Automatic decision-making method and system for marketing tasks, electronic equipment and readable medium

Abstract

The invention relates to an automatic decision-making method and system for marketing tasks, electronic equipment and a readable medium, and relates to the technical field of computers. According to the method, in a marketing task execution process, a start-stop control scheme is adopted to determine a start-stop state of a marketing task; wherein the start-stop control scheme comprises at least one of a start-stop scheme based on an effect index threshold value and a start-stop scheme based on a channel; the marketing task corresponds to a first user group, the first user group is ranked and determined by adopting a preset user screening strategy based on a user ranking model, the user ranking model performs parameter updating by adopting differential privacy based on attribute information of the marketing task, and the attribute information is extracted after execution of the marketing task is completed. According to the method, feedback information of marketing task execution can be fully utilized, and the response efficiency of automatic marketing and the optimization capability of model updating iteration are effectively improved; in addition, the user ranking model enhances user privacy protection by adopting differential privacy in parameter updating, and information security is improved.

Description

Background Art

[0002] Marketing tasks refer to the tasks of promoting information to potential user groups through channels based on the product party's request. The marketing party is responsible for screening potential user groups based on the product party's request, as well as planning and implementing specific marketing task execution strategies. Usually, the marketing party can perform feature analysis and identification based on the product information provided by the product party and the user information of the potential user group to improve the accuracy of information promotion, as well as the hit rate and conversion rate.

[0003] Currently, automated marketing can be used to reduce implementation costs and improve promotion efficiency. However, in the process of automated marketing, data analysis models are usually used to analyze potential user groups. Data analysis models generally have low privacy protection during training, which leads to the leakage and abuse of user personal data. At the same time, it is difficult to make full use of real-time feedback information. There are problems such as slow response and insufficient optimization capabilities in task decision-making and resource allocation adjustments in automated marketing, which affects the accuracy of potential user group screening and identification, and thus affects the hit rate and conversion rate of automated marketing. Summary of the invention

[0004] The purpose of the present disclosure is to provide an automated decision-making method for marketing tasks, an automated decision-making system for marketing tasks, an electronic device and a computer-readable medium, which can strengthen the protection of user privacy data, and make full use of feedback information in automated marketing task decisions and model updates, improve response efficiency and optimization capabilities, improve the accuracy of user screening and identification, and thereby improve the hit rate and conversion rate of automated marketing.

[0005] According to a first aspect of the present disclosure, an automated decision-making method for marketing tasks is provided, which may include: during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task; the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel; wherein the marketing task corresponds to a first user group, the first user group is determined by sorting based on a user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to update parameters based on attribute information of the marketing task, and the attribute information is extracted after the execution of the marketing task is completed.

[0006] Optionally, the start-stop control scheme is a start-stop scheme based on an effect indicator threshold. During the execution of the marketing task, the start-stop control scheme is used to determine the start-stop status of the marketing task, including: obtaining real-time effect indicators during the execution of the marketing task; the real-time effect indicators are used to evaluate the effect of the marketing task on the second user group, and the second user group includes at least part of the user group that has been reached by the marketing task in the first user group through at least one channel; determining the effect indicator threshold corresponding to the marketing task; when the real-time effect indicator is greater than or equal to the effect indicator threshold, determining the marketing task to be in the on state; when the real-time effect indicator is less than the effect indicator threshold, determining the marketing task to be in the paused state.

[0007] Optionally, real-time effect indicators during the execution of the marketing task are obtained, including: determining the business window period; when the real-time time is within the business window period, determining the marketing task in a paused state; when the real-time time is outside the business window period, determining the marketing task in an enabled state, and obtaining real-time effect indicators during the execution of the marketing task.

[0008] Optionally, obtaining real-time effect indicators during the execution of the marketing task includes: determining a second user group; when the number of users in the second user group is less than the initial user threshold, determining the marketing task to be in an on state; when the number of users in the second user group is greater than or equal to the initial user threshold, obtaining real-time effect indicators during the execution of the marketing task.

[0009] Optionally, determining the effect indicator threshold corresponding to the marketing task includes: obtaining the first historical effect indicator of the historical marketing task that has been executed within the first historical period, and determining the third user group corresponding to the historical marketing task by sorting based on a user sorting model using a preset user screening strategy, and the historical effect indicator threshold corresponding to the historical marketing task; determining the historical mean indicator according to the first historical period and the first historical effect indicator; when the historical mean indicator is greater than zero, lowering the historical effect indicator threshold as the effect indicator threshold; when the historical mean indicator is less than or equal to zero, increasing the historical effect indicator threshold as the effect indicator threshold.

[0010] Optionally, determining the effect indicator threshold corresponding to the marketing task includes: obtaining a second historical effect indicator for each time period within a second historical cycle during the execution of the marketing task; determining a historical trend indicator based on the second historical cycle and the second historical effect indicator, and using the historical trend indicator as the effect indicator threshold.

[0011] Optionally, the marketing task corresponds to a profit threshold, and when the real-time effect indicator is less than the effect indicator threshold, the marketing task is determined to be in a suspended state, including: determining the real-time total cost and the real-time total benefit during the execution of the marketing task; determining the real-time total profit based on the real-time total cost and the real-time total benefit; when the real-time effect indicator is less than the effect indicator threshold, and the real-time total profit is less than or equal to the profit threshold, the marketing task is determined to be in a suspended state.

[0012] Optionally, the start-stop control scheme is a channel-based start-stop scheme. During the execution of the marketing task, the start-stop control scheme is used to determine the start-stop status of the marketing task, including: during the execution of the marketing task, determining the channel effect indicator of each marketing task under the channel in the third historical period; determining the first channel mean indicator corresponding to the channel based on the channel effect indicator; when the channel mean indicator is less than the mean indicator threshold, determining the second channel mean indicator of all marketing tasks under the channel in the fourth historical period, as well as the total mean indicator of all marketing tasks, the fourth historical period is greater than the third historical period; when it is determined that there is a channel abnormality based on the deviation value of the second channel mean indicator relative to the total mean indicator, all marketing tasks in the execution process under the channel are determined to be in a suspended state.

[0013] Optionally, during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task, including: in an automated decision-making mode, during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task.

[0014] Optionally, the user ranking model is obtained by the product party and the marketing party jointly participating in training in federated learning. The training steps of the user ranking model are as follows: the product party performs local model training locally and transmits the local model parameters to the marketing party in an encrypted manner; the marketing party aggregates the local model parameters provided by each product party to obtain the user ranking model.

[0015] Optionally, the user ranking model adopts the following steps to update parameters: the marketing party determines the feature weights to be adjusted based on the attribute information of the marketing task, and adopts differential privacy to update the parameters of the feature weights to be adjusted.

[0016] Optionally, the user sorting model uses the following steps to update parameters: the marketing party determines the user preference label based on the attribute information of the marketing task, and encrypts and transmits the user preference label to the corresponding product party; the product party calculates derivative features of local data based on the user preference label, and encrypts and transmits the obtained derivative features to the marketing party; the marketing party performs relevance screening on the derivative features provided by each product party, obtains key derivative features, and uses differential privacy to update the parameters of the user sorting model based on the key derivative features.

[0017] Optionally, the attribute information is obtained by analyzing at least one of the following user intentions:

[0018] User intention analysis based on the completion status of marketing tasks;

[0019] User intention analysis based on the completion effect of marketing tasks;

[0020] User intention analysis based on channel comparison;

[0021] User intention analysis based on product comparison;

[0022] User intention analysis based on the completion status and completion effect of marketing tasks;

[0023] User intention analysis based on the completion status of marketing tasks and comparison with telecommunications channels;

[0024] User intention analysis based on the completion effect of marketing tasks and comparison with telecommunications channels.

[0025] According to a second aspect of the present disclosure, an automated decision-making system for marketing tasks is provided, which may include: a marketing task decision module, which is used to determine the start and stop status of the marketing task by adopting a start and stop control scheme during the execution of the marketing task, and to extract attribute information after the execution of the marketing task is completed; the start and stop control scheme includes at least one of a start and stop scheme based on an effect indicator threshold and a start and stop scheme based on a channel; a model prediction and update module, which is used to determine the first user group corresponding to the marketing task by adopting a preset user screening strategy through a user sorting model, and to update the parameters of the user sorting model based on the attribute information of the marketing task by adopting differential privacy.

[0026] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0027] Processor; and

[0028] a memory for storing a computer program for the processor;

[0029] The processor is configured to implement the above-mentioned automated decision-making method for implementing the marketing task as the first aspect by executing a computer program.

