A data processing method, related apparatus, device, and storage medium

Abstract

The application discloses a data processing method, which is used in the technical field of Internet. The method comprises the following steps: firstly, obtaining M objects and attribute labels of the M objects; then, receiving a delivery instruction carrying K target attribute labels; based on a label combination condition defined in the instruction, dynamically screening N completely matched candidate objects from the M objects to generate a positive list; finally, delivering information to the objects in the positive list based on the delivery instruction. The method provided by the application realizes deep coupling of data organization mode and query logic through the positive list of objects, solves the core defects of traditional technology, such as strong correlation between object package quantity and query performance, high iteration cost of rules, and the like, and has high concurrent processing capacity and real-time dynamic adaptability.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a data processing method, related apparatus, device, and storage medium. Background Technology

[0002] With the rapid development of internet advertising, precise targeting technology has become crucial for improving advertising effectiveness. Currently, advertisers need to dynamically construct target object sets based on object tags, supporting high-concurrency queries, real-time updates, and large-scale set management. However, in scenarios with ultra-large-scale object data, traditional technologies face three major challenges: First, high-concurrency query performance is strongly correlated with set size, resulting in query latency for sets of thousands that cannot meet the real-time bidding requirements of advertising, directly impacting the efficiency and accuracy of ad delivery; second, offline set generation methods struggle to support real-time tag updates, leading to insufficient timeliness in ad delivery; and third, hard-coded rule logic has poor scalability, requiring frequent adjustments to the underlying code and service redeployment for adding or modifying filtering conditions, increasing operational costs.

[0003] Existing technologies mainly employ three approaches: 1) Offline computation based on inverted indexes, which supports fast collection retrieval by pre-storing tags in the inverted index of objects, but requires periodic offline reconstruction of the collection and cannot respond to tag changes. This method struggles to guarantee the accuracy of ad delivery in scenarios with frequent tag changes; 2) Memory caching to accelerate queries, which caches high-frequency collection results in Redis, reducing query latency, but is limited by storage capacity, and requires a full cache refresh for rule changes, failing to support dynamic expansion; 3) Batch processing using distributed computing frameworks, utilizing Hadoop / Spark for offline calculation of object ownership, but the computation cycle is long and cannot meet the real-time requirements of advertising. None of these approaches effectively balance the needs of high concurrency, real-time performance, and scalability.

[0004] The core flaw in existing technologies lies in the inability of these three elements to work together: inverted index schemes sacrifice real-time performance, memory caching is limited by storage capacity, and distributed computing cannot guarantee low latency. Furthermore, filtering rules are typically embedded in the system in a hard-coded form, causing rule iterations to depend on code releases, resulting in poor flexibility. In scenarios with extremely large-scale objects, the query performance of traditional solutions increases linearly with the size of the collection, making it difficult to support advertisers' needs for refined operations, ultimately leading to decreased ad matching efficiency, lower conversion rates, and increased operating costs. Therefore, an effective method is urgently needed to address these issues. Summary of the Invention

[0005] This application provides a data processing method, related apparatus, device, and storage medium, which solves the problems of high real-time tag update latency, poor query performance of thousands of object packages, and insufficient rule scalability in existing data processing methods.

[0006] This application provides a data processing method, including:

[0007] Get M objects, where each object corresponds to at least one attribute label, and M > 1;

[0008] Receive information delivery instructions, wherein the information delivery instructions carry K target attribute labels and the logical relationship between the K target attribute labels, and the information delivery instructions are used to represent the first information delivery to objects that satisfy the K target attribute labels, where K≥1;

[0009] Based on K target attribute labels and M objects, generate an object forward sorting table. The object forward sorting table includes N object identifiers corresponding to N target objects and attribute labels corresponding to each of the N target objects. The object forward sorting table uses the N object identifiers as the primary key index. The N target objects are selected from the M objects. The attribute label of each target object is the same as at least one of the K target attribute labels, and the attribute label of each target object satisfies the logical relationship between the K target attribute labels, where M≥N≥1.

[0010] Based on the information delivery instruction, the first information delivery is performed on N target objects in the object forward list.

[0011] This application also provides a data processing apparatus, comprising:

[0012] The acquisition module is used to acquire M objects, where each object corresponds to at least one attribute label, and M > 1;

[0013] The receiving module is used to receive information delivery instructions, wherein the information delivery instructions carry K target attribute labels and the logical relationship between the K target attribute labels. The information delivery instructions are used to represent the first information delivery to objects that satisfy the K target attribute labels and the logical relationship between the K target attribute labels, where K≥1;

[0014] The generation module is used to generate an object forward list based on K target attribute tags and M objects. The object forward list includes N object identifiers corresponding to N target objects and attribute tags corresponding to each of the N target objects. The object forward list is indexed by the N object identifiers as the primary key. The N target objects are selected from the M objects. The attribute tag of each target object is the same as at least one of the K target attribute tags, and the attribute tag of each target object satisfies the logical relationship between the K target attribute tags, where M≥N≥1.

[0015] The delivery module is used to deliver the first information to N target objects in the object forward list based on the information delivery instruction.

[0016] In another aspect, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the methods described above.

[0017] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described above.

[0018] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the methods described above.

[0019] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0020] This application provides a data processing method. First, M objects and their attribute tags are acquired to form a multi-dimensional data source. A unique mapping relationship is established through object identifiers to ensure data traceability. Then, a delivery instruction carrying K target attribute tags and the logical relationships between them is received. Based on the tag combination conditions defined in the instruction, N perfectly matching candidate objects are dynamically selected from the M objects to generate a forward sorting table. The generated forward sorting table uses object identifiers as indexes and integrates multi-dimensional attribute tags to form structured data, supporting both fast retrieval and easy incremental updates. Based on this forward sorting table, the delivery execution process only needs to match object identifiers to directly locate the target group, avoiding the overhead of traversing the inverted index linked list or fully calculating object ownership in traditional solutions. This reduces the query latency for hundreds of billions of object packages from milliseconds to microseconds. The method provided in this application achieves deep coupling between data organization and query logic through an object forward sorting table, solving the core defects of traditional technologies such as strong correlation between the number of object packages and query performance and high rule iteration costs. It enables concurrent filtering of all objects, and performs object filtering in real time according to the delivery instructions to generate a forward sorting table. It has both high concurrency processing capabilities and real-time dynamic adaptability, and can support tens of millions of accurate delivery requests per day in scenarios with hundreds of millions of objects, thereby improving the advertising conversion rate. Attached Figure Description

[0021] Figure 1 An interface diagram of the data processing method provided in this application embodiment for ad placement in tweets;

[0022] Figure 2 An interface diagram for selecting target attribute tags provided in the embodiments of this application;

[0023] Figure 3 The data processing method provided in this application is applied to the scenario architecture diagram of advertising delivery;

[0024] Figure 4A flowchart illustrating the data processing method provided in the embodiments of this application;

[0025] Figure 5 A flowchart illustrating the data processing method provided in an embodiment of this application;

[0026] Figure 6 A schematic diagram illustrating the process of generating an object forward list as provided in an embodiment of this application;

[0027] Figure 7 A flowchart of the target object screening process provided in the embodiments of this application;

[0028] Figure 8 A flowchart illustrating another target object screening process provided in this application embodiment;

[0029] Figure 9 A flowchart illustrating another target object screening process provided in this application embodiment;

[0030] Figure 10 A flowchart illustrating another target object screening process provided in this application embodiment;

[0031] Figure 11 A flowchart for updating the object forward list provided in this application embodiment;

[0032] Figure 12 A schematic diagram illustrating the updating of the full object forward list provided in this application embodiment;

[0033] Figure 13 A schematic diagram illustrating a full object update provided in an embodiment of this application;

[0034] Figure 14 A schematic diagram illustrating differential object updates provided in an embodiment of this application;

[0035] Figure 15 A schematic diagram illustrating the updating of the label change object provided in an embodiment of this application;

[0036] Figure 16 A schematic diagram illustrating the display of first information provided in an embodiment of this application;

[0037] Figure 17 A schematic diagram illustrating the differentiated data processing method provided in an embodiment of this application;

[0038] Figure 18 A schematic diagram illustrating the process of generating an object forward list as provided in an embodiment of this application;

[0039] Figure 19 This is a structural diagram of the data processing apparatus provided in an embodiment of this application. Detailed Implementation

[0040] This application provides a data processing method that achieves millisecond-level real-time matching, high concurrency handling, and dynamic rule hot updating capabilities in scenarios with hundreds of millions of objects by dynamically constructing an object forward sorting table, streaming incremental computation, and a distributed indexing mechanism.

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below 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 of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0042] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0043] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0044] Before introducing the specific methods of this application, we will first provide an illustrative example of the application scenarios for the data processing methods in this application. It should be understood that the following application scenarios are merely examples and are not limited to these.

[0045] In one possible implementation, such as Figure 1 As shown, Figure 1This diagram illustrates the interface for ad placement in tweets using the data processing method provided in this application. The information placement method provided in this application is applied to ad placement scenarios within tweets. When a user creates a tweet ad task on the platform, the system obtains multi-dimensional attribute tags of the fan group in real time and dynamically updates the user tag forward ranking table through a streaming computing engine. For example, when a sports brand launches a limited-time promotion, the system, based on the three target attribute tags preset in the placement instruction, filters out eligible user groups from a million-level fan database within seconds, generating a forward ranking table containing user identifiers and corresponding tags. This forward ranking table achieves efficient retrieval through columnar storage and distributed indexing technology, avoiding the full table scan overhead of traditional inverted indexes, enabling tens of thousands of user attribution determinations to be processed in parallel per second. During ad placement, the system directly matches the tag combinations in the forward ranking table based on the user ID, accurately reaching the target user, and monitoring click-through rate and conversion rate data in real time. If the user updates their tags during ad exposure, the streaming engine will immediately trigger an incremental update of the forward ranking table, dynamically adjusting subsequent push strategies to ensure that ad accuracy is always synchronized with the user's real-time behavior, reducing query latency compared to traditional solutions and minimizing the consumption of ineffective exposure resources.

[0046] In another possible implementation, such as Figure 2 As shown, Figure 2 This diagram illustrates the target attribute tag selection interface provided in an embodiment of this application. The data processing method provided in this application achieves precise targeting and dynamic optimization in scenarios where advertisers dynamically construct object packages by integrating object behavior data and a tag system in real time. For example, when an e-commerce advertiser plans a "sports goods promotion" campaign, it captures changes in object attributes in real time based on the object tag forward index, immediately updates the object's attribute tags through a streaming processing engine, and triggers incremental calculations of object package rules. Advertisers can configure composite conditions for three target attribute tags to filter matching objects from a database of hundreds of millions of objects within seconds, generating object packages containing object IDs and real-time tags, and synchronizing them to a distributed index. During ad delivery, object identifiers are directly matched based on the object package ID, bypassing the full table scan of traditional inverted indexes, reducing query latency for thousands of object packages, and supporting tens of thousands of high-frequency delivery requests per hour. If subsequent object behavior triggers tag changes, the streaming engine will dynamically adjust the object package affiliation to ensure that ads only reach currently matching high-value objects. Real-world application data shows that this method can improve ad click-through rates, reduce invalid exposures, and advertisers do not need to wait for offline calculation cycles. After adjusting the object package rules in real time, the changes take effect in a short period of time, significantly improving the agility and ROI of marketing campaigns.

[0047] Please see Figure 3 , Figure 3This diagram illustrates a scenario architecture diagram of the data processing method provided in this application applied to advertising. The scenario includes a terminal 311 for object A (the advertiser), a terminal 312 for object B (the target object), a network 320, a server 330, and a database 340.

[0048] Both terminal 311 and terminal 312 include a human-computer interaction screen, a processor, and a memory.

[0049] The human-computer interaction screen of terminal 311 is used to display the attribute tag selection interface; it also provides a human-computer interaction interface to receive human-computer interaction operations from user A (advertiser) regarding the selection of K target attribute tags. The processor is used to generate interaction instructions in response to the aforementioned human-computer interaction operations and send these instructions to the server. The memory is used to store relevant attribute tags, object packages, etc.

