A method for managing advertising value based on intelligent delivery

Abstract

The present invention discloses an advertising delivery value management method based on intelligent delivery, which specifically relates to the field of advertising delivery and is used to solve the problem of user interest changes. The method analyzes the feature coverage of the current user portrait to determine whether content features are needed to supplement and assist in updating the portrait, then extracts the interest orientation index and content feature value index corresponding to the analysis value and actual value of the content features, conducts a comprehensive analysis to obtain the feature excellence index, screens content features that meet the requirements and incorporates them into historical behavior data, and further updates the user portrait model, thereby ensuring the quality of the user portrait model, ensuring the accuracy and intelligence of advertising delivery, and maximizing the value of advertising delivery; effectively screens high-value content features to update the user portrait, improves the comprehensiveness and accuracy of the user portrait, makes advertising delivery more accurate and efficient, and significantly improves the effect of advertising delivery and user experience.

Description

Technical Field

[0001] The present invention relates to the field of advertising delivery, and more specifically, to an advertising delivery value management method based on intelligent delivery. Background Art

[0002] With the rapid development of Internet technology and big data analysis, accurate advertising has become an important means of digital marketing. The user portrait model is the core component of the advertising delivery system. By analyzing the user's behavior data, interest preferences and historical records, a detailed user portrait is constructed to achieve personalized advertising push. However, the user's interests and behaviors are dynamically changing. Simply relying on static user portraits is difficult to capture the user's latest needs and interest changes, resulting in a gradual weakening of the effect of advertising delivery. In this context, dynamically updating the user portrait model to ensure that it can reflect the user's latest status in a timely manner has become the key to improving the accuracy and effectiveness of advertising delivery.

[0003] Existing advertising delivery systems have significant deficiencies in updating user portrait models. On the one hand, these systems usually rely on periodic data updates and fail to reflect the dynamic changes in user interests in real time, resulting in insufficient timeliness and accuracy of user portraits. On the other hand, when faced with a decline in user portrait quality, existing technologies lack an effective content feature supplementation mechanism, making it difficult to optimize the quality of user portrait models by screening high-value content features. This not only affects the accuracy of advertising delivery and user experience, but also leads to a waste of advertising resources and a decline in delivery effectiveness. Therefore, there is an urgent need for a technical solution that can evaluate the quality of user portraits in real time and dynamically optimize user portrait models through an efficient content feature screening and supplementation mechanism.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an advertising delivery value management method based on intelligent delivery, which analyzes the feature coverage of the current user portrait, determines whether content features are needed to supplement and assist in updating the portrait, and then extracts the interest orientation index and content feature value index corresponding to the analysis value and actual value of the content feature, performs a comprehensive analysis to obtain the feature excellence index, selects content features that meet the requirements and incorporates them into historical behavior data, and further updates the user portrait model, thereby ensuring the quality of the user portrait model, ensuring the accuracy and intelligence of advertising delivery, and maximizing the value of advertising delivery. By dynamically evaluating and updating user portraits, using nonlinear adjustment and comprehensive analysis mechanisms, effectively screening high-value content features, and improving the comprehensiveness and accuracy of user portraits, it is suitable for advertising delivery value management methods based on intelligent delivery, making advertising delivery more accurate and efficient, and significantly improving the effect of advertising delivery and user experience, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Collect and store user behavior feature data and content feature data to the cloud platform in real time; based on real-time behavior feature data, dynamically update user portrait models and user interest fluctuation models to reflect users' latest interests and behavior patterns; dynamically decide whether to trigger the use of content features for supplementation based on the matching degree between existing feature coverage and users' actual interaction frequency; use the updated interest fluctuation model to analyze and match content features with users' current interests in real time, and select content features with analytical value based on similarity and interest changes to include in real-time behavior data; comprehensively consider semantic relevance, user intent, and user interaction rate, use nonlinear mapping and adjustment mechanisms to analyze the value of content features, and screen content features with actual value to include in real-time behavior data; use content features with analytical value and actual value as part of behavioral features to further update the user portrait model; match the most relevant advertisements from the advertising database based on the latest user portrait model.

[0008] In a preferred embodiment, based on real-time behavior feature data, the user portrait model and the user interest fluctuation model are dynamically updated, including the following contents:

[0009] S2.1, call the behavior feature data in the cloud platform and extract key features from the behavior feature data.

[0010] S2.2, user portrait model update: integrate new behavior feature data with existing user portraits, use incremental learning algorithms to update user interests, preferences and behavior patterns; user interest fluctuation model update: use time series analysis to analyze new behavior feature data, capture the changing trends and periodic fluctuations of user interests, and update the parameters in the user interest fluctuation model.

