An advertisement question and answer method and system, a computer device and a medium

Abstract

The application provides an advertisement question and answer method and system, a computer device and a medium, comprising: acquiring a link fingerprint, cutting a reply knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and performing retrieval matching on each independent branch knowledge base and a question raised by a user to obtain an initial reply to the question; further binding a level to the link fingerprint to form a white list, performing secondary verification on the initial reply according to the white list to obtain a final reply to the question, and pushing the final reply to the user. The application can assign a unique link identifier to each log by designing a three-dimensional link fingerprint, and solve the problem of multiple link data confusion. By constructing a branch level knowledge base sliced according to the link fingerprint, accurate matching of a target link can be realized when searching for a reply to a question, and cross-link data pollution can be eliminated. Furthermore, by creating a branch level white list mechanism, secondary verification can be performed before a reply is pushed to a user, so that cross-link illusion can be inhibited at the source.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an advertising question-and-answer method and system, computer equipment and media. Background Technology

[0002] Product advertising aims to promote products that users want to buy according to an advertising plan. For example, product ads can be placed in media, and when users see the ads placed by advertisers in the media, they can take actions such as purchasing, registering, or installing.

[0003] To evaluate the effectiveness of advertising campaigns, ad attribution analysis is typically performed. Because the upstream of ad attribution involves dimensions such as carrier, ad placement, and attribution mode, the tracking fields, reporting timing, and link nodes vary, forming multiple independent data link branches. However, existing large-scale model question-answering systems treat all logs as "flat text," failing to distinguish link branches; they also mix click attribution logs from carrier A with channel attribution logs from carrier B; and they create the illusion of "cross-link" errors, leading to unreproducible operational results. Summary of the Invention

[0004] This invention provides an advertising question-and-answer method and system, computer equipment and medium to solve or alleviate the technical problems described above.

[0005] The present invention provides an advertising question-and-answer method, comprising the following steps: Obtain a link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad placement and attribution mode; The answer knowledge base is divided into multiple independent branch knowledge bases according to the link fingerprint, and each independent branch knowledge base is searched and matched with the question raised by the user to obtain the initial answer to the question; A whitelist is formed based on the link fingerprint binding level, and the initial answer is verified a second time according to the whitelist to obtain the final answer to the question, and the final answer is pushed to the user.

[0006] In one embodiment of the present invention, the process of forming a link fingerprint based on advertising parameters includes: Obtain the original event tracking logs, which include the advertising parameters; The link fingerprint generation tool is called on the tracking software development kit, and the advertising parameters are transmitted to the link fingerprint generation tool to form an advertising parameter string. The advertising parameter string is encrypted, and a portion of the encryption result is extracted as the link fingerprint; and the link fingerprint and preset exposure log are added to the original event tracking log, and the supplemented original event tracking log is sent to the message queue. The message queue is synchronized to a preset analytical database using a stream processing framework and stored in a log partition associated with the link fingerprint, so that the link fingerprint can be output through the preset analytical database.

[0007] In one embodiment of the present invention, the process of dividing the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and retrieving and matching each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question includes: The answer knowledge base is grouped according to the link fingerprint, and the logs, question answers and investigation plans within a preset time period are aggregated based on the grouping results to obtain aggregated question and answer slices; The aggregated question-and-answer slices are converted into vector codes of a preset dimension using a natural language model, and the vector codes are written into a preset vector database. A corresponding vector set is created based on the link fingerprint, and the vector set is associated with the log partition corresponding to the link fingerprint in a preset analytical database. Based on the user's question, an approximate nearest neighbor search is performed on the vector set of the preset vector database to find a matching answer and troubleshooting solution; and, Retrieve the attribution failure logs of the link fingerprint within a preset time period from the preset analytical database; The matched question answers and troubleshooting solutions, along with the attribution failure logs within a preset time period, are associated, and the corresponding association results are used as the initial answer to the question.

[0008] In one embodiment of the present invention, the process of performing secondary verification on the initial answer according to the whitelist to obtain the final answer to the question includes: The initial answer is then subjected to a secondary verification based on the level in the whitelist to determine whether the question answer and investigation plan corresponding to the initial answer exceed the preset level. If the initial response exceeds the preset level, it will be intercepted and corrected, and the corrected result of the initial response will be used as the final response. If the initial answer does not exceed the preset level, it will be used as the final answer.

[0009] In one embodiment of the present invention, the method further includes: The ad parameters at the current time are compared with the ad parameters at the previous time in the original event log to determine whether the ad parameters at the two times are the same. If the advertising parameters at the current moment are different from those at the previous moment, a new link fingerprint is regenerated based on the advertising parameters at the current moment. The answer knowledge base is then re-segmented into multiple independent branch knowledge bases according to the new link fingerprint. A new log partition associated with the new link fingerprint is created in a preset analytical database. A new vector set corresponding to the new link fingerprint is created in a preset vector database, and the new vector set is associated with the new log partition. A new whitelist is formed based on the binding level of the new link fingerprint, and subsequent questions raised by the user are verified according to the new whitelist. If the ad parameters at the current moment are the same as those at the previous moment, then the user's subsequent questions will be answered based on the link fingerprint generated at the previous moment.

[0010] In one embodiment of the present invention, before retrieving and matching each independent branch knowledge base with the questions raised by the user, the method further includes: The system receives questions submitted by users on the front end and transmits them to the gateway interface via the Hypertext Transfer Security Protocol (HTTP). The gateway interface then converts the questions into a preset data exchange format, which is used to search and match each independent branch knowledge base according to the preset data exchange format.

