Dynamic delivery strategy optimization method and system combined with big data model

Abstract

This invention provides a method and system for optimizing dynamic delivery strategies using a big data model, relating to the fields of data processing and information delivery technology. First, based on multiple types of preset data, a big data model is invoked to generate strategy feature encoding results, constructing a dynamic response cluster for user scenarios and generating an initial dynamic delivery strategy. Next, this strategy is applied to a target user group, collecting and analyzing real-time data to generate a strategy iteration initiation command. Upon receiving the strategy iteration initiation command, the strategy feature encoding results are reconstructed and the initial dynamic delivery strategy is optimized. Then, the optimized strategy is applied to a test user group, collecting verification data to construct a strategy evolution adaptation chain. Based on the strategy evolution adaptation chain, a scenario adaptation identifier is generated, and the strategy application or re-encoding is determined according to the identifier. This invention can dynamically optimize delivery strategies and improve the accuracy of information delivery.

Description

Technical Field

[0001] This invention relates to the field of data processing and information delivery technology, and more specifically, to a method and system for optimizing dynamic delivery strategies by combining big data models. Background Technology

[0002] In today's digital age, various information delivery services are widely used in numerous fields, such as advertising push, message notifications, and service reminders. Existing methods for formulating delivery strategies have several shortcomings. On the one hand, traditional methods often rely on fixed rules or simple statistical analysis to formulate delivery strategies, lacking in-depth mining and comprehensive analysis of multi-dimensional data such as user behavior and delivery scenarios. For example, in advertising push scenarios, content is determined solely based on basic user attributes (such as age and gender), ignoring real-time user behavioral preferences in different scenarios. This results in content that doesn't match actual user needs, reducing user acceptance and wasting resources. On the other hand, existing methods lack dynamic adjustment mechanisms during the formulation and execution of delivery strategies. Delivery scenarios are complex, dynamic, and constantly evolving; user interests and behaviors change with time and environment. However, traditional methods cannot optimize and adjust delivery strategies in a timely manner based on these dynamic changes, causing the strategies to gradually lose effectiveness and fail to meet actual business needs. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for optimizing dynamic delivery strategies incorporating a big data model, the method comprising:

[0004] Based on a preset set of user behavior data, a set of delivery scenario data, and a set of historical policy feedback data, a big data model is invoked to perform policy feature encoding processing to generate policy feature encoding results. A user scenario policy dynamic response cluster is constructed based on the policy feature encoding results, and an initial dynamic delivery policy is generated based on the user scenario policy dynamic response cluster.

[0005] The initial dynamic delivery strategy is applied to the delivery scenario corresponding to the target user group, and the real-time user behavior data set, the real-time scenario change data set, and the real-time strategy feedback data set generated during the strategy execution process are collected.

[0006] The big data model is invoked to perform response cluster adaptability analysis on the user real-time behavior data set, the scenario real-time change data set, and the strategy real-time feedback data set, generating response cluster adaptability analysis results, and generating a strategy iteration start command based on the response cluster adaptability analysis results;

[0007] Upon receiving the strategy iteration start instruction, the big data model is invoked to perform feature element reconstruction processing on the strategy feature encoding result, generating a reconstructed feature encoding result. The reconstructed feature encoding result is then used to optimize the initial dynamic distribution strategy, resulting in an optimized dynamic distribution strategy.

[0008] The optimized dynamic delivery strategy is applied to the delivery scenario corresponding to the test user group. The strategy verification data set generated during the strategy verification process is collected. The big data model is called to perform evolutionary adaptation chain construction on the strategy verification data set to generate the strategy evolutionary adaptation chain.

[0009] Based on the strategy evolution adaptation chain, a scenario adaptation identifier is generated. If the scenario adaptation identifier is compatible, the optimized dynamic delivery strategy is formally applied to the delivery scenario corresponding to the target user group. If the scenario adaptation identifier is not compatible, the strategy verification data set is input into the big data model to re-execute the strategy feature encoding process.

[0010] Furthermore, embodiments of the present invention also provide a dynamic distribution strategy optimization system combining big data models, comprising:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described dynamic delivery strategy optimization method incorporating a big data model by executing the machine-executable instructions.

[0012] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described dynamic distribution strategy optimization method combined with a big data model.

[0013] Based on the above, by comprehensively utilizing pre-set user behavior data sets, distribution scenario data sets, and historical strategy feedback data sets, a big data model is invoked for strategy feature encoding processing. This constructs a dynamic response cluster for user scenario strategies and generates an initial dynamic distribution strategy. After applying the initial dynamic distribution strategy to the target user group, real-time data during strategy execution is collected, and the big data model is invoked for response cluster adaptability analysis to generate a strategy iteration start command. This enables real-time monitoring and dynamic evaluation of strategy execution effects, allowing for timely detection of mismatches between the strategy and the actual scenario. Upon receiving the strategy iteration start command, the strategy feature encoding results are reconstructed to optimize the initial dynamic distribution strategy, enabling rapid adjustment and optimization based on real-time feedback, thus improving the strategy's adaptability and effectiveness. The optimized dynamic distribution strategy is applied to a test user group, and strategy verification data is collected. A strategy evolution adaptation chain is constructed, and a scenario adaptation identifier is generated. Based on the identifier, the decision to formally apply the optimized strategy is made, further ensuring the reliability and stability of the strategy in real-world scenarios. If the scenario is not suitable, the strategy verification data is re-input into the big data model for strategy feature encoding processing. This allows for continuous improvement and refinement of the distribution strategy, thereby significantly enhancing the quality and effectiveness of the distribution strategy, improving the accuracy of information distribution, and increasing user satisfaction. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the execution flow of the dynamic distribution strategy optimization method combined with big data model provided in the embodiments of the present invention.

[0015] Figure 2 This is a schematic diagram of exemplary hardware and software components of the dynamic policy optimization system combining a big data model provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a dynamic distribution strategy optimization method combining a big data model, provided in one embodiment of the present invention. The following is a detailed description of this dynamic distribution strategy optimization method combining a big data model.

[0017] Step S110: Based on the preset user behavior data set, distribution scenario data set and historical policy feedback data set, call the big data model to perform policy feature encoding processing, generate policy feature encoding results, construct a user scenario policy dynamic response cluster based on the policy feature encoding results, and generate an initial dynamic distribution policy based on the user scenario policy dynamic response cluster.

[0018] In this embodiment, the optimization of product recommendation strategies on an e-commerce platform is used as the application scenario. The preset user behavior data set contains a large number of user behavior records on the e-commerce platform. The scenario data set covers different product recommendation scenario information, and the historical strategy feedback data set contains feedback on past recommendation strategy implementations. First, the above data needs to be input into a big data model for processing to generate key information such as the strategy feature encoding results required subsequently.

[0019] Step S111: After inputting the user behavior data set into the big data model, perform feature decomposition on each user behavior record in the user behavior data set, extract the behavior type, behavior occurrence cycle, behavior-related content and subsequent feedback corresponding to each user behavior record, and generate user behavior feature units.

[0020] In e-commerce platform scenarios, each user behavior record in the user behavior dataset contains rich information. For example, a user behavior record might show that a user browsed a certain type of product within a specific time period, added it to their shopping cart, completed the purchase, and left a positive review afterward. After inputting these user behavior records into a big data model, the model will decompose their features. Specifically, the behavior type might be browsing, adding to cart, or purchasing; the behavior occurrence period could be a fixed time each day, or specific days each week; the associated content includes the specific product category, brand, and price range that the user browsed, added to cart, or purchased; and subsequent feedback includes the user's product reviews and whether they returned the product. Through this decomposition, a user behavior feature unit containing the extracted information is generated for each user behavior record.

[0021] Step S112: Perform feature element mapping for each user behavior feature unit, converting behavior type into behavior type feature element, behavior occurrence cycle into behavior cycle feature element, behavior-related content into behavior-related feature element, and subsequent behavior feedback into behavior feedback feature element, thereby generating a set of user behavior feature elements.

[0022] After obtaining the user behavior feature units, further feature element mapping is required. For behavior types, such as "browsing," a specific behavior type feature element might be mapped. This feature element could include browsing attribute information, such as browsing duration and depth. "Adding to cart" and "purchasing" would also correspond to different behavior type feature elements. Behavior occurrence cycles, such as users habitually browsing products on Friday evenings, would be transformed into behavior cycle feature elements, which might exhibit a periodic time pattern. Behavior-related content, such as browsing mobile phones within the electronics category, a specific price range, or a particular brand, would be transformed into behavior-related feature elements. Subsequent feedback, such as five-star reviews or no returns, would be transformed into behavior feedback feature elements. After processing all user behavior feature units through the above mapping, they are integrated to generate a set of user behavior feature elements.

[0023] Step S113: After inputting the distribution scenario data set into the big data model, perform attribute parsing on each distribution scenario record in the distribution scenario data set, extract the scenario type, scenario resource configuration, scenario user capacity and scenario operation constraints corresponding to each distribution scenario record, and generate scenario feature units.

[0024] Each distribution scenario record in the distribution scenario dataset describes a different recommendation scenario on the e-commerce platform. Examples include homepage recommendation scenarios, product detail page recommendation scenarios, and event page recommendation scenarios. After inputting these distribution scenario records into the big data model, the model performs attribute parsing. Scenario type refers to the different types mentioned above, such as homepage recommendation and detail page recommendation; scenario resource configuration involves resource information such as the number of products available for recommendation in that scenario, the number of display positions, and server processing capacity; scenario user capacity refers to the maximum number of user visits that the scenario can handle within the same time period; scenario operation constraints include compliance requirements for recommended products and display format restrictions. By parsing these attributes, scenario feature units are generated for each distribution scenario record.

[0025] Step S114: Perform feature element mapping for each scene feature unit, converting scene type into scene type feature element, scene resource configuration into scene resource feature element, scene user capacity into scene capacity feature element, and scene operation constraints into scene constraint feature element, thereby generating a set of scene feature elements.

[0026] For the generated scene feature units, feature element mapping is also required. Scene type feature elements will be assigned corresponding features according to different scene types. For example, the feature elements of the homepage recommendation scene may focus more on the diversity and popularity of products, while the feature elements of the details page recommendation scene may focus more on the relevance to the currently viewed product. Scene resource configuration is transformed into scene resource feature elements, which may include quantitative information such as the size of the pool of recommendable products and the number of recommendation slots. Scene user capacity is transformed into scene capacity feature elements, reflecting the user processing capacity of the scene. Scene operation constraints are transformed into scene constraint feature elements, clarifying the rules and restrictions that the recommendation strategy must follow under this scene. After all scene feature units are mapped, they are integrated to form a scene feature element set.

[0027] Step S115: After inputting the historical strategy feedback data set into the big data model, perform feedback decomposition on each historical feedback record in the historical strategy feedback data set, extract the strategy execution effect, user satisfaction, strategy adjustment suggestions and feedback generation scenario corresponding to each historical feedback record, and generate feedback feature units.

[0028] The historical strategy feedback dataset records various feedback responses after the implementation of past recommendation strategies. Each historical feedback record may contain data on the strategy's performance, such as click-through rate and conversion rate, user satisfaction ratings for recommended products, user suggestions for strategy adjustments, and the recommendation scenario in which the feedback occurred. After inputting these historical feedback records into the big data model, the model decomposes the feedback, extracting information from each of the aforementioned aspects to generate feedback feature units.

[0029] The feedback context refers to the specific application environment or interface location where a user provides strategy feedback. In an information delivery system, the same user's feedback on different interfaces (such as the information display page, user center, feedback pop-up, etc.) may exhibit different behavioral patterns and credibility. Clearly defining the feedback context helps to more accurately assess the contextual value and weight of feedback data, thereby enabling more rational use of this feedback information in strategy optimization. For example, feedback generated by a user actively clicking the "Not Interested" button on a product details page may have different meanings and weights than the "Ignore" action on a message center list page.