[0011] S2.3, verify the accuracy and effectiveness of the updated user portrait model and interest fluctuation model through cross-validation or A / B testing.

[0012] In a preferred embodiment, whether to trigger the use of content features for supplementation is dynamically determined based on the matching degree between the existing feature coverage and the actual user interaction frequency, including the following contents:

[0013] S3.1, vectorize each feature in the user portrait, set ; Vectorize the user's actual interactive behavior and set it as ; Calculate the coverage of each user portrait feature and the user's actual interactive behavior, and calculate the matching degree between each feature and user behavior through cosine similarity ; Take the average coverage of all features to get the average coverage of user portraits .

[0014] S3.2, User Profile Completeness The calculation formula is as follows: ;in, is the portrait threshold, indicating the ideal user portrait coverage.

[0015] S3.3, calculate the user portrait completeness once in each time period, and collect the user portrait completeness values ​​within a period of time; use moving average to smooth the collected user portrait completeness values; after calculating the moving average value, use the median to smooth the moving average value; use the median smoothed value as the most representative user portrait completeness value per unit time, mark it as the user portrait completeness index, and make a decision; when the user portrait completeness index is less than 1, it means that the user portrait is incomplete and needs to be supplemented with content features to generate a supplementary signal.

[0016] In a preferred embodiment, the updated interest fluctuation model is used to analyze and match content features with the user's current interests in real time, and content features with analysis value are selected based on similarity and interest changes and incorporated into real-time behavior data, including the following:

[0017] S4.1, after obtaining the supplementary signal, vectorize the latest behavior data, use NLP technology and behavior analysis methods to extract the user's current interest vector .

[0018] S4.2, vectorize the current content feature vector to obtain the content feature vector ; Calculate the cosine similarity between the content feature vector and the user's current interest vector , represents the similarity between two vectors.

[0019] S4.3, Analyze users’ short-term interests using interest fluctuation model and long-term interests ; Calculate the rate of change of short-term interest vector and long-term interest vector , indicating the degree of change of user interest: .

[0020] S4.4, comprehensively consider the matching degree between content features and current interests and the user's interest change rate, and calculate the interest orientation index : .

[0021] In a preferred embodiment, semantic relevance, user intent, and user interaction rate are integrated, and nonlinear mapping and adjustment mechanisms are used to analyze the value of content features, and screen content features with actual value to be included in real-time behavior data, including the following:

[0022] S5.1, convert content features and user historical behavior features into vector representation; let the content feature vector be , the user's historical behavior feature vector is ;in, An index representing content features; Represents the index of historical behavior features; calculates the cosine similarity between the content feature vector and the historical behavior feature vector ; Take the maximum similarity of all historical behavior features as the semantic relevance score : .

[0023] S5.2, use the pre-trained intent classification model to predict the user's intent category and strength when browsing the current content; let the intent category be , the intention strength is ;in, Indicates the index of the intent category; normalize the intent strength to get the user intent score : .

[0024] S5.3, Calculate content feature importance factor , which is used to adjust the weight of semantic relevance and user intent in value calculation: ;in, is the user interaction rate of the content feature; combining the semantic relevance score and the user intent score to calculate the content feature value index : .

[0025] In a preferred implementation, content features with analytical value and actual value are used as part of the behavior features to further update the user portrait model, specifically including the following contents:

[0026] The interest index and content feature value index are combined, and content features with analytical and practical value are screened out to calculate the feature excellence index. : ; Compare the feature excellence index with the excellence threshold. If the feature excellence index is greater than or equal to the excellence threshold, it means that the content feature has a higher comprehensive value in user profiling and advertising delivery, and is suitable for inclusion in real-time behavior data for further processing and recommendation; conversely, if the feature excellence index is less than the excellence threshold, it means that the content feature has a lower comprehensive value in user profiling and advertising delivery, and is not included in real-time behavior data.

[0027] The technical effects and advantages of the advertising value management method based on intelligent delivery of the present invention are as follows:

[0028] The present invention analyzes the feature coverage of the current user portrait, determines whether content features are needed to supplement and assist in updating the portrait, then extracts the interest-oriented index and content feature value index corresponding to the analytical value and actual value of the content features, conducts a comprehensive analysis to obtain the feature excellence index, screens the content features that meet the requirements and incorporates them into the historical behavior data, and further updates the user portrait model, thereby ensuring the quality of the user portrait model, ensuring the accuracy and intelligence of advertising delivery, and maximizing the value of advertising delivery. By dynamically evaluating and updating user portraits, using nonlinear adjustment and comprehensive analysis mechanisms, effectively screening high-value content features, and improving the comprehensiveness and accuracy of user portraits, it is suitable for advertising delivery value management methods based on intelligent delivery, making advertising delivery more accurate and efficient, and significantly improving the effect of advertising delivery and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention is a flowchart of an advertising delivery value management method based on intelligent delivery. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Example 1