[0050] The human-computer interaction screen of terminal 312 is used to display the tweet advertisement interface; it is also used to provide a human-computer interaction interface to receive click operations on the advertisement by user B (the target object). The processor is used to generate interaction instructions in response to the above human-computer interaction operations and send the interaction instructions to the server.

[0051] The terminals 311 and 312 involved in this application include, but are not limited to, mobile phones, tablets, laptops, desktop computers, intelligent voice interaction devices, virtual reality devices, smart home appliances, vehicle terminals, aircraft, etc.

[0052] The terminal 311 contains a running client 3111. Taking the advertising client as an example, the client 3111 is deployed on the terminal 311. The client 3111 can run on the terminal 311 via a browser, or as a standalone application (APP) or mini-program, etc.

[0053] Client 3121 runs on terminal 312. Taking the tweet client as an example, client 3121 is deployed on terminal 312. Client 3121 can run on terminal 312 through a browser, or through a standalone application (APP) or mini-program, etc.

[0054] Network 320 uses standard communication technologies and / or protocols, typically the Internet, but can also be any network, including but not limited to Bluetooth, local area network (LAN), metropolitan area network (MAN), wide area network (WAN), mobile, private network, or any combination of virtual private network. In some embodiments, custom or dedicated data communication technologies may be used to replace or supplement the aforementioned data communication technologies.

[0055] Server 330 includes a processor. The server 330 involved in this application can be a standalone physical server, a server cluster or distributed system consisting of at least one physical server, 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, content delivery networks (CDNs), and big data and artificial intelligence (AI) platforms.

[0056] User A selects K target attribute tags in the attribute tag selection interface displayed on terminal 311, as in step S301. Terminal 311 generates an information delivery instruction based on user A's human-computer interaction. As in step S302, terminal 311 sends the information delivery instruction to server 330 via network 320. As in step S303, server 330 retrieves M objects from database 340, each object corresponding to at least one attribute tag, and the M objects corresponding to M object identifiers. As in step S304, server 330 receives the information delivery instruction, which carries the K target attribute tags and the logical relationships between them. The information delivery instruction is used to indicate that targeted advertising will be delivered to objects that satisfy the K target attribute tags and the logical relationships between them. As in step S305, server 330 generates an object forward list based on K target attribute tags and M objects. The object forward list includes N object identifiers corresponding to N target objects and attribute tags corresponding to each of the N target objects. The object forward list uses the N object identifiers as the primary key index. The N target objects are selected from the M objects. Each target object's attribute tag is identical to at least one of the K target attribute tags, and each target object's attribute tag satisfies the logical relationship between the K target attribute tags. As in step S306, server 330 performs targeted advertising on the N target objects in the object forward list based on the information delivery instruction. As in step S307, the targeted advertisement is displayed on user B's terminal 312.

[0057] In advertising scenarios, the system directly matches tag combinations in the positive sorting table based on the object ID to accurately reach the target object. It also monitors click-through rate and conversion rate data in real time to ensure that the accuracy of the advertisement is always synchronized with the real-time behavior of the object. Compared with traditional solutions, it reduces query latency and reduces the consumption of invalid exposure resources.

[0058] Please see Figure 4 , Figure 4 A flowchart of a data processing method is shown. It should be noted that the data processing method provided in this application embodiment can be applied to a server, and this application embodiment does not impose any limitations. The method includes:

[0059] S410, Get M objects.

[0060] Each object corresponds to at least one attribute label, and M > 1.

[0061] As can be understood, an object refers to a potential audience entity (such as a device, device, or account) in an advertising scenario, attribute tags are identifiers describing the characteristics of an object, and object identifiers (such as object IDs) are information that uniquely identifies an object. Multiple objects and their attribute tags are retrieved in batches from a database or real-time stream, establishing a mapping relationship between objects and tags to form the original dataset.

[0062] S420, Receive information delivery instructions.

[0063] The information delivery instruction carries K target attribute labels and the logical relationship between the K target attribute labels. The information delivery instruction is used to represent the first information delivery to objects that satisfy the K target attribute labels and the logical relationship between the K target attribute labels, where K≥1.

[0064] It's understandable that the information delivery instructions are the delivery strategies configured by advertisers (such as targeting conditions and delivery channels). The core rules for filtering objects are the target attribute tags (K tags) and the logical relationships between them. Parsing the target attribute tags and their logical relationships in the instructions transforms them into target attribute tag filtering rules, triggering the subsequent forward ranking generation process. By supporting advertisers to flexibly define complex tag logic (such as intersection and union), without the need for pre-set rule templates, the customization capability of delivery strategies is significantly improved. The logical relationships between target attribute tags refer to the way objects combine multiple tags. For example, "female" and "age 25-35" must be satisfied simultaneously (logical "AND"), or "browsing sports categories" or "following running communities and sports equipment review accounts" only requires one of them to be satisfied (logical "OR"). By parsing the composite conditions configured by advertisers (such as "running ≥4 times in the last 30 days and collecting racing shoes"), the calculation logic between each tag is clarified, providing a rule basis for subsequent filtering. By supporting the flexible definition of complex object targeting conditions, it avoids the limitations of traditional solutions that only support single-condition filtering. For example, a sports brand can simultaneously require the object to have the dual characteristics of "high spending power" and "recently viewed competitor products" to accurately target high-intent objects.

[0065] In one possible implementation, advertisers use methods such as Figure 2 In the interface diagram showing the attribute label selection, check or select K target attribute labels (such as selecting female for gender, and selecting three age groups: 25-34, 5-44, and 45-54 for age). After selecting the K target attribute labels, click the "Submit" control. The terminal will then generate an information delivery instruction carrying the K target attribute labels. The terminal will then send the information delivery instruction to the server, and the server will receive the information delivery instruction.

[0066] S430. Generate an object forward list based on the logical relationships between the K target attribute tags and the M objects.

[0067] The object forward index includes N object identifiers corresponding to N target objects and attribute labels corresponding to each of the N target objects. The object forward index uses the N object identifiers as the primary key index. The N target objects are selected from M objects. The attribute label of each target object is the same as at least one of the K target attribute labels, and the attribute label of each target object satisfies the logical relationship between the K target attribute labels, where M≥N≥1.

[0068] As can be understood, an object forward sorting table is a structured data table indexed by object identifiers, storing object IDs and their corresponding attribute tags, where N is the number of objects that meet the target tag conditions. By traversing M objects, N matching objects are selected based on the target tags, and a forward sorting table is constructed with object IDs as keys and tags as values. For example, object IDs can be associated and stored with tags such as "interest in sports equipment" and "high spending power".

[0069] The forward index storage method, by directly associating object attributes through the primary key index, combined with columnar storage optimization and real-time streaming computing, can achieve efficient querying. Specifically:

[0070] By using structured query paths, access complexity is reduced. The forward index uses the object ID as the primary key and directly stores the complete set of tags for each object (e.g., {user_id:1001, tags:[“Sports Equipment Interest”, “High Spending Power”]}). During a query, the tags can be quickly located using the object ID, eliminating the need to traverse the multi-level linked list of the inverted index (e.g., “Sports Enthusiasts → Object ID List”), thus avoiding complex index jumps. The structured query path is simple, and latency primarily depends on the time of a single primary key access, rather than rule complexity. The latency for queries involving thousands of objects does not increase linearly with the number of rules.

[0071] Secondly, columnar storage and vectorized computation accelerate batch processing. Columnar storage groups high-frequency tags (such as "gender" and "region") into separate columns, significantly reducing disk I / O overhead through sparse storage and compression algorithms (such as BitMap). During queries, objects matching the criteria (such as "gender = female") can be directly filtered by column, and data can be processed in batches using CPU vectorized instructions, avoiding row-by-row parsing. With extremely large datasets, a single query can efficiently filter invalid objects, concentrating computational resources on valid data and significantly improving concurrent throughput.

[0072] Furthermore, real-time streaming updates ensure data consistency. When an object tag is added or modified (e.g., adding the tag "sports enthusiast" to an object), a streaming task is triggered in real time via Spark Streaming / Flink to update only the object's forward sorted table records, rather than performing a full reconstruction. Incremental updates avoid the heavy load of offline computation and ensure that object package rules take effect in real time (e.g., adding a new object package "sports equipment enthusiast" immediately includes the object).

[0073] In addition, a combination of caching and distributed storage is employed to optimize performance in a tiered manner. Frequently accessed objects or object package results are cached in Redis, allowing queries to directly access memory and reducing latency. Massive amounts of large object data are stored in distributed systems such as HBase, distributing the load through sharding and parallel queries (e.g., object ID range sharding). This approach balances low latency for hot data and high throughput for cold data, catering to diverse needs across different scenarios.

[0074] By pre-compiling tag matching results, real-time queries are transformed into direct indexing operations, avoiding the performance bottleneck of traditional solutions that require a full scan of the object library for each deployment.

[0075] S440. Based on the information delivery instruction, perform the first information delivery to N target objects in the object forward list.

[0076] Understandably, the first step in information delivery refers to pushing advertising content to N target objects in a forward ranking list, such as displaying promotional tweets or product links to selected objects. The second step involves directly accessing the object ID list in the forward ranking list and using a message queue or API to precisely deliver advertising content to the target objects, such as pushing personalized ads to the object ID list. Direct matching based on object IDs significantly reduces query latency, supports real-time delivery in high-concurrency scenarios, reduces invalid exposures, and improves ad conversion rates.

[0077] For example, please refer to Figure 5 , Figure 5 A schematic diagram of the method described in this application is shown. A sports brand plans to promote its new smart running shoes, targeting individuals who are passionate about running and have moderate to high-intensity exercise habits. First, 800,000 potential targets (M = 800,000) were retrieved from a target database. Each target is associated with a unique ID (e.g., target ID: 1001, 1002…) and multi-dimensional attribute tags. For example, the tags for target ID: 1002 are "female, 35 years old, resides in City A, runs more than 50 kilometers per week, follows running communities and sports equipment review accounts." When the advertising operations personnel configure the placement instructions, they are required to filter out target groups that simultaneously meet the three target attribute tags (K = 3): "age 25-45 years old (K1)," "runs ≥ 4 times in the past 30 days (K2)," and "has bookmarked the racing shoe product page (K3)." The screening process then begins, comparing object attributes one by one using a tag matching engine. For example, object ID 2055, tagged "female, 28 years old, runs twice a week, no sports products saved," is excluded; while object ID 3098, tagged "male, 38 years old, completed 6 running training sessions in the last 30 days, saved three racing shoes," is identified as a perfect match. Ultimately, 18,000 precise targets (N=18,000) are selected from 800,000 objects, generating a positive ranking list containing these object IDs and their complete attribute tags. Simultaneously, the ad delivery process is triggered, pushing new product ads with "limited-time discount + running data synchronization function" to the target audience. This embodiment, through dynamic tag screening and positive ranking list pre-calculation, shortens the traditionally time-consuming offline targeting task to a short time, significantly improving ad click-through rates compared to random targeting and effectively reducing the cost of ineffective exposure to non-target audiences.

[0078] The method provided in this application integrates the scattered tag filtering, object matching, and ad delivery processes of traditional solutions into a unified operation by constructing an object forward index. When advertisers dynamically configure object package rules, they only need to scan the pre-calculated forward index to complete the object attribution determination, avoiding the full table scan overhead of traditional inverted indexes and reducing the query time for thousands of object packages. At the same time, the incremental update mechanism of the forward index (such as automatic refresh after object tag changes) ensures the real-time nature of data and solves the defects of offline calculation lag and high rule iteration costs in traditional technologies. In scenarios with hundreds of millions of objects, this method supports high-frequency delivery requests. Advertisers do not need to wait for offline calculation results; they can adjust object package rules in real time and the changes will take effect in a short time, significantly improving marketing agility and resource utilization, resulting in a significant increase in ad click-through rate compared to traditional solutions and a reduction in invalid exposures.

[0079] In an optional embodiment of the data processing method provided in the above embodiments of this application, generating an object forward sorting table based on K target attribute tags and M objects includes:

[0080] Based on the K target attribute labels and the logical relationships between them, generate target attribute label filtering rules;

[0081] Based on the target attribute tag filtering rules, according to the attribute tags corresponding to each of the M objects, N target objects that meet the target attribute tag filtering rules are selected from the M objects;

[0082] Generate an object forward list based on the N object identifiers corresponding to the N target objects and the attribute labels corresponding to each of the N target objects.