[0030] Step S116: Perform feature element mapping for each feedback feature unit, transforming the strategy execution effect into effect feature elements, user satisfaction into satisfaction feature elements, strategy adjustment suggestions into adjustment suggestion feature elements, and feedback generation scenario into feedback scenario feature elements, thereby generating a set of feedback feature elements.

[0031] For each feedback feature unit, feature elements are mapped. Strategy execution effectiveness, such as click-through rate reaching a certain level or conversion rate reaching a certain percentage, is transformed into performance feature elements, reflecting the strategy's impact on improving user behavior. User satisfaction ratings and positive review rates are transformed into satisfaction feature elements, reflecting user acceptance of the strategy. Strategy adjustment suggestions, such as users wanting to see more recommendations of similar products or fewer recommendations of certain product categories, are transformed into adjustment suggestion feature elements, providing direction for strategy optimization. The feedback generation scenario is transformed into feedback scenario feature elements, clarifying the specific recommendation scenario corresponding to the feedback. After mapping, all feedback feature units are integrated to generate a set of feedback feature elements.

[0032] For example, a record in historical strategy feedback data might contain feedback from a user who, after a product push notification, submitted "I don't like this recommendation" through the "feedback entry" at the bottom of the product details page. Here, the "bottom of the product details page" represents the "feedback generation scenario." During feature mapping, this scenario information is transformed into "feedback scenario feature elements," used to distinguish feedback generated in other scenarios such as the "push notification page" and the "personal center settings page." This allows for a more refined understanding of the patterns and value of feedback behavior in different scenarios during subsequent correlation analysis.

[0033] Step S117: Calculate the co-occurrence frequency of each element in the user behavior feature element set and each element in the scene feature element set in the big data model, calculate the element correlation based on the co-occurrence frequency, calculate the co-occurrence frequency of each element in the scene feature element set and each element in the feedback feature element set, calculate the element correlation based on the co-occurrence frequency, and select combinations in the big data model where the correlation between user behavior feature elements, scene feature elements, and feedback feature elements all reaches a preset correlation threshold. Each combination contains a set of user behavior feature elements, a set of scene feature elements, and a set of feedback feature elements. All combinations that meet the conditions are integrated into a multi-source feature combination set.

[0034] Big data models perform correlation analysis on elements within sets of user behavior features, scenario features, and feedback features. For example, the model calculates the frequency of co-occurrence between "purchase behavior type features" (from the user behavior feature set) and "homepage recommendation scenario type features" (from the scenario feature set) in historical data, and then calculates the correlation between them based on this co-occurrence frequency. Similarly, it calculates the co-occurrence frequency between "homepage recommendation scenario type features" (from the scenario feature set) and "high satisfaction features" (from the feedback feature set) and derives the correlation. Then, combinations where the correlation between user behavior features, scenario features, and feedback features all reaches a preset correlation threshold are selected. For example, if the correlation between "purchase behavior type features," "homepage recommendation scenario type features," and "high satisfaction features" all meet the threshold, then they form a combination. All of these combinations together constitute a multi-source feature combination set.

[0035] Step S1171: In the big data model, each behavior type feature element in the user behavior feature element set is paired with each scene type feature element in the scene feature element set to form a behavior-scene pairing combination; calculate the co-occurrence frequency of each behavior-scene pairing combination, where the co-occurrence frequency is the number of times the behavior type corresponding to the behavior type feature element and the scene type corresponding to the scene type feature element co-occur in historical data, and calculate the correlation between the behavior type feature element and the scene type feature element based on the co-occurrence frequency.

[0036] In e-commerce platform scenarios, the set of user behavior feature elements includes various behavior type feature elements, such as browsing behavior type feature elements and purchasing behavior type feature elements; similarly, the set of scenario feature elements also includes different scenario type feature elements, such as homepage recommendation scenario type feature elements and product detail page recommendation scenario type feature elements. Big data models pair each behavior type feature element with each scenario type feature element. For example, browsing behavior type feature elements are paired with homepage recommendation scenario type feature elements, product detail page recommendation scenario type feature elements, etc., forming multiple behavior-scenario pairings. For each pairing, the model counts the number of times the behavior type and scenario type co-occur in historical data, i.e., co-occurrence frequency. Then, the correlation between the two elements is calculated based on the co-occurrence frequency; the higher the co-occurrence frequency, the stronger the correlation generally is.

[0037] Step S1172: Pair each behavior scenario pairing with each effect feature element in the feedback feature element set to form a behavior scenario effect pairing. Calculate the correlation between the behavior scenario pairing and the effect feature element. The correlation is the percentage of times the behavior scenario pairing and the strategy execution effect corresponding to the effect feature element appear together in historical data.

[0038] After obtaining the behavioral scenario pairings, the model pairs each pairing with each effect feature element in the feedback feature element set. For example, the behavioral scenario pairing "Browsing behavior type feature element - Homepage recommendation scenario type feature element" is paired with effect feature elements such as "High click-through rate effect feature element" and "High conversion rate effect feature element" to form behavioral scenario effect pairings. Next, the model calculates the percentage of times the behavioral scenario pairing and the strategy execution effect corresponding to the effect feature element co-occur in historical data, using this as the correlation between them. For example, in historical data, when "Browsing behavior type" and "Homepage recommendation scenario type" co-occur, the percentage of times "High click-through rate effect" appears out of the total number of times they co-occur is the correlation of this behavioral scenario effect pairing.

[0039] Step S1173: Based on the correlation between behavior type feature elements and scene type feature elements, and the correlation between behavior scene pairing and effect feature elements, select behavior type feature elements, scene type feature elements and effect feature elements whose correlation reaches the preset correlation threshold, and combine the behavior type feature elements, scene type feature elements and effect feature elements to form the first feature combination.

[0040] After calculating the correlation between behavioral type features and scene type features, as well as the correlation between behavioral scene pairings and effect features, the model sets a preset correlation threshold. It then selects three elements whose correlation with both behavioral type features and scene type features reaches this threshold, and whose correlation with the corresponding behavioral scene pairings and effect features also reaches this threshold. For example, if the correlation between "purchase behavioral type features" and "homepage recommendation scene type features" meets the threshold, and the correlation between the behavioral scene pairing "purchase behavioral type features - homepage recommendation scene type features" and "high conversion rate effect features" also meets the threshold, then these three elements will be combined to form the first feature combination.

[0041] Step S1174: Using the same method, in the big data model, each behavior cycle feature element in the user behavior feature element set is paired with each scene carrying feature element in the scene feature element set, and the correlation between the behavior cycle feature element and the scene carrying feature element is calculated. Then, the resulting pairing is paired with each satisfaction feature element in the feedback feature element set, and the correlation between the pairing is calculated. Behavior cycle feature elements, scene carrying feature elements, and satisfaction feature elements that meet the correlation criteria are selected, and the behavior cycle feature elements, scene carrying feature elements, and satisfaction feature elements are combined to form a second feature combination.

[0042] The same approach is used for behavioral cycle features and scenario-based features. For example, the behavioral cycle features in the user behavior feature set might include "weekday evening behavioral cycle features" or "weekend all-day behavioral cycle features," while the scenario-based features in the scenario feature set might include "high user load scenario features" or "low user load scenario features." These are paired and their correlation is calculated. Then, the paired combinations are paired with satisfaction features such as "high satisfaction features" or "medium satisfaction features" in the feedback feature set to calculate their correlation. Finally, three feature combinations with satisfactory correlation are selected to form the second feature combination.

[0043] Step S1175: In the big data model, each behavior-related feature element in the user behavior feature element set is paired with each scene resource feature element in the scene feature element set. The correlation between the behavior-related feature element and the scene resource feature element is calculated. Then, the resulting pairing is paired with each adjustment suggestion feature element in the feedback feature element set. The correlation between the pairing and the adjustment suggestion feature element is calculated. Behavior-related feature elements, scene resource feature elements, and adjustment suggestion feature elements that meet the correlation criteria are selected. The behavior-related feature elements, scene resource feature elements, and adjustment suggestion feature elements are combined to form a third feature combination.

[0044] Behavioral association features may involve information such as the product categories and brands associated with the user, such as "electronic product behavioral association features" or "clothing behavioral association features." Scenario resource features include "rich product resource scenario features" and "limited product resource scenario features." After pairing these features and calculating their correlation, they are then paired with adjustment suggestion features from the feedback feature set, such as "increase product diversity adjustment suggestion features" and "optimize product sorting adjustment suggestion features," to calculate their correlation. Only the qualifying feature combinations are selected to form the third feature combination.

[0045] Step S1176: In the big data model, each behavioral feedback feature element in the user behavior feature element set is paired with each scene constraint feature element in the scene feature element set. The correlation between the behavioral feedback feature element and the scene constraint feature element is calculated. Then, the resulting pairing is paired with each feedback scene feature element in the feedback feature element set. The correlation between the pairing and the feedback scene feature element is calculated. Behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements that meet the correlation criteria are selected. The behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements are combined to form a fourth feature combination.

[0046] Behavioral feedback features may include "positive feedback features" and "negative feedback features," while scenario constraint features may include "compliance scenario constraint features" and "format restriction scenario constraint features." After pairing these features and calculating their correlation, they are then paired with feedback scenario feature features from the feedback feature feature set, such as "homepage feedback scenario feature features" and "details page feedback scenario feature features," to calculate their correlation again. Only the elements that meet the criteria are selected to form the fourth feature combination.

[0047] Step S1177: Integrate the first feature combination, the second feature combination, the third feature combination and the fourth feature combination to form multiple feature combinations. Each feature combination contains three different types of feature elements: user behavior feature elements, scene feature elements and feedback feature elements. The correlation between each feature element in each feature combination reaches a preset correlation threshold.

[0048] By combining the first, second, third, and fourth feature combinations formed earlier, we obtain a multi-source feature combination set. Each of the above combinations contains three different types of feature elements: user behavior, scenario, and feedback, and the correlation between them meets the preset requirements.

[0049] Step S118: Perform encoding integration on the multi-source feature combination set to generate a policy feature encoding result containing multiple sets of feature combinations, wherein each set of feature combinations in the policy feature encoding result corresponds to a set of user scenario policy association relationships.

[0050] Each feature combination in the multi-source feature set represents a relationship between user behavior, scenario, and feedback. The big data model encodes and integrates these combinations, transforming each feature combination into a specific encoding form. This encoding clearly represents the user scenario strategy relationship corresponding to that feature combination. For example, for the aforementioned combination containing "purchase behavior type feature elements," "homepage recommendation scenario type feature elements," and "high satisfaction feature elements," after encoding, there will be a corresponding code in the strategy feature encoding result. This code corresponds to a strategy relationship that leads to high satisfaction in the homepage recommendation scenario for purchase behavior.

[0051] Step S119: In the big data model, based on the behavior type feature elements and scene type feature elements in the strategy feature encoding results, feature combinations with the same behavior type and the same scene type are divided into the same response cluster unit.

[0052] Each feature combination in the strategy feature encoding results contains both behavior type feature elements and scenario type feature elements. The model groups the feature combinations based on these two elements. For example, all feature combinations containing both "browsing behavior type feature elements" and "homepage recommendation scenario type feature elements" will be grouped into the same response cluster unit; while feature combinations containing both "purchase behavior type feature elements" and "details page recommendation scenario type feature elements" will be grouped into another response cluster unit. Through this division, feature combinations with the same behavior type and scenario type are grouped together to form multiple response cluster units.

