Real-time marketing decision optimization method and system based on dynamic user portrait

Abstract

The present application relates to the technical field of marketing, and particularly relates to a real-time marketing decision optimization method and system based on dynamic user portrait. The method comprises the following steps: obtaining multi-dimensional real-time data of a user, wherein the multi-dimensional real-time data comprises a plurality of behavior characteristic information; calculating a behavior characteristic value of the user according to the multi-dimensional real-time data; inputting the behavior characteristic value into a preset dynamic user portrait model, wherein the dynamic user portrait model is trained on the behavior characteristic value, and a training result is used as a replacement to update a dynamic user portrait of the user; inputting the updated dynamic user portrait into a preset marketing decision algorithm to calculate expected revenue values of different marketing schemes; and selecting a marketing scheme with the maximum expected revenue value as an optimal marketing scheme for implementation. In the present application, the multi-dimensional real-time data is used to train and update the dynamic user portrait of the user, so as to quickly capture changes in user behavior, thereby improving the accuracy of marketing information pushing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marketing, and in particular to a real-time marketing decision optimization method and system based on dynamic user portrait. BACKGROUND

[0002] In the traditional marketing decision-making process, enterprises usually develop marketing strategies based on static user portraits. Static user portraits are often constructed from limited data such as age, gender, consumption history, etc. over a period of time, and are difficult to fully and timely reflect the real-time needs and behavior changes of users.

[0003] With the popularity of the Internet and mobile devices, user behavior and preferences have shown a high degree of dynamic characteristics. Users' interests may change dramatically in a short period of time due to real-time hot events, social media influence, changes in personal life status, etc. For example, on e-commerce platforms, users may suddenly have strong purchase desires for the same goods in a popular TV series due to the airing of the series; in the tourism market, users may quickly change their travel plans and tourism consumption preferences due to sudden natural disasters in the local area.

[0004] In summary, the existing marketing decision-making method based on static user portraits cannot quickly perceive and adapt to these dynamic changes, resulting in low accuracy of marketing information push, serious waste of marketing resources, and difficulty in maximizing marketing effectiveness. SUMMARY

[0005] In order to overcome the technical defect of low accuracy of marketing information push in the prior art, the purpose of the present application is to provide a real-time marketing decision optimization method and system based on dynamic user portrait, which updates the dynamic user portrait of the user by training with the obtained multi-dimensional real-time data, so as to quickly capture the changes in user behavior and improve the accuracy of marketing information push.

[0006] The present application discloses a real-time marketing decision optimization method based on dynamic user portrait, comprising the steps of:

[0007] Obtaining multi-dimensional real-time data of the user, the multi-dimensional real-time data comprising a plurality of behavior characteristic information;

[0008] Calculating the behavior characteristic value of the user according to the multi-dimensional real-time data , the specific calculation formula being:

[0009]

[0010] wherein, represents the jth behavior characteristic information of the ith user, Let represent the weight of the j-th behavioral feature, and n represent the dimension of the behavioral feature.

[0011] behavioral feature values Input is fed into a preset dynamic user profile model, which then uses behavioral feature values... Train the system and use the training results as a replacement to update the user's dynamic user profile.

[0012] The updated dynamic user profiles are input into a pre-defined marketing decision-making algorithm to calculate the expected return on different marketing campaigns. The marketing decision-making algorithm is as follows:

[0013]

[0014] Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue that the company will receive after the i-th user accepts the k-th marketing plan;

[0015] Select expected return value The largest marketing plan is implemented as the optimal marketing plan.

[0016] Preferably, when acquiring multi-dimensional real-time data of users, the multi-dimensional real-time data is preprocessed based on differential privacy protection technology.

[0017] Preferably, the user's behavioral characteristic values ​​are calculated based on multi-dimensional real-time data. In this step, the weights of the behavioral feature information are dynamically adjusted based on the time-series characteristics of the user's behavioral feature information. .

[0018] Preferably, in the step of using the training results as a replacement to update the user's dynamic user profile, the training results are used as a replacement to update the user's dynamic user profile based on a weighted fusion formula, which is:

[0019]

[0020] in, This represents the updated dynamic user profile. For users before the update, dynamic user profiles. These are weighting coefficients, and 0 < <1.

