Object determination method and device, electronic equipment, computer readable storage medium and computer program product

Abstract

The invention provides an object determination method and device, electronic equipment, a computer program product and a computer readable storage medium. The method comprises the steps that based on a first coefficient of first recommendation information, a second coefficient of the first recommendation information is determined from a coefficient set, and the second coefficient is a coefficient meeting a first preset condition with the first coefficient in the coefficient set; based on the second coefficient and the object data of the first object, first conversion data is determined, and the first conversion data is statistical data of the first object implementing a preset behavior according to the first recommendation information; based on the first conversion data, the first coefficient, the second coefficient and a pre-constructed gain fitting function, determining a first conversion gain of the first object when the first recommendation information is received; and determining a second object meeting a second preset condition from the first object based on the first conversion gain of the first object. According to the object determination method and device, the adaptive capacity of the object determination method can be improved, and the object determination efficiency and accuracy are improved.

Description

Technical Field

[0001] The present application relates to computer technology, and in particular to an object determination method, device, electronic device, computer-readable storage medium, and computer program product. Background Art

[0002] In recommendation scenarios, in order to fully stimulate the latent needs of customer groups and convert potential customers into actual consumers, it is necessary to take targeted measures for specific customer groups through intervention behaviors, such as developing promotional campaigns, personalized recommendations, and targeted advertising. In related technologies, various data are usually analyzed by building gain models to formulate the required intervention plans. However, market recommendation strategies in real scenarios are dynamically adjusted, and various data deviations often exist in the actual data collection, processing, and model training processes, resulting in a large gap between the model's predicted values ​​and the actual market feedback values. Summary of the Invention

[0003] The embodiments of the present application provide an object determination method, apparatus, computer-readable storage medium, and computer program product, which can improve the adaptability of the object determination method and improve the efficiency and accuracy of object determination.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] This embodiment of the present application provides an object determination method, the method comprising:

[0006] Determining, based on the first coefficient of the first recommendation information, a second coefficient of the first recommendation information from a coefficient set, where the second coefficient is a coefficient in the coefficient set that satisfies a first preset condition with the first coefficient;

[0007] determining first conversion data based on the second coefficient and the object data of the first object, where the first conversion data is statistical data of the first object performing a preset behavior according to the first recommendation information;

[0008] determining, based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function, a first conversion gain of the first subject after receiving the first recommendation information;

[0009] Based on the first conversion gain of the first object, a second object that meets a second preset condition is determined from the first objects.

[0010] An embodiment of the present application provides an object determination device, including:

[0011] a determining module, configured to determine, based on the first coefficient of the first recommendation information, a second coefficient of the first recommendation information from a coefficient set, where the second coefficient is a value in the coefficient set that satisfies a first preset condition with the first coefficient;

[0012] The determining module is further configured to determine first conversion data based on the second coefficient and the object data of the first object, where the first conversion data is statistical data of the first object performing a preset behavior according to the first recommendation information;

[0013] The determining module is further configured to determine a first conversion gain of the first subject upon receiving the first recommendation information based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function;

[0014] The determining module is further configured to determine, based on the first conversion gain of the first object, a second object that meets a second preset condition from the first objects.

[0015] An embodiment of the present application provides an electronic device, comprising:

[0016] a memory for storing computer-executable instructions or computer programs;

[0017] The processor is used to implement the object determination method provided in the embodiment of the present application when executing the computer-executable instructions or computer program stored in the memory.

[0018] An embodiment of the present application provides a computer-readable storage medium storing a computer program or computer-executable instructions for implementing the object determination method provided in the embodiment of the present application when executed by a processor.

[0019] An embodiment of the present application provides a computer program product, including a computer program or computer-executable instructions. When the computer program or computer-executable instructions are executed by a processor, the object determination method provided in the embodiment of the present application is implemented.

[0020] The embodiments of the present application have the following beneficial effects:

[0021] In an embodiment of the present application, based on the first coefficient of the first recommendation information, a second coefficient of the first recommendation information is determined from a coefficient set; a prediction process is performed based on the second coefficient and the object data of the first object to determine the first conversion data. In this way, even if the first recommendation information is new, a second coefficient that satisfies a first preset condition with the first coefficient can be determined from the coefficient set, and the first conversion data can be obtained by performing a prediction process based on the second coefficient, thereby improving the efficiency and reliability of determining the first conversion data of the first object. Then, based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function, the first conversion gain of the first object after receiving the first recommendation information is determined. The gain fitting function can represent the fitting relationship between the coefficient and the conversion gain. The gain fitting function can improve the accuracy of determining the first conversion gain, making the first conversion gain more consistent with the actual gain. Finally, based on the first conversion gain of each first object, a second object that meets the second preset condition is determined from the first object. This ensures that the determined second object is suitable for recommending the first recommendation information, thereby improving the efficiency and accuracy of determining the second object. Therefore, the object determination method provided in this application can adapt to various new first recommendation information, improve the adaptability of the object determination method in different types of information recommendation scenarios, and ensure the efficiency and accuracy of information recommendation, thereby improving the conversion rate of information recommendation and user retention rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 1 is a schematic diagram of the architecture of the information recommendation system 100 provided in an embodiment of the present application;

[0023] Figure 2 2 is a schematic diagram of the structure of the server 200 provided in an embodiment of the present application;

[0024] Figure 3A Schematic diagram of the process of determining an object provided in an embodiment of the present application;

[0025] Figure 3B is a schematic diagram of a process for determining a first conversion gain provided in an embodiment of the present application;

[0026] Figure 3C is another schematic diagram of a process for determining a first conversion gain provided in an embodiment of the present application;

[0027] Figure 3D is a schematic diagram of a process for determining a second object provided in an embodiment of the present application;

[0028] Figure 3E This is a schematic diagram of a process for constructing a coefficient set provided in an embodiment of the present application;

[0029] Figure 3FSchematic diagram of the process of constructing a gain fitting function according to an embodiment of the present application;

[0030] Figure 3G This is a flow chart of constructing a conversion correction function according to an embodiment of the present application;

[0031] Figure 4 This is a flow chart of model training and function construction provided by the embodiment of the present application;

[0032] Figure 5 Schematic diagram of the flow of the coupon recommendation method provided in the embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0034] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0035] In the following description, the terms "first\second\third\fourth" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third\fourth" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0036] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0037] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meanings as those commonly understood by those skilled in the art. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0038] The relevant data collection and processing in the embodiments of this application should be strictly in accordance with the requirements of relevant laws and regulations when applied in examples, and the informed consent or separate consent of the personal information subject should be obtained. Subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0039] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0040] 1) Intervention: A series of proactive measures taken by a company or recommender to influence consumer behavior and improve the effectiveness of recommendations. These measures can be strategic or real-time, and are designed to guide consumers along the purchase path;

[0041] 2) Homogeneous customer groups: refers to target customer groups that are similar or identical in one or more specific attributes;

[0042] 3) Organic Initiation Rate: A metric used in recommendation and product analytics to describe how often users spontaneously initiate a behavior or activity without external incentives or intervention.

[0043] 4) Natural consumption amount: the amount of consumption made by consumers based on their own needs and preferences without the influence of external incentives or promotional activities, that is, the amount of consumption by users without intervention;

[0044] 5) Prediction correction: refers to the process of adjusting or correcting the prediction results generated by the model to improve the accuracy and reliability of the prediction;

[0045] 6) Conversion data: refers to data that converts potential customers or traffic into actual sales or business results, typically including user purchase rate data for products, purchase amount data, and coupon redemption rate data;

[0046] 7) Conversion gain: refers to the additional revenue or profit growth achieved through interventions such as referral activities, typically including purchase rate gain, consumption amount gain, and write-off rate gain;

[0047] 8) Verification restriction coefficient: refers to the restriction or control ratio of the verification process under specific conditions, for example, the difficulty of using recommendation tools such as coupons, discounts or points in recommendation activities;

[0048] 9) Gain fitting function: a function that characterizes the fitting relationship between the write-off restriction coefficient and the conversion gain, used to determine the corresponding conversion gain based on the write-off restriction coefficient;

[0049] 10) Conversion correction function: A function that adjusts and corrects the conversion data predicted by the model based on actual conditions, which can minimize the gap between the conversion data predicted by the model and the actual conversion data.

