a preset semantic vector space to obtain a first semantic vector and a second semantic vector set; Calculate the cosine similarity between the first semantic vector and each vector in the second set of semantic vectors, and determine the calculation result as the semantic similarity.

10. An internet advertising promotion strategy optimization system, characterized in that, include: The data collection unit is used to collect real-time operation data of the advertising audience on the advertising platform and external public opinion trend information. The advertising audience refers to the terminal users who receive the advertising push. The real-time operation data includes the platform interaction behavior sequence and the corresponding behavior context identifier. The platform interaction behavior sequence is used to record the browsing, clicking and searching actions of the advertising audience on the advertising platform. The external public opinion trend information includes hot topics on third-party content publishing platforms and the corresponding trend popularity values. The identification unit is used to identify the structural shift in the behavioral pattern of the advertising audience based on the interaction behavior sequence within the platform and the behavioral context identifier, and to obtain behavioral shift features. The behavioral shift features are used to characterize the degree of migration of the advertising audience in content consumption preferences or conversion path selection. A matching unit is used to associate and match the behavioral shift features with the external public opinion trend information to obtain an association verification result. The association verification result is used to characterize the causal association strength between the structural shift of the behavioral pattern and external hot topics. The correction unit is used to dynamically correct the profile tag set of the advertising audience based on the association verification result, so as to obtain an updated audience profile. The profile tag set includes interest dimension tags and media touchpoint tags. The generation unit is used to generate and distribute the adjusted advertising strategy based on the updated audience profile. The advertising strategy includes the display format of the advertising materials, the bid amount, and the channel allocation ratio.

Images