[0021] Preferably, the updated dynamic user profile is input into a preset marketing decision algorithm to calculate the expected return on different marketing strategies. In this step, a risk assessment factor is introduced, and the expected return after introducing the risk assessment factor is calculated based on the risk assessment calculation formula. The risk assessment calculation formula is as follows:

[0022]

[0023] in, This represents the expected return after incorporating risk assessment factors. Let be the risk assessment factor for the k-th marketing plan, and 0 ≤ 0. ≤1.

[0024] Preferably, risk assessment factors Based on the failure rate of this type of marketing campaign in historical marketing data and the market volatility index, the calculation formula is as follows:

[0025]

[0026] in, This represents the failure rate of the type to which the k-th marketing campaign belongs. Represented as a market volatility index, , This is represented by the corresponding weight coefficient, and ,0< <1, 0< <1.

[0027] Preferably, users are categorized according to their behavioral characteristics. Layering is performed using a layered calculation formula, which is:

[0028]

[0029] in, This represents the hierarchical level to which the i-th user belongs. and These are respectively represented as the behavioral feature values ​​of all users. The minimum and maximum values ​​are given, where N is the preset number of tiers. For users at different tier levels, a differentiated marketing plan weight adjustment formula is used:

[0030]

[0031] in, This represents the adjusted weight of the k-th marketing campaign for the user. Represented as initial weights, Represented as hierarchical levels The corresponding weight adjustment coefficient.

[0032] In view of this, a second objective of the present invention is to provide a real-time marketing decision optimization system based on dynamic user profiles, comprising:

[0033] Acquisition module: Used to acquire multi-dimensional real-time data of users, which includes several behavioral feature information;

[0034] Behavioral trait value Calculation module Used to calculate user behavioral feature values ​​based on multi-dimensional real-time data acquired by the acquisition module. The specific calculation formula is as follows:

[0035]

[0036] in, This is represented as the j-th behavioral feature information of the i-th user. Let represent the weight of the j-th behavioral feature, and n represent the dimension of the behavioral feature.

[0037] Training / Replacement Module: Users will input behavioral feature values The calculation module calculates the behavioral feature values. Input into a preset dynamic user profile model to evaluate behavioral feature values. Train the system and use the training results as a replacement to update the user's dynamic user profile.

[0038] Expected return Calculation module: Used to input the dynamic user profiles updated by the training / replacement module into a preset decision algorithm to calculate the expected revenue of different marketing plans. The marketing decision-making algorithm is as follows:

[0039]

[0040] Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue that the company will receive after the i-th user accepts the k-th marketing plan;

[0041] Selection module: Used to select the expected return value. The expected return value calculated by the calculation module The largest marketing plan is implemented as the optimal marketing plan.

[0042] Compared with existing technologies, the advantages of adopting the above technical solution are as follows: This solution uses multi-dimensional real-time data to train and update dynamic user profiles, enabling rapid capture of changes in user behavior. Compared to existing methods based on static user profiles that rely on limited historical data, this solution improves the accuracy of marketing message delivery. Furthermore, this solution uses computational... Calculate behavioral feature values And adopt a weighted fusion method Updating dynamic user profiles further improves the accuracy of marketing message delivery; this solution utilizes marketing decision-making algorithms based on dynamic user profiles. Calculate the expected return on different marketing plans. And introduce risk assessment factors Derive the expected return after taking risks into account This allows for the selection of the optimal marketing strategy, precise allocation of marketing resources, and avoidance of resource waste. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating the steps of a real-time marketing decision optimization method and system based on dynamic user profiles according to the present invention. Detailed Implementation

[0044] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0045] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0046] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0047] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0048] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0049] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0050] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0051] This embodiment discloses a real-time marketing decision optimization method based on dynamic user profiles, including the following steps: acquiring multi-dimensional real-time data of users, the multi-dimensional real-time data including several behavioral feature information; calculating the user's behavioral feature values ​​based on the multi-dimensional real-time data. The specific calculation formula is as follows: ,in, This is represented as the j-th behavioral feature information of the i-th user. The weight of the j-th behavioral feature is represented by n, where n represents the dimension of the behavioral feature value. Input is fed into a preset dynamic user profile model, which then uses behavioral feature values... The system is trained, and the training results are used to update the user's dynamic user profile. The updated dynamic user profile is then input into a pre-defined marketing decision-making algorithm to calculate the expected return on different marketing strategies. The marketing decision-making algorithm is as follows: Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The expected revenue will be generated for the company after the i-th user accepts the k-th marketing plan; select the expected revenue value. The largest marketing plan is implemented as the optimal marketing plan.

