Data processing method and device, computer device and storage medium

Abstract

The application discloses a data processing method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining object behavior data of a target object for a recommended object in multiple time intervals; generating object behavior characteristics of the target object for the recommended object in each time interval according to the object behavior data of the target object for the recommended object in each time interval; and predicting an order placing index of the target object for the recommended object according to the object behavior characteristics of the target object for the recommended object in each time interval. The application can improve the accuracy of the predicted order placing index of the target object for the recommended object.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more particularly to a data processing method, apparatus, computer equipment, and storage medium. Background Technology

[0002] For all types of e-commerce platforms, the question of how to identify products that users are interested in in order to achieve precise marketing to users is extremely important.

[0003] In existing applications, the number of times a user views product S is usually used to determine their desire to place an order for product S. The more times a user views a product, the greater their desire to place an order. However, this method cannot accurately predict a user's desire to place an order for product S, because even if a user views product S many times, they may still not place an order. Summary of the Invention

[0004] This application provides a data processing method, apparatus, computer equipment, and storage medium that can improve the accuracy of the predicted order index of the target object for the recommended object.

[0005] This application provides a data processing method, which includes:

[0006] Obtain object behavior data of the target object for the recommended object in multiple time intervals;

[0007] Based on the object behavior data of the target object towards the recommended object in each time interval, generate object behavior features of the target object towards the recommended object in each time interval;

[0008] Based on the target object's behavior characteristics towards the recommended object in each time interval, predict the target object's order index towards the recommended object.

[0009] This application provides a data processing apparatus, which includes:

[0010] The acquisition module is used to acquire object behavior data of the target object towards the recommended object in multiple time intervals;

[0011] The generation module is used to generate object behavior features of the target object in each time interval based on the object behavior data of the target object in each time interval in relation to the recommended object.

[0012] The prediction module is used to predict the order index of the target object for the recommended object based on the object behavior characteristics of the target object for the recommended object in each time interval.

[0013] Optionally, any one of multiple time intervals can be represented as the target time interval; the recommended object belongs to the object in the recommendation platform;

[0014] The acquisition module obtains object behavior data of the target object for the recommended object in multiple time intervals in the following ways:

[0015] Retrieve multiple sub-recommendation objects contained within the recommendation object;

[0016] Retrieve the associated objects of the recommended object on the recommendation platform, and retrieve the recommendation activities of the objects associated with the recommended object on the recommendation platform;

[0017] Obtain the target object's first browsing behavior data on the recommendation platform for the recommended object and related objects within the target time interval;

[0018] Obtain the second browsing behavior data of the target object on the recommendation platform for multiple recommended sub-objects within the target time interval;

[0019] Obtain third-party browsing behavior data of the target object on the recommendation platform within the target time interval;

[0020] Based on the first browsing behavior data, the second browsing behavior data, and the third browsing behavior data, determine the target object's object behavior data towards the recommended object within the target time interval.

[0021] Optionally, the first browsing behavior data includes the number of times and the duration of browsing the first object description page of the target object for the recommended object and the associated object respectively; the second browsing behavior data includes the number of times and the duration of browsing the second object description page of the target object for multiple recommended sub-objects respectively; the third browsing behavior data includes the number of times and the duration of browsing the activity page of the target object for the object recommendation activity; the first object description page, the second object description page and the activity page belong to the recommendation platform.

[0022] Optionally, the generation module generates object behavior features of the target object relative to the recommended object in each time interval based on the object behavior data of the target object relative to the recommended object in each time interval, including:

[0023] Generate a first feature vector for the target object based on the first browsing behavior data, targeting the recommended object and the associated object;

[0024] Generate a second feature vector for the target object targeting multiple recommendation sub-objects based on the second browsing behavior data;

[0025] Generate a third feature vector for the target object's recommendation activity based on the third browsing behavior data;

[0026] Based on the first feature vector, the second feature vector, and the third feature vector, generate the object behavior features of the target object against the recommended object within the target time interval.

[0027] Optionally, the generation module generates object behavior features of the target object relative to the recommended object within the target time interval based on the first feature vector, the second feature vector, and the third feature vector, including:

[0028] Retrieve multiple browsing pages of the target object within the target time interval on the recommendation platform;

[0029] The browsing pages other than the first object description page, the second object description page, and the activity page are identified as associated browsing pages.

[0030] Extract browsing keywords from associated browsing pages and generate a fourth feature vector based on the browsing keywords;

[0031] The first feature vector, the second feature vector, the third feature vector, and the fourth feature vector are concatenated to obtain the object behavior features of the target object in response to the recommended object within the target time interval.

[0032] Optionally, the prediction module predicts the order placement index of the target object in relation to the recommended object based on the target object's object behavior characteristics in each time interval, including:

[0033] The object behavior features of the target object for each time interval are concatenated to obtain the concatenated behavior features.

[0034] The recommendation model is invoked to predict the order index of the target object for the recommended object based on the splicing behavior characteristics.

[0035] Optionally, the above-mentioned device is also used for:

[0036] Acquire sample data; the sample data includes the sample splicing behavior features of the sample object for the sample recommendation object and the order placement behavior tags of the sample object for the sample recommendation object. The order placement behavior tags are either the no-order behavior tags or the ordered behavior tags.

[0037] The initial recommendation model is invoked to predict the sample order index for the sample recommended object based on the sample splicing behavior characteristics.

[0038] The model parameters of the initial recommendation model are updated based on the sample order index and order behavior labels to obtain the recommendation model.

[0039] This application provides a computer device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the method of this application.

[0040] This application provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described above.

[0041] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative embodiments described above.

[0042] This application acquires object behavior data of a target object towards a recommended object within multiple time intervals; based on the object behavior data of the target object towards the recommended object in each time interval, it generates object behavior features of the target object towards the recommended object in each time interval; based on the object behavior features of the target object towards the recommended object in each time interval, it predicts the order index of the target object towards the recommended object. Therefore, the method proposed in this application can acquire object behavior features of the target object towards the recommended object according to time partitions, taking into account the object behavior of the target object towards the recommended object in each time partition, thus making the predicted order index of the target object towards the recommended object more accurate. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application;

[0045] Figure 2 This is a schematic diagram of a data prediction scenario provided in this application;

[0046] Figure 3 This is a flowchart illustrating a data processing method provided in this application;

[0047] Figure 4 This is a schematic diagram illustrating a scenario for obtaining feature vectors provided in this application;

[0048] Figure 5This is a schematic diagram of a data prediction scenario provided in this application;

[0049] Figure 6 This is a schematic diagram of a data processing scenario provided in this application;

[0050] Figure 7 This is a flowchart illustrating a data processing method provided in this application;

[0051] Figure 8 This is a schematic diagram of the structure of a data processing device provided in this application;

[0052] Figure 9 This is a schematic diagram of the structure of a computer device provided in this application. Detailed Implementation

[0053] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] This application relates to technologies related to artificial intelligence (AI). AI is the theory, methods, technology, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities.

[0055] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0056] This application primarily concerns machine learning within artificial intelligence. Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0057] The machine learning involved in this application mainly refers to how to train a recommendation model to predict the order index of a target object for a recommended object. For details, please refer to the following... Figure 3 The description in the corresponding embodiments.

[0058] This application relates to cloud technology. Cloud technology refers to a managed technology that unifies hardware, software, network, and other resources within a wide area network (WAN) or local area network (LAN) to enable data computation, storage, processing, and sharing.