[0050] Among various intervention methods, sending coupons to users is the most widespread and common one. When sending coupons to users, in order to efficiently and accurately identify customer groups with high growth potential, a common approach in related technologies is to build an uplift model. This model analyzes various data factors to identify potential growth opportunities. However, these methods in related technologies have the following problems:

[0051] 1. In actual recommendation activities, it is necessary to maximize the return on investment (ROI) within a limited recommendation cost range. Using only the gain model cannot flexibly and comprehensively consider the allocation of various recommendation resources.

[0052] 2. The types of coupons will continue to iterate with the dynamic adjustment of market recommendation strategies. Specifically, there will be various types of coupons with different usage thresholds and discount values. This situation will bring huge challenges to the gain model that has been deployed online. This is because the gain model is trained on the basis of historical observation samples, and its learning and judgment capabilities are mainly based on data that has been "seen" in the past. Therefore, the gain model can only make relatively accurate judgments and evaluations for familiar coupons that already exist in the training data. When a brand new coupon appears, in order for the model to adapt to the new promotion strategy and still maintain effective predictive capabilities, it is necessary to retrain the model and redeploy it to the online recommendation system. This process is not only extremely time-consuming, but also comes with high costs, which will have a significant impact on recommendation efficiency and resource investment;

[0053] 3. In the actual process of data collection, processing and model training, there are various inevitable deviations, such as sample selection deviation, data measurement deviation, and deviation between model assumptions and the actual market, which leads to a large gap between the model's predicted value and the actual market feedback value.

[0054] The embodiments of the present application provide an object determination method, apparatus, device, computer-readable storage medium, and computer program product, which can accurately determine the first conversion gain of the first object after receiving the first recommendation information through the first conversion data, the first coefficient, the second coefficient, and the pre-constructed gain fitting function, thereby flexibly and comprehensively allocating the recommendation resources through the first conversion gain, thereby improving the rationality of resource allocation. Moreover, even if the first recommendation information is new recommendation information, a second coefficient that meets the first preset condition with the first coefficient can be determined from the coefficient set, and the first conversion data can be obtained by predicting and processing based on the second coefficient, thereby improving the adaptability of the object determination method in different types of information recommendation scenarios, avoiding the process of repeated model training, and reducing the implementation cost of the solution. In addition, since the gain fitting function can characterize the fitting relationship between the coefficient and the conversion gain, the data regularity between the second coefficient and the first conversion gain can be accurately captured through the gain fitting function, thereby improving the accuracy of determining the first conversion gain and making the first conversion gain more consistent with the actual gain. The following describes exemplary applications of the electronic devices provided in the embodiments of the present application. The electronic devices provided in the embodiments of the present application can be implemented as various types of terminals, such as laptops, tablet computers, desktop computers, set-top boxes, smartphones, smart speakers, smart watches, smart TVs, and in-vehicle terminals. They can also be implemented as servers. The following describes exemplary applications of the device implemented as a server.

[0055] See also Figure 1 , Figure 1 This is a schematic diagram of the architecture of the information recommendation system 100 provided in an embodiment of the present application. The terminal 400 is connected to the server 200 via the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two. When it is necessary to push the first recommendation information, in order to improve the efficiency and accuracy of object determination, first, based on the first coefficient of the first recommendation information, the server 200 determines the second coefficient of the first recommendation information from the coefficient set, and determines the first conversion data based on the second coefficient and the object data of the first object; then, based on the first conversion data, the first coefficient, the second coefficient and the pre-constructed gain fitting function, the first conversion gain of the first object after receiving the first recommendation information is determined; thereafter, based on the first conversion gain of each first object, the second object that meets the second preset condition is determined from the first object; finally, the server 200 sends the first recommendation information to the terminal 400 corresponding to the second object via the network 300, thereby pushing the first recommendation information to the second object.

[0056] For example, the electronic device that determines the object is the server mentioned above, see Figure 2 , Figure 2 is a structural diagram of the server 200 provided in an embodiment of the present application, Figure 2The server 200 shown includes: at least one processor 210, a memory 230 and at least one network interface 220. The various components in the server 200 are coupled together via a bus system 240. It is understood that the bus system 240 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 240 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 240 is not described in detail. Figure 2 Various buses are labeled as bus system 240 .

[0057] The processor 210 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0058] The memory 230 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, a hard drive, an optical drive, etc. The memory 230 may optionally include one or more storage devices that are physically remote from the processor 210.

[0059] The memory 230 includes volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 230 described in the embodiments of the present application is intended to include any suitable type of memory.

[0060] In some embodiments, memory 230 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0061] Operating system 231, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0062] The network communication module 232 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 220 . Exemplary network interfaces 220 include Bluetooth, Wireless Authentication (WiFi), and Universal Serial Bus (USB).

[0063] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 2 The object determination device 233 stored in the memory 230 is shown. This device can be software in the form of a program or plug-in, and includes at least one software module: a determination module 2331. Determination module 2331 is logical and can be arbitrarily combined with other modules or further separated depending on the functionality implemented. The functions of determination module 2331 will be described below.

[0064] In other embodiments, the apparatus provided in the embodiments of the present application may be implemented in hardware. As an example, the apparatus provided in the embodiments of the present application may be a processor in the form of a hardware decoding processor, which is programmed to execute the object determination method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor may be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0065] The object determination method provided in the embodiment of the present application will be described in conjunction with the exemplary application and implementation of the terminal provided in the embodiment of the present application.

[0066] The following describes the object determination method provided by the embodiment of the present application. As mentioned above, the electronic device that implements the object determination method of the embodiment of the present application can be a terminal, a server, or a combination of the two. Therefore, the execution entity of each step will not be repeated below.

[0067] It should be noted that in the following examples of object determination, the first recommendation information is a coupon in a product recommendation scenario. Based on their understanding of the following, those skilled in the art can apply the object determination method provided in the embodiment of the present application to other types of object determination processing. For example, taking a restaurant reservation scenario as an example, the first recommendation information may be a voucher. Through the object determination method provided in the embodiment of the present application, the recipients of the vouchers can be accurately determined. These recipients are users who have a higher probability of receiving vouchers for consumption, which can stimulate users' desire to consume and increase restaurant revenue. Taking an event registration scenario as an example, the first recommendation information may be an event ticket. Whether it is a popular concert, an industry summit, or an art exhibition, event tickets are essential credentials for users to participate in the event. Similarly, through the object determination method provided in the embodiment of the present application, the recipients of the event tickets can be accurately predicted, which can avoid wasting priority admission qualifications on invalid user groups and increase event attendance rates. Taking the service scenario as an example, the first recommendation information may be trial rights. Taking the education and training scenario as an example, the first recommendation information may be a redemption code. The object determination method provided in the embodiment of this application can enable trial rights and training courses to accurately reach target users, and concentrate limited resources on the group that is most likely to generate consumption behavior, thereby making the recommendation activities more targeted and improving resource utilization and conversion rate.

[0068] See also Figure 3A , Figure 3A FIG3 is a flow chart of the object determination method provided in an embodiment of the present application, which will be described in conjunction with the steps shown in FIG3 .

[0069] In step 101 , based on a first coefficient of first recommendation information, a second coefficient of the first recommendation information is determined from a coefficient set.