[0052] See Figure 1 As shown in this embodiment, a real-time marketing decision optimization method based on dynamic user profiles will be described in detail, specifically including the following steps:

[0053] Step S100: First, acquiring multi-dimensional real-time user data is fundamental. This acquisition includes, but is not limited to, obtaining data through e-commerce platform order management and user registration systems, as well as third-party data platforms such as professional market research institutions and industry data statistics platforms. Multi-dimensional real-time data includes several behavioral characteristic information. Specifically, it includes, but is not limited to, basic user information, browsing behavior information, purchasing behavior information, and social information. For example, in an e-commerce platform, basic user information helps businesses understand the user's general consumption level; browsing behavior information reflects the user's current interests; purchasing behavior information reflects the user's spending power and preferences; and social data can provide insights into the user's social influence and potential needs.

[0054] Step S200: In this step, the user's behavioral feature values ​​will be calculated based on the multi-dimensional real-time data obtained in step S100 above. The specific calculation formula is as follows: .in, This is represented as the j-th behavioral feature information of the i-th user. Let represent the weight of the j-th behavioral feature, and n represent the dimension of the behavioral feature.

[0055] This calculation formula comprehensively considers different behavioral characteristics and their weights. Taking e-commerce users as an example, if... The duration a user spends browsing a certain type of product. Weighting based on browsing time, The amount paid by the user for this type of product. Assuming the purchase amount is weighted, this formula can be used to weight and synthesize data from different dimensions to obtain a value that represents the user's behavioral characteristics in a certain aspect. .

[0056] Step S300: In this step, the behavioral feature values ​​calculated in step S200 are... The input is fed into a preset dynamic user profile model to analyze behavioral feature values. The system is trained and the training results are obtained. These results are then used to update the user's dynamic user profile. For example, in some embodiments, when a user frequently browses sports equipment and purchases related products, the dynamic user profile model will adjust the content of the dynamic user profile regarding sports interests, consumption preferences, etc., based on this information.

[0057] Step S400: In this step, the expected revenue of different marketing plans will be calculated based on the dynamic user profile updated in step S300 using a preset marketing decision-making algorithm. The formula for the marketing decision-making algorithm is: Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue generated for the company after the i-th user accepts the k-th marketing plan.

[0058] For example, in some embodiments, assuming the company has designed k marketing plans, for the i-th user, the probability of accepting the k-th marketing plan is assessed based on their updated dynamic user profile. And the benefits that the company will gain after accepting the plan. By summing up the relevant information of all users, the expected revenue value of each marketing plan can be obtained.

[0059] Step S500: In this step, the expected revenue values ​​calculated in step S400 for different marketing plans will be reviewed. The maximum value is selected as the optimal marketing plan for implementation. For example, in some embodiments, on a certain apparel e-commerce platform, the company designed three marketing plans. Plan 1 is a 20% discount on all items; Plan 2 is a discount for purchases over a certain amount; Plan 3 is a buy-one-get-one-free offer. After the above calculations, the expected revenue value of Plan 2 is obtained. If the result is the largest, then Option 2 will be selected for implementation.

[0060] Furthermore, when acquiring multi-dimensional real-time data from users, the multi-dimensional real-time data is preprocessed based on differential privacy protection technology.

[0061] In this embodiment, step S100 will be described in detail again. When acquiring multi-dimensional time data of users, the multi-dimensional real-time data is preprocessed based on differential privacy protection technology. This differential privacy protection technology adds specific noise to the metadata to anonymize user factor data without affecting the overall distribution and statistical characteristics of the multi-dimensional real-time data. For example, in some embodiments, when acquiring user purchase behavior data on e-commerce platforms, noise is added to sensitive information such as the user's purchase amount and the category of purchased goods, making it impossible for attackers to accurately identify specific user privacy information by analyzing multi-dimensional time data. This approach ensures the usability of multi-dimensional time data for constructing dynamic user profiles and making marketing decisions, while effectively preventing the risk of data leakage.

[0062] Furthermore, based on multi-dimensional real-time data, user behavioral characteristic values ​​are calculated. In this step, the weights of the behavioral feature information are dynamically adjusted based on the time-series characteristics of the user's behavioral feature information. .