[0059] Cloud technology is a collective term for network technology, information technology, integration technology, management platform technology, and application technology applied to the cloud computing business model. It can form resource pools, providing flexible and convenient on-demand access. Cloud computing technology will become a crucial support. Backend services of technical network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of the internet industry, every item may have its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data will all require robust system support, which can only be achieved through cloud computing.

[0060] The cloud technology mentioned in this application can refer to the ability to store or retrieve object behavior data of a target object for a recommended object through the "cloud".

[0061] First, it should be noted that this application may display a prompt interface or pop-up window before and during the collection of user-related data (such as user data on the recommendation platform, including object behavior data and sample data of the target object). This prompt interface or pop-up window is used to inform the user that their data is being collected. This application will only begin the steps of collecting user-related data after receiving confirmation from the user regarding the prompt interface or pop-up window; otherwise (i.e., without receiving confirmation from the user), the steps of collecting user-related data will end, meaning no user-related data will be collected. In other words, all user data collected in this application is collected with the user's consent and authorization, and the collection, use, and processing of related user data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0062] Please see Figure 1 , Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application. Figure 1 As shown, the network architecture may include server 200 and a cluster of terminal devices. The cluster of terminal devices may include one or more terminal devices; the number of terminal devices is not limited here. Figure 1 As shown, the multiple terminal devices may specifically include terminal device 100a, terminal device 101a, terminal device 102a, ..., terminal device 103a; as Figure 1 As shown, terminal devices 100a, 101a, 102a, ..., 103a can all connect to server 200 via the network, so that each terminal device can interact with server 200 via the network.

[0063] like Figure 1 The server 200 shown can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal device can be a smartphone, tablet, laptop, desktop computer, smart TV, in-vehicle terminal, or other smart terminal. The following description uses the communication between terminal device 100a and server 200 as an example to illustrate the specific implementation of this application.

[0064] Please see also Figure 2 , Figure 2 This is a schematic diagram illustrating a data prediction scenario provided in this application. For example... Figure 2As shown, the target user can possess terminal device 100a, and can access the recommendation platform through terminal device 100a. Therefore, the target user can be any user on the recommendation platform who has not placed an order for the recommended product. The recommendation platform can be any platform that allows browsing and ordering products (i.e., purchasing). The recommendation platform can be a web-based platform or an application-based platform, and server 200 can be the backend server of the recommendation platform. Therefore, this application aims to predict the target user's order index for the recommended product. The higher the order index, the more likely the target user is to purchase the recommended product; the lower the order index, the less likely the target user is to purchase the recommended product. Please refer to the following description for details.

[0065] Therefore, when the target object browses the corresponding page in the recommendation platform of the terminal device 100a, the server 200 can obtain the page browsed by the target object. That is, the server 200 can know which pages the server 200 has browsed and the relevant browsing parameters of the target object for the browsed pages (such as the number of page views and browsing duration).

[0066] Furthermore, server 200 can obtain object behavior data of the target object towards the recommended object according to time partitions and based on the pages viewed by the target object on the recommendation platform. Here, it is assumed that this can include five time partitions: time partition 1, time partition 2, time partition 3, time partition 4, and time partition 5. Therefore, server 200 can obtain object behavior data 1 of the target object towards the recommended object in time partition 1, object behavior data 2 of the target object towards the recommended object in time partition 2, object behavior data 3 of the target object towards the recommended object in time partition 3, object behavior data 4 of the target object towards the recommended object in time partition 4, and object behavior data 5 of the target object towards the recommended object in time partition 5.

[0067] Furthermore, the server 200 can generate the following based on the object behavior data 1 obtained above: object behavior feature 1 of the target object in time partition 1 for the recommended object; object behavior feature 2 of the target object in time partition 2 for the recommended object based on the object behavior data 2 obtained above; object behavior feature 3 of the target object in time partition 3 for the recommended object based on the object behavior data 3 obtained above; object behavior feature 4 of the target object in time partition 4 for the recommended object based on the object behavior data 4 obtained above; and object behavior feature 5 of the target object in time partition 5 for the recommended object based on the object behavior data 5 obtained above.

[0068] Finally, the server can concatenate the generated object behavior features 1, 2, 3, 4, and 5 to obtain concatenated behavior features, and input these concatenated behavior features into the recommendation model to predict the order index of the target object for the recommended object.

[0069] By using the method provided in this application, and by obtaining the object behavior characteristics of the target object in each time partition for the recommended object, the order index of the target object for the recommended object predicted by the object behavior characteristics can be more accurate.

[0070] Please see Figure 3 , Figure 3 This is a flowchart illustrating a data processing method provided in this application. The execution entity in the embodiments of this application can be a single computer device or a cluster of computer devices. The computer device can be a server or a terminal device. Therefore, the execution entity in the embodiments of this application can be a server, a terminal device, or a combination of both. Here, we will use a server as the execution entity in this application as an example for explanation. Figure 3 As shown, the method may include:

[0071] Step S101: Obtain object behavior data of the target object for the recommended object in multiple time intervals.

[0072] Optionally, the target audience can be any user on the recommendation platform. The recommendation platform can be any platform that allows product recommendations or order placement (i.e., purchasing), such as any e-commerce platform or any company's official website for related products. The recommendation platform can exist as a webpage or as an app (application). The server can be the backend server of the recommendation platform, and the target audience can access the recommendation platform through their own terminal devices.

[0073] Therefore, it can be understood that the recommended object is an object in the recommendation platform. The recommended object can be any product in the recommendation platform that needs to be recommended for ordering, such as a cloud server in the recommendation platform.

[0074] The number of time intervals and the specific number of time intervals can be determined according to the actual application scenario. The time interval can be based on the current time (such as the time when the object behavior data is obtained), that is, the multiple time intervals can be multiple time periods before the current time.

[0075] For example, multiple time intervals can include time intervals within 3 days (i.e., the most recent 3 days before the current time), time intervals within 4-7 days (i.e., the 4th to 7th day before the current time), time intervals within 8-15 days (i.e., the 8th to 15th day before the current time), time intervals within 16-30 days (i.e., the 16th to 30th day before the current time), and time intervals beyond 30 days (i.e., the time beyond the 30th day before the current time).

[0076] Therefore, the server can obtain object behavior data of the target object in response to the recommended object in multiple time intervals. This process can be as follows:

[0077] In this context, any one of the multiple time intervals can be represented as the target time interval. Therefore, this example illustrates how to obtain the object behavior data of the target object towards the recommended object within the target time interval. It can be seen that the principle of obtaining the object behavior data of the target object towards the recommended object in each time interval and the object behavior characteristics described below are the same, as described below.

[0078] The server can obtain multiple sub-objects of a recommended object. These sub-objects can belong to the recommended object but are not of the same model. For example, the recommended object can be a CVM (cloud server), and the models of this cloud server can include 9 different models: S2.SMALL1, S2.SMALL2, S5.SMALL2, S4.SMALL1, GN2.7XLARGE56, S3.SMALL1, SA2.SMALL1, S2.MEDIUM4, and SA1.SMALL1. Therefore, one model corresponds to one sub-object, resulting in a total of 9 sub-objects.