[0011] In one embodiment of the present invention, the process of encrypting the advertising parameter string and extracting a portion of the encryption result as the link fingerprint includes: using a hash encryption algorithm to perform hash encryption on the advertising parameter string, and extracting the first sixteen digits of the hash encryption result as the link fingerprint.

[0012] The present invention also provides an advertising question-and-answer system, the system comprising: The link fingerprint module is used to obtain a link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad position and attribution mode. The branch retrieval module is used to divide the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and to retrieve and match each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question; The verification module is used to bind the link fingerprint to a whitelist, perform secondary verification on the initial answer according to the whitelist, obtain the final answer to the question, and push the final answer to the user.

[0013] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the advertising question-and-answer method described in any one of the above.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the advertising question-and-answer method described in any one of the above.

[0015] The beneficial effects of this invention are as follows: This invention proposes an advertising question-and-answer method, system, computer device, and medium. It obtains a link fingerprint based on advertising parameters, then divides the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint. Each independent branch knowledge base is then searched and matched with the user's question to obtain an initial answer. Next, the link fingerprint is bound to a level to form a whitelist, and the initial answer is verified a second time according to the whitelist to obtain the final answer. The final answer is then pushed to the user. The advertising parameters include the carrier, the ad placement, and the attribution mode. Therefore, this invention, by designing a three-dimensional link fingerprint of "carrier-ad placement-attribution mode," can assign a unique link identifier to each log, solving the problem of multi-link data confusion. Simultaneously, by constructing a branch-level knowledge base sliced ​​according to the link fingerprint, this invention can achieve accurate matching of the target link during question-and-answer retrieval, preventing cross-link data pollution. Furthermore, by creating a branch-level whitelist mechanism, this invention can perform a second verification operation on the initial answer through the whitelist before pushing the final answer to the user, thereby suppressing cross-link illusions at the source. Furthermore, when the data structure of the original event logs is adjusted, the link fingerprint can be automatically updated to ensure that the link fingerprint and knowledge base are adjusted synchronously with business changes. The entire process requires no manual intervention, ensuring business continuity and the accuracy of answering questions. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] In the attached diagram: Figure 1 A schematic diagram illustrating the illusion generated by the absence of branch constraints in related technologies; Figure 2 This is a flowchart illustrating an advertising question-and-answer method provided in one embodiment of the present invention; Figure 3 A flowchart illustrating an advertising question-and-answer method provided in another embodiment of the present invention; Figure 4 This is a schematic diagram of a link fingerprint-driven branch-level knowledge base provided in one embodiment of the present invention; Figure 5This is a timing diagram illustrating an application scenario of the advertising question-and-answer method provided in one embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an advertising question-and-answer system provided in one embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of a computer device suitable for implementing one or more embodiments of the present invention. Detailed Implementation

[0018] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0021] like Figure 1 As shown, existing related technologies lack a link branch constraint mechanism in the question-and-answer retrieval stage, indiscriminately recalling data from the "global log pool." For example, when a user asks "No conversion from incentivized video ads in mini-programs," it may incorrectly recall "APP splash screen click attribution logs," resulting in a situation where "click attribution logs from carrier A are mixed with channel attribution logs from carrier B." This leads to the retrieved logs being completely unrelated to the user's actual question link, creating potential for subsequent incorrect answers. Furthermore, due to cross-link data pollution during the retrieval stage, when LLM (Large Language Model) generates answers based on incorrectly associated logs, it is prone to "cross-link illusion"—outputting erroneous conclusions unrelated to the current question link, such as... Figure 1The example incorrectly answered "Please check the splash screen tracking". Furthermore, existing technologies do not record the specific log sources of the response, making it impossible for operations personnel to trace the original data links corresponding to the conclusion, reproduce the conclusion derivation process, verify the accuracy of the response, and consequently, effectively investigate advertising attribution issues.

[0022] Based on the above description Figure 2 A flowchart illustrating an advertising question-and-answer method is shown. Specifically, in an exemplary embodiment, such as... Figure 2 As shown, this embodiment provides an advertising question-and-answer method, including the following steps: S210, obtain the link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad position and attribution mode; S220: The answer knowledge base is divided into multiple independent branch knowledge bases according to the link fingerprint, and each independent branch knowledge base is searched and matched with the question raised by the user to obtain the initial answer to the question; S230: Link fingerprints are bound to different levels to create a whitelist. Initial answers are then validated a second time against this whitelist to obtain the final answer to the question, which is then pushed to the user. In some examples, the levels for binding link fingerprints include, but are not limited to, Q&A level, investigation level, simple interface level, and complex script level.

[0023] In an exemplary embodiment, the process of generating a link fingerprint based on advertising parameters may include: obtaining raw event tracking logs, which include advertising parameters; calling a link fingerprint generation tool on the event tracking software development kit (SDK) and transmitting the advertising parameters to the link fingerprint generation tool, concatenating the advertising parameters to form an advertising parameter string; encrypting the advertising parameter string and extracting a portion of the encryption result as the link fingerprint; supplementing the raw event tracking logs with the link fingerprint and a preset exposure log, and sending the supplemented raw event tracking logs to a message queue; synchronizing the message queue to a preset analytical database through a stream processing framework and storing it in a log partition associated with the link fingerprint, so as to output the link fingerprint through the preset analytical database. The process of encrypting the advertising parameter string and extracting a portion of the encryption result as the link fingerprint may include: hashing the advertising parameter string using a hash encryption algorithm and extracting the first sixteen bits of the hash encryption result as the link fingerprint.