[0063] In this embodiment, the user's behavioral feature values ​​will be calculated based on multi-dimensional real-time data in step 300 above. At the same time, an attention mechanism is introduced to dynamically adjust the weight of each behavioral feature based on the time-series characteristics of the user's behavioral features. For example, in some embodiments, taking video website users as an example, if a user has recently frequently watched a certain type of video, the user's behavioral feature value is calculated. In this process, the attention mechanism automatically increases the weight of behavioral data related to that video type (such as viewing time and frequency), giving these recent behavioral characteristics a higher proportion in the calculation and thus accurately reflecting the user's current interest in that video type. Compared to fixed weight calculation, this dynamic weight adjustment method can better adapt to the dynamic changes in user behavior, improving the real-time performance and accuracy of dynamic user profiles.

[0064] It should be noted that the preset dynamic user profile model employs an autoencoder neural network model, incorporating a graph convolutional network (GCNN) during the encoding and decoding processes. The autoencoder neural network can extract and compress user behavioral features, while the GCNN can effectively process user social network data. For example, in some embodiments, within a social e-commerce scenario, complex social relationships exist between users. The GCNN can uncover potential connections between users, such as friend recommendations and the influence of social circles. Integrating this information into the dynamic user profile update process allows for accurate scanning of user interests and consumption tendencies, demonstrating a stronger capability in processing social data and updating dynamic user profiles compared to traditional neural network models.

[0065] It should be noted that this embodiment will describe in detail the number of layers in the autoencoder neural network model used. For example, the number of layers in the autoencoder is set to 3. The number of neurons in the input layer is determined according to the dimension of the behavioral feature information. Assuming the dimension of the behavioral feature information is 50, then the number of neurons in the input layer is 50. The number of neurons in the hidden layer can be set to 30 and 20 respectively. Key features are extracted by progressively compressing the behavioral feature information. The number of neurons in the output layer is determined according to the dimension of the dynamic user profile. For example, if the dynamic user profile includes information in 10 dimensions, then the number of neurons in the output layer is 10. At the same time, this embodiment will describe in detail the combination of graph convolutional network and autoencoder. The graph convolutional network is responsible for processing social relationship network data, and its output is used as a supplement to the input of the hidden layer of the autoencoder, thereby fusing social relationship features into the dynamic user profile.

[0066] It's also worth noting that the parameters of the dynamic user profile model during training will be described in detail. For example, the learning rate is set to 0.001, which controls the step size for parameter updates during training. This learning rate setting ensures that the dynamic user profile model won't miss the optimal solution due to an excessively large step size, nor will it slow down the training speed due to an excessively small step size. The batch size is set to 64, meaning that 64 samples are selected from the training data for calculation and parameter updates during each training iteration, thereby improving training efficiency and stability. The number of training epochs is set to 100. During these 100 training epochs, the dynamic user profile model continuously adjusts its parameters to optimize the learning of the relationship between user behavioral characteristics and the dynamic user profile.

[0067] It's important to note that the pre-defined marketing decision-making algorithm is based on a multi-agent competitive and collaborative reinforcement learning algorithm. Multiple agents correspond to different marketing sub-domains, such as product promotion, pricing strategy, and promotional activities. In complex market environments, these agents interact through a competitive and collaborative mechanism. For example, during major e-commerce promotions, the agent responsible for product promotion and the agent responsible for pricing strategy will continuously adjust their decisions based on market feedback and each other's decision information, achieving collaborative optimization of different marketing plans. This algorithm is better able to handle market uncertainties and complexities, significantly improving the accuracy and adaptability of marketing decisions compared to a single reinforcement learning algorithm.

[0068] It should be noted that this embodiment will describe the multi-agent competition and cooperation mechanism in detail. Each agent has its own independent policy network and value network for decision-making and evaluating the effectiveness of decisions. In terms of competition, when different agents are responsible for decisions in different marketing sub-domains (such as product promotion, pricing strategy, and promotional activities), they will compete for limited marketing resources based on their own goals and evaluation metrics. The agent responsible for product promotion aims to increase product exposure, while the agent responsible for pricing strategy pursues profit maximization; they compete for resource allocation. In terms of cooperation, agents collaborate by sharing some information and reward signals. For example, in some embodiments, after the product promotion agent discovers that the exposure of a certain type of product has increased in a specific area, it shares this information with the agent responsible for promotional activities. Based on this, the promotional activities agent launches a targeted promotional plan for that area, achieving collaborative optimization of the strategy.