[0079] The server can also retrieve associated objects of the recommended object on the recommendation platform. These associated objects typically refer to objects that have an order-related connection with the recommended object. These associated objects can also be products within the recommendation platform; for example, other products that are typically purchased when buying a cloud server, or vice versa, can all be considered associated products (i.e., associated objects) of the cloud server. The number of associated objects of the recommended object can be determined based on the actual application scenario and is not limited. Alternatively, associated objects can also be components (such as parts or accessories) of the recommended object.

[0080] For example, if the recommended object is a cloud server CVM, the associated objects of the recommended object may include dnspod (an intelligent DNS (Domain Name Server) product), cos (a product that provides object storage), lighthouse (an open-source automation tool), GPU (Graphics Processing Unit), cpm (a display pay advertising service), CDN (Content Delivery Network), and CBS (a block-level network storage service), for a total of 7 associated objects.

[0081] The server can also obtain the recommended object's associated recommended activities on the recommendation platform. These recommended activities can be related to discount activities or promotional activities that the recommended object participates in, such as promotional activities like Singles' Day (11.11).

[0082] Therefore, the server can obtain the browsing behavior data of the target object on the recommendation platform for recommended objects and related objects within the target time interval. This browsing behavior data can be referred to as the first browsing behavior data. One object (such as a recommended object or a related object) corresponds to one first browsing behavior data. Different related objects have their own different first browsing behavior data. For an explanation of the first browsing behavior data, please refer to the following description.

[0083] In this context, the product detail pages of both recommended objects and associated objects on the recommendation platform can be referred to as the first object description pages. Each object in the recommended objects and associated objects has its own first object description page, and the number of these first object description pages can be determined according to the actual application scenario.

[0084] The first object description page of a recommended object may not include product detail pages specific to each recommended sub-object. For example, the first object description page may include the main recommendation page of the recommended object, which may contain descriptive information related to all recommended sub-objects, rather than descriptive information mainly targeting one of the multiple recommended sub-objects. In other words, if the recommended object is a cloud server, the first object description page may be a description page for the cloud server as a whole, rather than a description page for a specific model of the cloud server. Therefore, the first object description page of a recommended object can be a page describing the recommended object without selecting a specific model.

[0085] The first object description page for related objects can include all pages describing related objects, with each related object having its own first object description page. For example, the first object description page for related objects could be the product details page or purchase page (i.e., the order placement page) of the related object on the recommendation platform.

[0086] The target object's first browsing behavior data for the recommended object within the target time interval may include the number of times the target object browses the first object description page of the recommended object within the target time interval and the browsing duration. The number of browsing indicates the total number of times the target object clicks (i.e., opens or accesses) the first object description page of the recommended object within the target time interval, and the browsing duration indicates the total time the target object stays on the first object description page after each click (i.e., opening) within the target time interval (i.e., the sum of the stay times for each click), which is the total duration for the first object description page to be displayed at the top of the screen (such as the screen of the target object's terminal device) within the target time interval.

[0087] Similarly, the target object's first browsing behavior data for the associated object within the target time interval can include the number of times the target object browses the first object description page of the associated object within the target time interval and the browsing duration. The number of browsing indicates the total number of times the target object clicks (i.e., opens or accesses) the first object description page of the associated object within the target time interval, and the browsing duration indicates the total time the target object stays on the first object description page after each click (i.e., opening) of the first object description page of the associated object within the target time interval (i.e., the sum of the stay times each time), which is the total duration for the first object description page to be displayed at the top of the screen (such as the screen of the target object's terminal device) within the target time interval.

[0088] Furthermore, the server can also obtain browsing behavior data of the target object towards multiple recommended sub-objects within the target time interval. This browsing behavior data can be referred to as second browsing behavior data, and each recommended sub-object has one set of second browsing behavior data. It can be understood that the second browsing behavior data of any recommended sub-object can include the number of times the target object viewed the object description page (which can be called the second object description page) of that recommended sub-object within the target time interval, and the browsing duration. Similarly, the number of views represents the total number of times the target object clicked (i.e., opened or accessed) the second object description page of that recommended sub-object within the target time interval, and the browsing duration represents the total time the target object stayed on the second object description page after each click (i.e., opening) of the second object description page within the target time interval (i.e., the sum of the dwell times for each instance), which is the total duration for the second object description page to be displayed at the top of the screen (such as the screen of the target object's terminal device) within the target time interval.

[0089] In this context, the second object description page of any recommended sub-object can be a product details page used only or primarily to describe the recommended sub-object. For example, if a recommended sub-object is a certain model of a recommended product, then the second object description page of the recommended sub-object can be a page used to describe that model of product, such as the product details page or the purchase page for that model of product.

[0090] In addition, the server can also obtain browsing behavior data of the target object regarding the recommended activities of the recommended object within the target time interval. This browsing behavior data can be referred to as third browsing behavior data. This third browsing behavior data can include the number of times the target object views the activity page of the recommended activity for the object within the target time interval and the browsing duration. Similarly, the number of views represents the total number of times the target object clicks (i.e., opens or visits) the activity page of the recommended activity for the object within the target time interval, and the browsing duration represents the total time the target object stays on the activity page after each click (i.e., opening) of the activity page of the recommended activity for the object within the target time interval (i.e., the sum of the dwell times of each click), that is, the total time that the activity page is displayed at the top of the screen (such as the screen of the target object's terminal device) within the target time interval.

[0091] The activity page for an object recommendation activity can be a page that describes the object recommendation activity, such as a page that describes the activity rules or displays the products involved in the activity.

[0092] The first object description page, the second object description page, and the activity page mentioned above are all pages within the recommendation platform. The server can distinguish which page a target object is visiting by detecting the links accessed by the target object within the recommendation platform (which the target object can access through a terminal device). A page can have corresponding links.

[0093] For example, if a link to a page accessed by a target object on the recommendation platform contains the `act` (activity) field (which can be customized), then that page can be considered the activity page of the object's recommendation activity. If a link to a page accessed by a target object on the recommendation platform contains model information of a recommended object (which can be customized), then that page can be considered the second object description page of the recommended sub-object for that model. If a link to a page accessed by a target object on the recommendation platform contains the identifier information of a recommended object (which can be customized), then that page can be considered the first object description page of the recommended object. If a link to a page accessed by a target object on the recommendation platform contains the identifier information of an associated object (which can be customized), then that page can be considered the first object description page of the associated object.

[0094] Therefore, the server can use the first browsing behavior data for the recommended object and associated object, the second browsing behavior data for multiple recommended sub-objects, and the third browsing behavior data for the object recommendation activity as the object behavior data of the target object for the recommended object within the target time interval.

[0095] Therefore, the server can obtain the target object's behavior data for the recommended object in each time interval by following the above process.

[0096] Step S102: Based on the object behavior data of the target object towards the recommended object in each time interval, generate object behavior features of the target object towards the recommended object in each time interval.

[0097] Optionally, this explanation will still take calculating the target object's behavioral characteristics towards the recommended object within the target time interval as an example. It can be seen that the process of calculating the target object's behavioral characteristics towards the recommended object within each time interval is the same, as described below.

[0098] The server can generate a corresponding feature vector based on the target object's first browsing behavior data for recommended and associated objects within the target time interval. This feature vector can be called the first feature vector. One of the recommended and associated objects corresponds to a feature value in the first feature vector, which can be an element (i.e., a value) in the first feature vector.