[0024] In some examples, the specific process of generating link fingerprints based on advertising parameters may include: 1. Input data.

[0025] Original event tracking log: {"asset":"APP(Android)","adslot":"Splash screen","attr_mode":"SKAN","user_id":"U78901","ad_id":"AD2345","event_type":"Exposure","timestamp":"2024-XX-XXXX:XX:XX"}. Here, asset represents the carrier, adslot represents the ad slot, attr_mode represents the attribution mode, user_id represents the user ID, ad_id represents the ad ID, event_type represents the event type, and timestamp represents the timestamp.

[0026] 2. Processing flow.

[0027] Fingerprint generation: On the event tracking SDK side, call the "link fingerprint generation tool" and pass in the asset (APP(Android)), adslot (launch screen), and attr_mode (SKAN) from the original event tracking logs to form the string "APP(Android)|launch screen|SKAN".

[0028] Fingerprint encryption: Encrypt the string "APP(Android)|Open screen|SKAN" using SHA256, and extract the first 16 bits to obtain the link fingerprint. For example, the extracted link fingerprint is: 3a5b7c9d1e2f4g6h.

[0029] Field additions: Add the link fingerprint (link_fingerprint: 3a5b7c9d1e2f4g6h) and the original exposure log blood_depth (set to L0) to the original event tracking log.

[0030] Persistent storage: The logs with supplemented fields are first sent to the message queue Kafka for traffic shaping, and then synchronized in real time to the analytical database ClickHouse through the stream processing framework Flink, and stored in the log partition associated with the link fingerprint link_fingerprint: 3a5b7c9d1e2f4g6h.

[0031] 3. Output results.

[0032] The key log information stored in the analytical database ClickHouse includes: link_fingerprint: 3a5b7c9d1e2f4g6h, original exposure log blood_depth: L0, asset: APP (Android), ad slot: splash screen, attribution mode attr_mode: SKAN, user ID: U78901, and event_type: exposure.

[0033] Therefore, by designing a three-dimensional link fingerprint of "carrier-ad placement-attribution mode", a unique link identifier can be assigned to each log, thus solving the problem of data confusion across multiple links.

[0034] In an exemplary embodiment, the process of dividing the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and retrieving and matching each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question may include: grouping the answer knowledge base according to the link fingerprint, and aggregating the logs, question answers, and investigation solutions within a preset time period based on the grouping results to obtain aggregated question-answer slices; converting the aggregated question-answer slices into vector codes of a preset dimension using a natural language model, and writing the vector codes into a preset vector database; creating a corresponding vector set according to the link fingerprint, and associating the vector set with the log partition corresponding to the link fingerprint in a preset analytical database; performing an approximate nearest neighbor (ANN) search in the vector set of the preset vector database according to the question raised by the user to find matching question answers and investigation solutions; retrieving the attribution failure logs of the link fingerprint within a preset time period from the preset analytical database; associating the matched question answers and investigation solutions with the attribution failure logs within the preset time period, and using the corresponding association results as the initial answer to the question.

[0035] In some examples, the specific process of obtaining an initial answer may include: 1. Offline build process (executed automatically daily).

[0036] Scheduled trigger: Start the Spark offline task at 02:00 every day (during off-peak business hours); Data aggregation: Grouped by link fingerprint: 3a5b7c9d1e2f4g6h, aggregate the logs of the past 24 hours, historical FAQs (Frequently Asked Questions), and troubleshooting solutions for the corresponding link fingerprint; historical FAQs may include "reasons for failure of splash screen ad SKAN attribution", and troubleshooting solutions may include "checking SKAN configuration parameters".

[0037] Vector encoding: The Sentence-BERT model is used as the natural language model to convert the aggregated FAQ and investigation plan into a 768-dimensional vector; Sub-database write: Write 768-dimensional vector data into the vector database Milvus, create a vector set named kb_3a5b7c9d1e2f4g6h, and simultaneously associate it with the log partition corresponding to the link fingerprint: 3a5b7c9d1e2f4g6h in the analytical database ClickHouse.

[0038] 2. Online search process (when users ask questions).

[0039] Intent recognition: When a user asks "Why did the SKAN attribute of the APP (Android) splash screen ad fail?", the corresponding link fingerprint is identified as: 3a5b7c9d1e2f4g6h; Precise retrieval: Based on the user's question, the branch retrieval tool performs an approximate nearest neighbor search in the vector set of the vector database. The branch retrieval tool only performs an ANN search in the vector set kb_3a5b7c9d1e2f4g6h of the Milvus vector database to find matching FAQs and troubleshooting solutions. Log correlation: Retrieve the attribution failure logs for the past 24 hours from the analytical database ClickHouse (link_fingerprint: 3a5b7c9d1e2f4g6h) as factual basis for the answer.

[0040] 3. Output results.