[0030] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the automated decision-making method for marketing tasks according to the first aspect is implemented.

[0031] According to a fifth aspect of the present disclosure, a computer program product is provided, which, when executed on an electronic device, enables the electronic device to implement the automated decision-making method for marketing tasks as in the first aspect.

[0032] The present disclosure provides an automated decision-making method for marketing tasks, an automated decision-making system for marketing tasks, an electronic device, and a computer-readable medium. In the process of executing a marketing task, the method adopts a start-stop control scheme to determine the start-stop status of the marketing task; wherein the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel; the marketing task corresponds to a first user group, the first user group is determined by sorting based on a user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to update parameters based on the attribute information of the marketing task, and the attribute information is extracted after the execution of the marketing task is completed. In this method, the task start-stop decision of automated marketing and the parameter update of the user sorting model can make full use of the feedback information of the execution of the marketing task, effectively improve the response efficiency of automated marketing and the optimization ability of model update iteration, and comprehensively improve the accuracy of user screening and identification, as well as the hit rate and conversion rate of automated marketing; moreover, the user sorting model also strengthens the protection of user privacy data by adopting differential privacy in parameter update, thereby improving information security.

[0033] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0035] Figure 1 A flowchart of the implementation architecture of the automated decision-making method for marketing tasks provided in an embodiment of the present disclosure.

[0036] Figure 2 This is one of the step flow charts of the method for automated decision making for marketing tasks provided in an embodiment of the present disclosure.

[0037] Figure 3 This is a second flowchart of the method for automated decision making for marketing tasks provided in an embodiment of the present disclosure.

[0038] Figure 4 A flowchart of the training and updating steps of the user ranking model provided in an embodiment of the present disclosure.

[0039] Figure 5 A structural block diagram of an automated decision-making system for marketing tasks provided in an embodiment of the present disclosure.

[0040] Figure 6A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0041] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0042] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0043] It should be noted that the data obtained in this disclosure are accessed, collected, stored and used for subsequent analysis and processing with the consent and authorization of the users or the parties to whom the relevant data belong, after the users or the parties to whom the relevant data belong are clearly informed of the data collection content, data purpose, processing method and other information. The users or the parties to whom the relevant data belong may be sent the ways to access, correct and delete the data, as well as the methods to revoke consent and authorization.

[0044] Figure 1 This is a flowchart of the implementation architecture of the automated decision-making method for marketing tasks provided in the embodiment of the present disclosure. The steps of the method may be shown in step A below:

[0045] In step A, during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task; the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel.

[0046] Among them, the marketing task corresponds to the first user group, the first user group is determined by sorting based on the user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to update parameters based on the attribute information of the marketing task, and the attribute information is extracted after the marketing task is completed.

[0047] like Figure 1 As shown, the marketing task is a task of delivering product information to the first user group through a channel. Usually, the product party provides the marketing party with product information and business requirements. The marketing party can determine a preset user screening strategy based on at least one factor such as product information and business requirements, or can adopt an existing preset user screening strategy. The marketing party can determine the first user group from the potential user group based on the user screening strategy, and then establish a marketing task of delivering the product information to the first user group. The number of marketing tasks can be one or more, and different marketing tasks can be executed serially or in parallel. Figure 1 For example, the exemplary marketing task includes marketing task 1, marketing task 2 and marketing task 3, and the preset user screening strategies of different marketing tasks may be the same or different; and the first user groups of different marketing tasks may be the same or different.

[0048] The preset user screening strategy determined by the marketing party may include screening criteria among potential user groups, such as age, region, consumption habits, etc. The marketing party may use the user sorting model to conduct comprehensive data analysis and sorting of product information, business needs, user attributes, user behaviors, etc. of potential user groups, so as to screen the first user group with high intention among the potential user groups as the delivery target of the marketing task.

[0049] Channels can correspond to entities that reach users, such as entities that make phone calls, send text messages, send emails, etc. Phone calls can be manual calls, AI (Artificial Intelligence) outbound calls, etc. Marketers can push product information to specific users in corresponding forms through channels.

[0050] The execution process of the marketing task includes gradually delivering product information to users in the first user group through the channel until the delivery of product information to each user is completed, and the marketing task can be considered to be completed. In step A, a start-stop control scheme can be used to determine the start-stop state of the marketing task. The start-stop control scheme is used to control the start and stop of the marketing task through feedback information during the execution process, so as to timely suspend marketing tasks that deviate from the delivery intention, have a delivery effect lower than expected, or have other problems, or keep the marketing tasks whose delivery effect meets or exceeds expectations in the open state, so as to timely respond to feedback information and dynamically adjust the execution of the marketing task.

[0051] Among them, based on the form and source of feedback information, the start-stop control scheme may include a start-stop scheme based on effect indicator thresholds, a start-stop scheme based on channels, etc.

[0052] The start-stop scheme based on the effect index threshold can be to evaluate the marketing tasks based on the effect index threshold. The effect index threshold can be set based on the effect index of the marketing task. The effect index includes user conversion rate, return on investment (ROI), user praise rate, etc. The effect index threshold is used to evaluate the quality of the effect index. The effect index threshold can be a static threshold set based on experience, or a dynamic threshold calculated based on the recent historical data at each comparison, or a predicted threshold determined by calculating the development trend based on the recent historical data at each comparison. The embodiments of the present disclosure do not impose specific restrictions on this.

[0053] The channel-based start-stop scheme can be to evaluate the delivery effect of each channel. The delivery effect can be determined based on the effect indicators of all marketing tasks corresponding to the channel. The effect indicators can correspond to the above-mentioned related instructions, and will not be repeated here to avoid repetition.

[0054] The above examples of start-stop control schemes are for illustrative purposes only. Those skilled in the art may select and design a suitable start-stop control scheme according to the status of actual marketing tasks and implementation conditions, and the embodiments of the present disclosure do not impose specific limitations on this.

[0055] like Figure 1 As shown, during the execution of marketing task 1, marketing task 2 and marketing task 3, step A is executed, and the start-stop state is determined using the start-stop control scheme; marketing task 1 and marketing task 3 remain in the on state, and marketing task 2 is converted to a paused state.

[0056] On this basis, the corresponding attribute information can also be extracted after the marketing task is completed. The attribute information may include the start and stop behavior, execution effect, user feedback, etc. of the marketing task during execution. Therefore, the user ranking model can use differential privacy to update parameters based on the attribute information. The parameter update may include the adjustment of the original feature weight, the embedding of derived features, etc. Differential privacy effectively protects the security of user privacy data by adding random noise.

[0057] In an exemplary embodiment of the present disclosure, Figure 2 This is one of the flow charts of the steps of the automated decision-making method for marketing tasks provided by the embodiment of the present disclosure. Figure 2 As shown, when the start-stop control scheme is a start-stop scheme based on an effect index threshold, the aforementioned step A may include the following steps 201 to 204.

[0058] In step 201, real-time effect indicators are obtained during the execution of the marketing task. The real-time effect indicators are used to evaluate the effect of delivering the marketing task to the second user group, and the second user group includes at least part of the user group that has been reached by the marketing task in the first user group through at least one channel.

[0059] In the disclosed embodiment, during the execution of the marketing task, only part of the users in the first user group are delivered. At this time, at least part of the user group in the first user group that has been reached by at least one channel can be identified as the second user group. For example, the first user group includes 100 users, and the marketing task needs to complete AI outbound call delivery to 50 of them and SMS delivery to 50 users during the execution process; when AI outbound call delivery to 10 users and SMS delivery to 30 users have been completed, the second user group includes 40 users; other situations can be deduced in the same way.

[0060] On this basis, real-time effect indicators can be used to evaluate the effect of marketing tasks on the second user group, such as real-time ROI, real-time conversion rate, real-time favorable comment rate, etc. Among them, the real-time conversion rate can be calculated by the ratio of the number of users in the second user group who interact with the recommended product to the total number of users. The interactive behavior can be collection, ordering, following, downloading, evaluation, etc.; the real-time favorable comment rate can be calculated by the ratio of the total number of users who generate evaluations in the second user group to the total number of users who give favorable comments; the real-time ROI (CurrentROI) can be calculated by the following formula (1):

[0061]

[0062] Among them, CurrentRevenue can refer to the real-time delivery benefit, that is, the total benefit generated by the marketing party delivering to the second user group in the process of executing the marketing task; and CurrentCost can refer to the real-time delivery cost, which refers to the total cost generated by the marketing party delivering to the second user in the process of executing the marketing task. The calculation of delivery benefit is related to factors such as the size of the second user group, and the calculation of delivery cost is related to factors such as the size of the second user group and the channel type.