[0070] Here, the first coefficient is a redemption restriction coefficient, which is used to characterize the evaluation value of the conditions that need to be met when using the first recommended information, and can reflect the difficulty of using the first recommended information. For example, when the first recommended information is a new type of coupon that needs to be issued to the user, the first coefficient is used to characterize the difficulty of redeeming the coupon, which is usually related to the threshold for using the coupon. The higher the first coefficient, the more difficult it is to redeem the coupon, and the higher the threshold for using the coupon; the lower the first coefficient, the easier it is to redeem the coupon, and the lower the threshold for using the coupon. Among them, the first coefficient can be determined manually, and the relevant technical personnel will determine the corresponding first coefficient based on the recommendation experience and the threshold for using the coupon; or, the number of conditions that need to be met when using the first recommended information can be set as the first coefficient, or, the first coefficient can be set according to the difficulty of the conditions that need to be met when using the first recommended information. In other service scenarios, the first recommended information may also be an electronic ticket, such as a movie ticket, concert ticket, or air ticket. In this case, the first coefficient may be determined based on the place of use and the validity period of the electronic ticket. The more convenient the transportation to the place of use and the longer the validity period, the lower the first coefficient. Alternatively, the first recommended information may also be trial benefits, such as a free trial period or number of trials provided by certain designated products or services. In this case, the first coefficient may be determined based on the type of product or service. The wider the audience of the product or service type, the lower the first coefficient. For example, the first recommended information may also be other information that can be redeemed, such as membership cards, event tickets, redemption codes, and so on.

[0071] The coefficient set includes coefficients corresponding to multiple existing recommendation information. The second coefficient is a coefficient in the coefficient set that satisfies a first preset condition with the first coefficient. The first preset condition may be that the difference between the coefficients is minimal. Therefore, the second coefficient is the coefficient in the coefficient set that is closest to the first coefficient of the first recommendation information. The second coefficient may be the same as or different from the first coefficient.

[0072] In some embodiments, step 101 may be implemented by the following process:

[0073] For each third coefficient in the coefficient set, a coefficient difference between the third coefficient and the first coefficient is determined; and the third coefficient whose coefficient difference satisfies a first preset condition is determined as the second coefficient.

[0074] Here, the coefficient set includes multiple third coefficients, each corresponding to an existing recommendation. The first pre-set condition typically involves minimizing the coefficient difference. To determine the coefficient closest to the first coefficient in the coefficient set, the coefficient difference between each third coefficient and the first coefficient is determined. The third coefficient with the smallest absolute coefficient difference is then determined as the second coefficient. This ensures that the second coefficient is the closest to the first coefficient in the coefficient set, thereby ensuring the accuracy and efficiency of the conversion data predicted for the first recommendation using the second coefficient.

[0075] In step 102 , first conversion data is determined based on the second coefficient and the object data of the first object.

[0076] Here, for each first object, the first conversion data is determined based on the second coefficient and the object data of the first object. The first object refers to the candidate user to whom the first recommendation information is initially intended to be pushed, and it is necessary to screen and select to determine the user to whom the first recommendation information will ultimately be pushed. Object data refers to various information collected during the interaction with the user, which can be used to analyze user behavior, optimize recommendation methods, and enhance user experience, thereby helping to make accurate information recommendations to users. For each first object, the first conversion data corresponding to the first object needs to be determined by the second coefficient and the object data of the first object. The first conversion data is used to describe the conversion data that the first object will generate after receiving the first recommendation information, including at least the purchase rate data of the first object for the product, the consumption amount data generated by the purchase, and the redemption rate data for the first recommendation information (that is, the probability of using the received new coupon), etc.

[0077] In some embodiments, the first conversion data can be determined through a trained first prediction model, including: performing prediction processing based on the second coefficient and the object data of the first object through the trained first prediction model to obtain a conversion prediction result of the first object; and correcting the conversion prediction result of the first object based on a pre-constructed conversion correction function to obtain the first conversion data.

[0078] Here, the trained first prediction model includes at least one of a trained purchase prediction model, a trained consumption amount prediction model, and a trained write-off rate prediction model. The trained purchase prediction model can predict the purchase rate of the first subject after receiving the first recommendation information, i.e., the purchase rate prediction result, based on the second coefficient and the subject data of the first subject. The trained consumption amount prediction model can predict the consumption amount of the first subject after receiving the first recommendation information, i.e., the consumption amount prediction result, based on the second coefficient and the subject data of the first subject. The trained write-off rate prediction model can predict the write-off rate of the first subject after receiving the first recommendation information, i.e., the write-off rate prediction result, based on the second coefficient and the subject data of the first subject. Therefore, the conversion prediction result of the first subject includes at least one of the purchase rate prediction result, the consumption amount prediction result, and the write-off rate prediction result. The conversion correction function is used to correct the conversion prediction result. By correcting the conversion prediction result, first conversion data can be obtained. Therefore, the conversion correction function includes at least one of the purchase rate correction function, the consumption amount correction function, and the write-off rate correction function. The purchase rate correction function can be used to correct the purchase rate prediction results to obtain purchase rate data; the consumption amount correction function can be used to correct the consumption amount prediction results to obtain consumption amount data; the write-off rate correction function can be used to correct the write-off rate prediction results to obtain write-off rate data.

[0079] In an embodiment of the present application, a conversion prediction result for the first object is obtained by performing prediction processing based on the second coefficient and the object data of the first object using a trained first prediction model. This improves the accuracy and efficiency of determining the conversion prediction result. Subsequently, the conversion prediction result is corrected using a conversion correction function. This can reduce the gap between the model prediction value and the actual market feedback value caused by sample selection bias, data measurement bias, and deviations between model assumptions and the actual market during data collection, processing, and model training, thereby improving the accuracy and effectiveness of the first conversion data.

[0080] In step 103 , a first conversion gain of the first object after receiving the first recommendation information is determined based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function.

[0081] Here, the gain fitting function characterizes the fitting relationship between the coefficient and the conversion gain, and is used to determine the conversion gain based on the coefficient. The gain fitting function includes at least one of: a purchase rate gain fitting function, a consumption amount gain fitting function, and a write-off rate gain fitting function. The purchase rate gain can be determined by the purchase rate gain fitting function; the consumption amount gain can be determined by the consumption amount gain fitting function; and the write-off rate gain can be determined by the write-off rate gain fitting function. The conversion gain refers to the additional conversion data generated by the user after receiving the recommendation information, that is, the difference between the conversion data generated by the user after receiving the recommendation information and the conversion data generated by the user when no recommendation information is received. Therefore, the first conversion gain is used to represent the difference between the conversion data generated by the first object after receiving the first recommendation information and the conversion data generated by the user when no recommendation information is received. First, determine the difference between the first conversion data and the conversion data of the first object when no recommendation information is received (i.e., no intervention) as the second conversion gain of the first object; then determine the third conversion gain corresponding to the first coefficient based on the first coefficient and the gain fitting function; determine the fourth conversion gain corresponding to the second coefficient based on the second coefficient and a pre-constructed gain fitting function; then add the difference between the third conversion gain and the fourth conversion gain to the second conversion gain to determine the first conversion gain.

[0082] In some embodiments, see Figure 3B , step 103 may be implemented through steps 1031 to 1033, including:

[0083] In step 1031 , second conversion data of the first object not receiving any recommendation information is determined.

[0084] Here, the second conversion data refers to the conversion data of the first subject when it receives any recommendation information, that is, the conversion data of the first subject when no intervention is made. Since the first subject cannot redeem any recommendation information when it receives it, the redemption rate is 0. The second conversion data may also include the purchase rate of the first subject for the product when no intervention is made, as well as the purchase amount data generated by the purchase.

[0085] In step 1032 , the difference between the first conversion data and the second conversion data is determined as a second conversion gain of the first object.

[0086] Here, the difference between the same type of data in the first conversion data and the second conversion data is determined as the second conversion gain for the first object. For example, the difference between the purchase rates in the first conversion data and the second conversion data is determined as the second purchase rate gain, the difference between the consumption amounts in the first conversion data and the second conversion data is determined as the second consumption amount gain, and the difference between the write-off rates in the first conversion data and the second conversion data is determined as the second write-off rate gain. The second purchase rate gain, the second consumption amount gain, and the second write-off rate gain are then determined as the second conversion gain.

[0087] In step 1033 , a first conversion gain is determined based on the second conversion gain, the first coefficient, the second coefficient, and the gain fitting function.

[0088] Here, based on the first coefficient and the gain fitting function, the third conversion gain corresponding to the first coefficient is determined; based on the second coefficient and the pre-constructed gain fitting function, the fourth conversion gain corresponding to the second coefficient is determined; then the difference between the third conversion gain and the fourth conversion gain is added to the second conversion gain to determine the first conversion gain.