[0041] The initial response (including traceability information) included: "APP (Android) splash screen ad SKAN attribution failed. Possible reasons: 1. Incorrect 'conversion event ID' in the SKAN configuration parameters (refer to FAQID: FAQ-SKAN-001); 2. In the past 24 hours, there have been 8 log entries showing 'parameter verification failed' in this link (log IDs: LOG-8765 to LOG-8772, link_fingerprint: 3a5b7c9d1e2f4g6h). It is recommended to check the SKAN conversion event ID configuration first." Therefore, by dividing the knowledge base into independent branch knowledge bases according to the link fingerprint, operators do not need to understand the complex data structure of the entire link. They only need to supplement FAQs and troubleshooting plans for specific links, which reduces the threshold for operation and maintenance and improves the efficiency of knowledge maintenance. Moreover, by building branch-level knowledge bases by link fingerprint slicing, it is possible to accurately match the target link during retrieval and prevent cross-link data pollution.

[0042] In an exemplary embodiment, the process of performing secondary verification on the initial answer according to the whitelist to obtain the final answer to the question may include: performing secondary verification on the initial answer according to the level in the whitelist, and determining whether the question answer and investigation plan corresponding to the initial answer exceed the preset level; if it exceeds the preset level, the initial answer is intercepted and corrected, and the corrected result of the initial answer is used as the final answer; if it does not exceed the preset level, the initial answer is directly used as the final answer.

[0043] In some examples, the specific process of performing secondary validation on the initial answer according to the whitelist to obtain the final answer may include: Whitelist configuration: The configuration center stores the whitelist for this link. Examples may include: {"link_fingerprint": "3a5b7c9d1e2f4g6h", "max_measure_level": "investigation level", "allow_measures": ["answer basic questions", "provide investigation steps", "call the attribution failure log query interface"]}; Verification process: LLM initially generates the response: "It is recommended to execute the SKAN data repair script (complex script level) to repair the attribution data"; The whitelist validator determines that "execute data repair script" exceeds the "investigation level" and automatically intercepts and corrects the response: "This link does not currently support data repair script operations. It is recommended to investigate using the following steps: 1. Verify the SKAN conversion event ID; 2. Check the connectivity of the attribution callback interface (details can be obtained by calling the 'attribution failure log query interface')."

[0044] Therefore, by using a "link fingerprint-driven branch-level knowledge base + measure whitelist dual verification," cross-link data pollution is blocked at the source of retrieval, preventing LLM from generating erroneous conclusions unrelated to the current link and significantly reducing the cross-link illusion rate. Thus, by creating a branch-level whitelist mechanism, the initial answer can be verified a second time before the final answer is pushed to the user, suppressing cross-link illusions at the source.

[0045] In an exemplary embodiment, the advertising question-answering method can also be dynamically updated. Specifically, the dynamic update process may include: comparing the advertising parameters of the original event log at the current moment with the advertising parameters at the previous moment to determine whether the advertising parameters at the two moments are the same; if the advertising parameters at the current moment are different from the advertising parameters at the previous moment, then regenerating a new link fingerprint based on the advertising parameters at the current moment, and re-segmenting the answer knowledge base into multiple independent branch knowledge bases according to the new link fingerprint, creating a new log partition associated with the new link fingerprint in a preset analytical database, and creating a new vector set corresponding to the new link fingerprint in a preset vector database, and associating the new vector set with the new log partition; forming a new whitelist based on the binding level of the new link fingerprint, and verifying the questions subsequently raised by the user according to the new whitelist; if the advertising parameters at the current moment are the same as the advertising parameters at the previous moment, then answering the questions subsequently raised by the user based on the link fingerprint generated at the previous moment.

[0046] In some examples, the dynamic update trigger scenario could be: a new "Quick App - Interstitial Screen - Click Attribution" link is added to the business logic, and the event tracking SDK detects a new "Quick App" value in the carrier asset field, triggering a dynamic update. Specifically, the dynamic update process can include: The dynamic updater automatically generates a new link fingerprint. An example of a new link fingerprint could be 8f6g4e2d0c9b7a5c. Create a new log partition in the analytical database ClickHouse, and create a new vector set kb_8f6g4e2d0c9b7a5c in the vector database Milvus; Write the whitelist of new link fingerprints (initially set to "Q&A level") into the configuration center; Results: Within 30 seconds, all online service nodes will pull the new configuration. When users subsequently ask questions about "click attribution issues for interstitial ads in Quick Apps", the new link fingerprint and the new independent branch knowledge base will be automatically matched without requiring a service restart.

[0047] Therefore, when the data structure schema changes, the system can automatically respond to the changes, add / merge link fingerprints and branch knowledge bases, and complete configuration synchronization within 30 seconds without restarting the service. This ensures business continuity and the accuracy of Q&A, and guarantees that the knowledge base is accurate as the business evolves.

[0048] In one exemplary embodiment, before searching and matching each independent branch knowledge base with the user's question, the process may further include: receiving the question submitted by the user on the front end, transmitting the question to a gateway interface (API gateway) via Hypertext Transfer Protocol Secure (HTTPS), and using the gateway interface to convert the question into a preset data exchange format for searching and matching each independent branch knowledge base according to the preset data exchange format. In some examples, the preset data exchange format may be JSON (JavaScript Object Notation).

[0049] In another exemplary embodiment of this application, an advertising question-and-answer method is provided, comprising the following steps: 1. Link Metadata Model: Construct a unique link identification system.

[0050] Core design: Define a three-dimensional key value of <carrier, ad placement, attribution mode> as a "link fingerprint", and attach "lineage depth" to each ad attribution log; the lineage depth can be represented by L0-L3, where L0 represents the original tracking log, and L1-L3 represent logs that have been processed and aggregated 1-3 times respectively, used to identify the upstream and downstream levels of data.