[0099] The process by which the server generates feature values ​​for a recommended object based on the target object's first browsing behavior data can be as follows: First, the initial feature value can be calculated using the formula a*n+b, where 'a' represents the increase in value per browsing, such as a = 10; 'n' represents the number of browsing sessions, such as the number of browsing sessions in the target object's first browsing behavior data (i.e., the number of times the target object browses the first object description page of the recommended object within the target time interval); and 'b' represents the browsing duration, such as the browsing duration in the target object's first browsing behavior data. The unit of 'b' can be seconds. For example, if a = 10, the number of browsing sessions in the target object's first browsing behavior data is 5, and the browsing duration is 180 seconds, then the initial feature value corresponding to the recommended object is 10*5+180 = 230.

[0100] Then, the initial feature values ​​of the recommended objects need to be normalized to obtain the feature values ​​corresponding to the recommended objects. The formula for normalizing the initial feature values ​​can be shown in the following formula (1):

[0101]

[0102] Where x represents the initial feature value, such as x can be substituted into the initial feature value of the recommended object, x′ represents the normalized value, that is, if the initial feature value corresponding to the recommended object is substituted into x in formula (1), then x′ is equal to the calculated feature value corresponding to the recommended object, and min represents taking The minimum value among 1 and 2.

[0103] Similarly, the process by which the server generates the feature value corresponding to the associated object based on the target object's first browsing behavior data for the associated object can be as follows: First, the initial feature value can be calculated using the formula a*n+b, where a represents the value that can be increased with each browsing, n can represent the number of browsing sessions (e.g., the number of browsing sessions by the target object for the first object description page of the associated object within the target time interval), and b represents the browsing duration (e.g., the browsing duration in the first browsing behavior data of the associated object). The unit of b can be seconds. For example, if a equals 10, the number of browsing sessions in the first browsing behavior data of the associated object is 10, and the browsing duration is 60 seconds, then the initial feature value corresponding to the associated object is 10*10+60, which equals 160.

[0104] Furthermore, it is necessary to normalize the initial feature values ​​of the associated objects to obtain the feature values ​​corresponding to the associated objects. The normalization of the initial feature values ​​of the associated objects can also be performed using the formula (1) above. If x in formula (1) can be substituted with the initial feature values ​​of the associated objects, then x′ will be equal to the calculated feature values ​​corresponding to the associated objects.

[0105] The first feature vector can be generated by combining the feature values ​​of the recommended object and the associated objects. For example, if the feature value of the recommended object is 0.6, and there are 5 associated objects with feature values ​​of 0.3, 0.5, 0.1, 0.3, and 0.7 respectively, then the first feature vector can be (0.6, 0.3, 0.5, 0.1, 0.3, 0.7).

[0106] Furthermore, the server can generate corresponding feature vectors based on the second browsing behavior data corresponding to each of the aforementioned multiple recommendation sub-objects. These feature vectors can be referred to as second feature vectors. Each recommendation sub-object corresponds to a feature value in the second feature vector, which can be an element (i.e., a value) within the second feature vector.

[0107] This section describes the process of calculating the feature value corresponding to a recommendation sub-object (e.g., recommendation sub-object t, which can be any one of multiple recommendation sub-objects). It can be understood that the process of calculating the feature value corresponding to each recommendation sub-object is the same.

[0108] Optionally, the process of calculating the feature value corresponding to the recommended sub-object t based on the second browsing behavior data of the recommended sub-object t can be as follows: First, the initial feature value corresponding to the recommended sub-object t can still be calculated using the formula a*n+b mentioned above, where a represents the value that can be increased with each browsing, n can represent the number of browsing, such as substituting the number of browsing in the second browsing behavior data of the recommended sub-object t (i.e., the number of times the target object browses the second object description page of the recommended sub-object t within the target time interval), and b represents the browsing duration, such as substituting the browsing duration in the second browsing behavior data of the recommended sub-object t, where the unit of b can be seconds. For example, if a equals 10, the number of browsing in the second browsing behavior data of the recommended sub-object t is equal to 1, and the browsing duration is 10 seconds, then the initial feature value corresponding to the recommended sub-object t is equal to 10*1+10, which equals 20.

[0109] Furthermore, it is necessary to normalize the initial feature values ​​of the recommended sub-object t to obtain the feature values ​​corresponding to the recommended sub-object t. The normalization of the initial feature values ​​of the recommended sub-object t can also be performed using the formula (1) above. If x in formula (1) can be substituted with the initial feature values ​​of the recommended sub-object t, then x′ is equal to the calculated feature values ​​corresponding to the recommended sub-object t.

[0110] The above process allows us to calculate the feature value corresponding to each recommendation sub-object, and then generate the second feature vector using these feature values. For example, if there are 5 recommendation sub-objects with feature values ​​of 0.1, 0.2, 0.3, 0.5, and 0.6 respectively, then the second feature vector would be (0.1, 0.2, 0.3, 0.5, 0.6).

[0111] Furthermore, the server can also generate corresponding feature vectors based on the third browsing behavior data corresponding to the object recommendation activity. This feature vector can be called a third feature vector, which may have only one feature value, namely the feature value corresponding to the object recommendation activity.

[0112] The process of generating the third feature vector corresponding to the object recommendation activity using the third browsing behavior data can be as follows: First, the initial feature value corresponding to the object recommendation activity can still be calculated using the formula a*n+b mentioned above. Here, 'a' represents the value that can be increased with each browsing session, 'n' can represent the number of browsing sessions (e.g., the number of times the target object browses the activity page of the object recommendation activity within the target time interval), and 'b' represents the browsing duration (e.g., the browsing duration in the second browsing behavior data of the object recommendation activity). The unit of 'b' can be seconds. For example, if 'a' equals 10, the number of browsing sessions in the third browsing behavior data of the object recommendation activity is 5, and the browsing duration is 20 seconds, then the initial feature value corresponding to the object recommendation activity is 10*5+20, which equals 70.

[0113] Furthermore, it is necessary to normalize the initial feature values ​​of the object recommendation activity to obtain the corresponding feature values. The normalization of the initial feature values ​​of the object recommendation activity can also be performed using the formula (1) above. If x in formula (1) can be substituted with the initial feature values ​​of the object recommendation activity, then x′ will be equal to the calculated feature values ​​corresponding to the object recommendation activity.

[0114] Then, a third feature vector can be generated based on the feature value corresponding to the object recommendation activity. For example, if the feature value corresponding to the object recommendation activity is 0.3, then the third feature vector could be (0.3).

[0115] The server can directly concatenate the first, second, and third feature vectors to obtain the target object's behavioral features (also part of the feature vector) in relation to the recommended object within the target time interval. For example, if the first feature vector is (0.1, 0.2, 0.3), the second feature vector is (0.4, 0.5, 0.6), and the third feature vector is (0.7), then the object behavioral features obtained by concatenating the first, second, and third feature vectors would be (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7).

[0116] Alternatively, the server can generate a corresponding feature vector based on the keywords in the pages viewed by the target object on the recommendation platform. This feature vector can be called the fourth feature vector. Then, the first, second, third, and fourth feature vectors can be concatenated to obtain the target object's behavioral features towards the recommended object within the target time interval.

[0117] The process of generating the fourth feature vector can be as follows: the server can obtain multiple browsing pages of the target object on the recommendation platform within the target time interval. These multiple browsing pages can include all the pages that the target object has browsed on the recommendation platform within the target time interval.

