Game issuing channel optimization method and system

Abstract

The invention relates to a game issuing channel optimization method and system. The game issuing channel optimization method comprises the following steps: acquiring original user behavior information and original event data from a channel party, adding a system receiving timestamp to the original event data, and temporarily storing the original event data in a storage database; key user behavior events in the original event data are extracted, and when preset final confirmation data are not received, the state of the original event data is judged to be to-be-confirmed; asynchronously scanning the storage database, identifying a key event which arrives in a delayed manner in the original event data, and extracting a user identifier and an original report timestamp in the key event; and tracing and associating the key user behavior event in a preset time period adjacent to the occurrence time of the key event according to the user identifier and the original reporting timestamp, and correcting the state and the attribution channel of the key event according to the key user behavior event.

Description

Technical Field

[0001] This application relates to the field of game publishing, and in particular to a method and system for optimizing game publishing channels. Background Technology

[0002] In the digital content distribution field, especially in game publishing, effectively launching game products to the market and ensuring their strong market performance has always been a key focus for the industry. Optimizing game publishing channels essentially involves combining user analysis with intelligent resource allocation based on channel data, and monitoring metrics such as conversion rates and user acquisition costs to improve marketing efficiency. However, under market competition, publishers need to explore emerging channels such as vertical social media and live streaming communities, and performance evaluation criteria have shifted towards long-term value metrics such as user initial payment and 7-day retention, posing challenges to existing technological systems.

[0003] Because emerging channels exhibit significant differences in user behavior and traffic composition compared to traditional channels, the feature weights and attribution logic of the original evaluation system are difficult to adapt, making evaluation bias highly likely. More critically, emerging channels lack standardized API data fields and reporting frequencies, leading to delays and formatting issues when game publishers compile statistics. Currently, to ensure smooth data transmission from channels, technical teams use fixed time windows to mark data not received on time as "not occurred." This results in a large number of genuine user behaviors being misjudged, causing game publishers' channel evaluation systems to use a large amount of erroneous data. This leads to negative assessments of emerging channels, resulting in reduced channel budget allocations and a vicious cycle of reverse optimization.

[0004] To address this, we provide a method and system for optimizing game distribution channels. Summary of the Invention

[0005] This application provides a method and system for optimizing game distribution channels, which at least addresses the problem that the limitations of related technologies in determining trunk line coordination and control schemes lead to poor control effects.

[0006] Firstly, this application provides a method for optimizing game distribution channels, the method comprising: Obtain raw event data from the channel provider, append a system receiving timestamp to the raw event data, and temporarily store it in the storage database; Extract key user behavior events from the raw event data, and if no preset final confirmation data is received, determine the status of the raw event data as pending confirmation; Asynchronously scan the storage database and identify delayed key events in the raw event data, and extract the user identifier and original reporting timestamp from the key events; Based on the user identifier and the original reporting timestamp, the key user behavior events in a preset time period adjacent to the occurrence time of the key event are traced and associated, and the status and attribution channel of the key event are corrected based on the key user behavior events.

[0007] Optionally, the step of tracing and associating the key user behavior events within a preset time period adjacent to the occurrence time of the key event based on the user identifier and the original reporting timestamp includes: The original user behavior information from different channels is standardized and processed to extract user identification clues and behavior time information, and a unique internal number is generated for each standardized user behavior information. The standardized user behavior information is transformed into a user behavior relationship graph, wherein the user behavior relationship graph includes: nodes representing user identity clues, channels, user behavior events and user behavior content, and association edges representing the relationships between the nodes; In the user behavior relationship graph, multiple user identity clues that are close in time and have common characteristics are marked as identification nodes and connected by the associated edges to form a user identity chain; When the key event is identified, the user behavior path from the initial touchpoint to the final conversion is identified by traversing backwards in the user behavior relationship graph, starting from the key event; By comprehensively judging the type influence, time proximity, interaction depth, and identity clue certainty of user touchpoints in each channel in the user behavior path, a contribution score is calculated for each channel in the user behavior path, and the user's channel affiliation weight is assigned according to the contribution score.