[0069] Furthermore, this embodiment will describe the agent's decision-making rules and interaction methods in detail. Based on the current market state and user profile information, the agent outputs decision-making behavior through a policy network. The decision-making rules can adopt... - Greedy strategy, with a certain probability (e.g.) With a probability of 0.1, a decision is randomly selected to explore new marketing strategies; otherwise, the network chooses the decision deemed optimal by the current strategy. Interaction between agents utilizes a message passing mechanism. When an agent makes a decision or receives significant market feedback, it encapsulates the relevant information into a message and sends it to other agents. Upon receiving the message, other agents update their state and decision-making strategies based on their own circumstances.

[0070] It should be noted that when some behavioral feature information is missing, resulting in an incomplete dynamic user profile, the intelligent agent will fill in the missing behavioral feature information with corresponding behavioral feature values ​​through data injection. To ensure the completeness and accuracy of dynamic user profiles. When corresponding numerical behavioral feature information is missing, if the missing percentage is low (e.g., below 10%), methods include, but are not limited to, using the mean or median to fill in the missing behavioral feature information; if the missing percentage is high (e.g., above 30%), methods include, but are not limited to, using machine learning algorithms for prediction and filling. For missing categorical behavioral feature information, methods include, but are not limited to, using the mode to fill in the missing information or using classification prediction based on data distribution characteristics. For outliers, methods include, but are not limited to, using statistical methods (e.g., 3D modeling). Anomalies can be identified and processed using principles or machine learning-based anomaly detection algorithms (such as the Isolation Forest algorithm). For identified anomalies, correction and deletion can be performed.

[0071] Additionally, it's important to discuss processing capabilities and optimization strategies as the scale of multi-dimensional time-based data increases. With the influx of multi-dimensional real-time data, intelligent agents employ distributed computing architectures to improve data processing efficiency and perform parallel computation of behavioral feature values. The system trains dynamic user profile models and marketing decision-making algorithms to improve processing efficiency. Simultaneously, it regularly updates and optimizes the dynamic user profile model, dynamically adjusting its parameters and structure based on the growth of multi-dimensional real-time data. For example, it increases the number of layers and neurons in the autoencoder to adapt to the complexity brought by large-scale data.

[0072] It's important to note that after selecting and implementing the optimal marketing plan, federated learning technology is introduced. This involves combining user data from multiple partner companies to train the model. For example, companies from different industries, without sharing original user behavioral feature information, can use federated learning to train their models on their respective ontologies and upload updated model parameters to collaboratively optimize the dynamic user profile model and marketing decision-making algorithm. This approach protects the data privacy of both companies and users while fully utilizing resources from multiple parties to improve the performance of the dynamic user profile model and marketing decision-making algorithm.

[0073] Furthermore, in the step of using the training results as a replacement to update the user's dynamic user profile, the training results are used as a replacement to update the user's dynamic user profile based on a weighted fusion formula, which is: ,in, This represents the updated dynamic user profile. For users before the update, dynamic user profiles. These are weighting coefficients, and 0 < <1.

[0074] In this embodiment, in step S300 above, the training results are used as replacements to update the user's dynamic user profile based on a weighted fusion formula. The weighted fusion formula is as follows: .in, This represents the updated dynamic user profile. For users before the update, dynamic user profiles. These are weighting coefficients, and 0 < <1. In this embodiment, the weighted fusion formula calculates the behavioral feature values. Historical dynamic user profiles To achieve fusion, the weighting coefficients are adjusted. This aims to balance the impact of multi-dimensional real-time data with historical multi-dimensional real-time data on dynamic user profiles. For example, in some implementations, on a short video platform, if a user has recently frequently watched knowledge-based videos, a higher user behavior characteristic value is calculated. ,like If the value is 0.6, then the new dynamic user profile It will tend to reflect users' interest in knowledge-based content, while retaining some information about other user interests from historical dynamic user profiles. This allows for rapid response to changes in user behavior without completely discarding user characteristics reflected in historical, multi-dimensional, real-time data, making the updated dynamic user profiles more accurate and comprehensive.