[0089] In this embodiment of the present application, the second conversion data of the first subject is determined when no recommendation information is received; the difference between the first conversion data and the second conversion data is determined as the second conversion gain of the first subject; and the first conversion gain is determined based on the second conversion gain, the first coefficient, the second coefficient, and a pre-constructed gain fitting function. In this way, the second conversion gain, the first coefficient, the second coefficient, and the pre-constructed gain fitting function can accurately analyze the impact of the intervention behavior on the first subject, thereby improving the accuracy of determining the first conversion gain.

[0090] In some embodiments, see Figure 3C Step 1033 may be implemented through steps 10331 to 10334, including:

[0091] In step 10331, a third conversion gain is determined based on the first coefficient and the gain fitting function.

[0092] Here, the first coefficient is used as the independent variable of the gain fitting function, and the third conversion gain is used as the dependent variable of the gain fitting function to determine the third conversion gain. The third conversion gain is determined by the gain fitting function and is the conversion gain generated by the first object when receiving the first recommendation information.

[0093] In step 10332, a fourth conversion gain is determined based on the second coefficient and the gain fitting function.

[0094] Here, the second coefficient is used as the independent variable of the gain fitting function, and the fourth conversion gain is used as the dependent variable of the gain fitting function to determine the fourth conversion gain. The fourth conversion gain is determined by the gain fitting function and is the conversion gain generated by the first object when receiving the recommendation information corresponding to the second coefficient.

[0095] In step 10333, the difference between the third conversion gain and the fourth conversion gain is determined as the gain difference.

[0096] Here, the difference between the third conversion gain and the fourth conversion gain is determined as the gain difference. Therefore, the gain difference is used to represent the difference between the conversion gains generated when the first object receives the first recommendation information and when it receives the recommendation information corresponding to the second coefficient.

[0097] In step 10334, the sum of the gain difference and the second conversion gain is determined as the first conversion gain.

[0098] Here, the gain difference and the second conversion gain are added together to form the first conversion gain. For example, substituting the first coefficient into the gain fitting function for calculation yields a third conversion gain of 40%. Substituting the second coefficient into the gain fitting function for calculation yields a fourth conversion gain of 36%. Next, the difference between the third conversion gain of 40% and the fourth conversion gain of 36% is determined to be 4%, i.e., the gain difference is 4%. This gain difference of 4% is then added to the second conversion gain of 20% for the first recommendation information, yielding a sum of 24%, thus determining the first conversion gain to be 24%.

[0099] In this embodiment of the present application, the third conversion gain is determined based on the first coefficient and the gain fitting function; the fourth conversion gain is determined based on the second coefficient and a pre-established gain fitting function; the difference between the third and fourth conversion gains is determined as the gain difference; and the sum of the gain difference and the second conversion gain is determined as the first conversion gain. In this way, the gain fitting function can be used to make the first conversion gain closer to the actual gain data, further improving the accuracy of determining the first conversion gain.

[0100] In step 104 , based on the first conversion gain of each first object, a second object that meets a second preset condition is determined from the first objects.

[0101] Here, the second target refers to the user who is ultimately selected to receive the first recommended information. By sending the first recommended information to the second target, the conversion gain can be maximized within the limited information recommendation cost. A conversion gain threshold can be pre-set, and then the first target whose first conversion gain exceeds the conversion gain threshold can be determined as the second target. Alternatively, an objective function can be constructed using the first conversion gain and the cost of information recommendation, and an allocation algorithm can be used to maximize the benefits of information recommendation within the limited information recommendation cost.

[0102] In an embodiment of the present application, based on the first coefficient of the first recommendation information, a second coefficient of the first recommendation information is determined from a coefficient set. For each first object, a prediction process is performed based on the second coefficient and the object data of the first object to determine the first conversion data. In this way, even if the first recommendation information is new, the corresponding second coefficient can be determined from the coefficient set, and the first conversion data can be obtained by prediction processing based on the second coefficient, thereby improving the efficiency and reliability of determining the first conversion data for the first object. Then, based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function, the first conversion gain of the first object after receiving the first recommendation information is determined. The gain fitting function represents the fitting relationship between the coefficient and the conversion gain. Using the gain fitting function, the accuracy of determining the first conversion gain can be improved, making the first conversion gain more consistent with the actual gain. Finally, based on the first conversion gain of each first object, a second object that meets a second preset condition is determined from multiple first objects, and the first recommendation information is pushed to the second object, thereby improving the efficiency and accuracy of the first recommendation information. Therefore, the object determination method provided by the present application can adapt to various new first recommendation information, improve the adaptability of the object determination method, and ensure the efficiency and accuracy of object determination.

[0103] In some embodiments, see Figure 3D In step 104, “determining a second object from a plurality of first objects based on the first conversion gain of each first object” can be implemented through steps 1041 to 1044, including:

[0104] In step 1041 , second conversion data of the first object not receiving any recommendation information is determined.

[0105] Here, the second conversion data refers to the conversion data of the first object without receiving any recommendation information, that is, without intervention.

[0106] In step 1042 , an objective function for revenue is determined based on the second conversion data of the first object, the first conversion gain, and the write-off cost.

[0107] Here, the verification cost refers to the cost incurred when the first recommendation information is verified, and the objective function is used to calculate the maximum benefit (i.e., the difference between the total amount of consumption of all users after the first recommendation information is pushed and the verification cost of the first recommendation information). For example, see formula (1):

[0108]

[0109] Among them, x ui is the decision variable, x ui ∈{0,1}, if the first object u receives the first recommendation information of category i, it takes 1, otherwise it takes 0; m i represents the recommendation cost of the first recommendation information of category i; p u0 represents the purchase rate of the first object u without intervention; q u0 represents the consumption amount of the first object u without intervention; U is the total number of first objects; I is the number of types of first recommendation information of all types; A is the verification cost of the first recommendation information.

[0110] In step 1043, the objective function is solved based on the preset constraint information to determine the decision result of the first object.

[0111] Here, the decision result represents whether to send the first recommendation information to the first object, that is, the decision variable in formula (1). The preset constraint information is used to constrain the parameters in the objective function. For example, see formulas (2) and (3):

[0112]

[0113] Among them, formula (2) and formula (3) are both constraint information. Formula (2) is used to constrain the sum of decision variables corresponding to all first objects to be less than or equal to the total number of first objects, and the decision variables are 0 or 1; Formula (3) is used to constrain that within the recommendation cost range of the first recommendation information, at most one coupon is issued to each user.

[0114] In step 1044 , the first object whose decision result is the first value is determined as the second object, wherein when the decision result is the first value, it indicates that the first object meets the second preset condition.

[0115] Here, by solving the objective function, a decision result corresponding to each first object can be calculated. The decision result is either a first value or a second value. When the decision result is the first value, the first recommendation information is sent to the first object; when the decision result is the second value, the first recommendation information is not sent to the first object, and the first value and the second value are different. For example, the first value can be 1, and the second value can be 0.

[0116] In an embodiment of the present application, the second conversion data of each first object before receiving any recommendation information is determined; based on the second conversion data, the first conversion gain and the write-off cost of each first object, an objective function for the benefit is determined; based on the preset constraint information, the objective function is solved to determine the decision result of each first object, the decision result representing whether to send the first recommendation information to the first object; the first object whose decision result is the first value is determined as the second object, wherein when the decision result is the first value, it represents that the first object meets the second preset condition. In this way, by constructing the objective function and solving it under the constraint information, it is possible to flexibly and comprehensively consider the allocation of various resources, maximize the benefit of object determination under a limited cost budget, improve the accuracy of determining the second object, and improve the effectiveness and rationality of object determination.

[0117] In some embodiments, see Figure 3E , a coefficient set can also be constructed through steps 201 to 204, including:

[0118] In step 201 , a tendency prediction is performed based on the object data of the third object to obtain tendency data corresponding to the third object.