[0063] In step 202, the effect indicator threshold corresponding to the marketing task is determined.

[0064] In the embodiment of the present disclosure, the effect indicator threshold can correspond to the aforementioned Figure 1The description of the start-stop scheme based on the effect indicator threshold in the article is not repeated here to avoid repetition. When the effect indicator threshold is a static threshold, the effect indicator threshold pre-set or configured for the marketing task can be obtained; when the effect indicator threshold is a dynamic threshold or a predicted threshold, the dynamic effect indicator threshold can be determined by statistically calculating the historical effect indicators.

[0065] In step 203, when the real-time effect index is greater than or equal to the effect index threshold, the marketing task is determined to be in an on state.

[0066] In step 204, when the real-time effect index is less than the effect index threshold, the marketing task is determined to be in a paused state.

[0067] On this basis, the marketing task can be started and stopped based on the comparison between the real-time effect index and the effect index threshold. For example, when the real-time effect index is greater than or equal to the effect index threshold, it can be considered that the effect of the marketing task on the second user group meets or exceeds expectations, and the marketing task is kept in the active state (Active); when the real-time effect index is less than the effect index threshold, it can be considered that the effect of the marketing task on the second user group is lower than expectations, and the marketing task is converted to the paused state (Paused). Specifically, taking ROI as an example, its execution logic can be expressed as the following formula (2) and formula (3):

[0068] If CurrentROI≥T roi then TaskState=Active (2)

[0069] If CurrentROI <T roi then TaskState=Paused (3)

[0070] Among them, T roi is the effect indicator threshold. When formula (2) is met, if the marketing task is in a paused state, it can be reactivated and turned on.

[0071] In an exemplary embodiment of the present disclosure, step 201 may include the following steps B1 to B3.

[0072] Step B1: Determine the business window period.

[0073] Step B2: When the real time is within the business window period, the marketing task is determined to be in a suspended state.

[0074] Step B3: When the real time is outside the business window period, the marketing task is determined to be in an open state, and real-time effect indicators during the execution of the marketing task are obtained.

[0075] In the embodiment of the present disclosure, before the start-stop control scheme is adopted to control the start-stop state of the marketing task, it can also be determined whether the real time is in the business window period. Among them, the business window period can be a time window in which the marketing task is not executed. The business window period can be set according to the business needs of the product party, according to the use conditions of the channel, according to the distribution of the behavior habits of the potential user group, or according to other factors, or by comprehensive analysis of multiple factors. The embodiment of the present disclosure does not make specific restrictions on this.

[0076] On this basis, during the execution of a marketing task, it can be determined whether the business window period has been entered. If the real-time time is within the business window period, the marketing task in the open state can be turned into the paused state; if the real-time time is outside the business window period, the marketing task is turned into the open state, and the real-time effect index during the execution of the marketing task is obtained to control the subsequent start and stop states. Alternatively, when activating a marketing task, if the real-time time is within the business window period, the marketing task is kept in the paused state; if the real-time time is outside the business window period, the marketing task is successfully activated and turned into the open state, and the real-time effect index during the execution of the marketing task is obtained to control the subsequent start and stop states.

[0077] In an exemplary embodiment of the present disclosure, step 201 may include the following steps C1 to C3.

[0078] Step C1: determine the second user group.

[0079] Step C2: when the number of users in the second user group is less than the initial user threshold, the marketing task is determined to be in an on state.

[0080] Step C3: when the number of users in the second user group is greater than or equal to the initial user threshold, obtain a real-time effect indicator during the execution of the marketing task.

[0081] In the disclosed embodiment, in order to avoid the situation where the number of users delivered to the marketing task is too small in the initial stage, resulting in the effect indicator threshold having low effectiveness and reliability in evaluating the real-time effect indicator. At this time, the initial stage of the marketing task can be judged. During the execution of the marketing task, the second user group is first counted in real time. The second user group can correspond to the relevant description of the aforementioned step 201. To avoid repetition, it will not be repeated here. On this basis, the initial stage can be judged based on the number of users that have been delivered, and the number of users in the second user group can be evaluated based on the initial user threshold. When the number of users in the second user group is less than the initial user threshold, it can be considered that the number of users delivered since the execution of the marketing task is too small and is still in the initial stage; when the number of users in the second user group is greater than or equal to the initial user threshold, it can be considered that the number of users delivered since the execution of the marketing task is not in the initial stage, and then the real-time effect indicator during the execution of the marketing task is obtained to control the subsequent start and stop status. Specifically, its execution logic can be expressed as the following formula (4):

[0082] If Sent <N min then TaskState=Active (4)

[0083] Where Sent represents the number of users in the second user group that have been delivered; N min Indicates the initial user threshold. When a marketing task is still in the initial stage, there is no need to evaluate the delivery effect and it should be kept in the open state so that the new marketing task will not be mistakenly turned into a paused state before accumulating enough delivered users.

[0084] In the disclosed embodiment, corresponding initial user thresholds may also be set for different channels, and further distinction may be made as to whether the number of users reached through different channels during the execution of different marketing tasks is less than the corresponding initial user thresholds, so as to make an initial stage judgment.

[0085] In an exemplary embodiment of the present disclosure, step 202 may include the following steps D1 to D4.

[0086] Step D1, obtaining the first historical effect indicator of the executed historical marketing task in the first historical period, the third user group corresponding to the historical marketing task is determined by sorting based on the user sorting model using a preset user screening strategy, and the historical marketing task corresponds to the historical effect indicator threshold.

[0087] In the embodiment of the present disclosure, the historical marketing task refers to the marketing task that has been executed and completed. The historical marketing task can be delivered to the third user group. The third user group is determined by the user ranking model using the aforementioned Figure 1The preset user screening strategy in the description of step A is screened and sorted, and the historical effect indicator threshold has been adopted during the execution. It should be noted that even if the same preset user screening strategy is adopted, the number, distribution and specific identity of the first user group and the second user group determined on the potential user group may be different. This is because the potential user group may change, or different product parties may specify or provide different potential customer groups.

[0088] On this basis, the first historical effect index of the executed historical marketing task in the first historical period can be obtained. Taking the historical effect index as PreviousROI as an example, the historical delivery benefit and historical delivery cost of each historical marketing task completed in the first historical period are determined, and then the corresponding PreviousROI is determined. Among them, the first historical period can be set according to actual business needs, calculation conditions, etc., such as 1 hour, 1 day, 1 week, 1 month, etc., and the embodiments of the present disclosure do not impose specific restrictions on this.

[0089] Step D2: determine the historical average indicator according to the first historical cycle and the first historical effect indicator.

[0090] In the disclosed embodiment, the historical average index may represent the average historical effect index of each marketing task, and the historical average index under the first historical effect index may be determined in the first historical period. Taking the historical effect index as PreviousROI as an example, the calculation of the historical average index AvgROI is shown in the following formula (5):

[0091]

[0092] Where n is the total number of historical marketing tasks that have been executed and completed in the first historical cycle, PreviousROI i is the PreviousROI of the i-th historical marketing task.

[0093] Step D3: when the historical mean index is greater than zero, lower the historical effect index threshold as the effect index threshold.

[0094] Step D4: when the historical average indicator is less than or equal to zero, increase the historical effect indicator threshold as the effect indicator threshold.

[0095] In the disclosed embodiment, relative to the first historical cycle in which the historical marketing tasks that have been executed are located, the effect indicator threshold used in the execution process of the marketing tasks in this execution cycle can be obtained by adjusting the historical effect indicator based on the historical mean indicator. When the historical mean indicator is greater than zero, it can be considered that the preset user screening strategy meets the expectations for user intention screening, thereby lowering the historical effect indicator threshold as the effect indicator threshold in this execution cycle to expand the investment in this execution cycle; when the historical mean indicator is less than or equal to zero, the historical effect indicator threshold can be increased as the effect indicator threshold in this execution cycle to avoid unnecessary investment. Specifically, taking ROI as an example, the execution logic can be shown in the following formulas (6) and (7):

[0096] IfAvgROI>0 then T ROI,new =T ROI,old -ΔT (6)

[0097] IfAvgROI≤0 then T ROI,new =T ROI,old +ΔT (7)

[0098] Among them, T ROI,old is the historical effect indicator threshold of the first historical cycle, T ROI,new is the effect indicator threshold in this execution cycle, and ΔT is the adjustment range of increase or decrease. The size of ΔT can be set and adjusted according to actual business needs, such as ΔT=0.05, and the embodiment of the present disclosure does not make specific restrictions on this.