[0083] Understandably, target attribute tag filtering rules transform logical relationships into executable query conditions. For example, "age 25-35 years old" is converted into a field range query (age BETWEEN 25 AND 35), and "logical AND" relationships are transformed into the AND operator in SQL or joint filtering conditions in a distributed computing framework. Through syntax parsing and a rule engine, natural language-described delivery instructions are transformed into machine-recognizable structured query statements. This enables rapid parsing and rule solidification of tag combinations. For instance, after configuring a rule for "favorited products and active periods at night" on an e-commerce platform, the corresponding query plan can be directly generated without manual coding, lowering the operational threshold and improving delivery efficiency.

[0084] Based on the generated filtering rules, the system iterates through the attribute tags of each of the original M objects. For example, it compares the attributes of object ID:456 ("age 30 years old" and "runned 5 times in the last 30 days") to see if they meet the condition of "age 25-35 years old and ran ≥4 times". Ultimately, it filters out N perfectly matching target objects. The core of this approach is to improve filtering efficiency through index acceleration or distributed parallel computing (such as MapReduce). This avoids the overhead of a full table scan required by traditional inverted indexes. For example, in scenarios with hundreds of millions of objects, pre-computed tag indexes can reduce the time spent on a single filtering process from minutes to seconds, while also reducing wasted computing resources.

[0085] The "Object Forward Ranking Table" is a structured data table with object IDs as keys and tags as values. For example, the tags "Male, 38 years old, completed 6 running training sessions in the last 30 days, and have saved three racing shoes" for object ID 3098 are bound to its ID and stored to form a mapping relationship that can be directly queried. Materialized view technology is used to solidify the filtering results, forming a reusable intermediate dataset. Subsequent ad placements can directly call the list of object IDs in the forward ranking table without repeating complex tag matching logic. For example, when a sports brand executes the same placement strategy for three consecutive days, it can directly reuse the forward ranking table generated the previous day, reducing real-time calculation pressure.

[0086] For example, please refer to Figure 6 , Figure 6This diagram illustrates the process of generating a forward sorting list for objects. A sports brand plans to promote its new smart running shoes, targeting individuals who are passionate about running and have moderate to high-intensity exercise habits. First, 800,000 potential objects (M = 800,000) are retrieved from the object database. Each object is associated with a unique ID (e.g., object ID: 1001, 1002, etc.) and multi-dimensional attribute tags. For example, object ID: 1002's tags are "female, 35 years old, resides in City A, runs more than 50 kilometers per week, follows running communities and sports equipment review accounts." When the advertising operations personnel configure the placement instructions, they are required to filter out the object group that simultaneously meets the three target attribute tags (K = 3): "age 25-45 years old (K1)," "runs ≥ 4 times in the last 30 days (K2)," and "has bookmarked the racing shoe product page (K3)." The logical relationship is set as "K1 AND (K2 OR K3)." The filtering process then begins, using a tag matching engine to parse logical relationships and generate filtering rules: first, it filters objects aged 25-40, then matches those who "have run ≥ 4 times in the last 30 days" or "have saved racing shoe product pages." For example, object ID:2244, tagged "female, 35 years old, 2 times in the last 30 days, no racing shoe product pages saved," is excluded because it does not meet the running frequency requirement; object ID:2378, tagged "male, 38 years old, 15 times in the last 30 days, no racing shoe product pages saved," is retained because it meets the running frequency requirement, even though it has not saved racing shoe product pages; object ID:2499, tagged "female, 18 years old, 19 times in the last 30 days, have saved racing shoe product pages," is excluded because it exceeds the age limit. Ultimately, 87,000 target objects (N=87,000) that meet the composite criteria are selected from 800,000 objects. A forward sorting table containing the object IDs and their complete attribute tags is generated, and the advertising delivery process is triggered simultaneously to push ads to the target objects.

[0087] Please see Figure 7 , Figure 7 A flowchart illustrating the target object filtering process is shown. Assume the target attribute label filtering rule is: satisfying target attribute label K1, target attribute label K2, and target attribute label K3. For a given object, first determine if its attribute labels satisfy target attribute label K1. If so, then determine if it satisfies target attribute label K2. If so, then determine if it satisfies target attribute label K3. If both are satisfied, the object is a target object. If the object's attribute labels do not satisfy any of the target attribute labels, then the object is not a target object.

[0088] Please see Figure 8 , Figure 8A flowchart illustrating another target object filtering process is shown. Assume the target attribute label filtering rule is: satisfy target attribute label K1 and (target attribute label K2 or target attribute label K3). For a given object, first determine if its attribute label satisfies target attribute label K1. If so, then determine if its attribute label satisfies target attribute label K2. If so, the object is a target object. If the object's attribute label satisfies target attribute label K1 but not target attribute label K2, then determine if its attribute label satisfies target attribute label K3. If so, the object is a target object. If the object's attribute label does not satisfy target attribute label K1, then the object is not a target object; if the object's attribute label satisfies target attribute label K1 but not target attribute label K2 or target attribute label K3, then the object is not a target object.

[0089] Please see Figure 9 , Figure 9 A flowchart illustrating another target object filtering process is shown. Assume the target attribute label filtering rule is: satisfying (target attribute label K1 or target attribute label K2) and target attribute label K3. For a given object, first determine if its attribute label satisfies target attribute label K1. If yes, then determine if its attribute label satisfies target attribute label K3; otherwise, determine if its attribute label satisfies target attribute label K2. If the object's attribute label satisfies target attribute label K2, then determine if its attribute label satisfies target attribute label K3; if yes, the object is a target object. If not, the object is not a target object.

[0090] Please see Figure 10 , Figure 10 A flowchart illustrating another target object filtering process is shown. Assume the target attribute label filtering rule is: satisfying target attribute label K1, target attribute label K2, or target attribute label K3. For a given object, first determine if its attribute label satisfies target attribute label K1. If yes, the object is a target object; if not, then determine if its attribute label satisfies target attribute label K2. If yes, the object is a target object; if not, then determine if its attribute label satisfies target attribute label K3. If yes, the object is a target object; if not, the object is not a target object.

[0091] The method provided in this application, by dynamically defining logical relationships between tags, generating structured filtering rules, efficiently filtering target objects, and solidifying the results into a forward sorting table, achieves rapid parsing and accurate execution of rules for complex object packages. This process design, centered on a forward sorting table, not only solves the shortcomings of traditional technologies such as lagging rule iteration and insufficient real-time performance, but also provides a data foundation for dynamically optimizing delivery strategies (such as real-time removal of churned objects), significantly improving the agility and ROI of advertisers' marketing campaigns.

[0092] In an optional embodiment of the data processing method provided in the above embodiments of this application, the target attribute label includes at least one of the following: a numeric range type attribute label, a string range type attribute label, a sliding window date range type attribute label, a numeric exact match type attribute label, a string exact match type attribute label, a string fuzzy match type attribute label, and a label existence type attribute label;

[0093] Based on the K target attribute labels, determine the logical relationships between the K target attribute labels, including:

[0094] Based on the attribute labels of the numeric range type, string range type, sliding window date range type, numeric exact match type, string exact match type, string fuzzy match type, and tag existence type, determine the logical relationships between the attribute labels of the numeric range type, string range type, sliding window date range type, numeric exact match type, string exact match type, string fuzzy match type, and tag existence type; among them, the logical relationships include intersection and union relationships.

[0095] It is understood that this application categorizes all attribute tags into 7 types: attribute tags of numerical range type, attribute tags of string range type, attribute tags of sliding window date range type, attribute tags of numerical exact match type, attribute tags of string exact match type, attribute tags of string fuzzy match type, and attribute tags of tag existence type.

[0096] Numerical range attribute tags: Tags that use numerical values ​​as a metric and set upper and lower limits for matching (e.g., spending amount, exercise duration). For example: an object's running mileage in the last 30 days is ≥50 kilometers. One possible implementation is to use the `range_number` function to filter the numerical range: select two values, a closed interval. Then check if the object's attribute tag falls within the specified interval; if so, it matches; otherwise, it doesn't. Example:

[0097] "range_number":[

[0098] {"name":"label","value":[0,10.0]} ]

[0100] This indicates that the label value must be greater than or equal to 0 and less than or equal to 10. If the condition is not met, return "missed"; otherwise, continue to check other conditions. In one possible implementation...

[0101] Attribute tags of string range type: Matching is based on the alphabetical or lexicographical range of text attributes (such as city, product model). For example: the city is A. One possible implementation is to use `range_string` to filter string ranges, such as filtering a fixed date range: selecting two values ​​within a closed interval. It checks if an object's attribute tag falls within a specified range; if so, it matches; otherwise, it doesn't. Example:

[0102] "range_string":[

[0103] {"name":"label","value":["2022-01-01","2022-12-01"]} ]

[0105] This means the date for the label should be greater than or equal to 2022-01-01 and less than or equal to 2022-12-01. If the condition is not met, return "missed"; otherwise, continue to evaluate other conditions.

[0106] Attribute labels for sliding window date ranges: Dynamically calculate behavior within a specified time window (e.g., the last 7 days, 30 days) based on the current time. For example: purchased sports equipment within the past 30 days. One possible implementation is to use the `range_string_days` function to filter the sliding window date range: select two values, a closed interval. Check if the object's attribute label date falls within the specified interval; if so, it's a match; otherwise, it's not. Example:

[0107] "range_string_days":[

[0108] {"name":"label","value":[-2,0]} ]

[0110] Negative numbers represent the current date minus the corresponding number of days, while positive numbers represent the current date plus the specified number of days. If the current date is 2022-12-27, then the label date should be greater than or equal to 2022-12-25 and less than or equal to 2022-12-27. If this condition is not met, return "missed"; otherwise, continue evaluating other conditions.

[0111] Numeric exact match attribute tags: Tags that require a value exactly equal to a specific value (such as membership level, points balance). For example: VIP level = 3. Points balance = 5000 points. One possible implementation is to use the `terms_number` function for numeric exact match filtering: When selecting, upload a variable-length value, and check if the value of the object's attribute tag matches the given value; if it does, it's a match; otherwise, it's not. Example:

[0112] "terms_number":[

[0113] {"name":"label","value":[0,1,2]} ]

[0115] This means the label value must be equal to 0, 1, or 2. If none of these conditions are met, return "missed"; otherwise, continue evaluating other conditions.

[0116] Attribute tags for exact string matching: require text attributes to be completely identical (e.g., city, device model). For example: City = "City A" (geographic targeting). One possible implementation is to use the `terms_string` function for exact string matching filtering: a variable-length value is uploaded during selection, and the value of the object's attribute tag is checked against the given string; if it is, a match is found; otherwise, no match is found. Example:

[0117] "terms_number":[

[0118] {"name":"label","value":["a","b","c"]} ]

[0120] This means the label value must be equal to "a", "b", or "c". If none of these conditions are met, it returns a false negative; otherwise, it continues to evaluate other conditions.

[0121] Attribute tags for fuzzy string matching: Matching partial text (such as search terms or interest tags) using keywords or regular expressions. For example: browsing history containing "basketball shoes" or "cushioned running shoes". Object descriptions containing "fitness enthusiast" or "marathon runner". One possible implementation uses the `match` function for fuzzy string matching filtering: a variable-length value is uploaded during selection; the function checks if the object's attribute tag value is in the given string; if so, it's a match; otherwise, it's not. Example:

[0122] "match":[

[0123] {"name":"label","value":["a","b","c"]} ]

[0125] This means the label value must contain "a", "b", or "c". If none of these conditions are met, it returns "not found"; otherwise, it continues to evaluate other conditions.

[0126] Tag existence type attribute tags: This only checks if an object possesses a certain tag, without considering the specific value (such as behavior markers or permission indicators). For example: whether the object is following a live stream; whether it's linked to a smartwatch. One possible implementation uses the `exist` function for tag existence filtering: When selecting, a variable-length value is uploaded, and the object's attribute tag is checked for existence; if present, it's a match; otherwise, it's not. Example:

[0127] "exist":["x","y","z"]

[0128] This indicates that the label "x" exists, or "y" exists, or "z" exists. If none of these conditions are met, return "no match"; otherwise, continue to evaluate other conditions.