[0118] Therefore, the browsing pages other than the first object description page (including the object description pages of recommended objects and associated objects), the second object description page (including the object description pages of each recommended sub-object), and the activity page of the object recommendation activity can be referred to as associated browsing pages. The number of associated browsing pages is determined according to the actual application scenario.

[0119] The server can extract keywords from the associated browsing pages; these keywords can be called browsing keywords. The technique for extracting keywords from associated browsing pages can be selected based on the actual application scenario. Here, the Jieba library (an open-source Python third-party library) can be used to extract browsing keywords from associated browsing pages. During extraction, the weight coefficient of each word can be calculated based on its frequency and other characteristics in the associated browsing pages. This weight coefficient indicates the importance of each word; the larger the weight coefficient, the more important it is, and vice versa. Then, the server selects the M words with the highest weight coefficients as the extracted browsing keywords, where M is a positive integer. The specific value of M can be determined based on the actual application scenario; for example, M could be equal to 3.

[0120] The above process extracts multiple browsing keywords. The server can then input these keywords into the BERT model (a word vector model) to generate a word vector for each keyword, with one word vector per keyword. Furthermore, the server can average the word vectors of each keyword to obtain an average word vector. Since the word vectors of each keyword have the same dimension (e.g., 1*768), each keyword's word vector contains feature values ​​at 768 positions. Therefore, averaging the word vectors of each keyword involves averaging the feature values ​​at the same positions across all keyword vectors, using this average as the feature value in the average word vector. Thus, the dimension of the average word vector is the same as the dimension of any single keyword's word vector, also 1*768.

[0121] Furthermore, the server can reduce the dimensionality of the average word vector to obtain the final fourth feature vector. For example, the dimensionality reduction of the average word vector can be achieved by inputting it into a fully connected layer. For instance, if the dimension of the average word vector is 1*768, then the dimension of the fourth feature vector obtained after dimensionality reduction can be 1*5, and the dimension of the feature matrix in the fully connected layer used for dimensionality reduction can be 768*5.

[0122] After obtaining the fourth feature vector, the server can concatenate the first, second, third, and fourth feature vectors to obtain the target object's behavioral features towards the recommended object within the target time interval.

[0123] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating a scenario for obtaining feature vectors provided in this application. For example... Figure 4 As shown, this describes the process of obtaining object behavior features within a time partition. The 1-dimensional feature of the object recommendation activity is the third feature vector mentioned above. The 8-dimensional features of the 7 associated objects and 1 recommended object (i.e., the 8-dimensional features of the 8 products) are the first feature vector mentioned above. The 9-dimensional features of the 9 device models (i.e., the 9 recommended sub-objects) are the second feature vector mentioned above. Therefore, the 5-dimensional feature of browsing keywords is the fourth feature vector mentioned above. Thus, the first, second, and third feature vectors can be concatenated, and then concatenated with the fourth feature vector to obtain the object behavior features (23-dimensional in this case) within the corresponding time partition.

[0124] Step S103: Based on the target object's object behavior characteristics towards the recommended object in each time interval, predict the target object's order index towards the recommended object.

[0125] Optionally, the server can sequentially concatenate the object behavior features of the target object for the recommended object in each time interval. The resulting feature vector can be called the concatenated behavior feature. For example, multiple time intervals include time interval 1, time interval 2, and time interval 3. The object behavior features corresponding to time interval 1 are (0.1, 0.2, 0.3, 0.4, 0.5), the object behavior features corresponding to time interval 2 are (0.2, 0.3, 0.4, 0.5, 0.6), and the object behavior features corresponding to time interval 3 are (0.3, 0.4, 0.5, 0.6, 0.7). Then the concatenated behavior feature could be (0.1, 0.2, 0.3, 0.4, 0.5, 0.2, 0.3, 0.4, 0.5, 0.6, 0.3, 0.4, 0.5, 0.6, 0.7).

[0126] The server can input the splicing behavior feature into a recommendation model, which can be composed of multiple fully connected neural networks (referred to as fully connected layers). Then, the server can call the recommendation model to predict the probability (range 0-1) of the target object placing an order for the recommended object based on the splicing behavior feature. This order probability can be used as the target object's order index for the recommended object. The order index indicates the strength of the predicted purchase intention of the target object for the recommended object; the larger the order index, the stronger the target object's purchase intention for the recommended object; the smaller the order index, the weaker the target object's purchase intention for the recommended object. The recommendation model is a pre-trained model that can be used to predict the order probability. The specific training process of the recommendation model can be found below. Figure 7 The description in the corresponding embodiment.

[0127] If the recommendation model is composed of three fully connected layers connected sequentially, then the order index P of the target object for the recommended object can be expressed as shown in the following formula (2):

[0128] P=sigmoid(W3·R(W2·R(W1·Y))) (2)

[0129] Where sigmoid represents the sigmoid function, W3 represents the network parameters of the third fully connected layer, R represents ReLU (an activation function), W2 represents the network parameters of the second fully connected layer, W1 represents the network parameters of the first fully connected layer, and Y represents the splicing behavior feature.

[0130] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating a data prediction scenario provided in this application. For example... Figure 5As shown, firstly, the server can categorize the target object's behavioral data (i.e., website behavioral data) on the official website (such as a recommendation platform) into four types: activity-based (such as the aforementioned object recommendation activity), product-based (such as the aforementioned recommended object and its associated objects), device-based (such as the aforementioned recommended sub-objects), and keyword-based (such as the aforementioned browsing keywords). Then, it retrieves the features (i.e., object behavioral features) for each time bucket (i.e., time partition) according to these four categories. Here, time buckets include buckets within 3 days, buckets from 3 to 7 days, buckets from 7 to 15 days, and buckets from 15 to 30 days. The system uses time-based bucketing, including bucketing for periods beyond 30 days. Therefore, the server can obtain features within 3 days (i.e., the target object's behavioral characteristics towards the recommended object within 3 days), features from 4 to 7 days (i.e., the target object's behavioral characteristics towards the recommended object in the previous 4 to 7 days), features from 8 to 15 days (i.e., the target object's behavioral characteristics towards the recommended object in the previous 8 to 15 days), features from 16 to 30 days (i.e., the target object's behavioral characteristics towards the recommended object in the previous 16 to 30 days), and features beyond 31 days (i.e., the target object's behavioral characteristics towards the recommended object in the previous 31 days or more).

[0131] The server can obtain a recommendation model that has been pre-trained using the object behavior features of sample objects in each time partition. This recommendation model has learned the browsing behavior features of sample objects (including sample objects that have purchased sample recommended objects and sample objects that have not purchased sample recommended objects) on relevant pages (which can include all browsed pages) for sample recommended objects in each time partition during the training process. Therefore, by calling this recommendation model, the order index of the target object for the recommended object can be predicted based on the features of the target object in each time bucket.

[0132] In this application, the object behavior characteristics of the target object for the recommended object are obtained by dividing the time into time partitions. The impact of time partitions at different distances from the current time on the target user's willingness to purchase the recommended object is taken into account. Therefore, the order index of the target object for the recommended object predicted by the object behavior characteristics corresponding to each time partition is more accurate.

[0133] Optionally, the target audience detected in this application can be users who have not yet purchased the recommended product on the recommendation platform. If the target audience's order index for the recommended product is greater than a certain threshold (which can be set by the user), the relevant telemarketing personnel can be notified to conduct a precise new user acquisition campaign (i.e., recommend the target audience to purchase the recommended product). This new user acquisition campaign can be achieved through telephone marketing or SMS marketing. Therefore, the method provided in this application can effectively acquire new users (i.e., recommend users to purchase products they have not yet purchased).