[0053] Step S1110: Calculate the association strength of feature combinations within each response cluster unit, sort the feature combinations within the response cluster unit based on the association strength, and generate a dynamic response cluster of user scenario strategy containing multiple sorted response cluster units.

[0054] For each feature combination within a response cluster unit, the model calculates the correlation strength between them. This correlation strength calculation may take into account factors such as the correlation between other elements within the feature combination. For example, within a response cluster unit, there may be multiple feature combinations containing different feedback feature elements. The model will calculate the correlation strength based on the degree of correlation between these feedback feature elements and behavior type, scenario type, and other feature elements. Then, the feature combinations within the response cluster unit are sorted in descending order of correlation strength. After all response cluster units have undergone this processing, a dynamic response cluster for user scenario strategies is generated, which contains multiple sorted response cluster units.

[0055] Step S1111: In the big data model, perform parameter extraction on the feature combination within each response cluster unit of the user scenario strategy dynamic response cluster, extract the corresponding strategy content parameters, strategy delivery time parameters, and strategy coverage user parameters, associate and bind the strategy content parameters, strategy delivery time parameters, and strategy coverage user parameters with the corresponding response cluster unit, and generate the initial dynamic delivery strategy.

[0056] Each feature combination within the dynamic response cluster of user scenario strategies corresponds to specific strategy information. The big data model extracts strategy content parameters from these feature combinations, such as the specific category, brand, and price range of the recommended products; strategy delivery time parameters, i.e., the appropriate time of day and day of the week for the strategy to be delivered; and strategy coverage user parameters, specifying the characteristics of the target user group, such as age, gender, and spending power. After extraction, these parameters are associated and bound to the corresponding response cluster units, ensuring that each response cluster unit has its own corresponding strategy parameters. Finally, all response cluster units associated with and bound to strategy parameters are integrated to generate the initial dynamic delivery strategy.

[0057] Step S120: Apply the initial dynamic distribution strategy to the distribution scenario corresponding to the target user group, and collect the set of real-time user behavior data, the set of real-time scenario change data, and the set of real-time strategy feedback data generated during the strategy execution process.

[0058] After the initial dynamic delivery strategy is generated, it needs to be applied to the delivery scenarios corresponding to the target user groups on the actual e-commerce platform. For example, a clothing recommendation strategy targeting young female users can be applied to the homepage recommendation scenario. During strategy execution, user behavior data, scenario change data, and real-time user feedback data are collected in real time to evaluate the effectiveness of the strategy and make subsequent optimizations.

[0059] Step S121: Extract the policy delivery time parameter of each policy entry from the initial dynamic delivery strategy, and determine the execution start time of each policy entry in the delivery scenario corresponding to the target user group based on the policy delivery time parameter.

[0060] The initial dynamic delivery strategy contains multiple strategy entries, each with its corresponding delivery time parameter. For example, a strategy entry targeting commuter users might have its delivery time parameters set to the morning commute and the evening after get off work on weekdays. After extracting these time parameters from the initial dynamic delivery strategy, the specific execution start time for each strategy entry within the delivery scenario corresponding to the target user group is determined based on these parameters. For instance, the aforementioned strategy entry targeting commuters would have its execution start time set to 8:00-9:00 AM and 7:00-9:00 PM on weekdays.

[0061] Step S122: When the execution start time is reached, extract the policy coverage user parameters of each policy entry, and select users who meet the policy coverage user parameters from the target user group to form a policy execution user group.

[0062] When the execution start time for each policy entry is reached, the model extracts the policy coverage user parameters for that policy entry. For example, the policy coverage user parameters for a certain policy entry might be female users aged 25-35, with monthly spending within a certain range, and who frequently browse beauty products. Then, all users matching these parameters are selected from the target user group and grouped into a policy execution user group. The policy entry will only be executed on users within this user group.

[0063] Step S123: Push the policy content parameters of each policy entry to the corresponding policy execution user group to start the initial dynamic policy delivery execution process.

[0064] Once the user groups for strategy execution are identified, the strategy content parameters for each strategy entry are pushed to the corresponding user groups. For example, if the strategy content parameter is to recommend several popular beauty products, then the recommendation information for these products will be displayed to users in that strategy execution user group. In this way, the initial dynamic strategy delivery execution process is officially launched, and the e-commerce platform begins to push relevant content to users according to the strategy.

[0065] Step S124: During the strategy execution process, real-time behavior data of the strategy execution user group is collected according to the preset behavior data collection interval. Each piece of real-time behavior data is associated and marked with the corresponding strategy entry and the corresponding execution time point to generate a user real-time behavior data unit. The real-time behavior data includes the behavior trigger time, behavior execution steps, content involved in the behavior, and behavior end status.

[0066] During strategy execution, real-time behavioral data is collected from the user group executing the strategy at preset intervals, such as every ten minutes. This real-time behavioral data includes the time a user triggers an action after receiving a recommendation, such as the exact time they clicked on the recommended product; the action steps, such as clicking to enter the product details page, adding it to their cart, and submitting the order; the content involved in the action, i.e., the specific product the user interacted with; and the action's completion status, such as whether the purchase was completed or the action was abandoned. After collecting each piece of real-time behavioral data, it is associated with the corresponding strategy entry and the execution time of the action, forming a user real-time behavioral data unit containing this associated information.

[0067] Step S125: Collect and distribute real-time change data of the scenario according to the preset scenario data collection interval, associate and mark each piece of real-time change data with the corresponding strategy entry and the corresponding execution time point to generate a scenario real-time change data unit. The real-time change data includes scenario resource usage, changes in the number of users in the scenario, scenario running status, and changes in scenario constraints.

[0068] Simultaneously, according to a preset scenario data collection interval, such as every five minutes, real-time change data of the distributed scenario is collected. Scenario resource usage includes the occupancy of recommendation slots and server load; changes in the number of users within the scenario refer to real-time changes in user access volume under that distributed scenario; scenario operational status involves whether the scenario is running normally and whether there are any lags; changes in scenario constraints may be due to system adjustments or other reasons causing changes in certain constraints of the scenario. Each piece of real-time change data is associated with and marked with the corresponding strategy entry and execution time point to generate a real-time change data unit for the scenario.

[0069] Step S126: Collect real-time feedback data submitted by the strategy execution user group during strategy execution through a preset feedback collection interface. Associate and mark each piece of real-time feedback data with the corresponding strategy entry, the corresponding user real-time behavior data unit, the corresponding scene real-time change data unit, and the corresponding execution time point to generate a strategy real-time feedback data unit. The real-time feedback data includes evaluations of the strategy content, opinions on the timing of strategy issuance, feelings about the effect of strategy execution, and suggestions for strategy adjustment.

[0070] E-commerce platforms set up pre-defined feedback collection interfaces, such as rating buttons and feedback forms below recommended content. These interfaces collect real-time feedback data submitted by user groups during strategy execution. Real-time feedback data includes user evaluations of recommended products (e.g., whether they like the recommended products); opinions on the timing of strategy deployment (e.g., whether they think the timing is appropriate); feelings about the effectiveness of strategy execution (e.g., whether they find the recommendations helpful); and suggestions for strategy adjustments (e.g., what types of products they would like to see more of). Each piece of real-time feedback data is associated and tagged with the corresponding strategy entry, related real-time user behavior data units, real-time scenario change data units, and execution time points to generate a strategy real-time feedback data unit.

[0071] Step S127: Sort all real-time user behavior data units according to the order of their execution time, integrate the sorted real-time user behavior data units, and generate a set of real-time user behavior data.

[0072] After collecting multiple real-time user behavior data units, they are sorted according to the chronological order of their corresponding execution times. For example, data units collected at 8:00 AM are placed first, and those collected at 8:10 AM are placed later. Then, the sorted real-time user behavior data units are integrated and combined to form a real-time user behavior dataset. This dataset comprehensively reflects the dynamic changes in user behavior during strategy execution.

[0073] Step S128: Sort all real-time scene change data units according to the order of their execution time, integrate the sorted real-time scene change data units, and generate a set of real-time scene change data.

[0074] Similarly, the real-time change data units of the scenario are sorted according to the execution time and then integrated to generate a real-time change data set of the scenario. This real-time change data set of the scenario records the real-time status of the scenario during the execution of the strategy.

[0075] Step S129: Sort all real-time policy feedback data units according to their execution time, integrate the sorted real-time policy feedback data units, and generate a real-time policy feedback data set.

[0076] The real-time feedback data units of the strategy are also sorted and integrated according to the execution time to generate a real-time feedback data set of the strategy. This real-time feedback data set of the strategy includes the user's feedback at different time points during the execution of the strategy.

[0077] Step S130: Call the big data model to perform response cluster adaptability analysis on the user real-time behavior data set, the scene real-time change data set, and the strategy real-time feedback data set, generate response cluster adaptability analysis results, and generate a strategy iteration start command based on the response cluster adaptability analysis results.

[0078] After acquiring the sets of real-time user behavior data, real-time scenario change data, and real-time policy feedback data, it is necessary to use a big data model to perform response cluster adaptability analysis on the above data. This involves analyzing whether the current real-time data is compatible with the previously constructed dynamic response clusters for user scenario policies, and determining whether policy iteration and optimization need to be initiated based on the analysis results.

[0079] Step S131: After inputting the user real-time behavior data set into the big data model, perform feature extraction on each user real-time behavior data unit in the user real-time behavior data set, extract the behavior trigger frequency, behavior completion ratio, the matching situation between the content involved in the behavior and the strategy content, and the distribution of the behavior end state, and generate a real-time behavior feature set.

[0080] Each real-time user behavior data unit in the user real-time behavior dataset contains rich behavioral information. The big data model extracts features from each data unit. Behavior trigger frequency refers to the number of times a user triggers a certain behavior (such as clicking or purchasing) within a certain time period; behavior completion rate refers to the proportion of users who start a certain behavior (such as adding to cart) and ultimately complete the behavior (such as purchasing); the matching of behavior-related content with strategy content determines whether the products involved in the user's actual behavior match the product content recommended by the strategy; and the distribution of behavior completion states represents the proportion of different behavior completion states (such as completed or abandoned). Integrating these extracted features generates the real-time behavior feature set.

[0081] Step S132: After inputting the real-time scene change data set into the big data model, perform feature extraction on each real-time scene change data unit in the real-time scene change data set to extract the scene resource usage growth rate, the fluctuation of the number of users in the scene, the stability of the scene operation status, and the frequency of scene constraint changes, and generate a real-time scene feature set.

[0082] For each data unit in the real-time changing data set of the scene, the model extracts the scene resource usage growth rate, that is, the growth of scene resource usage over time; the fluctuation of the number of users in the scene, reflecting the change in user access volume; the stability of the scene operation status, such as the stability of server response time; and the frequency of scene constraint changes, that is, the number of times scene constraints change. The above features constitute the real-time scene feature set.

[0083] Step S133: After inputting the real-time feedback data set of the strategy into the big data model, perform feature extraction on each real-time feedback data unit of the strategy in the real-time feedback data set of the strategy, extract the positive user evaluation ratio, the approval ratio of the timing of the release, the satisfaction ratio of the execution effect, and the adoption ratio of the adjustment suggestions, and generate a real-time feedback feature set.

[0084] Each data unit in the real-time feedback dataset contains user feedback information. The model extracts the following from this dataset: the percentage of positive user reviews (the proportion of positive reviews in the total reviews); the percentage of users who approve of the timing of the recommendation (the proportion of users who approve of the timing of the recommendation); the percentage of users who are satisfied with the execution effect (the proportion of users who are satisfied with the execution effect of the strategy); and the percentage of users whose adjustment suggestions are worth adopting (the proportion of adjustment suggestions with adoptable value in the total suggestions). These features constitute the real-time feedback feature set.