[0075] Furthermore, the updated dynamic user profiles are input into a pre-defined marketing decision-making algorithm to calculate the expected return on different marketing campaigns. In this step, a risk assessment factor is introduced, and the expected return after introducing the risk assessment factor is calculated based on the risk assessment calculation formula. The risk assessment calculation formula is as follows: in, This represents the expected return after incorporating risk assessment factors. Let be the risk assessment factor for the k-th marketing plan, and 0 ≤ 0. ≤1.

[0076] Furthermore, risk assessment factors Based on the failure rate of this type of marketing campaign in historical marketing data and the market volatility index, the calculation formula is as follows: ,in, This represents the failure rate of the type to which the k-th marketing campaign belongs. Represented as a market volatility index, , This is represented by the corresponding weight coefficient, and ,0< <1, 0< <1.

[0077] In this embodiment, in step S400 above, the updated dynamic user profile is input into a preset marketing decision algorithm to calculate the expected revenue of different marketing plans. When introducing risk assessment factors, the expected return after introducing risk assessment factors is calculated based on the risk assessment calculation formula, which is: Among them, among them, This represents the expected return after incorporating risk assessment factors. Let be the risk assessment factor for the k-th marketing plan, and 0 ≤ 0. ≤1. Risk assessment factor Based on the failure rate of this type of marketing campaign in historical marketing and the market volatility index, the specific calculation formula is as follows: .in, This represents the failure rate of the type to which the k-th marketing campaign belongs. Represented as a market volatility index, , This is represented by the corresponding weight coefficient, and ,0< <1, 0< <1.

[0078] Specifically, first, according to the calculation formula Risk assessment factors were calculated. Taking a clothing brand's limited-time discount event as an example, the failure rate of similar limited-time discount events in the past was analyzed. And the volatility index of the current apparel market Combined with weighting coefficients , Calculate the risk assessment factors for this activity. If market competition is intense and similar products frequently offer discounts, it can lead to market volatility. The failure rate for this type of activity is relatively high, and historically it has been low. The risk assessment factor is not low either, so the calculated risk assessment factor The value is relatively large, and the final expected return value after considering the risks is obtained. This will correspondingly reduce costs. Businesses can use this method to reasonably assess the feasibility of marketing plans, avoid making blind decisions, and reduce marketing risks.

[0079] Furthermore, users will be categorized according to their behavioral characteristics. Layering is performed using a layered calculation formula, which is: ,in, This represents the hierarchical level to which the i-th user belongs. and These are respectively represented as the behavioral feature values ​​of all users. The minimum and maximum values ​​are given, where N is the preset number of tiers. For users at different tier levels, a differentiated marketing plan weight adjustment formula is used: ,in, This represents the adjusted weight of the k-th marketing campaign for the user. Represented as initial weights, Represented as hierarchical levels The corresponding weight adjustment coefficient.

[0080] In this embodiment, the user is categorized according to their behavioral characteristic values. Layering is performed using a layered calculation formula, which is: ,in, This represents the hierarchical level to which the i-th user belongs. and These are respectively represented as the behavioral feature values ​​of all users. The minimum and maximum values ​​are given, where N is the preset number of tiers. For users at different tier levels, a differentiated marketing plan weight adjustment formula is used: ,in, This represents the adjusted weight of the k-th marketing campaign for the user. Represented as initial weights, Represented as hierarchical levels The corresponding weight adjustment coefficient. In this embodiment, users can be divided into N different levels. For example, in an e-commerce platform, users can be divided into 5 levels, such as those with frequent purchases and high purchase amounts, thereby adjusting the user's behavioral characteristic values. Higher-level users will be assigned to higher tiers. Differentiated marketing strategies and weighting formulas are applied to users at different user tiers. For high-value users (i.e., users with higher tiers), the weight of marketing programs targeting high-end products may be increased, such as increasing the weight of luxury goods recommendations, to better meet their needs and improve marketing effectiveness. For low-value users, the weight of such programs may be appropriately reduced, while the weight of marketing programs targeting cost-effective products may be increased, thereby achieving precise marketing and improving resource utilization efficiency.