[0099] In an exemplary embodiment of the present disclosure, step 202 may include the following steps E1 and E2.

[0100] Step E1: obtaining the second historical effect indicator in different time periods within the second historical period during the execution of the marketing task.

[0101] In the disclosed embodiment, the change trend of the second historical effect indicator of the marketing task in the second historical period can also be analyzed, such as growth, flat, negative growth, etc., and then the effect indicator threshold can be determined according to its change trend. Specifically, the second historical effect indicator of the time period in the second historical period during the execution of the marketing task can be obtained. The second historical period takes the start time of the marketing task as the starting point, and determines the second historical effect indicator of the time period since the launch. The second historical effect indicator can use the time point of the time period in the second historical period as the statistical moment, and the distribution and density of the time period can be set according to business needs.

[0102] For example, if the marketing task has been executed for 6 hours, the past 6 hours can be used as the second historical cycle, 1 hour as the time period, and the end of each hour as the statistical moment of the second historical effect indicator in the time period, to obtain 6 second historical effect indicators.

[0103] Step E2: determine a historical trend indicator according to the second historical period and the second historical effect indicator, and use the historical trend indicator as an effect indicator threshold.

[0104] On this basis, a historical trend indicator can be determined based on the second historical period and the second historical effect indicator. The historical trend indicator is used to compare with the real-time effect indicator to determine whether there is negative growth, that is, the historical trend indicator can be used as the effect indicator threshold. The historical trend indicator can be a prediction of the effect indicator of real time based on the change rate of the second historical effect indicator in the second historical period, such as predicting the real-time effect indicator of the 7th hour based on the second historical effect indicator of the past 6 hours. It can also be a time period average indicator determined based on the second historical effect indicator in the second historical period, such as calculating the average of each hour in the past 6 hours by obtaining the second historical indicator of 6 hours. For example, taking ROI as an example, the historical trend indicator TrendROI can be calculated and determined by the following formula (8):

[0105]

[0106] Where N is the number of time periods in the second historical period, ROI t-i is the ROI value of the ti-th period.

[0107] At this time, taking TrendROI as the effect indicator threshold, the start and stop state control logic of the marketing task can be shown as the following formula (9):

[0108] If CurrentROI≥TrendROI then TaskState=Active (9)

[0109] Among them, CurrentROI is the real-time effect indicator corresponding to this period. When CurrentROI is greater than or equal to TrendROI, the above formula (9) is established, indicating that the real-time effect indicator has not shown negative growth, and the marketing task can be determined to be in the open state; on the contrary, when the above formula (9) is not established, it can be considered that the real-time effect indicator has shown negative growth, and the marketing task is controlled to be switched to the paused state.

[0110] On this basis, by taking the performance of preset user screening strategies and marketing tasks as the basis for adjusting the threshold of effect indicators, it is possible to respond to marketing task decisions in a timely manner, reasonably allocate delivery resources, and improve delivery effects.

[0111] In an exemplary embodiment of the present disclosure, on the basis of step E1 to step E2, the aforementioned step 204 may include the following steps E3 to step E5.

[0112] Step E3: Determine the real-time total cost and real-time total benefit during the execution of the marketing task.

[0113] Step E4: determining the real-time total profit based on the real-time total cost and the real-time total benefit.

[0114] Step E5: When the real-time effect index is less than the effect index threshold and the real-time total profit is less than or equal to the profit threshold, the marketing task is determined to be in a suspended state.

[0115] In the disclosed embodiment, on the basis of the judgment of negative growth, a judgment can also be made in combination with the real-time total profit. The real-time total cost and the real-time total benefit during the execution of the marketing task are counted, and then the real-time total profit is determined based on the real-time total cost and the real-time total benefit. The real-time total profit may gradually increase with the delivery of the marketing task, or it may gradually decrease, or it may be negative. On this basis, a profit threshold can be set as an evaluation standard for the real-time total profit. When the real-time total profit is less than or equal to the profit threshold, it indicates that the accumulated profit has decreased or has not increased, and it can be considered that the accumulated profit does not meet expectations. On this basis, combined with the judgment of the real-time effect index and the effect index threshold, when the real-time effect index is less than the effect index threshold, and the real-time total profit is less than or equal to the profit threshold, the marketing task is determined to be in a suspended state. Among them, the calculation of the real-time total profit Profit can be shown in the following formula (10):

[0116] Profit=TotalRevenue-TotalCost (10)

[0117] Among them, TotalRevenue is the real-time total benefit, and TotalCost is the real-time total cost.

[0118] Furthermore, with the profit threshold T profit_threshold The execution logic of the judgment is shown in the following formula (11):

[0119] Profit <T profit_threshold (11)

[0120] As shown in formula (11), when the real-time total profit is less than or equal to the profit threshold, if CurrentROI has negative growth, the marketing task should be suspended even if the real-time effect indicator is positive to avoid further loss.

[0121] In an exemplary embodiment of the present disclosure, Figure 3 The second step flow chart of the automated decision-making method for marketing tasks provided by the embodiment of the present disclosure. Figure 3As shown, when the start-stop control scheme is a channel-based start-stop scheme, the aforementioned step A may include the following steps 301 to 304.

[0122] Step 301: During the execution of the marketing task, determine the channel effect index of each marketing task under the channel in the third historical period.

[0123] In the disclosed embodiment, the start and stop status of the marketing task can also be controlled in the channel dimension according to the overall performance of the effect index in the channel. Among them, a marketing task can be delivered through one or more channels, and each channel can also carry one or more marketing tasks, so that one or more marketing tasks can be comprehensively analyzed and started and stopped in the channel dimension. During the execution of the marketing task, the channel effect index of each marketing task under the channel within the third historical period can be determined. The third historical period can be the historical period of the marketing task delivered to the first user on the channel to the real-time time, or it can be a historical period of a preset time before the real-time time, which can be 1 hour, 1 day, etc.

[0124] Step 302: Determine a first channel mean index corresponding to the channel based on the channel effect index.

[0125] In the embodiment of the present disclosure, based on the channel effect indicators corresponding to different marketing tasks, the first channel average indicator corresponding to the channel can be further determined in all marketing tasks. Taking ROI as an example, the first channel average indicator AvgROI channel The calculation formula (12) is as follows:

[0126]

[0127] Where m is the total number of marketing tasks performed through this channel in the third historical period, ROI j is the ROI value of the jth marketing task.

[0128] Step 303, when the channel mean index is less than the mean index threshold, determine the second channel mean index of all marketing tasks under the channel in the fourth historical period, and the total mean index of all marketing tasks, and the fourth historical period is greater than the third historical period.

[0129] On this basis, the mean indicator threshold of the first channel can be used to evaluate the performance of all marketing tasks of the channel in the third historical period. When the mean indicator of the first channel is less than the mean indicator threshold, it can be considered that the performance of the channel in the third historical period is low. Taking ROI as an example, the mean indicator threshold can be set to T threshold , in AvgROI channel Less than T thresholdIn the case of T, it can be considered that the overall ROI of the channel is low and further resource status inspection is needed. threshold The specific value of can be set according to business requirements and calculation conditions, such as T threshold =0.1 (indicating 10%), which is not specifically limited in the embodiment of the present disclosure.

[0130] Furthermore, the resource status check of the channel is to evaluate the overall performance of the channel's effectiveness indicators in the fourth historical period. The fourth historical period is greater than the third historical period. For example, if the third historical period is 1 hour, the fourth historical period may be 3 hours, 6 hours, 12 hours, etc. If the third historical period is 1 day, the fourth historical period may be 3 days, 5 days, 7 days, etc. The disclosed embodiment does not impose specific restrictions on this. The second channel mean indicator of all marketing tasks under the channel in the fourth historical period and the total mean indicator of all marketing tasks may correspond to the relevant calculations of the aforementioned third historical period. To avoid repetition, they will not be repeated here. Taking ROI as an example, the OverallROI of the second channel mean indicator channel The calculation of is shown in the following formula (13):

[0131]

[0132] Where p is the total number of marketing tasks performed through this channel in the fourth historical period, ROI k is the ROI value of the kth marketing task.

[0133] Step 304: When it is determined that a channel abnormality exists based on the deviation value of the second channel mean index relative to the total mean index, all marketing tasks in execution process under the channel are determined to be in a suspended state.