[0085] Step S134: Extract the feature combination elements of each response cluster unit in the user scenario strategy dynamic response cluster in the big data model, compare the real-time behavior features with behavior type feature elements and behavior cycle feature elements, and calculate the behavior feature fit; compare the real-time scene features with scene type feature elements and scene resource feature elements, and calculate the scene feature fit; compare the real-time feedback features with effect feature elements and satisfaction feature elements, and calculate the feedback feature fit.

[0086] The big data model extracts feature combinations from the dynamic response clusters of user scenario strategies, including previously constructed feature elements such as behavior type, behavior cycle, scenario type, scenario resources, effect, and satisfaction. Then, the features in the real-time behavior feature set are compared with the behavior type feature elements and behavior cycle feature elements respectively, analyzing their similarity to calculate the behavior feature fit. Similarly, real-time scenario features are compared with scenario type feature elements and scenario resource feature elements to calculate scenario feature fit; and real-time feedback features are compared with effect feature elements and satisfaction feature elements to calculate feedback feature fit.

[0087] Step S135: Based on behavioral feature fit, scene feature fit, and feedback feature fit in the big data model, calculate the comprehensive fit of each response cluster unit and generate a response cluster fit analysis result that includes the comprehensive fit of each response cluster unit.

[0088] Each response cluster unit has three fitness scores: behavior, scenario, and feedback. The model calculates the overall fitness score of the response cluster unit based on these three fitness scores, possibly by weighted averaging or other methods. The overall fitness scores of all response cluster units are then aggregated to generate the response cluster fitness analysis results.

[0089] Step S136: In the big data model, count the number of response cluster units whose overall fit does not reach the preset fit benchmark in the response cluster fit analysis results.

[0090] A baseline for fitness is preset, such as an overall fitness score of 0.7 (this is just an example and not a specific value). The model will count the number of response cluster units with an overall fitness score lower than this baseline in the response cluster fitness analysis results.

[0091] Step S137: If the number of response cluster units that have not reached the preset adaptation benchmark exceeds the preset number threshold, a strategy iteration start instruction is generated. The strategy iteration start instruction contains the identifier of the response cluster unit that needs to be reconstructed in terms of feature elements. If the number of response cluster units that have not reached the preset adaptation benchmark does not exceed the preset number threshold, a strategy iteration start instruction is not generated, and the initial dynamic distribution strategy is continued to be executed.

[0092] Set a preset threshold, such as the number of non-compliant response cluster units accounting for 30% of the total number of response cluster units (this is just an example, not a specific value). If the number of response cluster units that fail to meet the preset adaptation benchmark exceeds this threshold, it indicates that the current initial dynamic delivery strategy is not well adapted to the real-time situation on multiple response cluster units and needs to be optimized. At this time, a strategy iteration start command is generated, which specifies which response cluster units need to have their feature elements reconstructed. If the threshold is not exceeded, it indicates that the overall adaptation of the strategy is acceptable, and the initial dynamic delivery strategy continues to be executed.

[0093] Step S140: Upon receiving the strategy iteration start instruction, the big data model is invoked to perform feature element reconstruction processing on the strategy feature encoding result, generating a reconstructed feature encoding result. The reconstructed feature encoding result is then used to optimize the initial dynamic distribution strategy to obtain an optimized dynamic distribution strategy.

[0094] Upon receiving the policy iteration start command, the policy feature encoding results need to be reconstructed. For the response cluster units that need reconstruction as identified in the command, their feature elements are adjusted to generate reconstructed feature encoding results. These results are then used to optimize the initial dynamic delivery policy, resulting in an optimized dynamic delivery policy that is more adaptable to real-time conditions.

[0095] Step S141: After receiving the strategy iteration start instruction, extract the response cluster unit identifier that needs to be reconstructed from the strategy iteration start instruction, and select the corresponding feature combination from the strategy feature encoding result based on the response cluster unit identifier to form a set of feature combinations to be reconstructed.

[0096] The strategy iteration initiation command includes identifiers of response cluster units that require feature reconstruction. Based on these identifiers, feature combinations corresponding to these identifiers are selected from the strategy feature encoding results. For example, if the identified response cluster unit corresponds to a combination that does not meet the adaptation criteria for browsing behavior in the homepage recommendation scenario, then all feature combinations related to that response cluster unit in the strategy feature encoding results are selected to form a set of feature combinations to be reconstructed.

[0097] Step S142: After inputting the set of features to be reconstructed into the big data model, perform defect identification on each feature combination to be reconstructed, including behavioral type feature elements, behavioral cycle feature elements, behavioral association feature elements, behavioral feedback feature elements, scene type feature elements, scene resource feature elements, scene carrying feature elements, scene constraint feature elements, effect feature elements, satisfaction feature elements, adjustment suggestion feature elements, and feedback scene feature elements, and determine the parts of each feature element that do not match the real-time features, and generate feature element defect information.

[0098] Each feature combination in the set of features to be reconstructed contains multiple types of feature elements. The big data model identifies defects in each element, comparing it with corresponding real-time features (such as real-time behavior features, real-time scene features, and real-time feedback features) to find the mismatches. For example, if the periodic pattern in the behavior periodic feature element does not match the user's current behavior periodic pattern in the real-time behavior features, then this part is the defective part of the behavior periodic feature element. All defective parts in all feature elements are recorded to generate feature element defect information.

[0099] Step S143: Based on the defect information of the feature elements, supplementary feature elements are selected from the user behavior feature element set, the scene feature element set, and the feedback feature element set. The selection criterion is that the feature matching ratio between the supplementary feature elements and the defective parts reaches the preset matching ratio requirement.

[0100] Based on the defect information of the feature elements, identify the types and requirements of the feature elements that need to be supplemented. Select supplementary feature elements that can compensate for the defects from the previously constructed sets of user behavior feature elements, scene feature elements, and feedback feature elements. The selection criterion is that the feature matching ratio between the supplementary feature element and the defective part must reach a preset matching ratio requirement, such as a matching ratio of 80% (this is just an example, not a specific value), to ensure that the supplemented feature elements can better adapt to the current real-time situation.

[0101] Step S144: Replace the defective parts in the supplementary feature elements and the feature combination to be reconstructed to generate a temporary feature combination.

[0102] The selected supplementary feature elements replace the defective parts in the feature combination to be reconstructed, forming a temporary feature combination. For example, replacing the behavior cycle feature elements in the original feature combination that do not match the real-time behavior cycle with the selected new behavior cycle feature elements results in a temporary feature combination.

[0103] Step S145: Perform correlation verification on the temporary feature combination, calculate the correlation between different types of feature elements in the temporary feature combination. If the correlation reaches the preset correlation benchmark, the temporary feature combination is determined as the reconstructed feature combination. If the correlation does not reach the preset correlation benchmark, the feature elements are re-selected and supplemented and replaced until a reconstructed feature combination with the correct correlation is generated.

[0104] After a temporary feature combination is formed, its correlation needs to be verified to ensure that the correlation between different types of feature elements in the combination meets the requirements. If the correlation meets the requirements, the temporary combination becomes the reconstructed feature combination; if it does not meet the requirements, feature elements need to be screened and supplemented for replacement, and the verification needs to be repeated until a reconstructed feature combination with the required correlation is generated.

[0105] Step S1451: Extract the behavior type feature elements, scene type feature elements, and effect feature elements from the temporary feature combination; calculate the correlation between the behavior type feature elements and the scene type feature elements; calculate the correlation between the scene type feature elements and the effect feature elements; calculate the correlation between the behavior type feature elements and the effect feature elements.

[0106] From the temporary feature combination, three feature elements are extracted: behavior type, scene type, and effect. The correlation between the behavior type feature element and the scene type feature element is calculated to determine the degree of correlation of the behavior type under the scene type; the correlation between the scene type feature element and the effect feature element is calculated to see if the expected effect can be achieved under the scene type; the correlation between the behavior type feature element and the effect feature element is calculated to determine whether the behavior type can bring about the effect.

[0107] Step S1452: Extract the behavioral cycle feature elements, scene carrying feature elements, and satisfaction feature elements from the temporary feature combination; calculate the correlation between the behavioral cycle feature elements and the scene carrying feature elements; calculate the correlation between the scene carrying feature elements and the satisfaction feature elements; and calculate the correlation between the behavioral cycle feature elements and the satisfaction feature elements.

[0108] Similarly, we extract the characteristics of behavior cycle, scenario capacity, and satisfaction, and calculate the pairwise correlation between them. For example, the correlation between behavior cycle and scenario capacity, i.e., whether the scenario capacity matches the behavior cycle; the correlation between scenario capacity and satisfaction, i.e. whether the scenario capacity affects user satisfaction; and the correlation between behavior cycle and satisfaction, i.e. how satisfied the user is during the behavior cycle.

[0109] Step S1453: Extract the behavior-related feature elements, scene resource feature elements, and adjustment suggestion feature elements from the temporary feature combination; calculate the correlation between the behavior-related feature elements and the scene resource feature elements; calculate the correlation between the scene resource feature elements and the adjustment suggestion feature elements; and calculate the correlation between the behavior-related feature elements and the adjustment suggestion feature elements.

[0110] Extract the feature elements of behavior association, scene resources, and adjustment suggestions. Calculate the correlation between the feature elements of behavior association and the feature elements of scene resources to see if the content of behavior association matches the scene resources; the correlation between scene resources and adjustment suggestions to see if the scene resources require corresponding adjustment suggestions; and the correlation between behavior association and adjustment suggestions to see if the user's behavior association content will generate specific adjustment suggestions.

[0111] Step S1454: Extract the behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements from the temporary feature combination; calculate the correlation between the behavioral feedback feature elements and the scene constraint feature elements; calculate the correlation between the scene constraint feature elements and the feedback scene feature elements; calculate the correlation between the behavioral feedback feature elements and the feedback scene feature elements.

[0112] Extract behavioral feedback, scenario constraints, and feedback scenario feature elements; calculate the correlation between behavioral feedback and scenario constraints to determine whether user behavioral feedback is affected by scenario constraints; calculate the correlation between scenario constraints and feedback scenarios to determine whether scenario constraints differ across different feedback scenarios; and calculate the correlation between behavioral feedback and feedback scenarios to determine whether user behavioral feedback differs across different feedback scenarios.

[0113] Step S1455: Count the number of correlations that do not reach the preset correlation benchmark among all calculated correlations.

[0114] Compare all the correlation degrees calculated above with the preset correlation degree benchmark, and count the number of correlation degrees that do not meet the benchmark.

[0115] Step S1456: If the number of correlations that does not meet the preset correlation benchmark is zero, then the correlation of the temporary feature combination is determined to meet the standard, and the temporary feature combination is determined as the reconstructed feature combination.

[0116] If all correlations reach the preset benchmark, it indicates that the correlation between the elements in the temporary feature combination is good, and it is determined as the reconstructed feature combination.

[0117] Step S1457: If the number of correlations that do not meet the preset correlation benchmark is greater than zero, then determine the feature element pairs corresponding to the non-compliant correlations. For each feature element pair corresponding to the non-compliant correlation, re-select supplementary feature elements from the corresponding feature element set. The selection criterion is that the matching ratio between the supplementary feature elements and the existing feature elements reaches the preset matching ratio requirement.

[0118] If there are unmet correlations, identify the corresponding feature element pairs. For example, if the correlation between a behavior type feature element and a scene type feature element is unmet, then these two elements form an unmet feature element pair. For each of these feature pairs, re-select and supplement feature elements from the corresponding feature element set (such as the user behavior feature element set or the scene feature element set). During the selection process, the matching ratio between the supplemented feature elements and the existing feature elements must meet a preset requirement.