[0032] Figure 1The present invention provides an advertising delivery value management method based on intelligent delivery, comprising:

[0033] Collect and store user behavior feature data and content feature data to the cloud platform in real time; based on real-time behavior feature data, dynamically update user portrait models and user interest fluctuation models to reflect users' latest interests and behavior patterns; dynamically decide whether to trigger the use of content features for supplementation based on the matching degree between existing feature coverage and users' actual interaction frequency; use the updated interest fluctuation model to analyze and match content features with users' current interests in real time, and select content features with analytical value based on similarity and interest changes to include in real-time behavior data; comprehensively consider semantic relevance, user intent, and user interaction rate, use nonlinear mapping and adjustment mechanisms to analyze the value of content features, and screen content features with actual value to include in real-time behavior data; use content features with analytical value and actual value as part of behavioral features to further update the user portrait model; match the most relevant advertisements from the advertising database based on the latest user portrait model.

[0034] In the advertising delivery system driven by big data and artificial intelligence, accurate data collection and storage are the basis for realizing intelligent advertising delivery. Users' operation behaviors and browsing content on various Internet platforms, whether through desktop devices or mobile devices, will generate a large amount of data. These data contain user behavior characteristics and content characteristics, and are the key basis for the advertising delivery system to conduct user portrait modeling, interest analysis, content matching and value evaluation. Therefore, real-time and efficient collection and storage of these data on the cloud platform can not only provide a reliable foundation for subsequent data analysis and processing, but also ensure that the advertising delivery system has the ability to respond dynamically and make accurate recommendations, significantly improving the effectiveness of advertising delivery and user experience.

[0035] Collect and store user behavior feature data and content feature data in real time on the cloud platform, including the following:

[0036] Behavioral characteristics refer to a series of operations and interactive data generated by users when using a platform or system. These characteristics include but are not limited to users' click records, browsing history, search queries, purchase behavior, dwell time, scrolling behavior, and interactive actions (such as likes, comments, and shares). Behavioral characteristics reflect users' interests, preferences, and behavior patterns, and are an important source of data for understanding user needs and predicting their future behavior.

[0037] Content features refer to the various information elements contained in the web page or application content that the user is currently browsing. These features include the text content, pictures, videos, links, and keywords, themes, sentiment, semantic information, etc. of the web page. Content features reflect the interests and concerns of users at a specific moment and are key data for achieving personalized content recommendations and accurate advertising delivery.

[0038] For behavioral feature data: use tracking scripts and API interfaces to record user operations on different devices and channels; monitor user operations on the platform in real time, such as clicks, browsing, searches, purchases, dwell time, interactive behaviors, etc.

[0039] For content feature data: extract keywords, themes, sentiment tendencies and other semantic information of the page; analyze the content of the web page currently browsed by the user in real time, including text, pictures, videos, etc.

[0040] The collected behavioral feature data and content feature data are uploaded to the cloud platform through a secure channel to ensure the real-time and security of data transmission.

[0041] On the cloud platform, distributed database and big data storage technologies are used to ensure the efficiency and scalability of data storage, and to classify and store the collected data.

[0042] The collected data are preliminarily cleaned and standardized to remove invalid and duplicate data.

[0043] Using data processing pipelines, the processed data is prepared for subsequent analysis and modeling.

[0044] Based on real-time behavioral feature data, dynamically update the user portrait model and user interest fluctuation model, including the following:

[0045] S2.1, call the behavior feature data in the cloud platform and extract key features from the behavior feature data, including click-through rate, browsing depth, dwell time, interaction frequency, etc. Use natural language processing (NLP) technology to extract keywords and semantic information from users' search queries and text content.

[0046] S2.2, User portrait model update:

[0047] Integrate new behavioral feature data with existing user portraits and use incremental learning algorithms to update user interests, preferences, and behavior patterns.

[0048] Use clustering algorithms (such as K-means) or dimensionality reduction algorithms (such as PCA) to adjust user grouping to ensure the accuracy and real-time performance of the portrait model.

[0049] User interest fluctuation model update:

[0050] Use time series analysis (such as ARIMA models or LSTM) to analyze new behavioral feature data and capture changing trends and cyclical fluctuations in user interests.

[0051] Update the parameters in the user interest fluctuation model to reflect short-term and long-term changes in user interests, ensuring that the model dynamically responds to changes in user behavior.