[0119] Here, through the trained second prediction model, a tendency prediction is performed based on the object data of each third object to obtain the tendency data corresponding to the third object. The third object refers to the user who participated in the recommendation activity in a certain historical recommendation activity, which can be obtained from the historical recommendation information. Object data refers to various information collected during the interaction with the user. This information can be used to analyze user behavior, optimize recommendation methods, and improve user experience, thereby helping to make accurate information recommendations to users. The trained second prediction model can predict the probability of an object receiving recommended information, and the tendency data indicates the probability of the third object receiving the recommended information. The tendency data is a numerical value between 0 and 1. The higher the value, the higher the probability of the third object receiving the recommended information. Alternatively, the tendency data can also be used to indicate the interest preference of the third object corresponding to the recommended information. The higher the numerical value corresponding to the tendency data, the higher the degree of interest of the third object in the recommended information.

[0120] In some possible implementations, the reception status of each third object for different recommendation information can be determined, and whether the recommendation information is received is used as a training label. If received, it is marked as 1, otherwise it is marked as 0. The object data of the third object is used as a training sample, and then the second prediction model to be trained is trained using these training samples and training labels to obtain a trained second prediction model.

[0121] In step 202 , the third objects are clustered based on the tendency data corresponding to the third objects to obtain N+1 first sets.

[0122] Here, there are multiple third objects, and based on the propensity data corresponding to each third object, the multiple third objects are clustered. The first set is a homogeneous customer group, which refers to a customer group with similar or identical propensity data. In other words, the third objects in the same first set have similar probabilities of receiving recommended information. Therefore, multiple third objects can be clustered based on the propensity data corresponding to each third object, thereby obtaining N+1 first sets. The value of N is an integer and can be adjusted according to actual needs to ensure that each first set has a preset number of third objects.

[0123] In some embodiments, step 202 may be implemented by the following process, including:

[0124] The tendency interval corresponding to the third object is determined from the preset tendency interval based on the tendency data corresponding to the third object; and the third objects belonging to the same tendency interval are constructed into a first set.

[0125] Here, multiple tendency intervals are pre-set, and the size of each tendency interval can be the same or different. Then, the third objects whose tendency data are in the same tendency interval are constructed into a first set. For example, multiple tendency intervals may include: [0, 0.2), [0.2, 0.4), [0.4, 0.8), [0.8, 1]. The tendency data of the third object 1 is 0.3, the tendency data of the third object 2 is 0.7, the tendency data of the third object 3 is 0.25, and the tendency data of the third object 4 is 0.5. Then, the third objects 1 and 3 belong to the same first set, and the third objects 2 and 4 belong to the same first set. In this way, the third objects with similar tendency data can be accurately divided into the same first set, thereby improving the efficiency and accuracy of constructing the first set.

[0126] In step 203, based on the third coefficients corresponding to the plurality of second recommendation information, the fourth coefficient corresponding to the first set is determined.

[0127] Here, each third object in the same first set has received recommendation information, namely, the second recommendation information. Although the reception probabilities of multiple third objects in a first set are similar, different third objects may receive different second recommendation information. In other words, the second recommendation information is the recommendation information received by each third object included in the first set. The second recommendation information corresponding to different third objects in the same first set may be the same or different. The third coefficient is used to describe the difficulty of verifying the second recommendation information. Each second recommendation information has a corresponding third coefficient. Here, the average value of each third coefficient corresponding to the first set can be calculated and determined as the fourth coefficient corresponding to this first set.

[0128] In step 204 , N+1 fourth coefficients are constructed as a coefficient set.

[0129] Here, since there are N+1 first sets, there are N+1 fourth coefficients in total, and these N+1 fourth coefficients are constructed as a coefficient set.

[0130] In an embodiment of the present application, a trained second prediction model is used to perform a tendency prediction based on the object data of each third object to obtain tendency data corresponding to the third object. Based on the tendency data corresponding to each third object, multiple third objects are clustered to obtain N+1 first sets. For each first set, a fourth coefficient corresponding to the first set is determined based on the third coefficients corresponding to multiple second recommendation information, where the second recommendation information is the recommendation information received by each third object included in the first set. The N+1 fourth coefficients are then used to construct a coefficient set. In this way, the coefficient set can be constructed by reusing the previously existing second recommendation information, thereby improving the efficiency of constructing the coefficient set and the comprehensiveness of the coefficient set.

[0131] In some embodiments, see Figure 3F , the gain fitting function can be obtained through steps 301 to 303, including:

[0132] In step 301 , third conversion data corresponding to a non-recommended set and fourth conversion data corresponding to each recommended set are determined.

[0133] Here, the N+1 first sets include a non-recommended set that has not received the second recommendation information and N recommended sets that have received the second recommendation information. The third conversion data is the average conversion data of all third objects in the non-recommended set, and the fourth conversion data is the average conversion data of all third objects in the recommended set. The average conversion data includes at least one of the average purchase rate, the average consumption amount, and the average verification rate. For the average purchase rate, first determine the number of purchasers of the third objects who purchased the goods in the non-recommended set, and then divide the number of purchasers by the total number of people in the non-recommended set to obtain the average purchase rate. For the average consumption amount, first determine the total consumption amount of all third objects in the non-recommended set, and then divide the total consumption amount by the total number of people in the non-recommended set to obtain the average consumption amount rate. For the average verification rate, first determine the total verification quantity of all third objects in the non-recommended set, and then divide the total verification quantity by the total number of people in the non-recommended set to obtain the average verification rate.

[0134] In step 302 , the difference between the fourth conversion data and the third conversion data corresponding to the recommended set is determined as the fifth conversion gain corresponding to the recommended set.

[0135] Here, since the non-recommended set is the customer group that has not received the second recommendation information, the third conversion data represents the conversion data of the third object when there is no intervention. Therefore, the difference between the fourth conversion data and the third conversion data can reflect the conversion gain of the recommended set when it receives the second recommendation information, that is, the fifth conversion gain corresponding to the recommended set.

[0136] In step 303, a fitting process is performed based on the fourth coefficient corresponding to the recommended set and the fifth conversion gain corresponding to the recommended set to obtain a gain fitting function.

[0137] Here, based on the fourth coefficient and fifth conversion gain corresponding to each recommended set, a function of the conversion gain with respect to the coefficient is constructed, thereby obtaining a gain fitting function. For example, first, based on the characteristics between the fourth coefficient and the fifth conversion gain corresponding to each recommended set, an appropriate function model, such as a linear function model, a polynomial function model, or an exponential function model, is selected. Then, using statistical methods, such as the least squares method, the parameters of the function model are estimated and adjusted to minimize the difference between the data predicted by the model based on the coefficients and the conversion gain, thereby obtaining a gain fitting function.

[0138] In an embodiment of the present application, the N+1 first sets include one unrecommended set that has not received the second recommendation information and N recommended sets that have received the second recommendation information. The third conversion data corresponding to the unrecommended set and the fourth conversion data corresponding to each recommended set are determined. For each recommended set, the difference between the fourth conversion data and the third conversion data corresponding to the recommended set is determined as the fifth conversion gain corresponding to the recommended set. Fitting is performed based on the fourth coefficient corresponding to each recommended set and the fifth conversion gain corresponding to each recommended set to obtain a gain fitting function. In this way, an accurate gain fitting function can be constructed using the fourth coefficient and the fifth conversion gain of each recommended set, facilitating determination of the conversion gain of newly appearing recommendation information using the gain fitting function, thereby improving the accuracy and efficiency of object identification.

[0139] In some embodiments, see Figure 3G , the conversion correction function can be obtained through steps 401 to 405, including:

[0140] In step 401, prediction processing is performed based on the fourth coefficient corresponding to the first set to obtain a conversion prediction result corresponding to the first set.

[0141] Here, for each first set, a prediction process is performed using the trained first prediction model based on the fourth coefficient corresponding to the first set to obtain a conversion prediction result corresponding to the first set. The fourth coefficient corresponding to the first set is used as input to the trained first prediction model, and prediction processing is performed using the trained first prediction model to obtain a conversion prediction result corresponding to the first set. Since the trained first prediction model includes at least one of a trained purchase prediction model, a trained consumption amount prediction model, and a trained write-off rate prediction model, the conversion prediction result corresponding to the first set includes at least one of a purchase prediction result, a consumption amount prediction result, and a write-off rate prediction result.