[0119] Step S1458: Replace the corresponding feature elements in the temporary feature combination with the newly selected supplementary feature elements to generate a new temporary feature combination.

[0120] The newly selected supplementary feature elements replace the corresponding elements in the non-compliant feature element pairs in the temporary feature combination, forming a new temporary feature combination.

[0121] Step S1459: Re-perform the correlation verification on the new temporary feature combination, repeat the above correlation calculation, standard judgment and supplementary feature element screening and replacement steps until a reconstructed feature combination with qualified correlation is generated.

[0122] The newly generated temporary feature combination is re-verified for correlation. The previous calculation, judgment and screening replacement process is repeated to continuously optimize the feature elements until a reconstructed feature combination with the required correlation is generated.

[0123] Step S146: Integrate all reconstructed feature combinations with unreconstructed feature combinations to generate reconstructed feature encoding results.

[0124] All reconstructed feature combinations that meet the correlation criteria are integrated with those feature combinations that do not require reconstruction to form a reconstructed feature encoding result. This reconstructed feature encoding result is an improvement in adapting to real-time data compared to the previous strategy feature encoding result.

[0125] Step S147: In the big data model, based on each feature combination in the reconstructed feature encoding results, extract the corresponding optimized strategy content parameters, optimized strategy delivery time parameters, and optimized strategy user coverage parameters.

[0126] Each feature combination in the reconstructed feature encoding results corresponds to optimized policy information. The model extracts optimized policy content parameters from each combination, which may be product recommendations that better match the user's current interests; optimized policy delivery time parameters, which may be delivery periods adjusted according to real-time behavior cycles; and optimized policy coverage user parameters, which may be more accurate user group segmentation.

[0127] Step S148: Select strategy entries corresponding to the response cluster unit identifiers that need feature element reconstruction from the initial dynamic distribution strategy, and replace the strategy content parameters, strategy distribution time parameters, and strategy coverage user parameters of the strategy entries corresponding to the response cluster unit identifiers that need feature element reconstruction from the initial dynamic distribution strategy with the optimized strategy content parameters, optimized strategy distribution time parameters, and optimized strategy coverage user parameters, respectively.

[0128] Based on the response cluster unit identifier that needs to be reconstructed in the strategy iteration start instruction, the corresponding strategy entry is found from the initial dynamic distribution strategy. The original strategy content parameters, distribution time parameters, and coverage user parameters of the above strategy entry are replaced with optimized parameters extracted from the reconstruction feature encoding results to optimize the above strategy entry.

[0129] Step S149: Perform correlation and coordination processing on all policy entries in the replaced initial dynamic distribution strategy to make the optimized policy content parameters, optimized policy distribution time parameters, and optimized policy coverage user parameters of each policy entry mutually compatible.

[0130] After replacing the parameters, all strategy entries need to be reconciled and coordinated. For example, ensure that the products recommended by the optimized strategy content parameter of a certain strategy entry are suitable for the user group targeted by its optimized strategy coverage user parameter, and that the optimized strategy delivery time parameter is also suitable for pushing the product content to that user group, to avoid contradictions or inconsistencies between parameters.

[0131] Step S1410: Determine the set of strategy entries that have completed the association coordination process as the optimized dynamic distribution strategy.

[0132] After the correlation and coordination process, the parameters of all strategy items are adapted to each other. By integrating the above strategy items together, the optimized dynamic distribution strategy is obtained.

[0133] Step S150: Apply the optimized dynamic delivery strategy to the delivery scenario corresponding to the test user group, collect the strategy verification data set generated during the strategy verification process, call the big data model to perform evolutionary adaptation chain construction on the strategy verification data set, and generate the strategy evolutionary adaptation chain.

[0134] The optimized dynamic delivery strategy needs to be validated in delivery scenarios corresponding to the test user group. A set of strategy validation data is collected during the validation process, and then a big data model is invoked to construct a strategy evolution and adaptation chain based on this data, analyzing the strategy's evolution and adaptation.

[0135] Step S151: Extract the optimized policy coverage user parameters for each policy entry from the optimized dynamic distribution strategy, and select qualified users from the test user group based on the optimized policy coverage user parameters to form a policy verification user group.

[0136] Each policy entry in the optimized dynamic delivery strategy has optimized policy coverage user parameters. Based on these parameters, users meeting the criteria are selected from the test user group to form a policy verification user group. For example, if the optimized coverage user parameter for a policy entry is young male users, then users matching this characteristic are selected from the test user group to form the verification user group for that policy entry.

[0137] Step S152: Extract the optimized strategy delivery time parameter of each strategy entry in the optimized dynamic delivery strategy, and determine the verification start time of each strategy entry in the delivery scenario corresponding to the test user group.

[0138] Extract the optimized policy delivery time parameter for each policy entry, and determine the verification start time in the delivery scenario corresponding to the test user group based on this parameter. For example, if the optimized delivery time parameter is a weekend afternoon, then the verification start time is set to a weekend afternoon.

[0139] Step S153: When the verification start time is reached, push the optimized policy content parameters of each policy entry to the corresponding policy verification user group to start the verification process of the optimized dynamic policy.

[0140] When the verification start time arrives, the optimized strategy content parameters for each strategy entry are pushed to the corresponding strategy verification user group, officially starting the verification process of the optimized strategy, and observing the user's reaction and strategy execution in the test scenario.

[0141] Step S154: During the verification process, the verification behavior data of the user group is collected according to the preset verification data collection interval. The verification behavior data includes the verification behavior trigger time, the verification behavior execution process, the verification behavior associated content, and the verification behavior result. Each piece of verification behavior data is associated with the corresponding strategy entry and the corresponding verification time point to generate a verification behavior data unit.

[0142] In the verification process, verification behavior data of the user group is collected according to a preset verification data collection interval strategy. The verification behavior trigger time is the specific time when the user triggers the behavior in the test scenario; the verification behavior execution process is the steps the user takes from start to finish executing the behavior; the verification behavior associated content is the specific product involved in the user's behavior; the verification behavior result is the final state of the behavior, such as successful purchase or abandoned purchase. Each piece of verification behavior data is associated with and marked with the corresponding strategy entry and verification time point to generate a verification behavior data unit.

[0143] Step S155: Collect verification scenario data during the verification process of the test deployment scenario. The verification scenario data includes the resource consumption of the verification scenario, the change in the number of users in the verification scenario, the running status of the verification scenario, and the change in the constraints of the verification scenario. Associate and mark each piece of verification scenario data with the corresponding strategy entry and the corresponding verification time point to generate a verification scenario data unit.

[0144] Simultaneously, verification scenario data for the test deployment scenarios are collected, including resource consumption (such as server resource usage), changes in the number of users, operational status (such as smoothness), and changes in constraints. Each verification scenario data item is associated with and marked with its corresponding policy entry and verification time point to generate a verification scenario data unit.

[0145] Resource consumption in the verification scenario refers to metrics such as the utilization and load rate of system resources like servers, network bandwidth, database connections, and computing resources during the policy verification process. Examples include CPU utilization, memory usage, and interface response time. Excessive resource consumption may affect the speed of policy deployment and user experience, thereby impacting the validity of the verification results.

[0146] The change in the number of users within the verification scenario refers to the real-time fluctuations in the number of concurrent or active users actually in the delivery scenario (such as a specific activity page or push channel) within the test user group during the strategy verification period. Sudden increases or decreases in the number of users are important dynamic characteristics of the scenario.

[0147] Verifying the operational status of a scenario refers to the health and stability of the service itself, such as service availability, the presence of error alerts, and the normality of logs. Abnormal operational status may prevent policies from being executed correctly.

[0148] Changes in verification scenario constraints refer to adjustments made to pre-defined rules or restrictions that may affect strategy execution during the verification process. For example, operations personnel might temporarily adjust the upper limit of push notification frequency, modify the tagging rules for target user filtering, or temporarily restrict the push of certain content due to compliance requirements. These changes directly impact the environmental boundaries of strategy execution.

[0149] Step S156: Collect the verification feedback data submitted by the strategy verification user group during the verification process. The verification feedback data includes evaluations of the optimized strategy content, evaluations of the optimized distribution timing, evaluations of the optimized strategy execution effect, and suggestions for further adjustments to the optimized strategy. Each piece of verification feedback data is associated and marked with the corresponding strategy item, the corresponding verification behavior data unit, the corresponding verification scenario data unit, and the corresponding verification time point to generate a verification feedback data unit.

[0150] Verification feedback data from the strategy verification user group is collected through feedback channels in the test scenario. This includes evaluations of the optimized strategy content, issuance timing, execution effectiveness, and suggestions for further adjustments. Each piece of verification feedback data is associated with and tagged with the corresponding strategy entry, verification behavior data unit, verification scenario data unit, and verification time point to generate a verification feedback data unit.

[0151] Step S157: Integrate all verification behavior data units, all verification scenario data units, and all verification feedback data units in chronological order of verification time points to generate a strategy verification data set.

[0152] The three types of data units mentioned above are integrated together in chronological order of verification time to form a strategy verification data set. This strategy verification data set comprehensively records various data of the optimized strategy during the verification process.

[0153] Step S158: After inputting the strategy verification data set into the big data model, extract the verification behavior features, verification scenario features, and verification feedback features from the strategy verification data set.

[0154] The big data model processes the strategy verification dataset to extract verification behavior features (such as the behavior patterns of test users), verification scenario features (such as the operational status of the test scenario), and verification feedback features (such as the feedback tendencies of test users).

[0155] Step S159: In the big data model, the feature combination in the reconstructed feature encoding result is used as the adaptation chain node, and the verification behavior feature, verification scenario feature and verification feedback feature are used as node attributes. The association strength between adjacent nodes is calculated, and the node connection relationship is constructed based on the association strength.

[0156] Each feature combination in the reconstructed feature encoding result is used as a node in the policy evolution adaptation chain, and the extracted verification behavior features, verification scenario features, and verification feedback features are used as attributes of these nodes. Then, the association strength between adjacent nodes is calculated, and the connection relationship between nodes is constructed based on the association strength to form the basic structure of the policy evolution adaptation chain.

[0157] For example, step S1591: extract all feature combinations from the reconstructed feature encoding results, use each feature combination as a node in the policy evolution adaptation chain, and assign a unique node identifier to each node.

[0158] The reconstructed feature encoding result contains multiple feature combinations. Each combination becomes a node in the policy evolution adaptation chain and is assigned a unique identifier for differentiation and management.

[0159] Step S1592: For each node, extract the behavior type feature elements, behavior cycle feature elements, scene type feature elements, scene resource feature elements, effect feature elements, and satisfaction feature elements from the corresponding feature combination.

[0160] For each node, multiple key feature elements are extracted from its corresponding feature combination. These elements form the basis for subsequent calculations of node attributes and association strength.

[0161] Step S1593: Associate and bind the verification behavior features, verification scenario features, and verification feedback features corresponding to the node in the strategy verification data set with the extracted feature elements, and set the verification behavior features, verification scenario features, and verification feedback features as the node attributes of the node.

[0162] The verification behavior features, verification scenario features, and verification feedback features related to the node in the strategy verification dataset are associated with the extracted feature elements, so that these verification features become the attributes of the node and enrich the node's information.

[0163] Step S1594: Sort all nodes according to the type order of the behavioral type feature elements in the feature combination corresponding to the node, and determine the adjacency relationship between the nodes.

[0164] The nodes are sorted according to the type of behavioral feature elements (such as browsing, purchasing, etc.) in the feature combination corresponding to the node, thereby determining the adjacent relationship between the nodes.