[0129] The intersection and union relationships in logical relations are the core rules for defining "which conditions are satisfied" when combining multiple screening criteria, and their meaning is consistent with set operations in mathematics:

[0130] Intersection relation (AND logic): All conditions must be met simultaneously; that is, the filtering result is the common intersection of all condition labels. Example:

[0131] Condition 1: Age between 25 and 35 years old

[0132] Condition 2: Running distance ≥ 50 km in the past 30 days

[0133] Logical relationship: Condition 1 AND Condition 2

[0134] Filtering results: Only those who meet both the "age requirement" and "sufficient running mileage" will be retained.

[0135] Union relation (OR logic): It is sufficient to satisfy any one of the conditions, that is, the filtering result is the combined set of all condition labels.

[0136] Example:

[0137] Condition 1: The city is "City A".

[0138] Condition 2: The city is "City B".

[0139] Logical relationship: Condition 1 OR Condition 2

[0140] Screening results: All subjects from City A or City B are retained, and the target groups from both locations are merged.

[0141] In one possible implementation, the rule logic expression is implemented using an operator function: intersection and union: logical operations between two rule parts. Example:

[0142] "operator": ["and"]

[0143] Intersection (and), Union (or).

[0144] The method provided in this application supports multiple types of tags, including numerical values, strings, dates, and fuzzy matching, as well as intersection and union logic combinations, enabling precise mapping of advertisers' complex targeting needs. This filtering mechanism, centered on multiple types of tags and flexible logical relationships, significantly improves the agility and resource utilization of ad placement, providing technical support for product advertising and marketing.

[0145] In an optional embodiment of the data processing method provided in the above embodiments of this application, an object forward sorting table is generated based on N object identifiers corresponding to N target objects and attribute tags corresponding to each of the N target objects, including:

[0146] Identify N object identifiers as N keys;

[0147] The attribute label corresponding to each of the N target objects is determined as a set of N values ​​corresponding to N keys, wherein the set of values ​​includes at least one value, and the value corresponds to the attribute label of the target object;

[0148] Generate an object forward list based on N keys and the set of N values ​​corresponding to those N keys.

[0149] Understandably, the object identifier is a key value (such as object ID or device ID) that uniquely identifies an object or device, used to distinguish different objects; in this application, the key is defined as the index field of a forward index. The unique identifier of each object is extracted as the key of the forward index, for example, using the object ID "12345" as the primary key of the table, ensuring that each object has a unique position in the table. By creating an index based on the uniqueness of the key, the target object can be quickly located directly by object ID, avoiding data redundancy. For example, an e-commerce platform can quickly retrieve historical behavior tags by object ID, improving query efficiency.

[0150] Attribute tags are metadata describing object characteristics (such as "male", "25-30 years old", "fitness enthusiast"); the value set here refers to the set of tags corresponding to the keys. Each object's attribute tags are bound to its ID for storage; for example, object ID "12345" corresponds to the tags "male, 28 years old, running enthusiast," forming a key-value pair structure. By storing tag data in a structured manner, on-demand retrieval and combined queries are supported. For example, a sports brand can quickly filter for the "25-30 years old and running enthusiast" group, reducing the overhead of full table scans.

[0151] An object forward index is a structured data table that uses object IDs as indexes and tags as values, similar to row-based storage in HBase or hash structures in Redis. Key-value pairs are arranged in order of object IDs to generate a directly queryable forward index. For example, object ID "12345" and tags "male, running enthusiast" can be stored in the table as the basis for subsequent ad targeting. By pre-compiling tag relationships, the overhead of real-time calculations is avoided. For instance, during ad targeting, the list of object IDs in the forward index can be directly accessed, significantly reducing the query latency for thousands of object packages.

[0152] For example, please refer to Table 1, which is an example of an object forward sorting table. In Table 1, the object ID refers to the key that uniquely identifies the object (e.g., 1001, 1002), serving as the primary key of the forward sorting table. Tag Types: Numerical Range Tag: such as "Recent 30-day spending = 800-1000 yuan", used to filter spending levels. String Exact Match Tag: such as "Follow 'Running Gear'", requiring an exact match of the tag name. Sliding Window Date Tag: such as "Visited the sports mall ≥ 3 times in the past 7 days", based on dynamic time window statistics. Logical Relationships: Intersection (AND): The object must simultaneously meet the conditions of "spending ≥ 800 yuan" and "following running gear". Union (OR): If the advertiser configures "follow running gear or fitness courses", then matching either tag is sufficient.

[0153] Table 1

[0154]

[0155] The method provided in this application constructs a forward sorting table by using object identifiers as keys and attribute tags as values. This method achieves a high degree of unity between data storage and query logic. Furthermore, the materialized view feature of the forward sorting table supports high-frequency reuse; for example, when the same strategy is executed for three consecutive days, the forward sorting table generated the previous day can be directly called, significantly reducing computational resource consumption.

[0156] In an optional embodiment of the data processing method provided in the above embodiments of this application, after generating the object forward sorting table, the method further includes:

[0157] Based on the object forward list, determine the object forward list identifier and the target attribute tag filtering rules corresponding to the object forward list;

[0158] The object's forward list identifier is determined as the key, and the target attribute label filtering rule is determined as the value corresponding to the key;

[0159] Store objects in a forward sorted list based on their keys and values.

[0160] Understandably, the object's forward ranking table identifier is a unique key-value pair (such as a hash value or UUID) used for quick location and management of the forward ranking table; the target attribute tag filtering rule is an expression defining the filtering logic (such as "age ≥ 25 years old and active in the last 7 days"). A unique identifier is assigned to each generated forward ranking table, and its associated filtering rules (such as "spending amount ≥ 1000 yuan") are extracted to form metadata pairs. For example, a forward ranking table identifier might be "HASH_1234," corresponding to the rule "gender = male and running distance ≥ 50 kilometers." This unique identifier enables rapid retrieval and management of the forward ranking table, supporting dynamic rule updates without affecting data storage. For example, if an advertiser modifies the "spending amount" threshold, only the rule corresponding to the identifier needs to be updated, without reconstructing the entire forward ranking table.

[0161] The key is the ascending list identifier, and the value is the filtering rule; together they form a key-value pair, such as "HASH_1234 → 'Gender = Male and running distance ≥ 50 km'". These key-value pairs are stored in a distributed key-value database (such as Redis) to ensure efficient read and write operations. For example, after binding the ascending list identifier of object ID "1001" with a rule, the corresponding rule can be directly retrieved using "HASH_1234". The key-value pair structure supports millisecond-level query response times; for example, in high-concurrency scenarios, an advertising system can retrieve identifiers and rules from thousands of ascending lists in parallel, improving throughput.

[0162] Storage refers to persisting key-value pairs to a distributed storage system (such as HBase or Cassandra) to ensure data reliability and scalability. The system distributes key-value pairs across multiple nodes according to a partitioning strategy (such as hash sharding). For example, "HASH_1234" is stored on node 3, and updates to its rules are only synchronized to that node. Distributed storage supports horizontal scaling; for example, a forward sorted table with hundreds of millions of records can be distributed across hundreds of nodes, avoiding single-point bottlenecks, while a replication mechanism ensures data disaster recovery capabilities.

[0163] In one possible implementation, the object forward index object package is stored in the object forward index storage table, such as HABSE. The storage format is that each object forward index ID corresponds to a JSON string corresponding to an object forward index rule. Since advertisers can choose any number of tags, formatted storage is not applicable. The rules can be serialized into a JSON string and stored as a value. In actual use, the JSON string can be retrieved and deserialized to obtain all the rules. The specific design of the object package rule is as follows: The object package rule includes: 1) The object package owner ID. 2) The object package rule content JSON string. The JSON string content of a single object package rule is in the following form: Rule1 Rule Protocol Logical Operator Rule2 Rule Protocol Logical Operator… A single Rule protocol includes the following tag types:

[0164] Numerical range filtering: range_number: [double, double]

[0165] String range filtering: rage_string: [string, string]

[0166] Relative date range filtering: range_string_days:[int,int]

[0167] Precise numerical filtering: terms_number:[int,…]

[0168] Precise string filtering: terms_string: [string,…]

[0169] Fuzzy string filtering: match:[string,…

[0170] The tag can be filtered: exist:[string,…

[0171] For example, please refer to Table 2, which is an example of object forward sorting table storage. In Table 2, the forward sorting table identifier is a hash value (e.g., HASH_1234) that uniquely identifies the forward sorting table and is used to quickly locate the storage location. In the example, the identifier is generated using MD5 (filtering rules) to ensure uniqueness. The target attribute tag filtering rule is an expression that defines the filtering logic, supporting intersection (AND), union (OR), and complex combinations. The example rules cover various tag types such as numeric, string, date, and existence. The storage location is the specific table name in distributed storage (e.g., user_tags_2023 in HBase), partitioned by rule type or time. In the example, all rules are stored in the same table, distributed to different Regions via a partition key (e.g., HASH_1234). The associated object ID range is the range of object IDs covered by the forward sorting table (e.g., 1001-1500), used for sharded query optimization. In the example, the range is divided into intervals of 500 IDs to reduce the amount of data in a single table. Caching strategy: Enable Redis caching for frequently accessed ascending lists (such as HASH_1234) and set TTL (Time To Live). Do not cache dynamic rules (such as HASH_3456) to ensure real-time performance.

[0172] Table 2

[0173]

[0174] The method provided in this application decouples data management from query logic by assigning a unique identifier to the object forward sorting table and binding filtering rules. Simultaneously, the key-value pair storage supports high-frequency concurrent access, such as tens of thousands of query requests per second, reducing query latency and significantly improving performance compared to traditional SQL databases.

[0175] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0176] Every first preset time interval, retrieve the attribute label corresponding to each of the N target objects in the object forward list;

[0177] Based on K target attribute labels and the attribute labels corresponding to each of the N target objects, determine T objects to be deleted. The T objects to be deleted are selected from the N target objects, and the attribute labels corresponding to the deleted objects are different from the K target attribute labels. N≥T≥1.

[0178] Delete the T object identifiers corresponding to the T deleted objects in the object forward list to obtain the updated object forward list.

[0179] Understandably, the first preset duration is a preset periodic check interval (e.g., hourly or daily) used to trigger the data update process; attribute tags are dynamic fields describing object characteristics (e.g., spending amount, active time period). The system periodically scans N target objects in the object ranking table, extracting their current attribute tags. For example, it checks every 10 minutes whether the "number of active days" for object ID:1001 has changed. This periodic update mechanism ensures that the object tags stored in the ranking table always reflect the latest status, avoiding delivery deviations caused by changes in object behavior. For example, if an object is downgraded from "high spending" to "low spending," it can be detected and marked promptly.

[0180] "K target attribute tags" are the advertising rules configured by the advertiser (e.g., "spending ≥ 500 yuan and active in the last 30 days"); "T deleted objects" are target objects that need to be removed from the positive ranking table due to mismatched attribute tags. The system compares the current object's attribute tags with the original advertising rules, filtering out objects that no longer meet the conditions. For example, if object ID:2002's "active times in the last 30 days" decreases from 5 to 2, a deletion operation is triggered. Dynamically removing invalid objects reduces invalid exposure, improves ad click-through rate, and lowers the cost of invalid traffic.

[0181] An object identifier is a key-value pair that uniquely identifies an object (e.g., object ID). The updated forward sorted table of objects is a streamlined dataset after removing invalid objects. T object IDs and their associated data are deleted in batches from the forward sorted table. For example, after cleaning up object ID:2002, only the remaining object IDs that meet the rules are retained in the forward sorted table. Storage efficiency is optimized through subtraction operations; for example, a social media platform cleans up millions of invalid object IDs daily, reducing the size of the forward sorted table and improving query performance.