[0142] In step 402 , based on the descending order of the conversion prediction results of the first sets, N+1 first sets are sorted to obtain sorted first sets, and the sorted first sets are divided to obtain K second sets.

[0143] Here, K is a positive integer less than or equal to N+1, and the size of K can be flexibly set according to actual needs. The N+1 first sets are sorted in descending order of conversion prediction results to obtain a sorted first set, and then the sorted first set is divided into K second sets.

[0144] In step 403 , fifth conversion data corresponding to the second set is determined.

[0145] Here, the fifth conversion data is the average real conversion data corresponding to the second set. The real conversion data corresponding to each third object in the second set can be collected from the database, and then the average value of the real conversion data of all third objects is calculated as the fifth conversion data.

[0146] In step 404 , based on the fifth transformed data corresponding to the second set, K second sets are merged to obtain L third sets.

[0147] Here, L is a positive integer less than or equal to K. The sixth conversion data corresponding to the L third sets are sorted in descending order, and the sixth conversion data is the average actual conversion data corresponding to the merged third sets. It should be noted that if the fifth conversion data corresponding to each second set is sorted in descending order, the second sets do not need to be merged. In this case, L is equal to K, and the third set is the second set. If the fifth conversion data corresponding to each second set does not conform to the descending order, the second sets need to be merged. In this case, L is less than K. Taking the purchase rate as an example, the fifth conversion data corresponding to each second set are 80%, 60%, 50%, 54%, 40%, 30%, and 15%, respectively. "54%" does not conform to the descending order. Therefore, the second set with a fifth conversion data of 54% can be merged with the previous second set (i.e., the second set with a fifth conversion data of 50%), or with the next second set (i.e., the second set with a fifth conversion data of 40%). For example, if the third set is merged with the previous second set, the sixth conversion data corresponding to the third set obtained after the two second sets are merged is (50% + 54%) / 2 = 52%. At this time, the sixth conversion data corresponding to the multiple merged third sets are: 80%, 60%, 52%, 40%, 30%, 15%, which conforms to the descending order.

[0148] In step 405, the conversion prediction result corresponding to the third set is determined, and a fitting process is performed based on the sixth conversion data corresponding to the third set and the conversion prediction result to obtain a conversion correction function.

[0149] Here, for each third set, the average value of the conversion prediction results of all first sets included in the third set is calculated as the conversion prediction result corresponding to the third set. Based on the sixth conversion data and the conversion prediction result corresponding to each third set, a function of the sixth conversion data with respect to the conversion prediction result is constructed to obtain a conversion correction function. For example, first, based on the characteristics between the sixth conversion data and the conversion prediction result corresponding to each third set, a suitable function model, such as a linear function model, a polynomial function model, or an exponential function model, is selected; then, a statistical method, such as the least squares method, is used to estimate and adjust the parameters of the model so that the difference between the data predicted by the model based on the conversion prediction result and the sixth conversion data is minimized, thereby obtaining a conversion correction function.

[0150] In an embodiment of the present application, for each first set, a prediction process is performed based on the fourth coefficient corresponding to the first set by using the trained first prediction model to obtain a conversion prediction result corresponding to the first set; based on the descending order of the conversion prediction results of the first set, N+1 first sets are sorted to obtain a sorted first set, and the sorted first set is divided to obtain K second sets; the fifth conversion data corresponding to each second set is determined; based on the fifth conversion data corresponding to each second set, the second sets are merged to obtain L third sets, and the sixth conversion data corresponding to the L third sets are sorted in descending order; the conversion prediction result corresponding to each third set is determined, and a fitting process is performed based on the sixth conversion data corresponding to each third set and the conversion prediction result to obtain a conversion correction function. In this way, by constructing a function between the actual conversion data and the conversion prediction result predicted by the model as a conversion correction function, it can help to adjust the prediction result of the model according to the actual situation, minimize the gap between the predicted conversion data and the actual conversion data, and improve the accuracy and effectiveness of the conversion data.

[0151] Below, an exemplary application of the embodiment of the present application in an actual recommendation scenario will be described, using the recommendation information as a coupon for illustration.

[0152] The embodiment of the present application proposes a coupon gain maximization method based on prediction value correction and intervention adaptation. Through precise prediction value correction and flexible intervention adaptation mechanism, it effectively copes with problems such as constantly changing coupon types and prediction deviations, thereby maximizing coupon gain and improving the efficiency and accuracy of issuing coupons to users.

[0153] See also Figure 4 , Figure 4 This is a flow chart of the model training and function construction provided by the embodiment of the present application, including:

[0154] In step 41, N+1 first sets are constructed.

[0155] Here, a sample x (corresponding to the third object in other embodiments) of a certain recommended activity is collected from the user database, and whether or not a coupon is received is used as a training label y. If received, it is marked as 1, otherwise it is marked as 0. Secondly, these training samples are used to train a tree model (such as RF model, LGBM model) or a deep learning model (such as DNN model) to obtain a trained second prediction model. Then, the trained second prediction model is used to score the sample x to obtain the tendency data. Finally, based on the output tendency data and the set tendency interval, the N+1 first set set C={c i|i=0,1,…,N}, where i=0 represents the customer group that did not receive the coupon (corresponding to the non-recommended set in other embodiments). The tendency interval can be adjusted according to actual needs to ensure that each first set has a sufficient sample size.

[0156] In step 42, a gain fitting function is constructed.

[0157] Here, through data analysis and combined with expert experience, coupons with different usage thresholds can be mapped to different usage difficulty coefficients d (corresponding to the coefficients in other embodiments). Therefore, the difficulty coefficient set of the N+1 customer groups that receive coupons is defined as D∈{d i |i=0,1,…,N,d0=0}(corresponding to the coefficient set in other embodiments). In addition, the purchase rate p of the i-th customer group is counted. i = Number of buyers / Total number of customers, spending amount q i =Total consumption amount / total number of customers, write-off rate w i = Number of coupons redeemed / (Number of coupons issued (i) ≠ 0), then the purchase rate gain is Δp i =p i -p0, consumption amount gain Δq i =q i -q0.

[0158] Based on the above statistical data, the functions of the purchase rate gain Δp, consumption amount gain Δq, and write-off rate w with respect to the difficulty coefficient d can be fitted, which are defined as Δp=P(d), Δq=Q(d) and w=W(d) respectively (corresponding to the gain fitting functions in other embodiments).

[0159] In step 43, the first prediction model to be trained is trained to obtain a trained first prediction model.

[0160] Here, the first prediction model includes a purchase rate prediction model, a consumption amount prediction model, and a write-off rate prediction model. Collect training samples x, including non-intervention features f u and intervention characteristics f t , where the value of the intervention feature is the coupon usage difficulty coefficient d. The purchase rate prediction model, consumption amount prediction model, and write-off rate prediction model to be trained are all trained using a single machine learning model (Single-Learner). The purchase rate prediction model and write-off rate prediction model to be trained are binary classification models, and the consumption amount prediction model to be trained is a regression model. For coupons with a usage difficulty coefficient of d, the model paradigm can be expressed as follows: y = f(f u ,f t=d). After the purchase rate prediction model, the consumption amount prediction model, and the write-off rate prediction model are trained, a trained first prediction model is obtained.

[0161] In step 44, a conversion correction function is constructed.

[0162] Here, the model's predicted values ​​are corrected using historical observation data to obtain a conversion correction function. The conversion correction function includes a purchase rate correction function, a consumption amount correction function, and a write-off rate correction function. Taking the write-off rate correction function as an example, the specific method is as follows:

[0163] For the i-th customer group, use the trained write-off rate prediction model to predict it, and sort each first set in descending order according to the prediction results. Divide the first set after descending order into K buckets (corresponding to the K second sets in other embodiments), and calculate the average model prediction result w of each bucket pred and the average true write-off rate w true , observe the write-off rate w of K buckets true Is it also a monotonically decreasing arrangement? If not, reduce the number of buckets until it shows a monotonically decreasing trend. true After satisfying the monotonically decreasing trend, the write-off rate mapping function w is fitted. true =f(w pred ), and obtain the write-off rate correction function in the conversion correction function. The construction methods of the purchase rate correction function and the consumption amount correction function are the same as those of the write-off rate correction function, and will not be repeated here.