[0165] Step S1595: For two adjacent nodes after sorting, extract the verification feedback features from the node attributes of the previous node and the verification behavior features and verification scenario features from the node attributes of the next node.

[0166] For the sorted adjacent nodes, extract the verification feedback features of the previous node and the verification behavior features and verification scenario features of the next node to calculate the correlation strength between them.

[0167] Step S1596: Calculate the similarity between the verification feedback features of the previous node and the verification behavior features of the next node, and calculate the similarity between the verification feedback features of the previous node and the verification scenario features of the next node.

[0168] Compare the similarity between the verification feedback features of the previous node and the verification behavior features of the next node, as well as the similarity between the verification feedback features of the previous node and the verification scenario features of the next node.

[0169] Step S1597: Calculate the association strength between two adjacent nodes based on two types of similarity. The association strength is the weighted average of the two types of similarity, and the weights are determined based on the association importance between feedback features and behavioral features, and between feedback features and scene features in historical data.

[0170] The calculated similarity is weighted and averaged to obtain the association strength between adjacent nodes. The weights are determined based on the association importance of feedback features with behavioral features and scene features in historical data, with higher similarity being assigned higher weights.

[0171] Step S1598: Compare the calculated association strength with the preset association strength threshold; if the association strength reaches the preset association strength threshold, then establish a bidirectional connection between two adjacent nodes to indicate that there is a valid association between the two nodes; if the association strength does not reach the preset association strength threshold, then do not establish a connection between two adjacent nodes, or establish a unidirectional connection to indicate that the degree of association between the two nodes does not meet the preset association strength threshold requirement.

[0172] The association strength is compared with a preset threshold. If the threshold is met, a bidirectional connection is established, indicating an effective association between nodes. If the threshold is not met, no connection is established or a unidirectional connection is established, indicating insufficient association.

[0173] Step S1599: Perform the above association strength calculation and connection relationship construction steps on all adjacent node pairs to form the node connection relationship of the strategy evolution adaptation chain.

[0174] The association strength is calculated and the connection relationship is constructed for all adjacent node pairs, and finally the node connection relationship of the policy evolution adaptation chain is formed, thus constructing the complete policy evolution adaptation chain.

[0175] Step S1510: Integrate the nodes, node attributes and node connection relationships to generate a policy evolution adaptation chain. Each node in the policy evolution adaptation chain corresponds to a set of feature combinations and corresponding verification features. The connection relationships between nodes reflect the association of different feature combinations in the verification process.

[0176] By integrating nodes, node attributes, and node connections, a policy evolution adaptation chain is generated. This chain displays each feature combination (node) and its verification feature (attribute), as well as the associations (connections) between different feature combinations during the verification process.

[0177] Step S160: Generate a scenario adaptation identifier based on the strategy evolution adaptation chain. If the scenario adaptation identifier is compatible, the optimized dynamic delivery strategy is formally applied to the delivery scenario corresponding to the target user group. If the scenario adaptation identifier is not compatible, the strategy verification data set is input into the big data model to re-execute the strategy feature encoding process.

[0178] Based on the constructed strategy evolution and adaptation chain, a scenario adaptation identifier is generated to determine whether the optimized dynamic delivery strategy is suitable for formal application to the delivery scenario corresponding to the target user group. If it is suitable, it is formally applied; if it is not suitable, the strategy feature encoding process needs to be re-processed.

[0179] Step S161: Extract all nodes and all node connections in the strategy evolution adaptation chain, and count the proportion of node pairs with connections to the total number of node pairs. This proportion is recorded as the node connection rate.

[0180] Extract all nodes and their connections from the strategy evolution adaptation chain. Count the number of node pairs with connections, and then calculate the proportion of each node pair to the total number of node pairs to obtain the node connection rate. This metric reflects the overall degree of association between nodes in the chain.

[0181] Step S162: For each node, extract the verification behavior features, verification scenario features, and verification feedback features from its node attributes, calculate the matching degree between each feature and the preset feature benchmark, and count the proportion of the number of features in each node that achieve the preset matching degree benchmark to the total number of features. This proportion is recorded as the node feature adaptation rate.

[0182] For each node, its verification behavior, scenario, and feedback features are extracted, and the matching degree of each feature is calculated with the preset feature benchmark. The proportion of features with a matching degree that meet the standard in each node is counted out to obtain the node feature adaptation rate, which reflects the adaptation of the node's features with the benchmark.

[0183] Step S163: Calculate the average node feature adaptation rate of all nodes, denoted as the average node feature adaptation rate.

[0184] The average node feature adaptation rate is obtained by averaging the node feature adaptation rates of all nodes, which reflects the overall feature adaptation level of the nodes.

[0185] Step S164: Calculate the scene adaptation comprehensive score based on the node connection rate and the average node feature adaptation rate. The scene adaptation comprehensive score is the weighted sum of the node connection rate and the average node feature adaptation rate. The weights are determined according to the correlation importance of the node connection relationship and the node feature adaptation rate in the scene adaptation judgment.

[0186] The overall scene adaptation score is calculated based on the node connectivity rate and the average node feature adaptation rate, using a weighted sum of the two. The weights are determined by the importance of node connectivity and node feature adaptation rate in judging scene adaptation, with higher-important metrics assigned higher weights.

[0187] Scenario adaptation assessment aims to determine whether the optimized dynamic delivery strategy matches and adapts to the real, complex, and dynamically changing business operating environment corresponding to the target user group. It goes beyond simply examining the strategy's content; it evaluates whether the strategy can stably, efficiently, and compliantly achieve the expected results under real-world resource conditions, user load, operational constraints, and real-time status. The assessment is typically based on a comprehensive calculation of multiple dimensions, including the stability of the correlation between strategy feature combinations reflected in the strategy evolution and adaptation chain, and the matching degree between verification data and expected features.

[0188] The specific judgment logic is as follows: First, by analyzing the strategy evolution adaptation chain, we determine whether the relationships (node ​​connection rate) among the various feature combinations (nodes) within the strategy are stable and close after testing and verification. This reflects the inherent consistency of the strategy logic. Second, we analyze the degree of matching between the actual characteristics (node ​​attributes) exhibited by each feature combination in the verification and the expected feature benchmark constructed based on historical data (node ​​feature adaptation rate). This reflects the external performance of the strategy in the test environment.

[0189] Finally, the "node connection rate," reflecting the stability of the internal logic, and the "average node feature adaptation rate," reflecting the matching degree of external performance, are comprehensively weighted and scored to obtain the "scenario adaptation comprehensive score." Here, "scenario adaptation" has a broad meaning, requiring the optimized strategy to simultaneously meet the following conditions: 1) the strategy's internal logic is self-consistent and can adapt to dynamic data changes; 2) in test scenarios containing real-world environmental factors such as resources, load, and constraints, the strategy's execution effect meets or closely approximates expectations. Only when the score reaches the target is the strategy considered to have the ability to operate stably and effectively in the more complex and varied real-world deployment scenarios corresponding to the target user group—that is, "adapted"—and can it be formally applied.

[0190] Step S165: Compare the scene adaptation comprehensive score with the preset adaptation score benchmark; if the scene adaptation comprehensive score reaches the preset adaptation score benchmark, generate a scene adaptation mark as adapted; if the scene adaptation comprehensive score does not reach the preset adaptation score benchmark, generate a scene adaptation mark as unsuitable.

[0191] The calculated scene adaptation score is compared with a preset adaptation score benchmark. If the score reaches the benchmark, it indicates that the optimized strategy is well adapted to the scene, and an adaptation label is generated; otherwise, an incompatibility label is generated.

[0192] Step S166: If the generated scenario adaptation identifier is adapted, associate and match each strategy entry in the optimized dynamic delivery strategy with the scenario type of the delivery scenario corresponding to the target user group, so that each strategy entry corresponds to its associated delivery scenario, and complete the formal application of the optimized dynamic delivery strategy in the delivery scenario corresponding to the target user group.

[0193] When the scenario adaptation flag is set to "adapted," each strategy item in the optimized dynamic delivery strategy is associated and matched with the scenario type corresponding to the target user group. For example, a strategy item targeting beauty products is matched with a beauty-themed recommendation scenario, ensuring that each strategy item is applied to the appropriate scenario, thus completing the formal application of the strategy.

[0194] Step S167: If the generated scenario adaptation identifier is not compatible, the policy verification data set is integrated with the preset user behavior data set, the preset delivery scenario data set, and the preset historical policy feedback data set to form a new multi-source data set.

[0195] If the scenario adaptation flag is "incompatible," it means the optimized strategy still needs improvement. The strategy verification dataset is then integrated with the previous user behavior dataset, scenario distribution dataset, and historical strategy feedback dataset to form a new multi-source dataset.

[0196] Step S168: Input the new multi-source data set into the big data model, re-execute the strategy feature encoding process, the processing steps include performing feature decomposition on the new multi-source data set, performing element mapping on the decomposed features, performing combination on the mapped elements, performing encoding on the combined features to generate strategy feature encoding results, reconstructing the user scenario strategy dynamic response cluster based on the newly generated strategy feature encoding results and generating a new initial dynamic distribution strategy.

[0197] The new multi-source dataset is input into the big data model, and the strategy feature encoding process is re-executed. Following the previous process, including feature decomposition, element mapping, combination, and encoding, new strategy feature encoding results are generated. Then, the user scenario strategy dynamic response cluster is reconstructed, a new initial dynamic distribution strategy is generated, and a new round of strategy optimization cycle begins.

[0198] Throughout the data processing, data potentially containing sensitive privacy information, such as user behavior data, is involved. To protect user privacy and prevent data leakage, data anonymization technology is employed to anonymize users' personal identification information (such as name, mobile phone number, address, etc.), replacing this sensitive information with meaningless identifiers. Simultaneously, data encryption technology is used to encrypt data during transmission and storage, ensuring that even if data is illegally obtained, it cannot be easily decrypted and used. Furthermore, strict access control policies are implemented, allowing only authorized personnel to access relevant data, and detailed logs are maintained for data access and operations to facilitate auditing and traceability, comprehensively safeguarding the security of user privacy data.

[0199] Based on the same inventive concept, please refer to Figure 2 This paper shows a schematic block diagram of a dynamic distribution strategy optimization system 100 based on a big data model, which is used to execute the above-described dynamic distribution strategy optimization method based on a big data model. The dynamic distribution strategy optimization system 100 based on a big data model may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0200] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located in the dynamic deployment strategy optimization system 100 incorporating a big data model and are separately configured. Alternatively, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the dynamic deployment strategy optimization method incorporating a big data model provided in the aforementioned method embodiments.

[0201] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for optimizing dynamic delivery strategies by combining big data models, characterized in that, The method includes: Based on a preset set of user behavior data, a set of delivery scenario data, and a set of historical policy feedback data, a big data model is invoked to perform policy feature encoding processing to generate policy feature encoding results. A user scenario policy dynamic response cluster is constructed based on the policy feature encoding results, and an initial dynamic delivery policy is generated based on the user scenario policy dynamic response cluster. The initial dynamic delivery strategy is applied to the delivery scenario corresponding to the target user group, and the real-time user behavior data set, the real-time scenario change data set, and the real-time strategy feedback data set generated during the strategy execution process are collected. The big data model is invoked to perform response cluster adaptability analysis on the user real-time behavior data set, the scenario real-time change data set, and the strategy real-time feedback data set, generating response cluster adaptability analysis results, and generating a strategy iteration start command based on the response cluster adaptability analysis results; Upon receiving the strategy iteration start instruction, the big data model is invoked to perform feature element reconstruction processing on the strategy feature encoding result, generating a reconstructed feature encoding result. The reconstructed feature encoding result is then used to optimize the initial dynamic distribution strategy, resulting in an optimized dynamic distribution strategy. The optimized dynamic delivery strategy is applied to the delivery scenario corresponding to the test user group. The strategy verification data set generated during the strategy verification process is collected, and the big data model is called to perform evolutionary adaptation chain construction on the strategy verification data set to generate the strategy evolutionary adaptation chain. Based on the strategy evolution adaptation chain, a scenario adaptation identifier is generated. If the scenario adaptation identifier is compatible, the optimized dynamic delivery strategy is formally applied to the delivery scenario corresponding to the target user group. If the scenario adaptation identifier is not compatible, the strategy verification data set is input into the big data model to re-execute the strategy feature encoding process.