[0134] In the disclosed embodiment, the second channel mean index describes the mean of the effect indexes of all marketing tasks under a single channel in the fourth historical period, and the total mean index describes the mean of the effect indexes of all marketing tasks under all channels in the fourth historical period. Based on this, the deviation value of the second channel mean index relative to the total mean index can be used to evaluate the resource anomaly of the channel, wherein the deviation value indicating that the second channel mean index is lower than the total mean index can be regarded as an anomaly, or the deviation value indicating that the second channel mean index exceeds the total mean index by too much can be regarded as an anomaly, or the deviation value indicating that the second channel mean index is lower than or exceeds the total mean index by too much can be regarded as an anomaly. For example, taking ROI as an example, the execution logic can be as shown in the following formula (14):

[0135] |Overall ROI channel -AvgROI previous |>0.2·AvgROI previous(14)

[0136] Among them, AvgROI previous is the overall mean indicator.

[0137] On this basis, by combining short-term and long-term evaluation, when the short-term channel average index of a channel is low, we can further evaluate whether the long-term channel average index of the channel deviates from the long-term total average index. When marketing tasks are delivered through multiple channels, the marketing tasks can be started and stopped in the channel dimension.

[0138] In an exemplary embodiment of the present disclosure, the aforementioned step 201 may further include the following step F.

[0139] Step F: In the automated decision-making mode, during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task.

[0140] In the embodiments of the present disclosure, the automated decision-making mode can be distinguished from other marketing task maintenance modes. Other marketing task maintenance modes may include a manual decision-making mode, in which the start and stop states are completely controlled manually; or a semi-manual decision-making mode, in which automated data statistical analysis is performed and the analysis results are provided manually, and the start and stop states of the marketing tasks are decided manually based on the analysis results; or a semi-manual review mode, in which automated decision-making on the start and stop states is performed, and the decision results are reviewed manually. If the review is passed, the decision is executed, and if the review is rejected, a new decision is made; or technical personnel in this field can design corresponding modes according to actual needs, and the embodiments of the present disclosure do not impose specific restrictions on this.

[0141] On this basis, when it is determined that the automatic decision-making mode is in place, the start-stop control scheme can be used to determine the start-stop status of the marketing task during the execution of the marketing task. Figures 1 to 3 To avoid repetition, any related description will not be repeated here.

[0142] In an exemplary embodiment of the present disclosure, Figure 4 A flowchart of the training and updating steps of the user ranking model provided in the embodiment of the present disclosure is shown in FIG. Figure 4 As shown in the figure, the user ranking model is obtained by the joint training of the product side and the marketing side in federated learning. Figure 4 The product side is only used as an example. Those skilled in the art can adjust the product side involved in the federated training according to actual needs, and the embodiments of the present disclosure do not impose specific limitations on this.

[0143] In the disclosed embodiment, the user ranking model can be implemented based on federated learning in which the product side and the marketing side participate, including feature discovery on the product side and feature screening and modeling training on the marketing side. The marketing side can also further perform model testing to verify the performance of the user ranking model.

[0144] The process may include the following steps 401 to 402 .

[0145] Step 401: The product side performs local model training locally and transmits the local model parameters to the marketing side in encrypted form.

[0146] In the disclosed embodiment, the product side can perform local model training based on its own product information, user data, etc., and upload local model parameters to the marketing side without uploading local original data. For each product side, the process of local model training can be expressed as the following formula (15):

[0147]

[0148] Among them, θ i new is the updated local model parameter of the i-th product side, θ i old is the existing local model parameter of the i-th product, η is the learning rate, L(θ i old ;D i ) is the i-th product based on the local data set D i The loss function of .

[0149] Step 402: The marketing party aggregates the local model parameters provided by each product party to obtain a user ranking model.

[0150] In the disclosed embodiment, the marketing party performs global model training, and the product party can encrypt and transmit local model parameters to the marketing party. The marketing party decrypts based on the encryption protocol to obtain the local model parameters provided by all product parties, and then aggregates the local model parameters provided by different product parties to generate a global model. Among them, the method of aggregating local model parameters can be selected according to business needs and calculation conditions, such as addition, average, weighted average, etc. Taking weighted average aggregation as an example, for the marketing party, the process of aggregating the global model can be expressed as the following formula (16):

[0151]

[0152] Among them, n i is the number of samples of the i-th product party, θ global is the feature weight of the global model.

[0153] On this basis, in an exemplary embodiment of the present disclosure, based on the attribute information of the marketing task, the user ranking model can adopt the following step G to update the parameters.

[0154] Step G: The marketing party determines the feature weights to be adjusted based on the attribute information of the marketing task, and uses differential privacy to update the parameters of the feature weights to be adjusted.

[0155] In the disclosed embodiment, feature tags can be used in the user ranking model to identify intentions in potential user groups to complete screening. Different feature tags correspond to specific feature weights, indicating their criticality and decisiveness in the screening process. On the basis of the completion of the marketing task, the marketing party can determine the feature weights to be adjusted locally based on the attribute information. The attribute information may include the start and stop behavior, execution effect, user feedback, etc. of the marketing task during the execution process. The attribute information represents the status of the marketing task in different attribute aspects. The attribute information can be used to analyze the association between the feature tags for user screening and the different attribute aspects of the marketing task, so that the marketing party can adjust the feature weights of the user ranking model based on the attribute information, such as, when the feature tag includes age, the association between age and pause behavior; or, when the feature tag includes consumption habits, the association between consumption habits and conversion rate, etc. On this basis, the user ranking model can be adjusted based on the feature weights to match or strengthen the adaptation of the marketing task to the business needs in the corresponding attribute aspects, such as by adjusting the feature weights to reduce the pause frequency of the screened user group when accepting the delivery, improve ROI, improve conversion rate, improve praise rate, etc.

[0156] In the embodiment of the present disclosure, based on the determination of the feature weights to be adjusted, differential privacy can be used to update the parameters of the feature weights to be adjusted. Taking the user ranking model as a linear regression model as an example, its expression (17) is as follows:

[0157]

[0158] Where y represents the prediction target; β0 represents the intercept of the user ranking model; β j represents f j The feature weight of ;∈ represents the error term. The prediction target can be selected according to the needs, such as the user's conversion rate.

[0159] On this basis, when updating the parameters of the user ranking model, differential privacy is used to introduce random noise N(0,σ 2 ) is shown in the following formula (18):

[0160]

[0161] Among them, σ 2 represents the variance of the noise.

[0162] Alternatively, the user ranking model may use the following steps H1 to H3 to update parameters.

[0163] Step H1: The marketing party determines the user preference tag based on the attribute information of the marketing task, and encrypts and transmits the user preference tag to the corresponding product party.

[0164] In the disclosed embodiment, the marketing party may also determine a user preference tag based on the attribute information of the marketing task, and the user preference tag may be a feature tag that meets or exceeds expectations in terms of the corresponding attribute aspects of the marketing task. The user preference tag may be provided to the corresponding product party through encrypted transmission, so that the product party can discover and mine derived features based on the user preference tag, thereby further enriching the intention tags for user group screening.

[0165] Step H2: The product party calculates derived features of the local data based on the user preference tags, and transmits the obtained derived features to the marketing party in encrypted form.

[0166] In the embodiment of the present disclosure, the product side can extract similar or close derivative features based on the user preference tags on the local data, thereby obtaining an initial set of derivative features. On this basis, the derivative features can be correlated to obtain the derivative features that are encrypted and transmitted to the marketing side. The initial set includes {f1, f2, ..., f k} as an example, for each derived feature f j , the correlation with the target variable T can be calculated as shown in the following formula (19):

[0167]

[0168] Among them, Corr(f j ,T) represents the feature f j The correlation between the target variable T and Cov(f j ,T) represents the feature f j Covariance with the target variable T; σ fj Represents feature f j The standard deviation of T Represents the standard deviation of the target variable T. The target variable T can be an effect indicator, such as conversion rate, ROI, etc. Through the calculation of formula (18), the product party selects appropriate derivative features based on relevance and provides them to the marketing party, ensuring the security of local original data.

[0169] Step H3: The marketing party screens the derivative features provided by each product party for relevance, obtains key derivative features, and uses differential privacy to update the parameters of the user ranking model based on the key derivative features.