[0051] Link fingerprinting: Different links are distinguished by a unique triple string. For example, "APP (iOS) - Incentivized Video - Click Attribution" and "Mini Program - Information Flow - Channel Attribution" correspond to different fingerprints, ensuring that each link has a unique identifier.

[0052] Generation method: On the tracking SDK side, the three-dimensional key value is encrypted using the SHA256 algorithm, and the first 16 bits of the encrypted result are extracted as the final link fingerprint to ensure the uniqueness and simplicity of the link fingerprint.

[0053] Implementation logic: "Link fingerprint + lineage depth" are used as fixed fields, embedded in each original log, and reported synchronously to the database storage system along with the log, laying the foundation for subsequent link-based processing.

[0054] 2. Branch-level Knowledge Base: Enables knowledge storage and retrieval with isolated data links. Core Design: Breaking away from the traditional flat storage model of a "global log pool," the knowledge required for answering questions (including original logs, FAQs, and troubleshooting solutions) is divided into independent sub-bases based on "link fingerprints." Each sub-base corresponds to only one link, avoiding data confuse across links. The corresponding storage and retrieval logic is as follows: Storage architecture: It adopts a combination of "ClickHouse (stores the original logs of the corresponding link) + Milvus (stores the vector encoding of the knowledge of the corresponding link)," with each link fingerprint corresponding to 1 ClickHouse log partition and 1 Milvus vector set; Sub-database naming rules: Milvus vector sets should be uniformly named kb_<link fingerprint> (e.g., kb_3a5b7c9d1e2f4g6h) to ensure accurate link matching during retrieval; Retrieval constraints: After a user asks a question, the "link fingerprint" corresponding to the question is first determined by the intent recognition model. Then, data is retrieved only in the ClickHouse partition and Milvus vector set corresponding to the fingerprint, thus preventing cross-link data pollution from the source.

[0055] 3. Branch-level measures whitelist: A dual verification mechanism before AI output.

[0056] Core design: Each "link fingerprint" is bound to an "executable measure level" to form a whitelist configuration. Before generating an answer, the large language model LLM must pass the verification and only allow the output of operation suggestions within the whitelist to block out-of-scope calls and avoid "cross-link illusion".

[0057] The measures are categorized by level (based on operational complexity and risk): Q&A level: basic Q&A (e.g., "What fields are included in the tracking points of this link?"); troubleshooting level: simple troubleshooting guidance (e.g., "Check the timing of conversion event reporting"); simple interface level: allows calling basic data query interfaces (e.g., "Query the conversion volume of this link in the past hour"); complex script level: only open to high-trust links (e.g., "Execute the link data repair script").

[0058] Verification process: After the Large Language Model (LLM) generates the initial response, the whitelist verifier extracts the operation type from the initial response and compares it with the "maximum measure level" bound to the link fingerprint. If the operation type exceeds the maximum measure level, it is automatically intercepted and corrected to ensure that the final response is compliant and matches the current link.

[0059] 4. Dynamic Link Fingerprint Updater: An automatic synchronization mechanism adapted to business evolution. Core Design: Addressing the issue of frequent adjustments to the tracking schema (data structure) in ad attribution scenarios (such as adding carrier types or adjusting ad placement fields), an automatic update mechanism is designed to ensure that the link fingerprint and knowledge base are synchronized with business changes without manual intervention. Corresponding update logic: Triggering condition: When the event tracking SDK detects a schema change (such as the addition of a "Quick App" carrier type), the "link fingerprint update event" is automatically triggered. Processing rules: If the change is a new link (such as adding "Quick App - Interstitial Screen - Click Attribution"), a new link fingerprint and corresponding sub-library will be created automatically; if the change is an optimization of an existing link (such as adding non-core tracking fields to an existing link), it will be merged into the original link fingerprint and the sub-library knowledge will be updated synchronously. Effective time: Hot reloading is achieved through the configuration center. All online service nodes will automatically pull the new configuration within 30 seconds after the update, without the need to restart the service, ensuring business continuity.

[0060] Therefore, this method forcibly binds {link fingerprint, lineage depth, and original log association information} to each answer. Operators can locate the specific links and data sources that the answer depends on with one click, ensuring that the answer is 100% traceable, easily reproducing the conclusion derivation process, and reducing the cost of review and problem investigation.

[0061] According to the above description, in one example, such as Figure 3 As shown, this demonstrates the entire process of an advertising Q&A method from "advertiser asking a question" to "outputting a compliant answer," including the front-end, API gateway, AI Q&A core (link fingerprint generator, branch retrieval device, whitelist validator), hierarchical log repository (ClickHouse), branch knowledge base (Milvus), and advertising attribution platform. Key steps are marked "only if fingerprint matches," reflecting the constraints of the entire process and ensuring that each step does not deviate from the current link, thus eliminating the illusion of cross-links from the architectural level.

[0062] According to the above description, in some examples, such as Figure 4 As shown, the independent sub-library architecture of three typical links (Mini Program - Information Flow - Channel Attribution, APP - Incentivized Video - Click Attribution, H5 - Splash Screen - SKAN) is demonstrated. Each link corresponds to the combination of "ClickHouse Log Repository + Milvus Vector Library". After a user asks a question, the link fingerprint is used to filter and only the data in the corresponding sub-library is retrieved. The answer "only related to this branch" is output, thereby achieving link isolation.