[0081] This embodiment also provides a real-time marketing decision optimization system based on dynamic user profiles, including: an acquisition module for acquiring multi-dimensional real-time data of users, the multi-dimensional real-time data including several behavioral feature information; and behavioral feature values. Calculation module Used to calculate user behavioral feature values ​​based on multi-dimensional real-time data acquired by the acquisition module. The specific calculation formula is as follows: ,in, This is represented as the j-th behavioral feature information of the i-th user. The weight of the j-th behavioral feature is represented by n, where n represents the dimension of the behavioral feature; Training / Replacement Module: The user assigns behavioral feature values... The calculation module calculates the behavioral feature values. Input into a preset dynamic user profile model to evaluate behavioral feature values. The training process is conducted, and the training results are used to update the user's dynamic user profile; expected revenue. Calculation module: Used to input the dynamic user profiles updated by the training / replacement module into a preset decision algorithm to calculate the expected revenue of different marketing plans. The marketing decision-making algorithm is as follows: Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The expected revenue for the enterprise after the i-th user accepts the k-th marketing plan; Selection module: used to select the expected revenue value. The expected return value calculated by the calculation module The largest marketing plan is implemented as the optimal marketing plan.

[0082] In this embodiment, a real-time marketing decision optimization system based on dynamic user profiles will be described in detail, specifically including:

[0083] The acquisition module is fundamental for obtaining multi-dimensional real-time user data. This acquisition includes, but is not limited to, data obtained through e-commerce platform order management and user registration systems, as well as third-party data platforms such as professional market research institutions and industry data statistics platforms. Multi-dimensional real-time data includes several behavioral characteristic information. Specifically, it includes, but is not limited to, basic user information, browsing behavior information, purchasing behavior information, and social information. For example, in an e-commerce platform, basic user information helps businesses understand the user's general consumption level; browsing behavior information reflects the user's current interests; purchasing behavior information reflects the user's spending power and preferences; and social data can provide insights into the user's social influence and potential needs.

[0084] Behavioral trait value Calculation module: used to calculate feature values ​​in this row. The calculation module will calculate the user's behavioral feature values ​​based on the multi-dimensional real-time data obtained by the acquisition module. The specific calculation formula is as follows: .in, This is represented as the j-th behavioral feature information of the i-th user. Let represent the weight of the j-th behavioral feature, and n represent the dimension of the behavioral feature.

[0085] This calculation formula comprehensively considers different behavioral characteristics and their weights. Taking e-commerce users as an example, if... The duration a user spends browsing a certain type of product. Weighting based on browsing time, The amount paid by the user for this type of product. By weighting the purchase amount, this formula can weight and synthesize data from different dimensions to obtain a value that represents the user's behavioral characteristics in a certain aspect. .

[0086] Training / Replacement Module: Used in this training / replacement module to process the above behavioral feature values. Behavioral feature values ​​calculated by the calculation module The input is fed into a preset dynamic user profile model to analyze behavioral feature values. The system is trained and the training results are obtained. These results are then used to update the user's dynamic user profile. For example, in some embodiments, when a user frequently browses sports equipment and purchases related products, the dynamic user profile model will adjust the content of the dynamic user profile regarding sports interests, consumption preferences, etc., based on this information.

[0087] Expected return Calculation module: used for calculating the expected return value In the calculation module, based on the dynamic user profiles updated by the training / replacement module, a pre-defined marketing decision-making algorithm is used to calculate the expected revenue of different marketing strategies. The formula for the marketing decision-making algorithm is: Where m is the number of users. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue generated for the company after the i-th user accepts the k-th marketing plan.

[0088] For example, in some embodiments, assuming the company has designed k marketing plans, for the i-th user, the probability of accepting the k-th marketing plan is assessed based on their updated dynamic user profile. And the benefits that the company will gain after accepting the plan. By summing up the relevant information of all users, the expected revenue value of each marketing plan can be obtained.

[0089] Selection Module: Used in this selection module to select the expected return value. The calculation module calculates the expected revenue for different marketing plans. The maximum value is selected as the optimal marketing plan for implementation. For example, in some embodiments, on a certain apparel e-commerce platform, the company designed three marketing plans. Plan 1 is a 20% discount on all items; Plan 2 is a discount for purchases over a certain amount; Plan 3 is a buy-one-get-one-free offer. After the above calculations, the expected revenue value of Plan 2 is obtained. If the result is the largest, then Option 2 will be selected for implementation.