[0170] In the disclosed embodiment, the marketing party can obtain the derived features provided by the product party, and then further perform correlation screening on the derived features to obtain high-weight and stable key derived features. Specifically, the correlation between each derived feature and the target variable can be analyzed, and the correlation threshold α can be set. The execution logic of the key derived feature screening can be expressed as the following formula (20):

[0171] S={f j |Corr(f j ,T)≥α} (20)

[0172] Among them, S represents the screened key derivative features.

[0173] Furthermore, time series analysis can be introduced to evaluate the temporal stability of derived features so that they have the expected stability in multiple time periods. j ) is shown in the following formula (21):

[0174]

[0175] The screening condition based on time series analysis can be expressed as the following formula (22):

[0176] Stability(f j )≤β (22)

[0177] Among them, σ fj is the derived feature f j The standard deviation of fj is the derived feature f j β is the stability threshold. On this basis, the derived features whose temporal stability is less than the stability threshold are screened out to ensure temporal stability.

[0178] The key derivative features embedded in the user ranking model can correspond to the aforementioned description of parameter updating using differential privacy. To avoid repetition, it will not be repeated here.

[0179] In an exemplary embodiment of the present disclosure, based on the parameter update of the user ranking model, the marketing party can also test the user ranking model to verify its performance. Taking the bucket test as an example for comparative analysis, assuming that the user ranking model before the parameter update is M baseline , the user ranking model after parameter update is M new In the process of random allocation by bucketing, the potential user group is randomly allocated to two groups, one of which uses M baseline To screen the target customer group, one group uses M newThe target customer groups are screened and the same marketing tasks are performed on the two target customer groups respectively. The model performance is evaluated using the following formula (23):

[0180]

[0181] Among them, CR new Indicates M new The conversion rate of the target customer group; CR baseline Indicates M baseline The conversion rate of the target customer group; new Indicates M new The number of successful conversions of the target customer group; C baseline Indicates M baseline The number of successful conversions of the target customer group; T new Indicates M new The number of times the target customer group is delivered; T baseline Indicates M baseline The number of contacts with the target customer group.

[0182] Furthermore, a significance test can be performed based on the conversion rate using a statistical test, such as a t-test as shown in the following formula (24):

[0183]

[0184] Where SE is the standard error, which can be calculated using the following formula (25):

[0185]

[0186] The above-mentioned parameter updates such as feature weight adjustment, derivative feature discovery, and key feature screening can be performed continuously and periodically to monitor whether there are high-weight features or the processing of invalid and useless features, so that the user ranking model can be continuously optimized. In the training and updating process of the user ranking model, privacy computing technologies such as federated learning and differential privacy are used to ensure data security.

[0187] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on the user intention analysis of the completion status of the marketing task.

[0188] In the disclosed embodiment, the extraction of attribute information can be obtained based on the user intention analysis of the completion status of the marketing task. The completion status of the marketing task can be evaluated by whether the process of delivering to the user is completed as expected, such as whether the AI ​​outbound call is connected, or whether it is hung up before the AI ​​voice playback ends, etc., and further, it can also evaluate whether the first call of the AI ​​outbound call is completed, whether the supplementary call is completed when the first call is not completed, etc.; or, the completion status of the marketing task can be evaluated by whether the marketing task is suspended during the execution process, such as whether the marketing task is suspended during the execution process, the stage of the suspended task, the frequency of suspension, whether it is restarted, etc. On this basis, user intention analysis can be performed in combination with attribute information according to actual business needs. For example, if the first call of the AI ​​outbound call is not completed, it means that the user's intention priority for marketing delivery is biased, otherwise it means that the intention priority is accurate; or the marketing task is suspended in the early stage, the frequency of suspension is high, or it is not restarted, etc., indicating that the user sorting model has a deviation in the user's intention recognition, otherwise it means that the intention recognition is accurate. On this basis, the feature weights of the feature labels related to the corresponding user group can be adjusted, or the accurate user intention labels can be encrypted and transmitted to the participating parties for derivative feature mining.

[0189] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on the user intention analysis of the completion effect of the marketing task.

[0190] In the disclosed embodiment, the extraction of attribute information can be obtained based on the user intention analysis of the completion effect of the marketing task. The completion effect of the marketing task can be determined by statistical analysis of the effect indicators, such as statistical analysis of the fluctuation trend of ROI, statistical analysis of the conversion rate of the user group, and statistical analysis of the user rejection or complaint rate. On this basis, user intention analysis can be performed in combination with attribute information according to actual business needs, such as according to the growth, stability or decay trend of ROI. When ROI grows, it indicates that the user sorting model accurately recognizes the user intention, and there may be deviations when it is stable or decayed; the conversion rate of the user group is lower than expected, which may indicate that the user sorting model has a deviation in recognizing the user intention, otherwise it is accurate; when the user rejection or complaint rate increases, there may be unreasonable problems in the distribution of feature weights. On this basis, the feature weights of the feature labels related to the corresponding user group can be adjusted, or the accurate user intention labels can be encrypted and transmitted to the participating parties for derivative feature mining.

[0191] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on user intention analysis of channel comparison.

[0192] In the disclosed embodiment, the extraction of attribute information can be obtained by comparing marketing tasks on different channels and performing user intention analysis. Channel comparison of marketing tasks can be determined by comparing start-stop control and statistical analysis of effect indicators, such as statistical analysis of ROI, conversion rate, etc. on the user group reached by AI outbound calls, and statistical analysis of ROI, conversion rate, etc. on the user group reached by SMS; or, comparing the suspension ratio of marketing tasks delivered by AI outbound calls and the suspension ratio of marketing tasks delivered by SMS. On this basis, derivative feature expansion of the channel dimension can be performed, such as whether the user group with poor effect indicators on AI outbound calls can improve the performance of effect indicators on AI outbound calls, so that when the user ranking model performs user screening, the relevant feature labels can be associated and expanded or the weights adjusted.

[0193] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on a user intention analysis of product comparison.

[0194] In the disclosed embodiment, the extraction of attribute information can be obtained by performing user intention analysis based on the comparison of marketing tasks on different products. Product comparison of marketing tasks can be determined by comparing the start-stop control and statistical analysis of effect indicators, such as statistical analysis of the start-stop ratio of user groups selected by the user sorting model based on different product information in the execution of marketing tasks, ROI fluctuation trends, etc. On this basis, user intention analysis can be performed in combination with attribute information according to actual business needs. For example, if the ROI growth trend is good in the marketing task of product A and the ROI is low in the marketing task of product B, it may indicate that the potential customer group has a higher demand for product A, so feature labels can be supplemented or weights can be adjusted to characterize the demand intention related to product A. In practical applications, taking the credit field as an example, feature labels can be supplemented and weights adjusted based on users' preferences for product amounts, interest rates, brands, etc.

[0195] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on a user intention analysis combining a completion status and a completion effect of a marketing task.

[0196] In the disclosed embodiment, the extraction of attribute information can be obtained by conducting user intention analysis based on the completion status of the marketing task in combination with the completion effect. For example, the changes in the completion status in different time periods during the execution of the marketing task can be analyzed, such as the changes in the completion degree of the first call and the completion degree of the supplementary call. Then, the potential trend of the change in the completion effect can be statistically analyzed based on the changes in the completion status. Taking the changes in the completion degree of the first call as an example, when the completion degree of the first call increases, the ROI increases significantly, and when the completion degree of the supplementary call increases, the ROI increases less or decreases. Therefore, it is determined that the user intention-related features of the first call completion have a higher priority, and the feature weights are adjusted, or the relevant user intention labels are encrypted and transmitted to the participants for derivative feature mining; the user intention-related features of the supplementary call completion can have their feature weights reduced or removed. On this basis, through the user intention analysis that combines the completion status of the marketing task with the completion effect, the timeliness of the features can be maintained, and some inefficient and outdated features in the user sorting model can be identified and updated in a timely manner.

[0197] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on a user intention analysis comparing the completion status of the marketing task with the telecommunication channel.

[0198] In the disclosed embodiment, the telecommunication channel is an entity that provides communication and network services to users. The potential user group may include sub-user groups served by multiple different telecommunication channels. Each telecommunication channel can reach the corresponding sub-user group through different channels based on the service type. For example, telecommunication channel A provides communication and network services to sub-user group 1, and can reach the corresponding sub-user group 1 through AI outbound calls, text messages or other channels; telecommunication channel B provides communication and network services to sub-user group 2, and can reach the corresponding sub-user group 2 through AI outbound calls, text messages or other channels. Since different telecommunication channels serve different user groups, marketing tasks may also be affected by telecommunication channels. When the performance of marketing tasks on different telecommunication channels is significantly different, it may mean that the feature labels of the user ranking model do not cover the sub-user groups served by the telecommunication channels.