[0134] Please see Figure 6 , Figure 6 This is a schematic diagram of a data processing scenario provided in this application. For example... Figure 6 As shown, the server can scan the database daily. The scanned databases include the purchase order database and the official website behavior database. The purchase order database records whether users have purchased the recommended object. Therefore, users who have not purchased the recommended object, as determined by the purchase order database, can be identified as target users.

[0135] Subsequently, the server can obtain the target object's behavior data towards the recommended object from the official website's behavior database (such as the target object's behavior data towards the recommended object in various time intervals). Then, using this behavior data, the server can generate the target object's behavior features towards the recommended object in various time intervals. By inputting these behavior features into a deep learning model (such as the recommendation model mentioned above), the server can identify target objects whose order index is greater than the order index threshold (which can be set by the user). These target objects are users with a strong willingness to purchase the recommended object, and marketing strategies can then be implemented for these users.

[0136] This application acquires object behavior data of a target object towards a recommended object within multiple time intervals; based on the object behavior data of the target object towards the recommended object in each time interval, it generates object behavior features of the target object towards the recommended object in each time interval; based on the object behavior features of the target object towards the recommended object in each time interval, it predicts the order index of the target object towards the recommended object. Therefore, the method proposed in this application can acquire object behavior features of the target object towards the recommended object according to time partitions, taking into account the object behavior of the target object towards the recommended object in each time partition, thus making the predicted order index of the target object towards the recommended object more accurate.

[0137] In addition, this application also conducted a related control experiment. The experimental group consisted of users identified by this application as having a strong purchase intention (i.e., target users whose order index was greater than the order index threshold), while the control group consisted of users active on the official website (e.g., users who browsed products frequently but had not yet purchased the recommended items). The experiment lasted for 15 days, and the results are as follows: the order conversion rate (number of user purchase orders / number of users) in the experimental group was 16%, while the order conversion rate in the control group was 2%. This clearly demonstrates that the method provided in this application significantly improves the effectiveness of new user acquisition activities (i.e., firstly, users with a strong purchase intention for the recommended items are detected using the method provided in this application, and then targeted marketing is achieved to these users, leading them to purchase the recommended items).

[0138] Please see Figure 7 , Figure 7This is a flowchart illustrating a data processing method provided in this application. The embodiments of this application mainly describe the training process of the aforementioned recommendation model; the execution entity in these embodiments can also be a server. Figure 7 As shown, the method includes:

[0139] Step S201: Obtain sample data; the sample data includes the sample splicing behavior features of the sample object for the sample recommendation object and the order placement behavior label of the sample object for the sample recommendation object. The order placement behavior label is either the no-order behavior label or the ordered behavior label.

[0140] Optionally, the multiple time partitions in this application are determined based on the time at which the corresponding data (such as the sample data here and the object behavior data of the target object mentioned above) is acquired. For example, multiple time intervals may include time intervals within 3 days, time intervals within 4-7 days, and time intervals within 8-15 days. The multiple time partitions corresponding to the sample data can be multiple time intervals determined based on the time at which the sample data is acquired (e.g., time 1). For example, these multiple time intervals may include time intervals within 3 days after time 1, time intervals from the 4th to the 7th day after time 1, and time intervals from the 8th to the 15th day after time 1. Similarly, the multiple time intervals corresponding to the object behavior data of the target object (i.e., the time intervals within 3 days after time 1, the time intervals from the 4th to the 7th day after time 1, and the time intervals from the 8th to the 15th day after time 1) are also determined. Figure 3 The multiple time intervals in the corresponding embodiments can be multiple time intervals determined based on the time when the object's behavior data is obtained (such as time 2). For example, the multiple time intervals can include the time interval within 3 days after time 2, the time interval from the 4th day to the 7th day after time 2, and the time interval from the 8th day to the 15th day after time 2.

[0141] The sample data here may include sample splicing behavior features of the sample object in relation to the sample recommended object, as well as order placement behavior tags of the sample object in relation to the sample recommended object. These order placement behavior tags are used to indicate whether the sample object has placed an order for the sample recommended object (i.e., whether it has purchased it).

[0142] Therefore, the order placement behavior label can be either a no-order behavior label or an order placement behavior label. The no-order behavior label indicates that the sample object has not purchased the sample recommended object, while the order placement behavior label indicates that the sample object has purchased the sample recommended object.

[0143] The process of obtaining the sample splicing behavior features of the sample object towards the sample recommendation object is the same as the process of obtaining the splicing behavior features of the target object towards the recommendation object. That is, the sample splicing behavior features are generated based on the object behavior data of the sample object towards the sample recommendation object in various time intervals. The sample recommendation object can be the aforementioned recommendation object, or it can be other related products. The sample object can be a user on the recommendation platform other than the target object.

[0144] Step S202: Call the initial recommendation model to predict the sample order index of the sample object for the sample recommendation object based on the sample splicing behavior characteristics.

[0145] Optionally, the recommendation model that has not yet been trained can be referred to as the initial recommendation model. The server can input sample data into the initial recommendation model, and then call the initial recommendation model to predict the probability of a sample object placing an order for a sample recommended object based on the sample concatenation behavior characteristics. This order probability can be called the sample order probability or sample order index. The process of predicting the sample order index is the same as the process of predicting the order index of the target object for the recommended object. This sample order index characterizes the strength of the sample object's purchase intention for the sample recommended object.

[0146] Step S203: Update the model parameters of the initial recommendation model based on the sample order index and order behavior labels to obtain the recommendation model.

[0147] Optionally, the server can update the model parameters of the initial recommendation model using the sample order index and order behavior label. The update principle can be: if the order behavior label in the sample data is the "no order" label, then the model parameters of the initial recommendation model need to be updated to make the predicted sample order index smaller; conversely, if the order behavior label in the sample data is the "order" label, then the model parameters of the initial recommendation model need to be updated to make the predicted sample order index larger.

[0148] The number of sample data points can be determined based on the actual application scenario. Ultimately, the model parameters of the initial recommendation model are continuously updated using a number of sample data points. When the model parameters are updated (e.g., the model parameters converge), the initial recommendation model at this point can be used as the final trained recommendation model. Figure 3 The recommendation model used in the corresponding embodiment is the recommendation model trained from the initial recommendation model.

[0149] The model parameters of the initial recommendation model can be updated using the following loss function (cross-entropy loss function), as shown in formula (3) below:

[0150]

[0151] The initial recommendation model can be trained in rounds using a number of sample data. For example, with 10,000 sample data points, 10 sample data points can be used in each round to train the initial recommendation model together. Therefore, N represents the number of sample data points used in one round of training the initial recommendation model, and n1 represents the number of negative samples in the sample data used in one round of training. Negative samples refer to sample data points whose order placement label is "no order placement". Therefore, the value of i can range from 1 to n1, and the i-th sample data point can refer to the i-th negative sample among n1 negative samples. Thus, y... i Let y represent the order placement behavior label of the i-th negative sample. i This can be represented by a vector, which can contain two feature values: one representing the behavior of not placing an order, and the other representing the behavior of placing an order. If the label for the order behavior is the same as the label for the behavior of not placing an order, then the feature value representing the behavior of not placing an order can be 1, and the feature value representing the behavior of placing an order can be 0. i This represents the sample order index for the predicted i-th negative sample.