[0063] According to the above description, in some examples, the advertising Q&A methods described in the above embodiments can be applied in various scenarios, such as e-commerce marketing attribution Q&A, local life order attribution Q&A, game customer acquisition attribution Q&A, and financial promotion attribution Q&A. Specifically, e-commerce marketing attribution Q&A: adapts to the scenario of "multi-channel (APP / mini-program / H5) × multiple activities (discounts / live streaming) × multiple conversions (clicks / orders)," accurately matching link Q&A and avoiding cross-channel conclusion confusion. Local life order attribution Q&A: covers the scenario of "multi-carrier (APP / Alipay mini-program) × multiple services (food delivery / in-store) × multiple attributions (clicks / verification)," isolating link data and preventing Q&A confusion. Game customer acquisition attribution Q&A: for the scenario of "multi-terminal (mobile games / web games) × multiple advertisements (trial play / splash screen) × multiple conversions (downloads / payments)," ensures that AI only calls target link data, eliminating cross-terminal interference. Financial promotion attribution Q&A: Applied to scenarios involving multiple platforms (APP / H5) × multiple promotions (financial management / credit card application) × multiple attributions (clicks / forms), dynamically adapting to changes in tracking points to ensure Q&A uses the latest link knowledge. Taking e-commerce marketing attribution Q&A as an example, such as... Figure 5 As shown, the process involves several steps: **Fingerprint Generation:** At 02:15 AM on June 18th, it was discovered that the conversion rate of the "Mini Program - Incentivized Video - Channel Attribution" link had fallen below the threshold. Immediately, the link fingerprint "mv_rew_channel" was generated and its lineage depth (L2) was marked, and written to the ClickHouse log partition. **Fingerprint Sub-library Storage:** Upon receiving a new link fingerprint, the knowledge base automatically creates an independent sub-library, Milvuscollection "kb_mv_rew_channel," which stores only the FAQs and log segments for that link fingerprint and is physically isolated from other link fingerprints. **Q&A Validation (Use):** When operations personnel input "Mini Program Incentivized Video No Conversion" into the Q&A interface, the link fingerprint "mv_rew_channel" is locked first, and then the segment is retrieved via vector within this independent sub-library to prevent cross-link contamination. **Measures Whitelist Validation:** Based on the "Investigation Script" permission level for that link fingerprint, the script is pushed to the operations platform after generation and can be executed with a click. Schema changes (evolution) include: if the deeplink_flag field is added to the event log the day after June 18, the original link fingerprint will be automatically expanded to "mv_rew_channel_d", and the old template from June 18 will be copied to rebuild the index. Subsequent Q&A will take effect according to the rebuilt index, thus eliminating the need for manual data transfer.

[0064] In summary, the advertising question-answering method proposed in this invention obtains a link fingerprint based on advertising parameters, then divides the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and searches and matches each independent branch knowledge base with the user's question to obtain the initial answer to the question. Next, a whitelist is formed by binding levels to the link fingerprints, and the initial answer is further verified according to the whitelist to obtain the final answer to the question, which is then pushed to the user. The advertising parameters include the carrier, the ad placement, and the attribution mode. Therefore, this method, by designing a three-dimensional link fingerprint of "carrier-ad placement-attribution mode," can assign a unique link identifier to each log, solving the problem of multi-link data confusion. Simultaneously, by constructing a branch-level knowledge base sliced ​​according to the link fingerprint, this method can achieve accurate matching of the target link during question answer retrieval, preventing cross-link data pollution. Furthermore, by creating a branch-level whitelist mechanism, this method can perform a secondary verification operation on the initial answer through the whitelist before pushing the final answer to the user, thereby suppressing cross-link illusions at the source. Furthermore, when the data structure of the original event logs is adjusted, the link fingerprint can be automatically updated to ensure that the link fingerprint and knowledge base are adjusted synchronously with business changes. The entire process requires no manual intervention, ensuring business continuity and the accuracy of answering questions.

[0065] In another exemplary embodiment of this application, such as Figure 6 As shown, an advertising question-and-answer system is provided, including: Link fingerprint module 610 is used to obtain a link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad position and attribution mode; The branch retrieval module 620 is used to divide the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and to retrieve and match each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question; The verification module 630 is used to bind the link fingerprint level, form a whitelist, perform secondary verification on the initial answer according to the whitelist, obtain the final answer to the question, and push the final answer to the user.

[0066] It is understood that the advertising question-answering system and the advertising question-answering method provided in the above embodiments belong to the same concept. The specific way in which the advertising question-answering method performs its operations has been described in detail in the above method embodiments and will not be repeated here. In practical applications, the advertising question-answering system provided in the above embodiments can allocate the above functions to different functional modules as needed. That is, the internal structure of the advertising question-answering system can be divided into different functional modules, and then all or part of the functions of the corresponding functional modules can be implemented through the advertising question-answering method described in the above embodiments. For example, all or part of the functions of the link fingerprint module 610 can be implemented through the relevant execution process of step S210, all or part of the functions of the branch retrieval module 620 can be implemented through the relevant execution process of step S220, and all or part of the functions of the verification module 630 can be implemented through the relevant execution process of step S230. For specific implementation processes, please refer to the above embodiments, and no specific limitations are imposed here.