[0164] See also Figure 5 , Figure 5 : is a flow chart of the coupon recommendation method provided in the embodiment of the present application, including:

[0165] In step 51 , a new coupon (corresponding to the first recommendation information in other embodiments) is received.

[0166] Here, the difficulty coefficient of using the new coupon i is defined as d new (corresponding to the first coefficient in other embodiments). Then, a coefficient corresponding to d is selected from the existing difficulty coefficient set D. new The closest difficulty coefficient d old (Corresponding to the second coefficient in other embodiments).

[0167] In step 52, the new coupon is predicted using the trained first prediction model to obtain a conversion prediction result.

[0168] Here, the conversion prediction result is obtained by performing prediction processing on user u (corresponding to the first object in other embodiments) through the trained purchase rate model, consumption amount model and verification rate model in the trained first prediction model. old =f 购买率 (f u ,d old )-f 购买率 (f u ,d0), consumption amount gain Δq old =f 消费金额 (f u ,d old )-f 消费金额 (f u ,d0), write-off rate purchase rate gain w old =f 核销率 (f u ,d old ).

[0169] In step 53 , a first conversion gain is determined by using a gain fitting function.

[0170] Here, the gain fitting function is used to determine the user u’s difficulty coefficient d. new The purchase rate gain Δp under the new coupon i ui =Δp old +P(d new )-P(d old ), consumption amount gain Δq ui =Δq old +Q(d new )-Q(d old ), write-off rate w ui =w old +W(d new )-W(d old ).

[0171] In step 54, the user to whom the new coupon needs to be sent (corresponding to the second object in other embodiments) is determined through the allocation algorithm.

[0172] Here, in order to maximize the benefits within a limited cost, the embodiment of the present application combines the allocation algorithm to find the optimal solution. The calculation logic is shown in formula (1), formula (2) and formula (3):

[0173]

[0174] Among them, x ui is the decision variable, x ui ∈{0,1}, if user u receives type i coupon, it takes 1, otherwise it takes 0; m irepresents the cost of coupon type i; p u0 represents the purchase rate of user u without coupon intervention; q u0 represents the amount of spending by user u without coupon intervention; U is the total number of users; I is the number of coupon types; and A is the coupon redemption cost. Formula (1) is the objective function, which is used to calculate the maximum benefit (total spending minus coupon redemption cost). Formulas (2) and (3) are the constraints, which require that at most one coupon be issued to each user within the coupon issuance cost range. By solving the above allocation algorithm using integer programming, the users to whom new coupons should be sent are determined.

[0175] In the embodiments of the present application, a correction algorithm is used to calibrate the model's predicted values, helping to adjust the model's output according to actual conditions, making it more consistent with real business scenarios and market demands, minimizing the gap between the model's predicted values ​​and the true values, and improving the model's accuracy and effectiveness. Through an allocation algorithm, the embodiments of the present application flexibly and comprehensively consider the allocation of various recommendation resources, optimizing the recommendation strategy within a limited cost budget, and thus better achieving the goal of maximizing recommendation revenue.

[0176] The embodiment of the present application has a mechanism for responding to changes in coupon types: when the coupon type is dynamically adjusted with the market recommendation strategy (new coupons with different usage thresholds and discount values ​​appear), there is no need to retrain the model and deploy it to the online recommendation system, so as to adapt to the new promotion strategy and maintain effective prediction capabilities; prediction value correction mechanism: through precise prediction value correction, the correction algorithm is used to adjust the output of the model according to the actual situation, thereby reducing the gap between the model prediction value and the actual market feedback value caused by sample selection bias, data measurement bias, deviation between model assumptions and actual market in the process of data collection, processing and model training, so that the model output is more in line with real business scenarios and market needs; intervention adaptive mechanism: with flexible intervention adaptive capabilities, it works in conjunction with the prediction value correction mechanism to jointly respond to problems such as the constant change of coupon types and prediction deviations, and combines the allocation algorithm to maximize the coupon gain, providing strong support for market recommendation activities.

[0177] The embodiments of the present application can effectively improve the efficiency of recommendations, avoid the need to retrain the model and then deploy it online due to changes in coupon types, so that recommendation activities can respond to new coupon strategies more quickly and reduce the time cost of model adjustment. Since there is no need to frequently retrain the model and deploy it, the cost investment in manpower, computing resources, etc. is also reduced. By correcting the model output results and intervening in the adaptive mechanism, the error between the model prediction value and the actual market feedback value can be reduced, so that the delivery and use effects of coupons are more in line with expectations, and the accuracy of recommendation activities is improved. At the same time, combined with the allocation algorithm, the coupon gain is maximized, so that coupons can play the greatest role in market recommendation activities and increase recommendation revenue.

[0178] The following continues to describe the exemplary structure of the object determination device 233 provided in the embodiment of the present application implemented as a software module. In some embodiments, such as Figure 2 As shown, the software modules stored in the object determination device 233 of the memory 230 may include:

[0179] Determination module 2331 is used to determine the second coefficient of the first recommendation information from a coefficient set based on the first coefficient of the first recommendation information, where the second coefficient is a coefficient in the coefficient set that satisfies a first preset condition with the first coefficient; determine first conversion data based on the second coefficient and the object data of the first object, where the first conversion data is statistical data of the first object performing a preset behavior according to the first recommendation information; determine the first conversion gain of the first object after receiving the first recommendation information based on the first conversion data, the first coefficient, the second coefficient and a pre-constructed gain fitting function; and determine a second object that meets a second preset condition from the first object based on the first conversion gain of the first object.

[0180] The determination module 2331 is further configured to determine, for each third coefficient in the coefficient set, a coefficient difference between the third coefficient and the first coefficient; and determine the third coefficient whose coefficient difference satisfies the first preset condition as the second coefficient.

[0181] The determination module 2331 is further configured to determine the second conversion data of the first object when the first object does not receive any recommendation information; determine the difference between the first conversion data and the second conversion data as the second conversion gain of the first object; and determine the first conversion gain based on the second conversion gain, the first coefficient, the second coefficient, and the gain fitting function.

[0182] The determination module 2331 is further used to determine a third conversion gain based on the first coefficient and the gain fitting function; determine a fourth conversion gain based on the second coefficient and the gain fitting function; determine the difference between the third conversion gain and the fourth conversion gain as the gain difference; and determine the sum of the gain difference and the second conversion gain as the first conversion gain.

[0183] The determination module 2331 is also used to perform tendency prediction based on the object data of each third object through the trained second prediction model to obtain tendency data corresponding to the third object; cluster multiple third objects based on the tendency data corresponding to each third object to obtain N+1 first sets; for each first set, determine the fourth coefficient corresponding to the first set based on the third coefficients corresponding to multiple second recommendation information, where the second recommendation information is the recommendation information received by each third object included in the first set; and construct the coefficient set with N+1 fourth coefficients.

[0184] The determining module 2331 is further configured to determine, from a preset tendency interval, a tendency interval corresponding to each third object based on the tendency data corresponding to the third object; and construct the third objects belonging to the same tendency interval into a first set.

[0185] The determination module 2331 is further used to determine the third conversion data corresponding to the non-recommended set and the fourth conversion data corresponding to each of the recommended sets; determine the difference between the fourth conversion data corresponding to the recommended set and the third conversion data as the fifth conversion gain corresponding to the recommended set; and perform fitting processing based on the fourth coefficient corresponding to each of the recommended sets and the fifth conversion gain corresponding to each of the recommended sets to obtain a gain fitting function.