[0199] In the disclosed embodiment, attribute information can be used to determine user intention analysis based on the comparison of the completion status of marketing tasks in different telecom channels. If the AI ​​outbound call completion rate of telecom channel A in the delivery of marketing tasks is significantly lower than the overall average, it means that there is a deviation in the intention recognition of the sub-user group served by telecom channel A, and feature migration or feature weight adjustment is required.

[0200] In an exemplary embodiment of the present disclosure, the attribute information is obtained based on a user intention analysis comparing the completion effect of the marketing task with the telecommunication channel.

[0201] In the disclosed embodiment, attribute information can be used to determine user intention analysis based on the comparison of the completion effects of marketing tasks in different telecom channels. If the ROI of telecom channel B in the delivery of marketing tasks is significantly lower than the overall average, it means that there is a deviation in the intention identification of the sub-user group served by telecom channel B, and feature migration or feature weight adjustment is required.

[0202] In the process of automated marketing in the disclosed embodiments, user ranking model training and parameter updating as well as marketing task start and stop control are core links that are interrelated and influence each other. The user ranking model provides a decision-making basis for the start and stop control of marketing tasks by predicting customer intentions, and then the start and stop status of the marketing tasks and execution feedback provide data support for the parameter update of the user ranking model, thereby presenting a data-driven closed loop between the parameter update of the user ranking model and the start and stop control of the marketing tasks.

[0203] The present disclosure provides an automated decision-making method for marketing tasks, an automated decision-making system for marketing tasks, an electronic device, and a computer-readable medium. In the process of executing a marketing task, the method adopts a start-stop control scheme to determine the start-stop status of the marketing task; wherein the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel; the marketing task corresponds to a first user group, the first user group is determined by sorting based on a user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to update parameters based on the attribute information of the marketing task, and the attribute information is extracted after the execution of the marketing task is completed. In this method, the task start-stop decision of automated marketing and the parameter update of the user sorting model can make full use of the feedback information of the execution of the marketing task, effectively improve the response efficiency of automated marketing and the optimization ability of model update iteration, and comprehensively improve the accuracy of user screening and identification, as well as the hit rate and conversion rate of automated marketing; moreover, the user sorting model also strengthens the protection of user privacy data by adopting differential privacy in parameter update, thereby improving information security.

[0204] Figure 5 A structural block diagram of an automated decision-making system 500 for marketing tasks is also provided. The system 500 is used to manage and optimize the execution of marketing tasks, including marketing task allocation, delivery progress tracking, and delivery effect evaluation. Figure 5As shown, the system 500 may include: a marketing task decision module 501, which is used to determine the start and stop status of the marketing task by using a start and stop control scheme during the execution of the marketing task, and to extract attribute information after the execution of the marketing task is completed; the start and stop control scheme includes at least one of a start and stop scheme based on an effect indicator threshold and a start and stop scheme based on a channel; a model prediction and update module 502, which is used to determine the first user group corresponding to the marketing task by using a preset user screening strategy through a user sorting model, and to update the parameters of the user sorting model based on the attribute information of the marketing task by using differential privacy.

[0205] In an exemplary embodiment of the present disclosure, the start-stop control scheme is a start-stop scheme based on an effect indicator threshold. The marketing task decision module 501 specifically obtains real-time effect indicators during the execution of the marketing task; the real-time effect indicators are used to evaluate the effect of the marketing task on the second user group, and the second user group includes at least part of the user group that has been reached by the marketing task in the first user group through at least one channel; the effect indicator threshold corresponding to the marketing task is determined; when the real-time effect indicator is greater than or equal to the effect indicator threshold, the marketing task is determined to be in an on state; when the real-time effect indicator is less than the effect indicator threshold, the marketing task is determined to be in a paused state.

[0206] In an exemplary embodiment of the present disclosure, the marketing task decision module 501 is specifically used to determine the business window period; when the real-time time is within the business window period, the marketing task is determined to be in a paused state; when the real-time time is outside the business window period, the marketing task is determined to be in an enabled state, and real-time effect indicators during the execution of the marketing task are obtained.

[0207] In an exemplary embodiment of the present disclosure, the marketing task decision module 501 is specifically used to determine the second user group; when the number of users in the second user group is less than the initial user threshold, the marketing task is determined to be in an on state; when the number of users in the second user group is greater than or equal to the initial user threshold, real-time effect indicators during the execution of the marketing task are obtained.

[0208] In an exemplary embodiment of the present disclosure, the marketing task decision module 501 is specifically used to obtain the first historical effect indicator of the historical marketing tasks that have been executed within the first historical period, and the third user group corresponding to the historical marketing tasks is determined by sorting based on the user sorting model using a preset user screening strategy, and the historical effect indicator threshold corresponding to the historical marketing tasks; the historical mean indicator is determined according to the first historical period and the first historical effect indicator; when the historical mean indicator is greater than zero, the historical effect indicator threshold is lowered as the effect indicator threshold; when the historical mean indicator is less than or equal to zero, the historical effect indicator threshold is increased as the effect indicator threshold.

[0209] In an exemplary embodiment of the present disclosure, the marketing task decision module 501 is specifically used to obtain the second historical effect indicator for each time period within the second historical cycle during the execution of the marketing task; determine the historical trend indicator based on the second historical cycle and the second historical effect indicator, and use the historical trend indicator as the effect indicator threshold.

[0210] In an exemplary embodiment of the present disclosure, a marketing task corresponds to a profit threshold, and a marketing task decision module 501 is specifically used to determine the real-time total cost and the real-time total benefit during the execution of the marketing task; determine the real-time total profit based on the real-time total cost and the real-time total benefit; when the real-time effect index is less than the effect index threshold, and the real-time total profit is less than or equal to the profit threshold, determine the marketing task to be in a paused state.

[0211] In an exemplary embodiment of the present disclosure, the start-stop control scheme is a channel-based start-stop scheme. During the execution of the marketing task, the marketing task decision module 501 is specifically used to determine the channel effect index of each marketing task under the channel within the third historical period during the execution of the marketing task; determine the first channel mean index corresponding to the channel based on the channel effect index; when the channel mean index is less than the mean index threshold, determine the second channel mean index of all marketing tasks under the channel within the fourth historical period, and the total mean index of all marketing tasks, and the fourth historical period is greater than the third historical period; when it is determined that there is a channel abnormality based on the deviation value of the second channel mean index relative to the total mean index, all marketing tasks in the execution process under the channel are determined to be in a suspended state.

[0212] In an exemplary embodiment of the present disclosure, the marketing task decision module 501 is specifically configured to determine the start and stop status of the marketing task by adopting a start and stop control scheme during the execution of the marketing task in an automated decision mode.

[0213] In an exemplary embodiment of the present disclosure, the user ranking model is obtained by the product party and the marketing party jointly participating in training in federated learning. The model prediction and update module 502 is also used by the product party to perform local model training locally and encrypt and transmit the local model parameters to the marketing party; the marketing party aggregates the local model parameters provided by each product party to obtain the user ranking model.

[0214] In an exemplary embodiment of the present disclosure, the model prediction and update module 502 is specifically used by the marketing party to determine the feature weights to be adjusted based on the attribute information of the marketing task, and to perform parameter update on the feature weights to be adjusted using differential privacy.

[0215] In an exemplary embodiment of the present disclosure, the model prediction and update module 502 is specifically used by the marketing party to determine the user preference label based on the attribute information of the marketing task, and to encrypt and transmit the user preference label to the corresponding product party; the product party calculates the derivative features of the local data based on the user preference label, and encrypts and transmits the obtained derivative features to the marketing party; the marketing party performs correlation screening on the derived features provided by each product party, obtains key derived features, and uses differential privacy to update the parameters of the user sorting model based on the key derived features.

[0216] In an exemplary embodiment of the present disclosure, the attribute information is obtained by analyzing at least one of the following user intentions:

[0217] User intention analysis based on the completion status of marketing tasks;

[0218] User intention analysis based on the completion effect of marketing tasks;

[0219] User intention analysis based on channel comparison;

[0220] User intention analysis based on product comparison;

[0221] User intention analysis based on the completion status and completion effect of marketing tasks;

[0222] User intention analysis based on the completion status of marketing tasks and comparison with telecommunications channels;

[0223] User intention analysis based on the completion effect of marketing tasks and comparison with telecommunications channels.