[0067] In summary, the advertising question-answering system proposed in this invention obtains a link fingerprint based on advertising parameters, then divides the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and searches and matches each independent branch knowledge base with the user's question to obtain the initial answer to the question. Next, a whitelist is formed by binding levels to the link fingerprints, and the initial answer is further verified according to the whitelist to obtain the final answer to the question, which is then pushed to the user. The advertising parameters include the carrier, the ad placement, and the attribution mode. Therefore, this system, by designing a three-dimensional link fingerprint of "carrier-ad placement-attribution mode," can assign a unique link identifier to each log, solving the problem of multi-link data confusion. Simultaneously, by constructing a branch-level knowledge base sliced ​​according to the link fingerprint, this system can achieve accurate matching of the target link during question answer retrieval, preventing cross-link data pollution. Furthermore, by creating a branch-level whitelist mechanism, this system can perform a secondary verification operation on the initial answer through the whitelist before pushing the final answer to the user, thereby suppressing cross-link illusions at the source. Furthermore, when the data structure of the original event logs is adjusted, the link fingerprint can be automatically updated to ensure that the link fingerprint and knowledge base are adjusted synchronously with business changes. The entire process requires no manual intervention, ensuring business continuity and the accuracy of answering questions.

[0068] In an exemplary embodiment of the present invention, a computer device is also provided. The computer device may include a memory, a processor, and a computer program stored in the memory. The processor can execute the computer program to cause the computer device to perform actions such as... Figure 2 or Figure 4 The steps of the advertising Q&A method are shown below. Figure 7 A schematic diagram of the structure of a computer device 1000 is shown. (See attached diagram.) Figure 7 As shown, the computer device 1000 includes: a processor 1010, a memory 1020, a power supply 1030, a display unit 1040, and an input unit 1060.

[0069] The processor 1010 is the control center of the computer device 1000. It connects various components via interfaces and lines, and performs various functions of the computer device 1000 by running or executing computer programs / instructions stored in the memory 1020, thereby providing overall monitoring of the computer device 1000. In some embodiments, when the processor 1010 calls a computer program stored in the memory 1020, it can execute, for example... Figure 2 or Figure 4 The steps of the advertising question-and-answer method are shown. Optionally, the processor 1010 may include one or more processing units; preferably, the processor 1010 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. In some embodiments, the processor 1010 and the memory 1020 may be implemented on a single chip; in other embodiments, they may be implemented on separate chips.

[0070] The memory 1020 mainly includes a program storage area and a data storage area. The program storage area can store the operating system, various applications, etc.; the data storage area can store instruction data created according to the use of the computer device 1000. In addition, the memory 1020 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0071] The computer device 1000 also includes a power supply 1030 (such as a battery) that supplies power to various components. The power supply can be logically connected to the processor 1010 through a power management system, thereby enabling the management of functions such as charging, discharging, and power consumption through the power management system.

[0072] The display unit 1040 can be used to display information input by the user or information provided to the user, and can also be used to display various menus of the computer device 1000, etc. In this embodiment of the invention, it is mainly used to display the display interfaces of various applications in the computer device 1000, as well as text, pictures, and other objects displayed in the display interfaces. The display unit 1040 may include a display panel 1050. The display panel 1050 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0073] The input unit 1060 can be used to receive information such as numbers or characters input by the user. The input unit 1060 may include a touch panel 1070 and other input devices 1080. The touch panel 1070 can also be referred to as a touch screen, and the touch panel 1070 can collect touch operations on or near the user (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 1070).

[0074] Specifically, the touch panel 1070 can detect user touch operations and the signals generated by these operations, convert these signals into touch point coordinates and send them to the processor 1010, and receive and execute commands transmitted by the processor 1010. Furthermore, the touch panel 1070 can employ various input methods such as resistive, capacitive, infrared, and surface acoustic waves to achieve interaction. Other input devices 1080 include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, and joystick.

[0075] Of course, the touch panel 1070 can also cover the display panel 1050. When the touch panel 1070 detects a touch operation on or near it, it can transmit the information to the processor 1010 to determine the type of touch event. Subsequently, the processor 1010 provides corresponding visual output on the display panel 1050 based on the type of touch event. Although in Figure 7 In this embodiment, the touch panel 1070 and the display panel 1050 are two separate components to realize the input and output functions of the computer device 1000. However, in some embodiments, the touch panel 1070 and the display panel 1050 can be integrated to realize the input and output functions of the computer device 1000.

[0076] The computer device 1000 may also include one or more sensors, such as pressure sensors, gravity acceleration sensors, proximity sensors, etc. Of course, depending on the specific application scenario, the computer device 1000 may also include other components such as cameras.

[0077] In an exemplary embodiment of the present invention, a computer-readable storage medium is also provided, which stores a computer program / instructions. When executed by a processor, the computer program / instructions enable the computer device to perform the functions described in the present invention. Figure 2 or Figure 4 The steps of the advertising Q&A method are shown below.