[0052] The user interest fluctuation model is a model that captures the changing patterns and trends of user interests by analyzing user behavior data in different time periods, thereby dynamically reflecting users' short-term and long-term interests. It is used to accurately predict users' future behaviors and needs.

[0053] S2.3, verify the accuracy and effectiveness of the updated user portrait model and interest fluctuation model through cross-validation or A / B testing.

[0054] Based on the verification results, model optimization and parameter adjustment are carried out to improve the model's predictive ability and adaptability.

[0055] The User Profile Model is a multi-dimensional user data model established by analyzing and processing user behavior data, demographic information, interest preferences, etc. The model aims to comprehensively describe the user's characteristics, behavior habits, and interest tendencies, thereby providing a basis for personalized services and targeted advertising.

[0056] Functions include: providing personalized content and product recommendations based on user portrait models to enhance user experience; using user portraits to deliver targeted advertising to improve advertising effectiveness and return on investment; dividing users into different groups through user portrait models to develop targeted marketing strategies and service plans.

[0057] The User Interest Fluctuation Model is a dynamic model that captures the changing patterns of user interests by analyzing user behavior data in different time periods. The model aims to identify short-term fluctuations and long-term trends in user interests, thereby more accurately predicting users' future behaviors and needs.

[0058] Functions include: adjusting recommended content and advertising strategies in real time according to the user interest fluctuation model to improve response speed and accuracy; identifying long-term trends in user interests through the interest fluctuation model and formulating long-term content and product strategies; using the user interest fluctuation model to predict users' future behavior and make personalized recommendations and marketing in advance.

[0059] In the process of advertising value management, accurate user portraits are the key to ensuring effective advertising. However, user behavior and interests are dynamically changing, and relying solely on historical data may not reflect the latest needs of users in a timely manner. In order to improve the accuracy and real-time nature of user portraits, it is necessary to dynamically evaluate the matching degree between the existing feature coverage and the actual frequency of user interaction. When the completeness of the user portrait is lower than the set standard, the content feature supplement mechanism is intelligently triggered to ensure the comprehensiveness and accuracy of the user portrait. Through this dynamic adjustment and supplementation, the advertising delivery system can more accurately capture the real-time interests and preferences of users, thereby optimizing advertising matching, improving the effectiveness of advertising delivery and user experience. This process requires not only efficient data processing and analysis capabilities, but also relies on innovative algorithms and models for real-time evaluation and decision-making.

[0060] Dynamically decide whether to trigger the use of content features for supplementation based on the matching degree between the existing feature coverage and the actual user interaction frequency, including the following:

[0061] S3.1, vectorize each feature in the user portrait, set ; Vectorize the user's actual interactive behavior (click, browse, purchase, etc.) and set it as ; Calculate the coverage of each user portrait feature and the user's actual interactive behavior, and calculate the matching degree between each feature and user behavior through cosine similarity : ; Take the average coverage of all features to get the average coverage of user portraits : ;in, Represents the total number of features in the user profile. In other words, this is the number of all features included in the user profile; Indicates the index of the current feature. From 1 to Each user portrait feature is traversed in turn to calculate the matching degree between each feature and the user's actual interactive behavior.

[0062] S3.2, User Profile Completeness The calculation formula is as follows: ;in, is the portrait threshold, indicating the ideal user portrait coverage.

[0063] S3.3, in order to calculate the most representative user profile completeness (UPCI) per unit time and avoid decision errors due to chance, methods based on statistical analysis and time series can be used to smooth and optimize the user profile completeness value. This goal can be achieved through moving average and median smoothing methods.

[0064] Calculate the user portrait completeness once in each time period (for example, every minute or every hour), and collect the user portrait completeness values ​​within a period of time, such as the user portrait completeness values ​​every minute in the past hour; use Simple Moving Average (SMA) to smooth the collected user portrait completeness values ​​to reduce the impact of accidental fluctuations.

[0065] After calculating the moving average, median smoothing is used to process the moving average to further reduce the impact of extreme values.

[0066] Finally, the median smoothed value is used as the most representative user portrait completeness value per unit time, which is marked as the user portrait completeness index for decision making.

[0067] When the user portrait completeness index is less than 1, it means that the user portrait is incomplete and needs to be supplemented with content features to generate a supplementary signal.

[0068] The present invention vectorizes user portrait features and actual interactive behaviors, calculates their coverage, and dynamically determines whether to trigger the use of content features for supplementation through cosine similarity matching calculation, moving average smoothing and median smoothing, thereby ensuring the comprehensiveness and accuracy of user portraits and improving the accuracy and effectiveness of advertising. Statistical analysis and time series smoothing technology are used to evaluate the completeness of user portraits in real time, reduce the impact of accidental fluctuations, intelligently trigger content feature supplementation, and ensure that the user portrait model always reflects the user's latest interests and behaviors.