[0090] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A real-time marketing decision optimization method based on dynamic user profiles, characterized in that, Including the following steps: Acquire multi-dimensional real-time data of users, including several behavioral feature information; Calculate the user's behavioral characteristic value based on the multi-dimensional real-time data. The specific calculation formula is as follows: in, This is represented by the j-th behavioral feature information of the i-th user. The weight of the j-th behavioral feature information is represented by n, where n represents the dimension of the behavioral feature information. The behavioral feature value The data is input into a preset dynamic user profile model, which uses the behavioral feature values... The training is performed, and the training results are used as replacements to update the dynamic user profile of the user based on a weighted fusion formula, which is: in, This represents the updated dynamic user profile of the user. This is the dynamic user profile of the user before the update. These are weighting coefficients, and 0 < <1; The updated dynamic user profile is input into a preset marketing decision algorithm to calculate the expected return on different marketing strategies. The marketing decision-making algorithm is as follows: Where m is the number of users who obtained the multi-dimensional real-time data. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue that the company receives after the i-th user accepts the k-th marketing plan; Select the expected return value The largest marketing plan is implemented as the optimal marketing plan.

2. The real-time marketing decision optimization method based on dynamic user profiles according to claim 1, characterized in that, When acquiring the user's multi-dimensional real-time data, the multi-dimensional real-time data is preprocessed based on differential privacy protection technology.

3. The real-time marketing decision optimization method based on dynamic user profiles according to claim 1, characterized in that, Calculate the user's behavioral characteristic values ​​based on the multi-dimensional real-time data. In this step, the weights of the behavioral feature information are dynamically adjusted based on the time-series characteristics of the user's behavioral feature information. .

4. The real-time marketing decision optimization method based on dynamic user profiles according to claim 1, characterized in that, The updated dynamic user profile is then input into a preset marketing decision-making algorithm to calculate the expected return on different marketing strategies. In this step, a risk assessment factor is introduced, and the expected return value after introducing the risk assessment factor is calculated based on the risk assessment calculation formula, which is: in, This represents the expected return after incorporating the aforementioned risk assessment factor. Let be the risk assessment factor for the k-th marketing plan, and 0 ≤ 0. ≤1.

5. The real-time marketing decision optimization method based on dynamic user profiles according to claim 4, characterized in that, The risk assessment factors According to historical marketing The failure rate and market volatility index of the marketing plan type are calculated using the following formula: in, This represents the failure rate of the type to which the k-th marketing campaign belongs. This is represented as the market volatility index. , This is represented by the corresponding weight coefficient, and ,0< <1, 0< <1.

6. The real-time marketing decision optimization method based on dynamic user profiles according to claim 1, characterized in that, The user is determined according to the user's behavioral characteristics. Layering is performed using a layered calculation formula, which is: in, This represents the hierarchical level to which the i-th user belongs. and These are respectively represented as the behavioral feature values ​​of all the users. The minimum and maximum values ​​are given, where N is the preset number of tiers. For users at different tier levels, a differentiated marketing plan weight adjustment formula is used: in, This represents the adjusted weight of the k-th marketing campaign for the user. Represented as initial weights, Represented as hierarchical levels The corresponding weight adjustment coefficient.

7. A real-time marketing decision optimization system based on dynamic user profiles, characterized in that, include: Acquisition module: used to acquire multi-dimensional real-time data of users, including several behavioral feature information; Behavioral trait value Calculation module Used to calculate the user's behavioral feature value based on the multi-dimensional real-time data acquired by the acquisition module. The specific calculation formula is as follows: in, This is represented by the j-th behavioral feature information of the i-th user. The weight of the j-th behavioral feature information is represented by n, where n represents the dimension of the behavioral feature information. Training / Replacement Module: The user assigns the behavioral feature values The behavioral feature value calculated by the calculation module Input into a preset dynamic user profile model to evaluate the behavioral feature values. The training is performed, and the training results are used as replacements to update the dynamic user profile of the user based on a weighted fusion formula, which is: in, This represents the updated dynamic user profile of the user. This is the dynamic user profile of the user before the update. These are weighting coefficients, and 0 < <1; Expected return Calculation module: Used to input the dynamic user profile updated by the training / replacement module into a preset marketing decision algorithm to calculate the expected revenue of different marketing plans. The marketing decision-making algorithm is as follows: Where m represents the number of users whose multi-dimensional real-time data is acquired by the acquisition module. Let be the probability that the i-th user accepts the k-th marketing offer. The revenue that the company receives after the i-th user accepts the k-th marketing plan; Selection module: Used to select the expected return value. The expected return value calculated by the calculation module The largest marketing plan is implemented as the optimal marketing plan.

Images