[0224] The present disclosure provides an automated decision-making system for marketing tasks. During the execution of the marketing tasks, the system adopts a start-stop control scheme to determine the start-stop status of the marketing tasks; wherein the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel; the marketing task corresponds to a first user group, and the first user group is determined by sorting based on a user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to perform parameter updates based on attribute information of the marketing task, and the attribute information is extracted after the execution of the marketing task is completed. The task start-stop decision of automated marketing and the parameter update of the user sorting model in the system can make full use of the feedback information of the execution of the marketing tasks, effectively improve the response efficiency of automated marketing and the optimization capability of the model update iteration, and comprehensively improve the accuracy of user screening and identification, as well as the hit rate and conversion rate of automated marketing; moreover, the user sorting model also strengthens the protection of user privacy data by adopting differential privacy in parameter updates, thereby improving information security.

[0225] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0226] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0227] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0228] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0229] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0230] Refer to the following Figure 6 The electronic device 600 according to this embodiment of the present disclosure is described. Figure 6 The electronic device 600 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0231] like Figure 6As shown, the electronic device 600 is in the form of a general computing device. The components of the electronic device 600 may include but are not limited to: at least one processing unit 610, at least one storage unit 620, and a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610).

[0232] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 executes the steps described in the above “Exemplary Method” section of this specification according to various exemplary embodiments of the present disclosure.

[0233] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0234] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0235] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0236] The electronic device 600 may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed through a display unit 640 and an input / output (I / O) interface 650 connected to the display unit 640. In addition, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 660. As shown, the network adapter 660 communicates with other modules of the electronic device 600 through a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0237] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0238] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above "Exemplary Method" section of the present specification.

[0239] In the embodiments of the present disclosure, a program product for implementing the above method is also provided, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0240] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0241] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0242] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0243] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0244] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0245] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. An automated decision-making method for marketing tasks, characterized in that: The method comprises: During the execution of the marketing task, a start-stop control scheme is adopted to determine the start-stop status of the marketing task; the start-stop control scheme includes at least one of a start-stop scheme based on an effect indicator threshold and a start-stop scheme based on a channel; Among them, the marketing task corresponds to a first user group, the first user group is determined by sorting based on a user sorting model using a preset user screening strategy, and the user sorting model uses differential privacy to update parameters based on attribute information of the marketing task, and the attribute information is extracted after the marketing task is executed.

2. The method according to claim 1, characterized in that The start-stop control scheme is a start-stop scheme based on an effect indicator threshold. During the execution of the marketing task, the start-stop control scheme is used to determine the start-stop status of the marketing task, including: Acquire a real-time effect indicator during the execution of the marketing task; the real-time effect indicator is used to evaluate the effect of the marketing task on the second user group, the second user group including at least part of the user group that has been reached by the marketing task through at least one channel in the first user group; Determine the effect indicator threshold corresponding to the marketing task; When the real-time effect index is greater than or equal to the effect index threshold, determining the marketing task to be in an enabled state; When the real-time effect indicator is less than the effect indicator threshold, the marketing task is determined to be in a paused state.

3. The method according to claim 2, characterized in that The real-time effect indicators obtained during the execution of the marketing task include: Determine the business window period; When the real time is within the business window period, the marketing task is determined to be in a suspended state; When the real-time time is outside the business window period, the marketing task is determined to be in an open state, and the real-time effect indicator during the execution of the marketing task is obtained.

4. The method according to claim 2, characterized in that: The real-time effect indicators obtained during the execution of the marketing task include: determining a second user group; When the number of users in the second user group is less than the initial user threshold, determining the marketing task to be in an on state; When the number of users in the second user group is greater than or equal to the initial user threshold, the real-time effect indicator during the execution of the marketing task is obtained.

5. The method according to claim 2, characterized in that: Determining the effect indicator threshold corresponding to the marketing task includes: Obtaining a first historical effect indicator of a completed historical marketing task in a first historical period, wherein a third user group corresponding to the historical marketing task is determined by sorting based on a user sorting model using the preset user screening strategy, and the historical effect indicator threshold corresponding to the historical marketing task; Determine a historical average indicator according to the first historical period and the first historical effect indicator; When the historical mean indicator is greater than zero, lowering the historical effect indicator threshold as the effect indicator threshold; When the historical mean indicator is less than or equal to zero, the historical effect indicator threshold is increased as the effect indicator threshold.

6. The method according to claim 2, characterized in that Determining the effect indicator threshold corresponding to the marketing task includes: Obtaining a second historical effect indicator for each time period within a second historical period during the execution of the marketing task; A historical trend indicator is determined according to the second historical period and the second historical effect indicator, and the historical trend indicator is used as the effect indicator threshold.

7. The method according to claim 6, characterized in that The marketing task corresponds to a profit threshold, and when the real-time effect index is less than the effect index threshold, the marketing task is determined to be in a suspended state, including: Determine the real-time total cost and the real-time total benefit during the execution of the marketing task; Determine the real-time total profit based on the real-time total cost and the real-time total benefit; When the real-time effect index is less than the effect index threshold, and the real-time total profit is less than or equal to the profit threshold, the marketing task is determined to be in a paused state.

8. The method according to claim 1, characterized in that The start-stop control scheme is a channel-based start-stop scheme. During the execution of the marketing task, the start-stop control scheme is used to determine the start-stop status of the marketing task, including: During the execution of the marketing tasks, determining the channel effect index of each of the marketing tasks under the channel in the third historical period; Determine a first channel mean index corresponding to the channel based on the channel effect index; In the case where the channel mean index is less than the mean index threshold, determining the second channel mean index of all the marketing tasks under the channel in a fourth historical period, and the total mean index of all marketing tasks, the fourth historical period being greater than the third historical period; When it is determined that there is a channel abnormality based on the deviation value of the second channel mean index relative to the total mean index, all the marketing tasks in the execution process under the channel are determined to be in a suspended state.

9. The method according to claim 1, characterized in that: During the execution of the marketing task, the start-stop control scheme is used to determine the start-stop state of the marketing task, including: In the automated decision-making mode, during the execution of the marketing task, a start-stop control scheme is used to determine the start-stop status of the marketing task.

10. The method according to claim 1, characterized in that The user ranking model is obtained by the product party and the marketing party jointly participating in the training in federated learning. The training steps of the user ranking model are as follows: The product party performs local model training locally and transmits the local model parameters to the marketing party in encrypted form; The marketing party aggregates the local model parameters provided by each of the product parties to obtain the user ranking model.

11. The method according to claim 10, characterized in that The user ranking model uses the following steps to update parameters: The marketing party determines the feature weights to be adjusted based on the attribute information of the marketing task, and uses differential privacy to perform parameter update on the feature weights to be adjusted.

12. The method according to claim 10, characterized in that The user ranking model uses the following steps to update parameters: The marketing party determines a user preference tag based on the attribute information of the marketing task, and encrypts and transmits the user preference tag to the corresponding product party; The product party calculates derived features of the local data based on the user preference tags, and transmits the obtained derived features to the marketing party in encrypted form; The marketing party performs correlation screening on the derived features provided by each of the product parties to obtain key derived features, and uses differential privacy to update parameters of the user ranking model based on the key derived features.

13. The method according to claim 11 or 12, characterized in that: The attribute information is obtained by analyzing at least one of the following user intentions: User intention analysis based on the completion status of the marketing task; User intention analysis based on the completion effect of the marketing task; User intention analysis based on channel comparison; User intention analysis based on product comparison; User intention analysis based on the completion status and completion effect of the marketing task; User intention analysis based on the completion status of the marketing task and the comparison with the telecommunications channel; Analysis of user intention based on the completion effect of the marketing task and comparison with the telecommunications channel.

14. An automated decision-making system for marketing tasks, characterized in that: The system comprises: A marketing task decision module, used to determine the start and stop status of the marketing task by using a start and stop control scheme during the execution of the marketing task, and to extract attribute information after the execution of the marketing task is completed; the start and stop control scheme includes at least one of a start and stop scheme based on an effect indicator threshold and a start and stop scheme based on a channel; The model prediction and update module is used to determine the first user group corresponding to the marketing task by using a preset user screening strategy through a user sorting model, and to update the parameters of the user sorting model based on the attribute information of the marketing task by using differential privacy.

15. An electronic device, characterized in that: include: Processor; and a memory for storing a computer program for the processor; The processor is configured to execute the automated decision-making method for the marketing task according to any one of claims 1 to 13 by executing a computer program.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the automated decision-making method for marketing tasks as claimed in any one of claims 1 to 13 is implemented.

Images