[0069] In the intelligent advertising delivery system, ensuring the accuracy and comprehensiveness of the user portrait model is the key to improving the effectiveness of advertising delivery. As user behaviors and interests continue to change, relying solely on historical data is difficult to reflect the latest needs of users. By evaluating the feature coverage and actual interaction frequency of the user portrait in real time, it is dynamically determined whether new content features need to be added. When the completeness of the user portrait is lower than the set standard, the content feature supplement mechanism is triggered to ensure that the user portrait model is always kept up to date and comprehensive. Through this mechanism, the system can quickly respond to changes in user interests, continuously optimize user portraits, and lay a solid foundation for subsequent precision advertising delivery.

[0070] Use the updated interest fluctuation model to analyze and match content features with users' current interests in real time, and select content features with analysis value based on similarity and interest changes to include in real-time behavior data, including the following:

[0071] S4.1, after obtaining the supplementary signal, vectorize the latest behavior data, use NLP technology and behavior analysis methods to extract the user's current interest vector .

[0072] S4.2, vectorize the current content feature vector to obtain the content feature vector ; Calculate the cosine similarity between the content feature vector and the user's current interest vector , represents the similarity between two vectors.

[0073] S4.3, Analyze users’ short-term interests using interest fluctuation model and long-term interests ; Calculate the rate of change of short-term interest vector and long-term interest vector , indicating the degree of change of user interest: .

[0074] S4.4, comprehensively consider the matching degree between content features and current interests and the user's interest change rate, and calculate the interest orientation index : ; The product of matching degree and interest change rate reflects the relevance of content features to users’ current interests and the dynamics of users’ interest changes.

[0075] The interest-directed index is used to indicate and reflect the degree of match between the current content features and the user's current interests and the dynamic changes in user interests. In advertising, a larger interest-directed index means that the current content features are highly matched with the user's current interests and the user's interests have changed significantly, indicating that the content features have higher analytical value and are worthy of being included in real-time behavior data. Conversely, a smaller interest-directed index means that the current content features are less matched with the user's current interests or the user's interests have not changed much, indicating that the content features have lower analytical value and are not suitable for inclusion in real-time behavior data.

[0076] The following is an example of using the interest fluctuation model to analyze a user's short-term interests and long-term interests:

[0077] Assume that there is user behavior data for the last 6 months, and each month's data contains user clicks, browsing, purchases, and other behavior records. Divide the data into two parts: short-term (last month) and long-term (last 6 months).

[0078] Short-term data (1 month): represents the user’s behavior data within the last month.

[0079] Long-term data (6 months): refers to the user's behavior data in the last six months.

[0080] Using natural language processing (NLP) and behavior analysis technology, users’ interest keywords and frequencies are extracted from the data of each time period to construct an interest feature vector.

[0081] Assume that the dimension of each vector is the number of keywords, and the value of each keyword is its occurrence frequency in the time period.

[0082] Short-term interest vector:

[0083] Keywords and frequencies extracted from short-term data constitute short-term interest vectors.

[0084] Example: STV={"sports":10,"music":8,"technology":5,"cooking":2}.

[0085] Long-term interest vector:

[0086] Keywords and frequencies extracted from long-term data constitute long-term interest vectors.

[0087] LTV={"sports":50,"music":30,"technology":25,"cooking":15}.

[0088] The vector is normalized so that the value of each keyword is between 0 and 1.

[0089] The present invention utilizes the cosine similarity between the user's current interest vector and the content feature vector in combination with the user's short-term and long-term interest change rates to accurately evaluate the matching degree between the current content features and the user's interests and their dynamic changes. This processing method ensures that only those content features that are highly relevant and reflect the user's real-time interest fluctuations are included in the real-time behavior data, thereby enhancing the accuracy and real-time nature of user portraits, improving the accuracy and effectiveness of advertising delivery, and effectively screening out content features with higher analytical value, thereby avoiding irrelevant or low-value content from interfering with the decision-making of the advertising recommendation system.

[0090] In the process of advertising, extracting and incorporating high-value content features is the core task of building accurate user portraits. When using the updated interest fluctuation model, content features and users' current interests are analyzed and matched in real time, and content features with high analytical value are screened out by calculating the interest orientation index (IOI). The interest orientation index not only considers the matching degree between content features and users' current interests, but also evaluates the dynamic changes of users' interests to ensure that the screened content features have actual analytical value. By incorporating these high-value content features into historical behavior data, the user portrait model can be further enriched and updated to improve the accuracy of advertising and user experience.