[0186] The determination module 2331 is also used to perform prediction processing based on the fourth coefficient corresponding to the first set to obtain a conversion prediction result corresponding to the first set; sort the N+1 first sets based on the descending order of the conversion prediction results of the first set to obtain a sorted first set, and divide the sorted first set to obtain K second sets; determine the fifth conversion data corresponding to the second set; merge the K second sets based on the fifth conversion data corresponding to the second set to obtain L third sets, and the sixth conversion data corresponding to the L third sets are sorted in descending order; determine the conversion prediction result corresponding to the third set, perform fitting processing based on the sixth conversion data corresponding to the third set and the conversion prediction result to obtain a conversion correction function.

[0187] The determination module 2331 is further configured to perform prediction processing based on the second coefficient and the object data of the first object to obtain a conversion prediction result of the first object; and to correct the conversion prediction result based on a pre-constructed conversion correction function to obtain the first conversion data.

[0188] The determination module 2331 is also used to determine the second conversion data of the first object before receiving any recommendation information; determine the objective function for profit based on the second conversion data, first conversion gain and write-off cost of the first object; solve the objective function based on preset constraint information to determine the decision result of the first object, and the decision result represents whether the first object meets the second preset condition; determine the first object whose decision result is the first value as the second object, wherein when the decision result is the first value, it represents that the first object meets the second preset condition.

[0189] An embodiment of the present application provides a computer program product, which includes a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions from the computer-readable storage medium and executes the computer-executable instructions, causing the electronic device to perform the object determination method described in the embodiment of the present application.

[0190] The embodiment of the present application provides a computer-readable storage medium in which computer-executable instructions or computer programs are stored. When the computer-executable instructions or computer programs are executed by a processor, the processor will be caused to execute the object determination method provided by the embodiment of the present application, for example, Figure 3A The object determination method shown.

[0191] In some embodiments, the computer-readable storage medium may be a memory such as RAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

[0192] In some embodiments, computer-executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0193] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored in part of a file that stores other programs or data, e.g., in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0194] By way of example, computer-executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed across multiple sites and interconnected by a communication network.

[0195] In summary, through the embodiments of the present application, based on the first coefficient of the first recommendation information, the second coefficient of the first recommendation information is determined from a coefficient set; prediction processing is performed based on the second coefficient and the object data of the first object to determine the first conversion data. In this way, even if the first recommendation information is new, a second coefficient that satisfies a first preset condition with the first coefficient can be determined from the coefficient set, and the first conversion data can be obtained by performing prediction processing based on the second coefficient, thereby improving the efficiency and reliability of determining the first conversion data of the first object. Then, based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function, the first conversion gain of the first object after receiving the first recommendation information is determined. The gain fitting function can represent the fitting relationship between the coefficient and the conversion gain. The gain fitting function can improve the accuracy of determining the first conversion gain, making the first conversion gain more consistent with the actual gain. Finally, based on the first conversion gain of each first object, a second object that meets the second preset condition is determined from multiple first objects, and the first recommendation information is pushed to the second object, thereby improving the recommendation efficiency and accuracy of the first recommendation information. Therefore, the object determination method provided in this application can adapt to various new first recommendation information, improve the adaptability of the object determination method, and ensure the efficiency and accuracy of information recommendation, thereby improving the conversion rate of information recommendation and user retention rate.

[0196] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for determining an object, characterized in that: The method comprises: Determining, based on the first coefficient of the first recommendation information, a second coefficient of the first recommendation information from a coefficient set, where the second coefficient is a coefficient in the coefficient set that satisfies a first preset condition with the first coefficient; determining first conversion data based on the second coefficient and the object data of the first object, where the first conversion data is statistical data of the first object performing a preset behavior according to the first recommendation information; determining, based on the first conversion data, the first coefficient, the second coefficient, and a pre-constructed gain fitting function, a first conversion gain of the first subject after receiving the first recommendation information; Based on the first conversion gain of the first object, a second object that meets a second preset condition is determined from the first objects.

2. The method according to claim 1, characterized in that The determining, based on the first coefficient of the first recommendation information, the second coefficient of the first recommendation information from a coefficient set includes: determining, for each third coefficient in the set of coefficients, a coefficient difference between the third coefficient and the first coefficient; A third coefficient whose coefficient difference satisfies the first preset condition is determined as the second coefficient.

3. The method according to claim 1, characterized in that The determining, based on the first conversion data, the first coefficient, the second coefficient, and a pre-built gain fitting function, a first conversion gain of the first object after receiving the first recommendation information includes: Determining second conversion data of the first subject when the first subject does not receive any recommendation information; determining a difference between the first conversion data and the second conversion data as a second conversion gain of the first object; The first conversion gain is determined based on the second conversion gain, the first coefficient, the second coefficient, and the gain fitting function.

4. The method according to claim 3, characterized in that The determining the first conversion gain based on the second conversion gain, the first coefficient, the second coefficient, and the gain fitting function includes: determining a third conversion gain based on the first coefficient and the gain fitting function; determining a fourth conversion gain based on the second coefficient and the gain fitting function; determining a difference between the third conversion gain and the fourth conversion gain as a gain difference; The sum of the gain difference and the second conversion gain is determined as the first conversion gain.

5. The method according to claim 1, wherein The method further comprises: performing a tendency prediction based on the object data of the third object to obtain tendency data corresponding to the third object; Clustering the third object based on the tendency data corresponding to the third object to obtain N+1 first sets; determining a fourth coefficient corresponding to the first set based on a third coefficient corresponding to second recommendation information, where the second recommendation information is recommendation information received by a third object included in the first set; The coefficient set is constructed using N+1 of the fourth coefficients.

6. The method according to claim 5, characterized in that The clustering of the third objects based on the tendency data corresponding to the third objects to obtain N+1 first sets includes: Determining a tendency interval corresponding to the third object from preset tendency intervals based on the tendency data corresponding to the third object; The third objects belonging to the same tendency interval are constructed into a first set.

7. The method according to claim 5, characterized in that The N+1 first sets include one non-recommended set that has not received the second recommendation information and N recommended sets that have received the second recommendation information, and the method further includes: Determining third conversion data corresponding to the non-recommended set and fourth conversion data corresponding to the recommended set; determining a difference between the fourth conversion data corresponding to the recommended set and the third conversion data as a fifth conversion gain corresponding to the recommended set; A fitting process is performed based on the fourth coefficient corresponding to the recommended set and the fifth conversion gain corresponding to the recommended set to obtain a gain fitting function.

8. The method according to claim 5, characterized in that The method further comprises: Performing prediction processing based on the fourth coefficient corresponding to the first set to obtain a conversion prediction result corresponding to the first set; Sorting the N+1 first sets based on the descending order of the conversion prediction results of the first sets to obtain sorted first sets, and dividing the sorted first sets to obtain K second sets; determining fifth conversion data corresponding to the second set; Based on the fifth transformed data corresponding to the second set, the K second sets are merged to obtain L third sets, and the sixth transformed data corresponding to the L third sets are sorted in descending order; The conversion prediction result corresponding to the third set is determined, and a fitting process is performed based on the sixth conversion data corresponding to the third set and the conversion prediction result to obtain a conversion correction function.

9. The method according to claim 8, characterized in that The determining first conversion data based on the second coefficient and the object data of the first object includes: performing prediction processing based on the second coefficient and the object data of the first object to obtain a conversion prediction result of the first object; Based on a pre-built conversion correction function, the conversion prediction result of the first object is corrected to obtain the first conversion data.

10. The method according to claim 1, characterized in that The determining, based on the first conversion gain of the first object, a second object that meets a second preset condition from the first object includes: Determining second conversion data of the first subject when the first subject does not receive any recommendation information; determining an objective function for revenue based on the second conversion data of the first object, the first conversion gain, and the write-off cost; Solving the objective function based on preset constraint information to determine a decision result of the first object, where the decision result indicates whether the first object satisfies a second preset condition; A first object whose decision result is a first value is determined as a second object, wherein when the decision result is the first value, it represents that the first object meets a second preset condition.

11. An electronic device, characterized in that: The electronic device comprises: a memory for storing computer-executable instructions or computer programs; A processor, configured to implement the method according to any one of claims 1 to 10 when executing computer-executable instructions or computer programs stored in the memory.

12. A computer program product comprising computer executable instructions or a computer program, characterized in that When the computer executable instructions or computer program are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images