[0182] For example, please refer to Figure 11 , Figure 11The flowchart illustrates the process of updating an object forward ranking table. An e-commerce platform needs to target active objects that have made purchases within the last 30 days with a total spending of ≥100 yuan. The process automatically triggers an update to the object forward ranking table every hour (a first preset duration): First, the attribute tags of the 1000 objects (N=1000) currently stored in the object forward ranking table are scanned. For example, the "Recent Purchase Time" and "Cumulative Spending Amount" fields for object ID:456 are retrieved to obtain its latest data (e.g., object ID:456's last purchase was 35 days ago, with an amount of 80 yuan). Next, the real-time tags of each object are compared with the original rules (K=2 target attribute tags: "Purchased within the last 30 days" and "Spending ≥100 yuan"), filtering out objects that do not meet the criteria (e.g., object ID:456 is marked as one of the T=50 objects to be deleted due to insufficient spending and expired time). Finally, the IDs and associated data of these objects are removed from the forward ranking table in batches, and the updated forward ranking table retains only 950 objects that meet the rules. For example, if object ID: 789 spends 120 yuan within 30 days and the time is still valid, its ID will remain in the forward ranking list. This process ensures that ads only reach active objects, improves ad click-through rate after a certain execution, reduces invalid exposure, and saves storage resources through an automated cleanup mechanism, thus reducing the amount of data stored in the forward ranking list.

[0183] The method provided in this application constructs a self-healing object forward ranking maintenance mechanism through periodic detection, dynamic filtering, and data cleaning. This reduces invalid exposure rates, improves advertising budget utilization, and supports hundreds of thousands of tag update operations per hour in scenarios with hundreds of millions of objects. Through automated data lifecycle management, this method solves the shortcomings of traditional technologies, such as frequent manual intervention and poor data real-time performance, providing advertisers with high-precision, low-cost dynamic delivery capabilities.

[0184] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0185] Every second preset time interval, obtain the attribute tags corresponding to each of the M objects, and obtain the P attribute tag filtering rules corresponding to the forward list of P objects, wherein the forward list of P objects includes M object identifiers corresponding to the M objects, and P>1;

[0186] Based on the attribute labels corresponding to each of the M objects and the filtering rules of the P attribute labels corresponding to the forward list of P objects, the object identifiers in the forward list of P objects are updated to obtain the updated forward list of P objects.

[0187] Understandably, the second preset duration is a preset periodic check interval (e.g., every 30 minutes) used to trigger the data synchronization process; the M objects are all objects in the original dataset (e.g., objects, devices), each with dynamically changing attribute tags (e.g., spending amount, activity status); the P object forward ranking tables are multiple independent datasets (e.g., forward ranking tables grouped by region or interest), each storing object IDs and their tags that conform to specific rules. All objects and their latest attribute tags are periodically scanned (e.g., updating the "number of active days in the last 7 days" for object ID:1001 every 30 minutes), while simultaneously loading the filtering rules of the P forward ranking tables (e.g., "high-spending objects," "nighttime active objects"), for example, reading the rule configurations of 10 forward ranking tables from distributed storage. By periodically synchronizing the global object status and rule base, consistency between the forward ranking tables and the original data is ensured. For example, when an object upgrades from "low spending" to "high spending," its corresponding forward ranking table can be detected and marked in a timely manner, avoiding delivery deviations caused by data lag.

[0188] Object identifier updates refer to adding or removing object IDs from the forward list based on the latest attribute tags and rule matching results (e.g., adding object ID: 2002 to the "High-Consumption Objects" forward list, or removing object ID: 3003 from the "Nighttime Active Objects" table). Parallel processing of filtering tasks across P forward lists (e.g., simultaneously updating 10 forward lists using a MapReduce framework) compares the real-time tags of each object with the matching status of the corresponding rules. For example, if object ID: 4004's "Active Time" changes from "Nighttime" to "Daytime," it is removed from the "Nighttime Active Objects" forward list. Batch updates of multiple forward lists reduce manual intervention and delays. For instance, an e-commerce platform needs to simultaneously update five forward lists during promotional periods, including "High-Consumption," "Registered Members," and "Recently Viewed Products," processing millions of object data points per hour, thus improving update efficiency.

[0189] For example, please refer to Figure 12 , Figure 12This diagram illustrates the updating of the entire object forward ranking list. A short video platform needs to target active users who have watched fitness videos ≥5 times and liked ≥3 times in the past 7 days with sports equipment advertisements. The object forward ranking list is automatically updated every 30 minutes (a second preset duration): First, the latest attribute tags of all objects (M = 50 million) are scanned. For example, for object ID: 789, the "number of times fitness videos were watched in the past 7 days" is updated from 3 to 6, and the "number of likes" increases from 2 to 4. Simultaneously, the filtering rules for P = 3 forward ranking lists are loaded: "Highly Active Objects" (watches ≥5 times and likes ≥3 times), "Newly Registered Objects" (registration time ≤ 7 days), and "High Spending Potential Objects" (shopping cart items ≥ 10). Next, the update tasks for three forward lists were processed in parallel: For the "Highly Active Objects" table, object ID: 789 was added because it met the criteria, while object ID: 1001 was removed because it didn't have enough likes; for the "Newly Registered Objects" table, object ID: 2002 (registered on the 3rd day) was added, and object ID: 3003 (registered more than 7 days ago) was removed; in the "High Spending Potential Objects" table, object ID: 4004 was added because its shopping cart contained 12 items. Ultimately, the three forward lists were updated as follows: the Highly Active Objects table added 23,000 objects and removed 17,000 objects; the Newly Registered Objects table added 5,000 objects; and the High Spending Potential Objects table added 8,000 objects. After this update, the click-through rate of the advertising system effectively improved in the next time slot, reduced resource consumption, and ensured that changes to object tags took effect quickly.

[0190] The method provided in this application, through periodic global scanning and multi-table parallel update mechanism, achieves continuous alignment between dynamic data and static rules. Through automated multi-table collaborative updates, this method solves the defects of high manual maintenance cost and lag in multi-dimensional rule synchronization in traditional technology, and provides advertisers with high-precision and low-latency dynamic delivery capabilities.

[0191] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0192] For each of the M objects:

[0193] Get the attribute label filtering rules of the object's forward list;

[0194] If the attribute label of an object matches the attribute label filtering rules, the object identifier of the object in the positive sorting table is retained;

[0195] If the attribute label of an object does not match the attribute label filtering rules, delete the object identifier of the object in the forward list.

[0196] Understandably, the attribute tag filtering rules for the object's forward ranking table are predefined logical conditions (such as "age ≥ 25 years old and active in the last 7 days"), used to determine whether an object should be retained in the forward ranking table. The forward ranking table to which an object belongs is the target object package dataset associated with that object (such as "high-spending object table" and "nighttime active object table"). For each object (e.g., object ID: 1001), all its forward ranking tables are queried (e.g., belonging to both "high-spending object table" and "male object table"), and the filtering rules of each table are loaded (e.g., "spending amount ≥ 1000 yuan" and "gender = male"). By dynamically binding objects and rules, it is ensured that each object is always matched according to the latest strategy. For example, after an object's spending is upgraded, it can be automatically included in a higher-level forward ranking table, avoiding targeting deviations caused by rule lag.

[0197] Attribute tag matching refers to an object's current tag perfectly matching the forward ranking table's filtering rules (e.g., object ID:1001's "gender = male" and "spending amount ≥ 1000 yuan"). It iterates through all objects and their associated forward ranking tables, retaining those that match the rules (e.g., object ID:1001 is retained in the "high-spending object table"), ensuring that the forward ranking table only stores valid target groups. By using positive filtering to retain high-value objects, for example, a luxury goods advertisement might only reach objects with "spending ≥ 50,000 yuan and browsing the product ≥ 10 times in the past 30 days," increasing the ad click-through rate by 45%.

[0198] Attribute tag mismatch means that the current tag of an object does not conform to the forward sorting table rules (e.g., "Consumption amount < 1000 yuan" for object ID: 1001), and it needs to be removed from the forward sorting table. Batch detection and deletion of invalid object identifiers (e.g., object ID: 1001 was removed from the "high-consumption object table" due to consumption downgrade) prevents invalid data from consuming storage resources. Data quality is optimized through subtraction operations; for example, after an e-commerce platform cleaned up object IDs that "have not logged in for the past 30 days," the storage size of the forward sorting table was reduced, thus reducing invalid exposure.

[0199] For example, please refer to Figure 13 , Figure 13This diagram illustrates a full object update. An advertiser needs to target credit card ads to high-net-worth clients who have spent ≥ 50,000 RMB in the past 3 months and hold investment products. The full object update process is triggered hourly (at a second preset interval): For all objects (M = 1 million), each object's corresponding forward list and its filtering rules are checked one by one. For example, object ID: 1001 belongs to both the "High-Spending Object Table" (rule: monthly spending ≥ 50,000 RMB) and the "Active Investment Object Table" (rule: holding ≥ 3 investment products). Its current attribute tags are loaded—object ID: 1001 has spent 62,000 RMB in the past 3 months and holds 4 investment products, meeting both rules, so its ID is retained in both forward lists; while object ID: 2002, having spent only 30,000 RMB in the past 3 months and not holding any investment products, is deleted from both forward lists. Furthermore, object ID: 3003 originally belonged to the "Active Investment Object Table," but was removed due to a rule mismatch caused by redeeming all investment products. Following this update, the "High-Spending Target List" added 12,000 targets and removed 8,000 targets, while the "Active Investment Target List" added 5,000 targets and removed 15,000 targets. The advertising system can then precisely target its ads, improving credit card application conversion rates and reducing invalid exposures. Simultaneously, a dynamic verification mechanism ensures that changes to target tags take effect within one hour.

[0200] The method provided in this application constructs an adaptive forward sorting table maintenance mechanism by dynamically binding objects and rules, retaining matching items in the forward direction, and cleaning up invalid items in the reverse direction. Through automated dynamic verification and cleanup, this method solves the defects of high manual maintenance costs and lag in multi-dimensional rule synchronization in traditional technologies, providing advertisers with high-precision, low-latency dynamic delivery capabilities and improving advertising conversion rates.

[0201] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0202] When the first time is obtained, the attribute tags corresponding to M objects are obtained, resulting in M ​​sets of first attribute tags, where each set of first attribute tags includes at least one attribute tag;

[0203] When the second time is obtained, the attribute tags corresponding to M objects are obtained, resulting in M ​​sets of second attribute tags. Each set of second attribute tags includes at least one attribute tag, and the first time is earlier than the second time.

[0204] Based on the M sets of first attribute labels and M sets of second attribute labels corresponding to M objects, determine W label change objects, where the W label change objects are selected from the M objects, and the attribute labels in the first attribute label set and the attribute labels in the second attribute label set corresponding to the label change objects are not completely the same, M≥W≥1;

[0205] When the first time is obtained, the L attribute tag filtering rules corresponding to the L objects in the forward sorting table to which the W tag change objects belong;

[0206] Based on the set of W second attribute tags corresponding to the W tag-changed objects and the L attribute tag filtering rules, the forward sorting table of L objects is updated to obtain the updated forward sorting table of L objects.

[0207] Understandably, the first time point is the initial time point (e.g., 00:00 on October 1, 2023), and the first attribute tag set is all attribute tags of the object at that time point (e.g., "Age = 30 years old" and "Active in the last 7 days = 5 times" for object ID: 1001). Batch collection of attribute tags for all objects at the initial time point is performed, for example, extracting tag data for object IDs from 1 to 1 million from a database snapshot. By solidifying the initial state, baseline data is provided for subsequent change detection, avoiding calculation errors caused by data backtracking.

[0208] The second time point is a subsequent time point (e.g., 01:00 on October 1, 2023), and the second attribute tag set is the updated tag of the object at that time point (e.g., "Age = 30 years old" and "Active in the last 7 days = 6 times" for object ID: 1001). The latest attribute tags of all objects are collected again, for example, by capturing changes in object behavior data through a real-time stream processing engine (e.g., Flink). By dynamically updating the tags, it is ensured that decisions are always based on the latest data; for example, changes in object activity can be reflected in the delivery strategy in real time.

[0209] A changed tag object is an object whose attribute tag set has changed (e.g., object ID: 1001 is marked as a changed object because its "active times in the last 7 days" increased from 5 to 6). By comparing the tag sets at two different time points, objects with differences are filtered out, for example, by using hash verification or field-by-field comparison to discover changes in the activity count of object ID: 1001. This accurate identification of changed objects avoids the overhead of a full scan and reduces computational resource consumption.