[0091] Comprehensive semantic relevance, user intent and user interaction rate, use nonlinear mapping and adjustment mechanisms to analyze the value of content features, select content features with actual value and incorporate them into real-time behavior data, including the following:

[0092] S5.1, use Word2Vec or BERT to convert content features and user historical behavior features into vector representation; let the content feature vector be , the user's historical behavior feature vector is ;in, An index representing content features; Represents the index of historical behavior features; calculates the cosine similarity between the content feature vector and the historical behavior feature vector : ; Take the maximum similarity of all historical behavior features as the semantic relevance score : .

[0093] S5.2, use a pre-trained intent classification model (such as a BERT-based classifier) ​​to predict the user's intent category and strength when browsing the current content; let the intent category be , the intention strength is ;in, Indicates the index of the intent category; normalize the intent strength to get the user intent score : .

[0094] S5.3, Calculate content feature importance factor , which is used to adjust the weight of semantic relevance and user intent in value calculation: ;in, It is the user interaction rate of content features, such as click-through rate or dwell time; combine the semantic relevance score and user intent score to calculate the content feature value index : ; The weights of the two scores are adjusted by the importance factor, and the Sigmoid function is used to smooth the scores to ensure that the two scores are balanced in different ranges.

[0095] The feature value index (CFVI) is used to represent and reflect the actual value of the current content features in user profiling and advertising. It combines semantic relevance, user intent, and user interaction rate, and evaluates the relevance and influence of content features through nonlinear mapping and adjustment mechanisms. In this scenario, the larger the feature value index, the more relevant the content feature is to the user's historical behavior, the user's current intent is strong, and the interaction frequency is high, indicating that it has a high analytical value and is suitable for inclusion in real-time behavior data for further processing and recommendation; conversely, the smaller the feature value index, the less relevant the content feature is to the user, the intention is not strong, or the interaction rate is low, which means that its analytical value is low and it is not suitable for inclusion in real-time behavior data.

[0096] The present invention vectorizes content features and user historical behavior features by using Word2Vec or BERT, calculates the cosine similarity and takes the maximum similarity as the semantic relevance score, predicts user intent in combination with the pre-trained intent classification model and normalizes it to obtain the user intent score, calculates the content feature importance factor and smoothes the score through the Sigmoid function, and finally calculates the content feature value index (CFVI) by comprehensively considering semantic relevance, user intent and user interaction rate, so as to accurately evaluate the actual value of content features in user portraits and advertising delivery. The larger the feature value index, the more relevant the content features are, the stronger the intent is, and the higher the interaction frequency is, which is suitable for inclusion in real-time behavior data for further processing and recommendation; conversely, the smaller the index is, the lower its analytical value is. The core invention is that by integrating multi-dimensional features and nonlinear adjustment mechanisms, the actual value of content features is dynamically evaluated to improve the accuracy and effect of user portraits and advertising delivery.

[0097] In the intelligent advertising delivery system, accurately evaluating the value of content features is the key to ensuring accurate advertising delivery and improving user experience. Content features not only need to be highly matched with the user's current interests, but also should have sufficient interaction rate and actual value to ensure their importance and persistence in the user portrait. By comprehensively analyzing the semantic relevance, user intent and user interaction rate of content features, and using nonlinear mapping and adjustment mechanisms, the content feature value index (CFVI) is calculated. This index can accurately reflect the actual value of content features in user portraits and advertising delivery, thereby screening out the most analytical and actual value of content features, incorporating historical behavior data, and further optimizing the user portrait model. This process not only improves the accuracy and comprehensiveness of user portraits, but also provides a solid data foundation for intelligent advertising delivery, ensuring that advertisements can reach target users more accurately and maximize the value of advertising delivery.

[0098] The content features with analytical and practical value are used as part of the behavioral features to further update the user portrait model, including the following:

[0099] The feature excellence index is calculated by combining the interest orientation index (IOI) and the content feature value index (CFVI), and screening out content features with analytical and practical value. : .

[0100] It is used to smooth the interest-directed index so that larger values ​​do not excessively dominate the calculation results. At the same time, the growth characteristics of the logarithmic function can effectively amplify the differences in low values, ensuring that interest fluctuations are sensitively captured.

[0101] Used to process the content feature value index, the hyperbolic tangent function maps the index value to the interval [−1,1][-1,1][−1,1], moderately amplifies the intermediate values, suppresses the influence of extreme values, and ensures that the relevance of content features and user interaction rate are reasonably reflected.