2. The dynamic distribution strategy optimization method combining big data models according to claim 1, characterized in that, The process involves using a pre-defined set of user behavior data, a set of delivery scenario data, and a set of historical policy feedback data to call a big data model for policy feature encoding processing, generating policy feature encoding results, constructing a dynamic response cluster for user scenario policies based on the policy feature encoding results, and generating an initial dynamic delivery policy based on the dynamic response cluster for user scenario policies, including: After inputting the user behavior data set into the big data model, feature decomposition is performed on each user behavior record in the user behavior data set to extract the behavior type, behavior occurrence cycle, behavior-related content and subsequent feedback corresponding to each user behavior record, and generate user behavior feature units. Perform feature element mapping for each user behavior feature unit, converting behavior type into behavior type feature element, behavior occurrence cycle into behavior cycle feature element, behavior-related content into behavior-related feature element, and subsequent feedback into behavior feedback feature element, thereby generating a set of user behavior feature elements. After inputting the distribution scenario data set into the big data model, attribute parsing is performed on each distribution scenario record in the distribution scenario data set to extract the scenario type, scenario resource configuration, scenario user capacity and scenario operation constraints corresponding to each distribution scenario record, and to generate scenario feature units. For each scene feature unit, perform feature element mapping to transform scene type into scene type feature element, scene resource configuration into scene resource feature element, scene user capacity into scene capacity feature element, and scene operation constraints into scene constraint feature element, thereby generating a set of scene feature elements. After inputting the historical strategy feedback data set into the big data model, feedback decomposition is performed on each historical feedback record in the historical strategy feedback data set. The strategy execution effect, user satisfaction, strategy adjustment suggestions and feedback generation scenario corresponding to each historical feedback record are extracted to generate feedback feature units. Perform feature element mapping for each feedback feature unit, transforming the strategy execution effect into effect feature elements, user satisfaction into satisfaction feature elements, strategy adjustment suggestions into adjustment suggestion feature elements, and feedback generation scenario into feedback scenario feature elements, thereby generating a set of feedback feature elements. In the big data model, the co-occurrence frequency of each element in the user behavior feature element set and each element in the scene feature element set is calculated, and the element correlation is calculated based on the co-occurrence frequency. In addition, the co-occurrence frequency of each element in the scene feature element set and each element in the feedback feature element set is calculated, and the element correlation is calculated based on the co-occurrence frequency. Furthermore, in the big data model, combinations in which the correlation between user behavior feature elements, scene feature elements, and feedback feature elements all reach a preset correlation threshold are selected. Each combination contains a set of user behavior feature elements, a set of scene feature elements, and a set of feedback feature elements. All combinations that meet the conditions are integrated into a multi-source feature combination set. Encoding and integration are performed on the multi-source feature combination set to generate a policy feature encoding result containing multiple sets of feature combinations. Each set of feature combinations in the policy feature encoding result corresponds to a set of user scenario policy association relationships. In big data models, based on the behavioral type feature elements and scene type feature elements in the policy feature coding results, feature combinations with the same behavioral type and the same scene type are divided into the same response cluster unit. Calculate the association strength of feature combinations within each response cluster unit, sort the feature combinations within the response cluster unit based on the association strength, and generate a dynamic response cluster of user scenario strategy containing multiple sorted response cluster units. In the big data model, parameter extraction is performed on the feature combination within each response cluster unit of the user scenario strategy dynamic response cluster. The corresponding strategy content parameters, strategy delivery time parameters, and strategy coverage user parameters are extracted. The strategy content parameters, strategy delivery time parameters, and strategy coverage user parameters are associated and bound with the corresponding response cluster unit to generate the initial dynamic delivery strategy.

3. The dynamic distribution strategy optimization method combining big data models according to claim 1, characterized in that, The step of applying the initial dynamic delivery strategy to the delivery scenario corresponding to the target user group includes collecting a set of real-time user behavior data, a set of real-time scenario change data, and a set of real-time strategy feedback data generated during the strategy execution process, including: Extract the policy delivery time parameter of each policy entry from the initial dynamic delivery strategy, and determine the execution start time of each policy entry in the delivery scenario corresponding to the target user group based on the policy delivery time parameter; When the execution start time is reached, the policy coverage user parameters of each policy entry are extracted, and users who meet the policy coverage user parameters are selected from the target user group to form a policy execution user group. Push the policy content parameters of each policy entry to the corresponding policy execution user group to start the initial dynamic policy delivery execution process; During the strategy execution process, real-time behavior data of the strategy execution user group is collected according to the preset behavior data collection interval. Each piece of real-time behavior data is associated and marked with the corresponding strategy entry and the corresponding execution time point to generate a user real-time behavior data unit. The real-time behavior data includes the behavior trigger time, behavior execution steps, content involved in the behavior, and behavior end status. Real-time change data of the scenario is collected and distributed according to the preset scenario data collection interval. Each piece of real-time change data is associated and marked with the corresponding strategy entry and the corresponding execution time point to generate a scenario real-time change data unit. The real-time change data includes scenario resource usage, changes in the number of users in the scenario, scenario running status, and changes in scenario constraints. The system collects real-time feedback data submitted by the strategy execution user group during the strategy execution process through a preset feedback collection interface. Each piece of real-time feedback data is associated and marked with the corresponding strategy entry, the corresponding user real-time behavior data unit, the corresponding scene real-time change data unit, and the corresponding execution time point to generate a strategy real-time feedback data unit. The real-time feedback data includes evaluations of the strategy content, opinions on the timing of strategy issuance, feelings about the effect of strategy execution, and suggestions for strategy adjustment. All real-time user behavior data units are sorted according to the order of their execution time. The sorted real-time user behavior data units are then integrated to generate a set of real-time user behavior data. Sort all real-time change data units of the scene according to the order of their execution time, integrate the sorted real-time change data units of the scene, and generate a set of real-time change data of the scene. All real-time policy feedback data units are sorted according to their execution time, and then integrated to generate a real-time policy feedback data set.

4. The dynamic distribution strategy optimization method combining big data models according to claim 2, characterized in that, The step involves calling the big data model to perform response cluster adaptation analysis on the user's real-time behavior data set, the scene's real-time change data set, and the strategy's real-time feedback data set, generating response cluster adaptation analysis results, and generating a strategy iteration start instruction based on the response cluster adaptation analysis results, including: After inputting the user real-time behavior data set into the big data model, feature extraction is performed on each user real-time behavior data unit in the user real-time behavior data set to extract the behavior trigger frequency, behavior completion ratio, matching of behavior-related content with strategy content, and behavior end state distribution, thereby generating a real-time behavior feature set. After inputting the real-time scene change data set into the big data model, feature extraction is performed on each real-time scene change data unit in the real-time scene change data set to extract the scene resource usage growth rate, the fluctuation of the number of users in the scene, the stability of the scene operation status, and the frequency of changes in scene constraints, thereby generating a real-time scene feature set. After inputting the real-time feedback data set of the strategy into the big data model, feature extraction is performed on each real-time feedback data unit of the strategy in the real-time feedback data set of the strategy to extract the positive user evaluation ratio, the approval ratio of the timing of the release, the satisfaction ratio of the execution effect, and the adoption ratio of adjustment suggestions, and generate a real-time feedback feature set. In the big data model, feature combination elements of each response cluster unit in the dynamic response cluster of the user scenario strategy are extracted. Real-time behavior features are compared with behavior type features and behavior cycle features to calculate the behavior feature fit. Real-time scene features are compared with scene type features and scene resource features to calculate the scene feature fit. Real-time feedback features are compared with effect features and satisfaction features to calculate the feedback feature fit. In the big data model, based on the adaptability of behavioral features, scene features, and feedback features, the comprehensive adaptability of each response cluster unit is calculated, and the response cluster adaptability analysis results containing the comprehensive adaptability of each response cluster unit are generated. In the big data model, the number of response cluster units whose overall fit did not reach the preset fit benchmark was counted in the response cluster fit analysis results. If the number of response cluster units that have not reached the preset adaptation benchmark exceeds a preset number threshold, a strategy iteration start command is generated. The strategy iteration start command includes the identifier of the response cluster unit that needs to be reconstructed in terms of feature elements. If the number of response cluster units that have not reached the preset adaptation benchmark does not exceed the preset number threshold, a strategy iteration start command is not generated, and the initial dynamic distribution strategy continues to be executed.

5. The dynamic distribution strategy optimization method combining big data models according to claim 2, characterized in that, Upon receiving the strategy iteration start instruction, the process involves invoking the big data model to perform feature element reconstruction processing on the strategy feature encoding result, generating a reconstructed feature encoding result, and using the reconstructed feature encoding result to optimize the initial dynamic delivery strategy, resulting in an optimized dynamic delivery strategy, including: Upon receiving the strategy iteration start instruction, the response cluster unit identifier that needs to be reconstructed in the strategy iteration start instruction is extracted. Based on the response cluster unit identifier, the corresponding feature combination is selected from the strategy feature encoding result to form a set of feature combinations to be reconstructed. After inputting the set of features to be reconstructed into the big data model, defect identification is performed on the behavioral type feature elements, behavioral cycle feature elements, behavioral association feature elements, behavioral feedback feature elements, scene type feature elements, scene resource feature elements, scene carrying feature elements, scene constraint feature elements, effect feature elements, satisfaction feature elements, adjustment suggestion feature elements, and feedback scene feature elements in each feature combination to be reconstructed. The part of each feature element that does not match the real-time feature is determined, and feature element defect information is generated. Based on the defect information of the feature elements, supplementary feature elements are selected from the user behavior feature element set, the scene feature element set, and the feedback feature element set. The selection criterion is that the feature matching ratio between the supplementary feature element and the defective part reaches a preset matching ratio requirement. Replace the defective parts in the feature combination to be reconstructed with the supplementary feature elements to generate a temporary feature combination; The temporary feature combination is subjected to correlation verification. The correlation between different types of feature elements in the temporary feature combination is calculated. If the correlation reaches the preset correlation benchmark, the temporary feature combination is determined as the reconstructed feature combination. If the correlation does not reach the preset correlation benchmark, feature elements are re-selected and supplemented and replacement is performed until a reconstructed feature combination with the correct correlation is generated. All reconstructed feature combinations are integrated with unreconstructed feature combinations to generate reconstructed feature encoding results; In the big data model, based on each feature combination in the reconstructed feature encoding results, the corresponding optimized strategy content parameters, optimized strategy delivery time parameters, and optimized strategy user coverage parameters are extracted. From the initial dynamic distribution strategy, select the strategy entries corresponding to the response cluster unit identifiers that need feature element reconstruction. Then, replace the strategy content parameters, strategy distribution time parameters, and strategy coverage user parameters of the strategy entries corresponding to the response cluster unit identifiers that need feature element reconstruction from the initial dynamic distribution strategy with the optimized strategy content parameters, optimized strategy distribution time parameters, and optimized strategy coverage user parameters, respectively. Perform correlation and coordination processing on all policy entries in the replaced initial dynamic distribution strategy to make the optimized policy content parameters, optimized policy distribution time parameters, and optimized policy coverage user parameters of each policy entry mutually compatible. The set of strategy entries that have completed the association and coordination processing is determined as the optimized dynamic distribution strategy.