[0152] Similarly, n² represents the number of positive samples in the sample data used for one round of training of the initial recommendation model. Positive samples refer to sample data whose order placement label is "order placed". Therefore, the value of j can range from 1 to n², and the j-th sample data can refer to the j-th positive sample out of n² positive samples. Therefore, y j Let y represent the order placement behavior label of the j-th positive sample. j This can be represented by a vector, which can contain two feature values: one representing the behavior of not placing an order, and the other representing the behavior of placing an order. If the order behavior label is "placed," then the feature value representing the behavior of placing an order can be 1, and the feature value representing the behavior of not placing an order can be 0. j This represents the sample order index of the predicted j-th positive sample.

[0153] It should also be noted that, in reality, the number of negative samples is much greater than the number of positive samples, meaning that negative samples account for a large proportion of all sample data. Therefore, the loss function described above can be multiplied by a weight coefficient q for negative samples to weaken the influence of the large number of negative samples and highlight the positive samples, resulting in better training performance. This weight coefficient q can be equal to the ratio of the number of positive samples to the number of negative samples in one round of training; for example, the weight coefficient q can be equal to n2 / n1.

[0154] Please see Figure 8 , Figure 8This is a schematic diagram of a data processing apparatus provided in this application. The data processing apparatus can be a computer program (including program code) running on a computer device; for example, the data processing apparatus is application software. This data processing apparatus can be used to execute the corresponding steps in the methods provided in the embodiments of this application. Figure 8 As shown, the data processing device 1 may include: an acquisition module 11, a generation module 12, and a prediction module 13;

[0155] Module 11 is used to acquire object behavior data of the target object for the recommended object in multiple time intervals;

[0156] The generation module 12 is used to generate object behavior features of the target object against the recommended object in each time interval based on the object behavior data of the target object against the recommended object in each time interval.

[0157] Prediction module 13 is used to predict the order index of the target object for the recommended object based on the object behavior characteristics of the target object for the recommended object in each time interval.

[0158] Optionally, any one of multiple time intervals can be represented as the target time interval; the recommended object belongs to the object in the recommendation platform;

[0159] The acquisition module 11 acquires object behavior data of the target object for the recommended object in multiple time intervals in the following ways:

[0160] Retrieve multiple sub-recommendation objects contained within the recommendation object;

[0161] Retrieve the associated objects of the recommended object on the recommendation platform, and retrieve the recommendation activities of the objects associated with the recommended object on the recommendation platform;

[0162] Obtain the target object's first browsing behavior data on the recommendation platform for the recommended object and related objects within the target time interval;

[0163] Obtain the second browsing behavior data of the target object on the recommendation platform for multiple recommended sub-objects within the target time interval;

[0164] Obtain third-party browsing behavior data of the target object on the recommendation platform within the target time interval;

[0165] Based on the first browsing behavior data, the second browsing behavior data, and the third browsing behavior data, determine the target object's object behavior data towards the recommended object within the target time interval.

[0166] Optionally, the first browsing behavior data includes the number of times and the duration of browsing the first object description page of the target object for the recommended object and the associated object respectively; the second browsing behavior data includes the number of times and the duration of browsing the second object description page of the target object for multiple recommended sub-objects respectively; the third browsing behavior data includes the number of times and the duration of browsing the activity page of the target object for the object recommendation activity; the first object description page, the second object description page and the activity page belong to the recommendation platform.

[0167] Optionally, the generation module 12 generates object behavior features of the target object relative to the recommended object in each time interval based on the object behavior data of the target object relative to the recommended object in each time interval, including:

[0168] Generate a first feature vector for the target object based on the first browsing behavior data, targeting the recommended object and the associated object;

[0169] Generate a second feature vector for the target object targeting multiple recommendation sub-objects based on the second browsing behavior data;

[0170] Generate a third feature vector for the target object's recommendation activity based on the third browsing behavior data;

[0171] Based on the first feature vector, the second feature vector, and the third feature vector, generate the object behavior features of the target object against the recommended object within the target time interval.

[0172] Optionally, the generation module 12 generates object behavior features of the target object against the recommended object within the target time interval based on the first feature vector, the second feature vector, and the third feature vector, including:

[0173] Retrieve multiple browsing pages of the target object within the target time interval on the recommendation platform;

[0174] The browsing pages other than the first object description page, the second object description page, and the activity page are identified as associated browsing pages.

[0175] Extract browsing keywords from associated browsing pages and generate a fourth feature vector based on the browsing keywords;

[0176] The first feature vector, the second feature vector, the third feature vector, and the fourth feature vector are concatenated to obtain the object behavior features of the target object in response to the recommended object within the target time interval.

[0177] Optionally, the prediction module 13 predicts the order placement index of the target object for the recommended object based on the object behavior characteristics of the target object for the recommended object in each time interval, including:

[0178] The object behavior features of the target object for each time interval are concatenated to obtain the concatenated behavior features.

[0179] The recommendation model is invoked to predict the order index of the target object for the recommended object based on the splicing behavior characteristics.

[0180] Optionally, the above-mentioned device 1 is also used for:

[0181] Acquire sample data; the sample data includes the sample splicing behavior features of the sample object for the sample recommendation object and the order placement behavior tags of the sample object for the sample recommendation object. The order placement behavior tags are either the no-order behavior tags or the ordered behavior tags.

[0182] The initial recommendation model is invoked to predict the sample order index for the sample recommended object based on the sample splicing behavior characteristics.

[0183] The model parameters of the initial recommendation model are updated based on the sample order index and order behavior labels to obtain the recommendation model.

[0184] According to one embodiment of this application, Figure 3 The steps involved in the data processing method shown can be derived from... Figure 8 The data processing apparatus 1 shown is executed by each module. For example, Figure 3 Step S101 shown can be performed by Figure 8 The acquisition module 11 in the middle is used to execute, Figure 3 Step S102 shown can be performed by Figure 8 The generation module 12 in the middle is used to execute; Figure 3 Step S103 shown can be performed by Figure 8 The prediction module 13 in the middle is used to perform the operation.

[0185] This application acquires object behavior data of a target object towards a recommended object within multiple time intervals; based on the object behavior data of the target object towards the recommended object in each time interval, it generates object behavior features of the target object towards the recommended object in each time interval; based on the object behavior features of the target object towards the recommended object in each time interval, it predicts the order index of the target object towards the recommended object. Therefore, the device proposed in this application can acquire object behavior features of the target object towards the recommended object according to time partitions, taking into account the object behavior of the target object towards the recommended object in each time partition, thus making the predicted order index of the target object towards the recommended object more accurate.

[0186] According to one embodiment of this application, Figure 8The modules in the data processing device 1 shown can be individually or entirely combined into one or more units, or some of the units can be further divided into multiple functionally smaller sub-units to achieve the same operation without affecting the technical effects of the embodiments of this application. The above modules are based on logical function division. In practical applications, the function of one module can also be implemented by multiple units, or the function of multiple modules can be implemented by one unit. In other embodiments of this application, the data processing device 1 may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0187] According to one embodiment of this application, a general-purpose computer device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM), can perform operations such as... Figure 3 The computer program (including program code) for each step involved in the corresponding method shown, to construct such... Figure 8 The data processing apparatus 1 shown herein, and the data processing method for implementing the embodiments of this application, are described. The computer program described above may be recorded on, for example, a computer-readable recording medium, loaded onto the aforementioned computing device via the computer-readable recording medium, and run therein.