[0102] The inverse sine function (inverse sine function) is introduced to combine the interest orientation index and the content feature value index, and further ensure the coordination and influence of the two indexes through complex nonlinear combination. The design of this item is inspired by the characteristics of trigonometric functions, which can balance and adjust the relative influence of the two indexes to avoid excessive dominance of a single indicator.

[0103] Smoothing and amplification: The logarithmic function and hyperbolic tangent function are used to smooth and amplify numerical changes in different ranges, respectively, to ensure that the calculation results are sensitive to data in different numerical ranges.

[0104] Balance and adjustment: The inverse sine function term achieves balance and adjustment of the relative influence of IOI and CFVI through nonlinear combination, ensuring that both are reasonably reflected in the comprehensive calculation.

[0105] Avoid extreme effects: Through nonlinear mapping and complex combinations, the excessive impact of extreme values ​​on the final results is suppressed, improving the robustness and stability of the calculation.

[0106] Scenario applicability: This formula combines actual scenarios and can dynamically reflect the comprehensive value of user interests and content features, ensuring that the screened content features are of practical significance in user profiling and advertising delivery.

[0107] The Content Excellence Index (CEI) is used to represent and reflect the comprehensive value of current content features in user portraits and advertising delivery. It combines the Interest Orientation Index (IOI) and the Content Feature Value Index (CFVI) to evaluate the relevance of content features, user interest matching and interaction rate through a nonlinear combination function. The larger the Feature Excellence Index, the more closely the content features match the user's current interests and have a high interaction rate and actual value, making them suitable for inclusion in real-time behavior data for further processing and recommendation. Conversely, the smaller the Feature Excellence Index, the less relevant the content features are to the user, the less intent they have, or the lower the interaction rate, meaning that their comprehensive value is low and they are not suitable for inclusion in real-time behavior data.

[0108] The feature excellence index (Content Excellence Index, CEI) is compared with the excellence threshold. If the feature excellence index is greater than or equal to the excellence threshold, it means that the content feature has a higher comprehensive value in user profiling and advertising delivery, and is suitable for inclusion in real-time behavior data for further processing and recommendation; conversely, if the feature excellence index is less than the excellence threshold, it means that the content feature has a lower comprehensive value in user profiling and advertising delivery, and is not included in real-time behavior data.

[0109] Based on the latest user portrait model, the most relevant advertisements are matched from the advertisement database, including the following: First, the feature vectors in the latest user portrait model are extracted. These vectors include the user's interests, preferences, behavior patterns, and recent interest dynamics. Next, these user portrait features are matched with the advertisement feature vectors in the advertisement database. The advertisement feature vectors are also processed by vectorization technology and contain data such as keywords of the advertisement content, target audience information, and delivery history. Then, the similarity between the user portrait features and the advertisement feature vectors is calculated. Commonly used methods include cosine similarity and Euclidean distance to quantify the matching degree between the user portrait and the advertisement features. In order to ensure the accuracy of the matching results, machine learning algorithms such as collaborative filtering and matrix decomposition can be applied to optimize the similarity calculation. According to the calculated similarity, the advertisements that best match the user portrait are screened out to form an advertisement recommendation list. Finally, the effect of advertisement matching is evaluated through verification mechanisms such as A / B testing and user feedback data, and the user portrait model and advertisement matching algorithm are continuously optimized to ensure the accuracy of advertisement delivery and the improvement of user experience.