[0078] It will be understood by those skilled in the art that Figure 7This is merely an example of a computer device and does not constitute a limitation on the device. The device may include more or fewer components than illustrated, or a combination of certain components, or different components. For ease of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, in implementing this invention, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0079] Those skilled in the art will understand that the present invention can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described in accordance with flowcharts and / or block diagrams of advertising question-and-answer methods, advertising question-and-answer systems, and computer program products according to some embodiments. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be applied to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowcharts and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] It is understood that the above embodiments, in collecting, storing, using, processing, transmitting, providing, disclosing, and deleting relevant data (such as raw event tracking logs, advertising parameters, etc.), are carried out with or with the user's consent. For example, raw event tracking logs, advertising parameters, etc., are obtained with the user's knowledge and consent; or are provided voluntarily by the user after reading the relevant instructions; or are actively authorized / provided / uploaded by the user when using some or all of the functions described in the above embodiments; or are obtained through other means or channels with the user's consent.

[0081] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An advertising question-and-answer method, characterized in that, The method includes: Obtain a link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad placement and attribution mode; The answer knowledge base is divided into multiple independent branch knowledge bases according to the link fingerprint, and each independent branch knowledge base is searched and matched with the question raised by the user to obtain the initial answer to the question; A whitelist is formed based on the link fingerprint binding level, and the initial answer is verified a second time according to the whitelist to obtain the final answer to the question, and the final answer is pushed to the user.

2. The advertising question-and-answer method according to claim 1, characterized in that, The process of generating link fingerprints based on advertising parameters includes: Obtain the original event tracking logs, which include the advertising parameters; The link fingerprint generation tool is called on the tracking software development kit, and the advertising parameters are transmitted to the link fingerprint generation tool to form an advertising parameter string. The advertising parameter string is encrypted, and a portion of the encryption result is extracted as the link fingerprint; and the link fingerprint and preset exposure log are added to the original event tracking log, and the supplemented original event tracking log is sent to the message queue. The message queue is synchronized to a preset analytical database using a stream processing framework and stored in a log partition associated with the link fingerprint, so that the link fingerprint can be output through the preset analytical database.

3. The advertising question-and-answer method according to claim 2, characterized in that, The process of dividing the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and then searching and matching each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question includes: The answer knowledge base is grouped according to the link fingerprint, and the logs, question answers and investigation plans within a preset time period are aggregated based on the grouping results to obtain aggregated question and answer slices; The aggregated question-and-answer slices are converted into vector codes of a preset dimension using a natural language model, and the vector codes are written into a preset vector database. A corresponding vector set is created based on the link fingerprint, and the vector set is associated with the log partition corresponding to the link fingerprint in a preset analytical database. Based on the user's question, an approximate nearest neighbor search is performed on the vector set of the preset vector database to find a matching answer and troubleshooting solution; and, Retrieve the attribution failure logs of the link fingerprint within a preset time period from the preset analytical database; The matched question answers and troubleshooting solutions, along with the attribution failure logs within a preset time period, are associated, and the corresponding association results are used as the initial answer to the question.

4. The advertising question-and-answer method according to any one of claims 1 to 3, characterized in that, The process of performing secondary verification on the initial answer according to the whitelist to obtain the final answer to the question includes: The initial answer is then subjected to a secondary verification based on the level in the whitelist to determine whether the question answer and investigation plan corresponding to the initial answer exceed the preset level. If the initial response exceeds the preset level, it will be intercepted and corrected, and the corrected result of the initial response will be used as the final response. If the initial answer does not exceed the preset level, it will be used as the final answer.

5. The advertising question-and-answer method according to claim 4, characterized in that, The method further includes: The ad parameters at the current time are compared with the ad parameters at the previous time in the original event log to determine whether the ad parameters at the two times are the same. If the advertising parameters at the current moment are different from those at the previous moment, a new link fingerprint is regenerated based on the advertising parameters at the current moment. The answer knowledge base is then re-segmented into multiple independent branch knowledge bases according to the new link fingerprint. A new log partition associated with the new link fingerprint is created in a preset analytical database. A new vector set corresponding to the new link fingerprint is created in a preset vector database, and the new vector set is associated with the new log partition. A new whitelist is formed based on the binding level of the new link fingerprint, and subsequent questions raised by the user are verified according to the new whitelist. If the ad parameters at the current moment are the same as those at the previous moment, then the user's subsequent questions will be answered based on the link fingerprint generated at the previous moment.

6. The advertising question-and-answer method according to claim 1 or 3, characterized in that, Before retrieving and matching each independent branch knowledge base with the questions raised by the user, the method further includes: The system receives questions submitted by users on the front end and transmits them to the gateway interface via the Hypertext Transfer Security Protocol (HTTP). The gateway interface then converts the questions into a preset data exchange format, which is used to search and match each independent branch knowledge base according to the preset data exchange format.

7. The advertising question-and-answer method according to claim 2, characterized in that, The process of encrypting the advertising parameter string and extracting a portion of the encryption result as the link fingerprint includes: using a hash encryption algorithm to perform hash encryption on the advertising parameter string and extracting the first sixteen digits of the hash encryption result as the link fingerprint.

8. An advertising question-and-answer system, characterized in that, The system includes: The link fingerprint module is used to obtain a link fingerprint formed in advance or in real time based on advertising parameters, including carrier, ad position and attribution mode. The branch retrieval module is used to divide the answer knowledge base into multiple independent branch knowledge bases according to the link fingerprint, and to retrieve and match each independent branch knowledge base with the question raised by the user to obtain the initial answer to the question; The verification module is used to bind the link fingerprint to a whitelist, perform secondary verification on the initial answer according to the whitelist, obtain the final answer to the question, and push the final answer to the user.

9. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the advertising question-and-answer method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the advertising question-and-answer method as described in any one of claims 1 to 7.

Images