[0188] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in this application. For example... Figure 9 As shown, the computer device 1000 may include a processor 1001, a network interface 1004, and a memory 1005. Furthermore, the computer device 1000 may also include a user interface 1003 and at least one communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 9 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a device control application program.

[0189] exist Figure 9 In the computer device 1000 shown, the network interface 1004 provides network communication functionality; the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve:

[0190] Obtain object behavior data of the target object for the recommended object in multiple time intervals;

[0191] Based on the object behavior data of the target object towards the recommended object in each time interval, generate object behavior features of the target object towards the recommended object in each time interval;

[0192] Based on the target object's behavior characteristics towards the recommended object in each time interval, predict the target object's order index towards the recommended object.

[0193] It should be understood that the computer device 1000 described in the embodiments of this application can execute the foregoing text. Figure 3 The data processing method described in the corresponding embodiments can also be executed as described above. Figure 8 The description of the data processing apparatus 1 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0194] Furthermore, it should be noted that this application also provides a computer-readable storage medium storing a computer program executed by the aforementioned data processing apparatus 1. The computer program includes program instructions, which, when executed by a processor, enable the execution of the aforementioned... Figure 3 The description of the data processing method in the corresponding embodiments is already provided and will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the embodiments of the computer storage medium involved in this application, please refer to the description of the method embodiments of this application.

[0195] As an example, the above program instructions can be deployed and executed on a single computer device, or deployed and executed on multiple computer devices located in one location, or executed on multiple computer devices distributed across multiple locations and interconnected via a communication network. Multiple computer devices distributed across multiple locations and interconnected via a communication network can form a blockchain network.

[0196] The aforementioned computer-readable storage medium can be an internal storage unit of the data processing apparatus or computer device provided in any of the foregoing embodiments, such as a hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0197] This application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned... Figure 3 The data processing method described in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application.

[0198] The terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0199] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0200] The methods and related apparatuses provided in this application are described with reference to the method flowcharts and / or structural diagrams provided in this application. Specifically, each block of the method flowchart and / or structural diagram, as well as combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to create a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 A process or multiple processes and / or structures illustrate the steps of the functions specified in one or more boxes.

[0201] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A data processing method, characterized in that, The method includes: Obtain object behavior data of the target object for the recommended object in multiple time intervals; Based on the object behavior data of the target object towards the recommended object in each time interval, generate object behavior features of the target object towards the recommended object in each time interval; Based on the object behavior characteristics of the target object towards the recommended object in each time interval, predict the order index of the target object towards the recommended object; The order index is predicted by the recommendation model based on the splicing behavior features, which are obtained by splicing the object behavior features of the target object for the recommended object in each time interval; the method further includes: Acquire sample data; the sample data includes sample splicing behavior features of the sample object for the sample recommendation object and order placement behavior tags of the sample object for the sample recommendation object, wherein the order placement behavior tags are either no-order behavior tags or ordered-order behavior tags. The initial recommendation model is invoked to predict the sample order index of the sample object for the sample recommendation object based on the sample splicing behavior characteristics. The model parameters of the initial recommendation model are updated based on the sample order index and the order behavior label to obtain the recommendation model.

2. The method according to claim 1, characterized in that, Any one of the multiple time intervals is represented as the target time interval; the recommended object belongs to the object in the recommendation platform; The acquisition of object behavior data of the target object in relation to the recommended object in multiple time intervals includes: Obtain multiple recommendation sub-objects contained in the recommendation object; Obtain the associated objects of the recommended object on the recommendation platform, and obtain the object recommendation activities associated with the recommended object on the recommendation platform; Obtain the first browsing behavior data of the target object on the recommendation platform for the recommended object and the associated object during the target time interval; Obtain the second browsing behavior data of the target object on the recommendation platform for the multiple recommended sub-objects during the target time interval; Obtain third browsing behavior data of the target object on the recommendation platform for the object recommendation activity during the target time interval; Based on the first browsing behavior data, the second browsing behavior data, and the third browsing behavior data, the object behavior data of the target object towards the recommended object within the target time interval is determined.

3. The method according to claim 2, characterized in that, The first browsing behavior data includes the number of times and the duration of browsing the first object description pages of the target object for the recommended object and the associated object, respectively; the second browsing behavior data includes the number of times and the duration of browsing the second object description pages of the target object for the plurality of recommended sub-objects, respectively; The third browsing behavior data includes the number of times and the duration of browsing the activity page of the recommended activity for the target object; The first object description page, the second object description page, and the activity page belong to the recommendation platform.

4. The method according to claim 3, characterized in that, The step of generating object behavior features of the target object relative to the recommended object in each time interval based on the object behavior data of the target object relative to the recommended object in each time interval includes: Generate a first feature vector of the target object for the recommended object and the associated object based on the first browsing behavior data; A second feature vector of the target object for the plurality of recommended sub-objects is generated based on the second browsing behavior data; A third feature vector for the target object's recommendation activities is generated based on the third browsing behavior data. Based on the first feature vector, the second feature vector, and the third feature vector, the object behavior features of the target object relative to the recommended object are generated within the target time interval.

5. The method according to claim 4, characterized in that, The step of generating object behavior features of the target object for the recommended object within the target time interval based on the first feature vector, the second feature vector, and the third feature vector includes: Obtain multiple browsing pages of the target object on the recommendation platform within the target time interval; The browsing pages other than the first object description page, the second object description page, and the activity page among the plurality of browsing pages are identified as associated browsing pages; Extract browsing keywords from the associated browsing pages, and generate a fourth feature vector based on the browsing keywords; The first feature vector, the second feature vector, the third feature vector, and the fourth feature vector are concatenated to obtain the object behavior features of the target object towards the recommended object within the target time interval.

6. The method according to claim 1, characterized in that, The step of predicting the order index of the target object for the recommended object based on the object behavior characteristics of the target object for the recommended object in each time interval includes: The object behavior features of the target object for the recommended object are concatenated in each time interval to obtain the concatenated behavior features; The recommendation model is invoked to predict the order index of the target object for the recommended object based on the splicing behavior characteristics.

7. A data processing apparatus, characterized in that, The device includes: The acquisition module is used to acquire object behavior data of the target object towards the recommended object in multiple time intervals; The generation module is used to generate object behavior features of the target object in each time interval for each time interval based on the object behavior data of the target object in each time interval for the recommended object; The prediction module is used to predict the order index of the target object for the recommended object based on the object behavior characteristics of the target object for the recommended object in each time interval. The order index is predicted by the recommendation model based on the splicing behavior features, which are obtained by splicing the object behavior features of the target object for the recommended object in each time interval; the device is also used for: Acquire sample data; the sample data includes sample splicing behavior features of the sample object for the sample recommendation object and order placement behavior tags of the sample object for the sample recommendation object, wherein the order placement behavior tags are either no-order behavior tags or ordered-order behavior tags. The initial recommendation model is invoked to predict the sample order index of the sample object for the sample recommendation object based on the sample splicing behavior characteristics. The model parameters of the initial recommendation model are updated based on the sample order index and the order behavior label to obtain the recommendation model.

8. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-6.

9. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-6.

Images