[0210] The L forward sorted tables represent the object datasets to which the changed objects belonged at the time of the change (e.g., "Highly Active Objects Table," "Highly Consuming Objects Table"), and each table is associated with specific filtering rules. The system queries the historical attribution tables of the changed objects and their corresponding rules. For example, object ID: 1001 previously belonged to the forward sorted table for "≥5 active times in the last 7 days," with the rule "≥5 active times." By associating with historical rules, the system ensures that update operations are based on the original strategy logic, avoiding accidental deletion or retention due to forgotten rules.

[0211] The updated forward ranking list is based on a recalculated dataset using the latest tags and original rules (e.g., object ID: 1001 remains in the "Highly Active Objects" table because it meets the activity threshold). Rule matching is re-executed based on the latest tags; for example, comparing the activity count of object ID: 1001 with the rules determines whether to retain or remove its ID. By dynamically adjusting the forward ranking list content, it ensures that ad delivery always targets the eligible demographics.

[0212] For example, please refer to Figure 14 , Figure 14 This diagram illustrates the updating of differential objects. An online education platform needs to distribute paid course coupons to high-value objects who have "watched courses ≥ 5 times and completed at least 1 post-course test in the past 30 days." On October 1, 2023 (first time), the initial attribute tags of object ID: 1001 were recorded, including "registered duration 7 days," "number of course views 0," and "number of post-course test completions 0." The next day (second time), its tags were updated to "number of course views 3" and "number of post-course test completions 1," and "recent login time 20:00-22:00" was added. By comparing the two tag data, it was detected that object ID: 1001's "number of course views" increased from 0 to 3 and "number of post-course test completions" increased from 0 to 1, thus determining it to be an object with changed tags. Subsequently, querying the positive sorting table to which this object belonged in the first time revealed that it was classified into the "potential active object table" (rule: ≥ 3 course views) and the "new object table" (rule: registration ≤ 7 days). Based on the latest tags, object ID: 1001 still meets the filtering criteria of the "Potential Active Objects Table" and is therefore retained in that table. However, it has been removed from the "New Objects Table" because its registration period exceeds 7 days. This update covered 1 million objects, detecting 120,000 objects with changed tags. Of these, 120,000 new objects were added and 80,000 were removed from the "Potential Active Objects Table," while 150,000 expired objects were cleaned up in the "New Objects Table." After the advertising system accurately reached the target audience based on this data, it improved the click-through rate of paid course coupons, reduced invalid exposures, and improved the efficiency of updating the forward ranking data, significantly optimizing the allocation of advertising resources.

[0213] The method provided in this application, through dual-timepoint tag snapshot comparison and rule backtracking mechanism, achieves automated monitoring of object attribute changes and dynamic updating of the forward ranking table. By employing an incremental update strategy across spatiotemporal dimensions, this method overcomes the shortcomings of traditional technologies, such as wasted full-scale computational resources and strong rule lag, providing advertisers with highly real-time, low-cost, and precise targeting capabilities.

[0214] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0215] For each of the W label change objects:

[0216] Determine the attribute tag filtering rules corresponding to the forward sorting table of each tag-modified object;

[0217] If the attribute tags in the second attribute tag set of the tag-changed object match the attribute tag filtering rules, retain the object identifier of the tag-changed object in the object's positive list.

[0218] If the attribute tags in the second attribute tag set of the object being tagged do not match the filtering rules for the attribute tags, delete the object identifier of the object being tagged in the object forward list.

[0219] Understandably, the attribute tag filtering rules corresponding to the object's positive ranking table are predefined logical conditions (such as "spending amount ≥ 1000 yuan" or "active ≥ 3 times in the last 7 days"), used to determine whether an object with a changed tag should be retained in the positive ranking table. For each object with a changed tag (e.g., object ID: 1001), the filtering rules of its corresponding positive ranking table are queried (e.g., the rule for the "high-spending object table" is "spending amount ≥ 5000 yuan"), and the historical configuration of that rule is loaded. For example, object ID: 1001 belongs to the "high-spending object table" in the first instance, and its rule is "monthly spending ≥ 5000 yuan". By dynamically binding objects and rules, it is ensured that the evaluation of each object with a changed tag is based on its historical positive ranking table strategy, avoiding accidental deletion or retention due to rule omissions or mismatches. For example, if an object is included in the high-value table due to an upgrade in spending amount, its eligibility can still be verified according to the original rules, ensuring the consistency of the targeting strategy.

[0220] Attribute tag matching refers to the requirement that the current tag set of an object undergoing tag changes (e.g., "Monthly spending = 6000 yuan" for object ID: 1001) completely matches the forward ranking table filtering rules (e.g., "Monthly spending ≥ 5000 yuan"). All object objects undergoing tag changes are iterated through, and their latest tags are checked to see if they meet the rules of their respective forward ranking tables. For example, if object ID: 1001's "monthly spending" increases from 4000 yuan to 6000 yuan, it still meets the "≥ 5000 yuan" condition and is therefore retained in the "High-spending Objects Table". This forward filtering retains high-value objects, ensuring that ads accurately reach the target audience. For example, an advertiser might only push new product information to objects with "monthly spending ≥ 10,000 yuan" to increase ad click-through rates while reducing invalid exposures.

[0221] Attribute tag mismatch means that the current tag set of the object whose tag has been changed (e.g., "Monthly consumption = 4000 yuan" for object ID: 1001) cannot meet the filtering rules of the forward sorting table (e.g., "Monthly consumption ≥ 5000 yuan"). Batch detection and deletion of invalid object identifiers are performed. For example, object ID: 1001 no longer meets the rules of the "high-consumption object table" because its consumption level has been downgraded to 4000 yuan, and its ID is removed. Data quality is optimized through subtraction operations. For example, after an e-commerce platform cleaned up object IDs that "have not logged in for the past 30 days," the storage size of the forward sorting table was reduced, thus reducing invalid exposure.

[0222] For example, please refer to Figure 15 , Figure 15 This diagram illustrates the update of object tags. An e-commerce platform needs to distribute limited-time discount coupons to high-intent customers who have "viewed products ≥ 5 times and collected ≥ 2 items in the past 7 days." On October 1, 2023 (first time), the initial attribute tags of object ID: 4004 were recorded as "3 views" and "0 collections." The next day (second time), its tags were updated to "6 views" and "1 collection." By comparing the two tags, it was determined that object ID: 4004's "collection count" increased from 0 to 1, but its "view count" increased from 3 to 6, still meeting the "view ≥ 5 times" rule. Therefore, its ID was retained in the "high-intent customer table." At the same time, object ID: 5005 was removed from the table because its "view count" decreased from 4 to 2, triggering step S430. This update covered 100,000 objects, detecting 12,000 objects with changed tags. Among them, 3,000 people were added and 9,000 people were removed from the "high-intent customer table," reducing invalid exposure. After the advertising system accurately reaches the target audience, it increases the redemption rate of limited-time discount coupons, improves the conversion rate of the target audience, and at the same time improves the efficiency of positive sorting updates, significantly reducing resource consumption.

[0223] The method provided in this application achieves efficient maintenance and accurate updates of the forward sorting table through a two-way verification mechanism of dynamic rule binding, forward retention, and reverse cleanup. By automating dynamic verification and cleanup, this method overcomes the shortcomings of traditional technologies, such as high manual maintenance costs and lag in multi-dimensional rule synchronization. It provides advertisers with high-precision, low-latency dynamic delivery capabilities, thereby improving ad conversion rates.

[0224] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0225] The terminal receives an information display instruction, which indicates that the terminal's interface includes an information display area and that the terminal has a corresponding relationship with any one of N target objects.

[0226] The first piece of information is displayed in the information display area.

[0227] It's understandable that information display instructions are operation requests triggered by the terminal (such as an object opening an app's homepage or clicking on an ad slot), used to indicate the area in the interface where information needs to be displayed (such as the homepage ad slot or an interstitial ad in the news feed). The correspondence between the terminal and the target object refers to the association between the terminal device (such as the object's mobile phone) and the advertiser's specified target audience (such as a specific target group or a list of device IDs) (e.g., object ID: 1001 belongs to the "high-spending target table"). The system receives instructions from the terminal, confirms the existence of an available information display area in the current interface, and verifies whether the terminal belongs to the target object set (e.g., the app homepage ad slot for object ID: 1001 is triggered). For example, when an object is browsing the homepage of an e-commerce app, if their device ID matches the "high-spending target table," the ad display process is triggered. By dynamically binding the terminal and the target object, the accuracy of information display is ensured, and invalid exposure is avoided. For example, a luxury goods ad is only displayed to high-spending targets, increasing the click-through rate by 50% while reducing interference from non-target objects.

[0228] The first piece of information consists of the specific content to be displayed after filtering (such as promotional advertisements and product recommendations), which highly matches the target audience's attribute tags (such as spending level and interests). After verifying the association between the terminal and the target audience, advertisements or information that meet the rules are populated into designated areas (e.g., displaying a high-end watch advertisement in the homepage featured image for object ID: 1001). For example, object ID: 1001 sees product recommendations related to its interests because it meets the rule of "spending ≥ 10,000 yuan in the past 30 days". By matching rules with object status in real time, information relevance is improved. For example, an e-commerce platform increased the conversion rate by 35% by dynamically displaying advertisements for products viewed by the object.

[0229] For example, please refer to Figure 16 , Figure 16 A schematic diagram of the first information display is shown. (For example...) Figure 16 In (A), when the target user scrolls down while browsing tweets, and information display area 1601 appears, as shown... Figure 16 (B) in the information display area displays the first piece of information 1602.

[0230] The method provided in this application achieves a precise information delivery loop through terminal triggering mechanism and dynamic information matching; by real-time association between the terminal and the target object and dynamic content filling, this method solves the defects of information delivery lag and coarse object matching in traditional technology, and provides advertisers with a high-conversion-rate and low-cost intelligent delivery solution.

[0231] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0232] Secondary information delivery is performed on the MN objects that do not belong to the object forward list.

[0233] Understandably, the object forward ranking list stores the identifiers of target objects that meet the initial filtering rules (such as a list of object IDs), for example, a set of objects filtered by the rule of "spending ≥ 100 yuan in the last 30 days". MN objects: The remaining objects from the total number of objects (M) that did not enter the forward ranking list (such as objects filtered due to insufficient spending or low activity levels). Secondary information delivery: For objects not covered by the forward ranking list, differentiated content is pushed based on supplementary rules or secondary strategies (such as "low-spending potential objects" or "newly registered objects").

[0234] The N objects already covered by the forward sorting table are removed from the full set of objects. The remaining MN objects are then further filtered according to preset supplementary rules. For example, object ID:2002 is pushed a low-price promotional advertisement because its consumption amount does not reach the threshold of "high-consumption object", but it meets the secondary rule of "viewing products ≥ 3 times in the past 7 days".

[0235] By employing a tiered targeting strategy, potential audiences not reached by the main advertising list can be covered. For example, an e-commerce platform increased the conversion rate of new audiences by pushing first-order discount ads to "low-spending active audiences" while avoiding fatigue caused by repeated exposure to the main advertising list audience.

[0236] For example, please refer to Figure 17 , Figure 17 A schematic diagram of a differentiated information delivery method is shown. The first information is delivered to target terminals A and B, while the second information is delivered to non-target terminal C.

[0237] The method provided in this application achieves precise coverage of the entire target audience through a two-tiered mechanism of primary and supplementary ad placement. This solution overcomes the shortcomings of traditional technologies, such as incomplete coverage and coarse audience segmentation due to single strategies, providing advertisers with intelligent ad placement capabilities that balance efficiency and accuracy.

[0238] For easier understanding, please refer to Figure 18 , Figure 18This is a schematic diagram provided for an embodiment of this application. An e-commerce platform needs to target high-value individuals who have spent ≥500 yuan in the past 30 days and browsed maternal and infant products ≥3 times in the past 7 days with milk powder promotional advertisements, and achieve real-time and precise ad delivery. First, create target object rules using the object package creation function: After logging into the object platform, the advertiser selects the numerical range filter tag "spending amount ≥ 500 yuan" (type: range_number, value range [500, +∞]) and the sliding window date range filter tag "number of views of maternal and infant products ≥ 3 times in the past 7 days" (type: range_string_days, value: [-7, 0]), sets the logical relationship to "intersection (AND)", generates a unique object package ID (e.g., UUID: 550e8400-e29b-41d4-a716-446655440000), and serializes the rules into a JSON string (e.g., {"operator":"and","rules":[{"range_number":[{"name":"spending amount","value":[500, +∞)}]},{"range_string_days":[{"name":"number of views","value":[-7, 0]}]}]}) and stores it in HBase.