[0110] The present invention analyzes the feature coverage of the current user portrait, determines whether content features are needed to supplement and assist in updating the portrait, then extracts the interest-oriented index and content feature value index corresponding to the analytical value and actual value of the content features, conducts a comprehensive analysis to obtain the feature excellence index, screens content features that meet the requirements and incorporates them into historical behavior data, and further updates the user portrait model, thereby ensuring the quality of the user portrait model, ensuring the accuracy and intelligence of advertising delivery, and maximizing the value of advertising delivery. The beneficial effect of the present invention is that by dynamically evaluating and updating user portraits, using nonlinear adjustment and comprehensive analysis mechanisms, high-value content features are effectively screened, the comprehensiveness and accuracy of user portraits are improved, and it is suitable for advertising delivery value management methods based on intelligent delivery, making advertising delivery more accurate and efficient, and significantly improving the effect of advertising delivery and user experience.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0113] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0114] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for managing advertising value based on intelligent delivery, characterized in that: Collect and store user behavior feature data and content feature data in real time on the cloud platform; Based on real-time behavioral feature data, dynamically update the user portrait model and user interest fluctuation model to reflect the user's latest interests and behavior patterns; Dynamically decide whether to trigger the use of content features for supplementation based on the match between the existing feature coverage and the actual user interaction frequency; Use the updated interest fluctuation model to analyze and match content features with users' current interests in real time, and select content features with analytical value based on similarity and interest changes to include in real-time behavior data; Comprehensive semantic relevance, user intent and user interaction rate, use nonlinear mapping and adjustment mechanisms to analyze the value of content features, and select content features with actual value to be included in real-time behavior data; Use content features with analytical and practical value as part of behavioral features to further update the user portrait model; Based on the latest user portrait model, match the most relevant ads from the ad database; Dynamically decide whether to trigger the use of content features for supplementation based on the matching degree between the existing feature coverage and the actual user interaction frequency, including the following: S3.1, vectorize each feature in the user portrait, set ; Vectorize the user's actual interactive behavior and set it as ; Calculate the coverage of each user portrait feature and the user's actual interactive behavior, and calculate the matching degree between each feature and user behavior through cosine similarity ; Take the average coverage of all features to get the average coverage of user portraits ; S3.2, User Profile Completeness The calculation formula is as follows: ; in, is the portrait threshold, which indicates the ideal user portrait coverage; S3.3, calculate the user portrait completeness once in each time period and collect the user portrait completeness values ​​within a period of time; Use moving average to smooth the collected user portrait completeness values; After calculating the moving average, use the median to smooth the moving average; The median smoothed value is used as the most representative user portrait completeness value per unit time, and is marked as the user portrait completeness index for decision making: when the user portrait completeness index is less than 1, it means that the user portrait is incomplete and needs to be supplemented with content features to generate a supplementary signal; Use the updated interest fluctuation model to analyze and match content features with users' current interests in real time, and select content features with analysis value based on similarity and interest changes to include in real-time behavior data, including the following: S4.1, after obtaining the supplementary signal, vectorize the latest behavior data, use NLP technology and behavior analysis methods to extract the user's current interest vector ; S4.2, vectorize the current content feature vector to obtain the content feature vector ; Calculate the cosine similarity between the content feature vector and the user's current interest vector , represents the similarity between two vectors; S4.3, Analyze users’ short-term interests using interest fluctuation model and long-term interests ; Calculate the rate of change of short-term interest vector and long-term interest vector , indicating the degree of change of user interest: ; S4.4, comprehensively consider the matching degree between content features and current interests and the user's interest change rate, and calculate the interest orientation index : ; Comprehensive semantic relevance, user intent and user interaction rate, use nonlinear mapping and adjustment mechanisms to analyze the value of content features, select content features with actual value and incorporate them into real-time behavior data, including the following: S5.1, converting content features and user historical behavior features into vector representation; Assume that the content feature vector is , the user's historical behavior feature vector is ; in, An index representing content features; An index representing historical behavior characteristics; Calculate the cosine similarity between the content feature vector and the historical behavior feature vector ; Take the maximum similarity of all historical behavior features as the semantic relevance score : ； S5.2, use the pre-trained intent classification model to predict the user's intent category and intensity when browsing the current content; Assume the intent category is , the intention strength is in, An index representing the intent category; Normalize the intent strength to get the user intent score : ; S5.3, Calculate content feature importance factor , which is used to adjust the weight of semantic relevance and user intent in value calculation: ; in, is the user interaction rate of content features; Combine the semantic relevance score and user intent score to calculate the content feature value index : ； The content features with analytical and practical value are used as part of the behavioral features to further update the user portrait model, including the following: The interest index and content feature value index are combined, and content features with analytical and practical value are screened out to calculate the feature excellence index. : ； Compare the feature excellence index with the excellence threshold. If the feature excellence index is greater than or equal to the excellence threshold, it means that the corresponding content feature has a higher comprehensive value in user profiling and advertising delivery, and is suitable for inclusion in real-time behavior data for further processing and recommendation. Conversely, if the feature excellence index is less than the excellence threshold, it means that the corresponding content feature has a lower comprehensive value in user profiling and advertising delivery, and is not included in real-time behavior data.

2. The advertising value management method based on intelligent delivery according to claim 1, characterized in that: Based on real-time behavioral feature data, dynamically update the user portrait model and user interest fluctuation model, including the following: S2.1, calling the behavior feature data in the cloud platform and extracting key features from the behavior feature data; S2.2, User portrait model update: Integrate new behavior feature data with existing user portraits and use incremental learning algorithms to update user interests, preferences, and behavior patterns; User interest fluctuation model update: Use time series analysis to analyze new behavior feature data, capture the changing trends and periodic fluctuations of user interests, and update the parameters in the user interest fluctuation model; S2.

3. Verify the accuracy and effectiveness of the updated user portrait model and interest fluctuation model through cross-validation or A / B testing.

Images