6. The dynamic distribution strategy optimization method combining big data models according to claim 1, characterized in that, The step involves applying the optimized dynamic delivery strategy to the delivery scenario corresponding to the test user group, collecting the strategy verification data set generated during the strategy verification process, and calling the big data model to perform evolutionary adaptation chain construction on the strategy verification data set to generate a strategy evolutionary adaptation chain, including: Extract the optimized policy coverage user parameters for each policy entry from the optimized dynamic delivery strategy, and select qualified users from the test user group based on the optimized policy coverage user parameters to form a policy verification user group. Extract the optimized strategy delivery time parameter of each strategy entry in the optimized dynamic delivery strategy, and determine the verification start time of each strategy entry in the delivery scenario corresponding to the test user group; When the verification start time is reached, the optimized policy content parameters of each policy entry are pushed to the corresponding policy verification user group to start the verification process of the optimized dynamic policy delivery. During the verification process, verification behavior data of the user group is collected according to the preset verification data collection interval. The verification behavior data includes the verification behavior trigger time, verification behavior execution process, verification behavior associated content and verification behavior result. Each piece of verification behavior data is associated with the corresponding strategy entry and the corresponding verification time point to generate a verification behavior data unit. The verification scenario data is collected during the verification process of the test deployment scenario. The verification scenario data includes the resource consumption of the verification scenario, the change in the number of users in the verification scenario, the running status of the verification scenario, and the change in the constraint conditions of the verification scenario. Each piece of verification scenario data is associated with the corresponding strategy item and the corresponding verification time point to generate a verification scenario data unit. The verification feedback data submitted by the user group during the verification process includes evaluations of the optimized strategy content, evaluations of the optimized distribution timing, evaluations of the optimized strategy execution effect, and suggestions for further adjustments to the optimized strategy. Each piece of verification feedback data is associated with and marked with the corresponding strategy item, the corresponding verification behavior data unit, the corresponding verification scenario data unit, and the corresponding verification time point to generate a verification feedback data unit. All verification behavior data units, all verification scenario data units, and all verification feedback data units are integrated in chronological order of verification time points to generate a strategy verification data set; After inputting the strategy verification dataset into the big data model, the verification behavior features, verification scenario features, and verification feedback features in the strategy verification dataset are extracted. In the big data model, the feature combination in the reconstructed feature encoding result is used as the adaptation chain node, and the verification behavior feature, verification scenario feature and verification feedback feature are used as node attributes. The association strength between adjacent nodes is calculated, and the node connection relationship is constructed based on the association strength. By integrating nodes, node attributes, and node connection relationships, a strategy evolution adaptation chain is generated. Each node in the strategy evolution adaptation chain corresponds to a set of feature combinations and corresponding verification features. The connection relationships between nodes reflect the association between different feature combinations in the verification process.

7. The dynamic distribution strategy optimization method combining big data models according to claim 2, characterized in that, The process of calculating the co-occurrence frequency of each element in the user behavior feature element set and each element in the scene feature element set in the big data model, calculating the element correlation based on the co-occurrence frequency, calculating the co-occurrence frequency of each element in the scene feature element set and each element in the feedback feature element set, calculating the element correlation based on the co-occurrence frequency, and selecting combinations in the big data model where the correlation between user behavior feature elements, scene feature elements, and feedback feature elements all reaches a preset correlation threshold includes: In the big data model, each behavior type feature element in the user behavior feature element set is paired with each scene type feature element in the scene feature element set to form a behavior scene pairing combination. Calculate the co-occurrence frequency of each behavioral scenario pairing combination. The co-occurrence frequency is the number of times the behavioral type corresponding to the behavioral type feature element and the scene type corresponding to the scene type feature element co-occur in historical data. Calculate the correlation between the behavioral type feature element and the scene type feature element based on the co-occurrence frequency. Each behavior scenario pairing is paired with each effect feature element in the feedback feature element set to form a behavior scenario effect pairing. The correlation between the behavior scenario pairing and the effect feature element is calculated. The correlation is the percentage of times the behavior scenario pairing and the strategy execution effect corresponding to the effect feature element appear together in historical data. Based on the correlation between behavioral type feature elements and scene type feature elements, and the correlation between behavioral scene pairing and effect feature elements, behavioral type feature elements, scene type feature elements, and effect feature elements whose correlation all reach the preset correlation threshold are selected, and the three behavioral type feature elements, scene type feature elements, and effect feature elements are combined to form the first feature combination. Using the same method, in the big data model, each behavior cycle feature element in the user behavior feature element set is paired with each scene carrying feature element in the scene feature element set, and the correlation between the behavior cycle feature element and the scene carrying feature element is calculated. Then, the resulting pairing is paired with each satisfaction feature element in the feedback feature element set, and the correlation between the pairing is calculated. Behavior cycle feature elements, scene carrying feature elements, and satisfaction feature elements that meet the correlation criteria are selected, and the behavior cycle feature elements, scene carrying feature elements, and satisfaction feature elements are combined to form a second feature combination. In the big data model, each behavior-related feature element in the user behavior feature element set is paired with each scene resource feature element in the scene feature element set. The correlation between the behavior-related feature element and the scene resource feature element is calculated. Then, the resulting pairing is paired with each adjustment suggestion feature element in the feedback feature element set. The correlation between the pairing and the adjustment suggestion feature element is calculated. Behavior-related feature elements, scene resource feature elements, and adjustment suggestion feature elements that meet the correlation criteria are selected. The behavior-related feature elements, scene resource feature elements, and adjustment suggestion feature elements are combined to form a third feature combination. In the big data model, each behavioral feedback feature element in the user behavior feature element set is paired with each scene constraint feature element in the scene feature element set. The correlation between the behavioral feedback feature element and the scene constraint feature element is calculated. Then, the resulting pairing is paired with each feedback scene feature element in the feedback feature element set. The correlation between the pairing and the feedback scene feature element is calculated. Behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements that meet the correlation criteria are selected. The behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements are combined to form a fourth feature combination. The first feature combination, the second feature combination, the third feature combination, and the fourth feature combination are integrated to form multiple feature combinations. Each feature combination contains three different types of feature elements: user behavior feature elements, scene feature elements, and feedback feature elements. The correlation between each feature element in each feature combination reaches a preset correlation threshold.

8. The dynamic distribution strategy optimization method combining big data models according to claim 5, characterized in that, The process involves performing correlation verification on the temporary feature combination, calculating the correlation between different types of feature elements in the temporary feature combination, and determining the temporary feature combination as the reconstructed feature combination if the correlation reaches a preset correlation benchmark. If the correlation does not reach the preset correlation benchmark, feature elements are re-selected and supplemented, and replacement is performed until a reconstructed feature combination with satisfactory correlation is generated. This includes: Extract the behavior type feature elements, scene type feature elements, and effect feature elements from the temporary feature combination; calculate the correlation between the behavior type feature elements and the scene type feature elements; calculate the correlation between the scene type feature elements and the effect feature elements; calculate the correlation between the behavior type feature elements and the effect feature elements. Extract the behavioral cycle feature elements, scene carrying feature elements, and satisfaction feature elements from the temporary feature combination; calculate the correlation between the behavioral cycle feature elements and the scene carrying feature elements; calculate the correlation between the scene carrying feature elements and the satisfaction feature elements; calculate the correlation between the behavioral cycle feature elements and the satisfaction feature elements. Extract the behavioral association feature elements, scene resource feature elements, and adjustment suggestion feature elements from the temporary feature combination; calculate the correlation degree between the behavioral association feature elements and the scene resource feature elements; calculate the correlation degree between the scene resource feature elements and the adjustment suggestion feature elements; calculate the correlation degree between the behavioral association feature elements and the adjustment suggestion feature elements. Extract behavioral feedback feature elements, scene constraint feature elements, and feedback scene feature elements from the temporary feature combination; calculate the correlation between behavioral feedback feature elements and scene constraint feature elements; calculate the correlation between scene constraint feature elements and feedback scene feature elements; calculate the correlation between behavioral feedback feature elements and feedback scene feature elements. Count the number of correlations that did not reach the preset correlation benchmark among all calculated correlations; If the number of correlations that does not meet the preset correlation benchmark is zero, then the correlation of the temporary feature combination is determined to meet the standard, and the temporary feature combination is determined as the reconstructed feature combination. If the number of correlations that do not meet the preset correlation benchmark is greater than zero, then the feature element pairs corresponding to the non-compliant correlations are identified. For each feature element pair corresponding to the non-compliant correlation, supplementary feature elements are re-screened from the corresponding feature element set. The screening criterion is that the matching ratio between the supplementary feature elements and the existing feature elements meets the preset matching ratio requirement. Replace the corresponding feature elements in the temporary feature combination with the newly selected supplementary feature elements to generate a new temporary feature combination; Re-perform the correlation verification on the new temporary feature combination, repeat the above correlation calculation, standard judgment and supplementary feature element screening and replacement steps until a reconstructed feature combination with qualified correlation is generated.

9. The dynamic distribution strategy optimization method combining big data models according to claim 1, characterized in that, The scenario adaptation identifier is generated based on the strategy evolution adaptation chain. If the scenario adaptation identifier is adapted, the optimized dynamic delivery strategy is formally applied to the delivery scenario corresponding to the target user group. If the scenario adaptation flag is not suitable, the policy verification data set is input into the big data model to re-execute the policy feature encoding process, including: Extract all nodes and all node connections in the strategy evolution adaptation chain, and count the proportion of node pairs with connections to the total number of node pairs. This proportion is recorded as the node connection rate. For each node, extract the verification behavior features, verification scenario features, and verification feedback features from its node attributes, calculate the matching degree between each feature and the preset feature benchmark, and count the proportion of the number of features in each node that achieve the preset matching degree benchmark to the total number of features. This proportion is recorded as the node feature adaptation rate. Calculate the average node feature fit rate of all nodes, and denote it as the average node feature fit rate. The scene adaptation comprehensive score is calculated based on the node connection rate and the average node feature adaptation rate. The scene adaptation comprehensive score is a weighted sum of the node connection rate and the average node feature adaptation rate. The weights are determined according to the correlation importance of node connection relationship and node feature adaptation rate in scene adaptation judgment. The scene adaptation score is compared with the preset adaptation score benchmark. If the scene adaptation score reaches the preset adaptation score benchmark, a scene adaptation mark is generated as adapted. If the scene adaptation score does not reach the preset adaptation score benchmark, a scene adaptation mark is generated as unsuitable. If the generated scenario adaptation identifier is adapted, each strategy entry in the optimized dynamic delivery strategy is associated and matched with the scenario type of the delivery scenario corresponding to the target user group, so that each strategy entry is associated with its associated matching delivery scenario, and the optimized dynamic delivery strategy is formally applied in the delivery scenario corresponding to the target user group. If the generated scenario adaptation identifier is not suitable, the policy verification data set is integrated with the preset user behavior data set, the preset distribution scenario data set, and the preset historical policy feedback data set to form a new multi-source data set. The new multi-source data set is input into the big data model, and the strategy feature encoding process is re-executed. The processing steps include performing feature decomposition on the new multi-source data set, performing element mapping on the decomposed features, performing combination on the mapped elements, and performing encoding on the combined features to generate strategy feature encoding results. Based on the newly generated strategy feature encoding results, the user scenario strategy dynamic response cluster is reconstructed and a new initial dynamic distribution strategy is generated.

10. A dynamic distribution strategy optimization system combining big data models, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the dynamic distribution strategy optimization method combining a big data model as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images