[0239] Subsequently, object attribute tags are updated in real time using Spark Streaming: when object ID: 1001 completes a maternal and infant product order, its "Consumption Amount" tag is updated from 300 yuan to 600 yuan, and the "Number of Maternal and Infant Browsing Visits in the Last 7 Days" tag increases from 2 to 4. The stream processing engine constructs filtering rules based on 7 preset tag types (such as range_number and range_string_days) to detect in real time whether the object meets the target object package conditions. If it does, the object package ID 550e8400-e29b-41d4-a716-446655440000 is written to the profile forward sorting table (indexed by object ID, with column storage of object package ID sets), and the object package update process is triggered: on the one hand, historical rules are queried through HBase, and on the other hand, combined with real-time tag changes, a concurrent verification mechanism (Scala multi-threading) is used to match object attributes with rule protocols (such as parsing range_number and range_string_days conditions in JSON), and finally the matching object package IDs are updated to the forward sorting table.

[0240] During the personalized ad delivery phase, when object ID: 1001 opens the app ad slot, the backend service queries its corresponding object package ID set and finds it contains 550e8400-e29b-41d4-a716-446655440000. It then retrieves the corresponding milk powder promotional material from the ad library and displays it. If the object does not match any object package, the default ad is displayed. Through high-concurrency queries using forward sorting storage (direct indexing by object ID) and real-time updates using streaming computing (processing millions of tag changes per second), ad clicks are improved, and invalid exposures are reduced. Furthermore, a scheduled full update (triggered daily at midnight) is set up to iterate through all object attribute tags and object package rules, ensuring that updates missed due to streaming processing failures (such as object ID: 2002 not being updated in real-time due to network latency) are batch-corrected, guaranteeing data consistency. This solution achieves a dual breakthrough in the accuracy and real-time performance of ad delivery through a multi-level dynamic update mechanism and efficient storage design.

[0241] In an optional embodiment of the data processing method provided in the above embodiments of this application, the method further includes:

[0242] Based on the streaming computing framework, obtain the attribute label change identifier for each of the M objects;

[0243] When the attribute label change indicator indicates that the attribute label of an object has changed, the forward list of objects to which the object belongs is updated.

[0244] Understandably, streaming computing frameworks (such as real-time data pipeline tools like Flink or Kafka Streams) are used to capture change events of object attribute tags (such as modifying personal information or adding followed accounts), and a dynamic "change flag" (such as a boolean value or timestamp) is generated for each object to indicate whether its attributes have been updated. For example, when user ID:1001's "Recently Viewed Category" changes from "Electronic Products" to "Sports Equipment," the system immediately marks its change flag as "Updated." By replacing the traditional polling mechanism with event-driven streaming processing, only objects that have actually changed are captured, avoiding invalid scans of all objects, reducing computational resource consumption, and ensuring that the forward sorting table is always generated based on the latest state.

[0245] When the change indicator shows a change in the object's attribute tag, the corresponding forward index is adjusted accordingly. For example, if object ID 1001 is updated due to the addition of the "Sports Equipment" browsing tag, the system removes its ID from the "Electronic Products Interest Groups" table and adds it to the "Sports Preference Groups" table, while simultaneously updating the index information in the Redis cache. This incremental update mechanism, rather than a full recalculation, ensures precise operations only on affected objects. This reduces the maintenance latency of the forward index, avoids resource contention issues caused by full table rebuilds, and guarantees the real-time accuracy of user grouping during ad delivery.

[0246] The method provided in this application addresses the shortcomings of traditional technologies, such as wasted computing resources and insufficient real-time performance, and provides advertisers with highly flexible and low-latency dynamic delivery capabilities.

[0247] This application also provides a data processing apparatus; please refer to [link to relevant documentation]. Figure 19 In (A), the data processing device 1900 includes:

[0248] Module 1901 is used to obtain M objects, where each object corresponds to at least one attribute label, and M > 1;

[0249] The receiving module 1902 is used to receive an information delivery instruction, wherein the information delivery instruction carries K target attribute labels and the logical relationship between the K target attribute labels, and the information delivery instruction is used to represent the first information delivery to an object that satisfies the K target attribute labels and the logical relationship between the K target attribute labels, where K≥1;

[0250] The generation module 1903 is used to generate an object forward list based on K target attribute tags and M objects. The object forward list includes N object identifiers corresponding to N target objects and attribute tags corresponding to each of the N target objects. The object forward list is indexed by the N object identifiers as the primary key. The N target objects are selected from the M objects. The attribute tag of each target object is the same as at least one of the K target attribute tags. Furthermore, the attribute tag of each target object satisfies the logical relationship between the K target attribute tags, where M≥N≥1.

[0251] The delivery module 1904 is used to deliver the first information to N target objects in the object forward list based on the information delivery instruction.

[0252] The device provided in this application integrates the scattered tag filtering, object matching, and ad delivery processes of traditional devices into a single operation by constructing an object forward index. When advertisers dynamically configure object package rules, they only need to scan the pre-calculated forward index to complete the object attribution determination, avoiding the full table scan overhead of traditional inverted indexes and reducing the query time for thousands of object packages. At the same time, the incremental update mechanism of the forward index (such as automatic refresh after object tag changes) ensures the real-time nature of data and solves the defects of offline calculation lag and high rule iteration cost in traditional technologies. In scenarios with hundreds of millions of objects, this device supports high-frequency delivery requests. Advertisers do not need to wait for offline calculation results; they can adjust object package rules in real time and the changes will take effect in a short time, significantly improving marketing agility and resource utilization, resulting in a significant increase in ad click-through rate compared to traditional devices and reducing invalid exposures.

[0253] In an optional embodiment of the data processing apparatus provided in the above embodiments of this application, please refer to... Figure 19 In (B), the data processing device 1900 further includes:

[0254] Module 1905 is used to determine the logical relationship between K target attribute labels based on K target attribute labels;

[0255] The generation module 1903 is also used to generate target attribute label filtering rules based on the K target attribute labels and the logical relationship between the K target attribute labels;

[0256] The filtering module 1906 is used to filter N target objects that meet the target attribute label filtering rules based on the attribute labels corresponding to each of the M objects.

[0257] The generation module 1903 is also used to generate an object forward list based on the N object identifiers corresponding to the N target objects and the attribute tags corresponding to each target object in the N target objects.

[0258] The apparatus provided in this application, by dynamically defining logical relationships between tags, generating structured filtering rules, efficiently filtering target objects, and solidifying the results into a forward sorting table, achieves rapid parsing and accurate execution of rules for complex object packages. This process design, centered on a forward sorting table, not only solves the shortcomings of traditional technologies such as lagging rule iteration and insufficient real-time performance, but also provides a data foundation for dynamically optimizing delivery strategies (such as real-time removal of churned objects), significantly improving the agility and ROI of advertisers' marketing activities.

[0259] In an optional embodiment of the data processing apparatus provided in the above embodiments of this application, the target attribute label includes at least one of the following: a numerical range type attribute label, a string range type attribute label, a sliding window date range type attribute label, a numerical exact match type attribute label, a string exact match type attribute label, a string fuzzy match type attribute label, and a label existence type attribute label;

[0260] The determination module 1905 is further used to determine the logical relationships between attribute labels of the following types based on attribute labels of the following types: numeric range type, string range type, sliding window date range type, numeric exact match type, string exact match type, string fuzzy match type, and tag existence type. The logical relationships include intersection and union relationships.

[0261] The apparatus provided in this application supports multiple types of tags, including numerical values, strings, dates, and fuzzy matching, as well as combinations of intersection and union logic, enabling precise mapping of advertisers' complex targeting needs. This filtering mechanism, centered on multiple types of tags and flexible logical relationships, significantly improves the agility and resource utilization of ad placement, providing technical support for product advertising and marketing.

[0262] In one optional embodiment of the data processing apparatus provided in the above embodiments of this application,

[0263] The determination module 1905 is also used to determine N object identifiers as N keys;

[0264] The determination module 1905 is further configured to determine the attribute label corresponding to each of the N target objects as a set of N values ​​corresponding to N keys, wherein the set of values ​​includes at least one value, and the value corresponds to the attribute label of the target object;

[0265] The generation module 1903 is also used to generate an object forward list based on N keys and the set of N values ​​corresponding to the N keys.

[0266] The apparatus provided in this application constructs a forward sorting table by using object identifiers as keys and attribute tags as values, achieving a high degree of unification between data storage and query logic. Furthermore, the materialized view characteristic of the forward sorting table supports high-frequency reuse; for example, when the same strategy is executed for three consecutive days, the forward sorting table generated the previous day can be directly called, significantly reducing computational resource consumption.

[0267] In an optional embodiment of the data processing apparatus provided in the above embodiments of this application, please refer to... Figure 19 In (C), the data processing device 1900 further includes:

[0268] The determination module 1905 is also used to determine the object forward list identifier and the target attribute tag filtering rules corresponding to the object forward list based on the object forward list;

[0269] The determination module 1905 is also used to determine the object's forward list identifier as the key and the target attribute label filtering rule as the value corresponding to the key;

[0270] Storage module 1907 is used to store objects in a forward list based on keys and values.

[0271] The apparatus provided in this application decouples data management from query logic by assigning unique identifiers to object forward sorting tables and binding filtering rules. Simultaneously, the key-value pair storage supports high-frequency concurrent access, such as tens of thousands of query requests per second, reducing query latency and significantly improving performance compared to traditional SQL databases.

[0272] In an optional embodiment of the data processing apparatus provided in the above embodiments of this application, please refer to... Figure 19 In (D), the data processing device 1900 further includes:

[0273] The acquisition module 1901 is also used to acquire the attribute label corresponding to each of the N target objects in the object positive list every first preset time interval;

[0274] The determination module 1905 is also used to determine T objects to be deleted based on K target attribute labels and the attribute labels corresponding to each target object in N target objects. The T objects to be deleted are selected from the N target objects, and the attribute labels corresponding to the deleted objects are different from the K target attribute labels, and N≥T≥1.

[0275] Update module 1908 is used to delete the T object identifiers corresponding to the T deleted objects in the object forward list to obtain the updated object forward list.

[0276] The device provided in this application embodiment constructs a self-healing object forward ranking maintenance mechanism through periodic detection, dynamic filtering, and data cleaning. This reduces invalid exposure rates, improves advertising budget utilization, and supports hundreds of thousands of tag update operations per hour in scenarios with hundreds of millions of objects. Through automated data lifecycle management, this device solves the shortcomings of traditional technologies, such as frequent manual intervention and poor data real-time performance, providing advertisers with high-precision, low-cost dynamic delivery capabilities.

[0277] In one optional embodiment of the data processing apparatus provided in the above embodiments of this application,

[0278] The acquisition module 1901 is also used to acquire the attribute label corresponding to each of the M objects every second preset time interval, and to acquire the P attribute label filtering rules corresponding to the positive sorting table of P objects, wherein the positive sorting table of P objects includes M object identifiers corresponding to the M objects, and P>1;

[0279] The update module 1908 is also used to update the object identifiers in the P objects forward list based on the attribute labels corresponding to each of the M objects and the filtering rules of the P attribute labels corresponding to the P objects forward list, so as to obtain the updated P objects forward list.

[0280] The device provided in this application embodiment achieves continuous alignment between dynamic data and static rules through periodic global scanning and multi-table parallel update mechanism. Through automated multi-table collaborative update, the device solves the defects of high manual maintenance cost and multi-dimensional rule synchronization lag in traditional technology, and provides advertisers with high-precision and low-latency dynamic delivery capabilities.

[0281] In one optional embodiment of the data processing apparatus provided in the above embodiments of this application,

[0282] The 1901 module is also used to obtain the attribute label filtering rules of the object's forward list to which the object belongs;

[0283] Update module 1908 is also used to retain the object identifier of the object in the positive sort table when the attribute label corresponding to the object matches the attribute label filtering rule;

[0284] Update module 1908 is also used to delete the object identifier of the object in the object's forward list when the attribute label corresponding to the object does not match the attribute label filtering rules.

[0285] The device provided in this application embodiment constructs an adaptive forward ranking table maintenance mechanism by dynamically binding objects and rules, retaining matching items in the forward direction, and cleaning up invalid items in the reverse direction. Through automated dynamic verification and cleanup, the device solves the defects of high manual maintenance costs and multi-dimensional rule synchronization lag in traditional technologies, providing advertisers with high-precision and low-latency dynamic delivery capabilities and improving advertising conversion rates.

[0286] In one optional embodiment of the data processing apparatus provided in the above embodiments of this application,

[0287] The acquisition module 1901 is also used to acquire the attribute tags corresponding to the M objects at the first time, and obtain M sets of first attribute tags, wherein each set of first attribute tags includes at least one attribute tag;

[0288] The acquisition module 1901 is also used to acquire the attribute tags corresponding to M objects at the second time, and obtain M sets of second attribute tags, wherein each set of second attribute tags includes at least one attribute tag, and the first time is earlier than the second time.

[0289] The determination module 1905 is also used to determine W tag change objects based on the M sets of first attribute tags and M sets of second attribute tags corresponding to M objects, wherein the W tag change objects are selected from the M objects, and the attribute tags in the first attribute tag set corresponding to the tag change objects are not completely the same as the attribute tags in the second attribute tag set, and M≥W≥1;

[0290] The acquisition module 1901 is also used to acquire the L attribute tag filtering rules corresponding to the L object positive sorting table to which the W tag change objects belong at the first time.

[0291] The update module 1908 is also used to update the forward sorting table of L objects based on the set of W second attribute tags corresponding to the W tag change objects and the L attribute tag filtering rules, so as to obtain the updated forward sorting table of L objects.

[0292] The device provided in this application embodiment achieves automated monitoring of object attribute changes and dynamic updating of the forward ranking table through a dual-timepoint tag snapshot comparison and rule backtracking mechanism. By employing an incremental update strategy across spatiotemporal dimensions, this device overcomes the shortcomings of traditional technologies, such as wasted full-scale computing resources and strong rule lag, providing advertisers with highly real-time, low-cost, and precise targeting capabilities.

[0293] In one optional embodiment of the data processing apparatus provided in the above embodiments of this application,

[0294] Module 1905 is also used to determine the attribute tag filtering rules corresponding to the object forward list of each tag change object;

[0295] Update module 1908 is also used to retain the object identifier of the tag change object in the object positive list when the attribute tags in the second attribute tag set of the tag change object match the attribute tag filtering rules;

[0296] Update module 1908 is also used to delete the object identifier of the object in the object forward list when the attribute tags in the second attribute tag set of the tag change object do not match the attribute tag filtering rules.

[0297] The device provided in this application embodiment achieves efficient maintenance and accurate updates of the forward sorting table through a two-way verification mechanism of dynamic rule binding, forward retention, and reverse cleanup. By automating dynamic verification and cleanup, this device overcomes the shortcomings of traditional technologies, such as high manual maintenance costs and lag in multi-dimensional rule synchronization, providing advertisers with high-precision, low-latency dynamic delivery capabilities and improving ad conversion rates.

[0298] In an optional embodiment of the data processing apparatus provided in the above embodiments of this application, please refer to... Figure 19 In (E), the data processing device 1900 further includes:

[0299] The receiving module 1909 is used to receive information display instructions sent by the terminal, wherein the information display instructions indicate that the terminal's interface includes an information display area, and the terminal has a corresponding relationship with any one of N target objects;

[0300] Display module 1910 is used to display the first information in the information display area.

[0301] The device provided in this application embodiment achieves a precise information delivery closed loop through terminal triggering mechanism and dynamic information matching; through real-time association between the terminal and the target object and dynamic content filling, the device solves the defects of information delivery lag and coarse object matching in traditional technology, and provides advertisers with a smart delivery solution with high conversion rate and low cost.

[0302] In one optional embodiment of the information delivery device provided in the above embodiments of this application,

[0303] The delivery module 1904 is also used to deliver second information to MN objects that do not belong to the object forward list.

[0304] The device provided in this application embodiment achieves precise coverage of the entire target audience through a two-layer mechanism of primary and supplementary placement using a forward-ordering table. This device solves the shortcomings of traditional technologies, such as incomplete coverage and coarse target segmentation due to a single strategy, providing advertisers with intelligent placement capabilities that balance efficiency and accuracy.

[0305] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0306] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0307] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0308] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0309] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0310] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a server or terminal device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing computer programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0311] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A data processing method, characterized in that, include: Obtain M objects, where each object corresponds to at least one attribute label, and M > 1; Receive information delivery instruction, wherein the information delivery instruction carries K target attribute tags and the logical relationship between the K target attribute tags, and the information delivery instruction is used to characterize the first information delivery to an object that satisfies the K target attribute tags and the logical relationship between the K target attribute tags, where K≥1; Based on the K target attribute tags and the M objects, an object forward sorting table is generated. The object forward sorting table includes N object identifiers corresponding to the N target objects, and attribute tags corresponding to each of the N target objects. The object forward sorting table uses the N object identifiers as the primary key index. The N target objects are selected from the M objects. The attribute tag of each target object is the same as at least one of the K target attribute tags, and the attribute tag of each target object satisfies the logical relationship between the K target attribute tags, where M≥N≥1. Based on the information delivery instruction, the first information delivery is performed on N target objects in the object forward list.

2. The data processing method as described in claim 1, characterized in that, The step of generating a forward sorted list of objects based on the K target attribute tags and the M objects includes: Based on the K target attribute labels and the logical relationships between them, generate target attribute label filtering rules; Based on the target attribute tag filtering rules, N target objects that satisfy the target attribute tag filtering rules are selected from the M objects according to the attribute tags corresponding to each of the M objects; A forward sorting table of objects is generated based on the N object identifiers corresponding to the N target objects and the attribute tags corresponding to each of the N target objects.

3. The data processing method as described in claim 2, characterized in that, The target attribute label includes at least one of the following: attribute label of numerical range type, attribute label of string range type, attribute label of sliding window date range type, attribute label of numerical exact match type, attribute label of string exact match type, attribute label of string fuzzy match type, and attribute label of label existence type; Determining the logical relationship between the K target attribute tags based on the K target attribute tags includes: Based on the attribute labels of the numerical range type, the string range type, the sliding window date range type, the numerical exact match type, the string exact match type, the string fuzzy match type, and the tag existence type, determine the logical relationships among the attribute labels of the numerical range type, the string range type, the sliding window date range type, the numerical exact match type, the string exact match type, the string fuzzy match type, and the tag existence type; wherein, the logical relationships include intersection relationships and union relationships.

4. The data processing method as described in claim 2, characterized in that, The step of generating an object forward sorting table based on the N object identifiers corresponding to the N target objects and the attribute tags corresponding to each of the N target objects includes: The N object identifiers are identified as N keys; The attribute label corresponding to each of the N target objects is determined as a set of N values ​​corresponding to the N keys, wherein the set of values ​​includes at least one value that corresponds to the attribute label of the target object; Generate an object forward list based on the N keys and the N sets of values ​​corresponding to the N keys.

5. The data processing method as described in claim 2, characterized in that, After generating the forward sorted list of objects, the following is also included: Based on the object forward list, determine the object forward list identifier and the target attribute tag filtering rules corresponding to the object forward list; The object's forward list identifier is determined as the key, and the target attribute label filtering rule is determined as the value corresponding to the key; The object is stored in a forward list based on the key and value.

6. The data processing method according to any one of claims 1-5, characterized in that, The method further includes: Every first preset time interval, obtain the attribute label corresponding to each of the N target objects in the object positive list; Based on K target attribute labels and the attribute labels corresponding to each of the N target objects, T objects to be deleted are determined, wherein the T objects to be deleted are selected from the N target objects, and the attribute labels corresponding to the deleted objects are different from the K target attribute labels, and N≥T≥1; Delete the T object identifiers corresponding to the T deleted objects in the object forward list to obtain the updated object forward list.

7. The data processing method according to any one of claims 1-5, characterized in that, The method further includes: Every second preset time interval, obtain the attribute tags corresponding to each of the M objects, and obtain the P attribute tag filtering rules corresponding to the P object forward list, wherein the P object forward list includes M object identifiers corresponding to the M objects, and P>1; Based on the attribute labels corresponding to each of the M objects and the filtering rules of the P attribute labels corresponding to the P object forward list, the object identifiers in the P object forward list are updated to obtain the updated P object forward list.

8. The data processing method as described in claim 7, characterized in that, The method further includes: For each of the M objects: Obtain the attribute label filtering rules of the object's forward sorting list; If the attribute label corresponding to the object matches the attribute label filtering rule, the object identifier of the object in the positive sorting table is retained; If the attribute label corresponding to the object does not match the attribute label filtering rule, delete the object identifier of the object in the object's forward list.

9. The data processing method according to any one of claims 1-5, characterized in that, The method further includes: When the first time is obtained, the attribute tags corresponding to the M objects are obtained to obtain M sets of first attribute tags, wherein each set of first attribute tags includes at least one attribute tag; When the second time is obtained, the attribute tags corresponding to the M objects are used to obtain M sets of second attribute tags, wherein each set of second attribute tags includes at least one attribute tag, and the first time is earlier than the second time; Based on the M sets of first attribute tags and M sets of second attribute tags corresponding to the M objects, W tag change objects are determined, wherein the W tag change objects are selected from the M objects, and the attribute tags in the first attribute tag set corresponding to the tag change object are not completely the same as the attribute tags in the second attribute tag set, and M≥W≥1; When the first time is obtained, the L attribute tag filtering rules corresponding to the L object forward list to which the W tag change objects belong; Based on the set of W second attribute tags corresponding to the W tag change objects and the L attribute tag filtering rules, the L object forward sorting table is updated to obtain the updated L object forward sorting table.

10. The data processing method as described in claim 9, characterized in that, The method further includes: For each of the W label change objects: Determine the attribute tag filtering rules corresponding to the forward list of objects for each tag change object; If the attribute tags in the second attribute tag set of the tag-changing object match the attribute tag filtering rules, the object identifier of the tag-changing object in the positive list of objects is retained; If the attribute tags in the second attribute tag set of the object whose tag has been changed do not match the attribute tag filtering rules, delete the object identifier of the object whose tag has been changed from the object's positive list.

11. The data processing method according to any one of claims 1-10, characterized in that, The method further includes: For MN objects that do not belong to the object positive list, a second information is delivered, wherein the first information is information that corresponds to the attribute tags of the N target objects, and the second information is information that is unrelated to the attribute tags of the MN objects.

12. The data processing method according to any one of claims 1-10, characterized in that, The method further includes: Based on the streaming computing framework, the attribute label change identifier corresponding to each of the M objects is obtained, wherein the attribute label change identifier is determined based on the attribute label change event; If the attribute label change indicator indicates that the attribute label of the object has changed, the forward list of objects to which the object belongs is updated.

13. A data processing apparatus, characterized in that, include: The acquisition module is used to acquire M objects, wherein each object corresponds to at least one attribute label, and M > 1; A receiving module is used to receive an information delivery instruction, wherein the information delivery instruction carries K target attribute tags and the logical relationship between the K target attribute tags, and the information delivery instruction is used to characterize the first information delivery to an object that satisfies the K target attribute tags and the logical relationship between the K target attribute tags, where K≥1; The generation module is used to generate an object forward list based on the K target attribute tags and the M objects. The object forward list includes N object identifiers corresponding to the N target objects and attribute tags corresponding to each of the N target objects. The object forward list uses the N object identifiers as the primary key index. The N target objects are selected from the M objects. The attribute tag of each target object is the same as at least one of the K target attribute tags, and the attribute tag of each target object satisfies the logical relationship between the K target attribute tags, where M≥N≥1. The delivery module is used to deliver first information to N target objects in the object forward list based on the information delivery instruction.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the data processing method according to any one of claims 1 to 12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the data processing method according to any one of claims 1 to 12.

16. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the data processing method according to any one of claims 1 to 12.

Images