[0008] Optionally, in the user behavior relationship graph, marking multiple user identity clues that are close in time and have common characteristics as identification nodes and connecting them through the associated edges to form a user identity chain includes: Evaluate the reliability weights of the common features, wherein the reliability weights are used to indicate the importance of the common features in identity association; Based on the continuity pattern of the user behavior events, the channel to which the user identity clues belong is determined. The continuity pattern includes continuous clicks, browsing, or downloads on different channels within a short period of time. If a single user logs in through different channels, the user's identity information will be attributed to the channel of the login account. If there are multiple users with a history of logging into different accounts on the same device, then based on the time point of the user behavior event, the user identity clues are attributed to the channel party of the active account at that time point; If there is a scenario involving shared devices or accounts, the attribution channel of the user identity clues will be comprehensively determined based on the type and frequency of the user behavior events, wherein the type includes in-game payment behavior or ad click behavior. Based on the reliability weights of the common features, the continuity patterns of the user behavior events, and the attribution channels of the user identity clues, the association confidence score of the user identity clues is calculated comprehensively. When the association confidence score reaches a first preset threshold, it is marked as an identification node. Based on the identification node, association edges are created to form the user identity chain.

[0009] Optionally, the step of comprehensively calculating the association confidence score of the user identity clue based on the reliability weight of the common features, the continuity pattern of the user behavior event, and the attribution channel of the user identity clue includes: If the channel is a newly launched channel or a channel with a small amount of data, the weight of the channel to which the user identity clue belongs will be increased, and the identity will be associated in combination with short-term high-frequency behavior patterns to calculate the association confidence score. If the channel is a mature channel, the reliability weight of the common features is dynamically adjusted based on the historical data analysis results, and the association is performed by combining long-term behavioral patterns and user lifecycle stages to calculate the association confidence score. If it is during a game version update or a major marketing event, the weight of the continuity pattern of the user behavior event will be temporarily adjusted, and the association confidence score will be calculated. Monitor the fluctuations in the associated confidence score and the breakage rate of the user identity chain, and trigger an early warning when abnormal fluctuations occur or the breakage rate of the user identity chain increases.

[0010] Optionally, the monitoring of fluctuations in the association confidence score and the breakage rate of the user identity chain, triggering an early warning when abnormal fluctuations occur or the breakage rate of the user identity chain rises to a second preset threshold, includes: Collect and analyze abnormal feature information of the abnormal fluctuations or the increased breakage rate of the identity chain, wherein the abnormal feature information includes abnormal time point, abnormal duration and abnormal impact range; Extract the original event data transmission log from the abnormal feature information and analyze the data transmission quality of the original event data transmission log to obtain data transmission quality diagnostic results; In the user behavior relationship graph, the user behavior pattern during the abnormal duration is analyzed to diagnose whether the user behavior pattern is a real change in user behavior pattern and to obtain the user behavior pattern diagnosis result. Based on the channel provider's historical user data and user profile information, identify whether the user has characteristics of suspected malicious traffic fraud and obtain a diagnosis result of malicious traffic fraud. Based on the data transmission quality diagnosis results, the user behavior pattern diagnosis results, and the malicious traffic fraud diagnosis results, an anomaly diagnosis report is generated, and an alert level and processing strategy are selected based on the anomaly diagnosis report. The anomaly diagnosis report includes the anomaly cause type and anomaly severity to indicate the possible causes of the anomaly.

[0011] Optionally, selecting the warning level and handling strategy based on the abnormal diagnosis report includes: Based on the anomaly diagnosis report, the system matches the corresponding warning level from a preset warning level matrix and matches the corresponding processing strategy from a preset processing strategy library. If the abnormal diagnosis report points out multiple abnormal causes with different priorities, the multiple abnormal causes are sorted according to the preset priority rules, and the warning level and handling strategy corresponding to the abnormal cause with the highest priority after sorting are selected. If the anomaly diagnosis report points out multiple anomalies with the same priority, then the multiple anomalies are ranked according to their impact on the accuracy of channel value assessment, and the warning level and handling strategy corresponding to the anomaly with the greatest impact after ranking are selected.

[0012] Optionally, the step of ranking the various abnormal causes according to their impact on the accuracy of channel value assessment, and selecting the warning level and handling strategy corresponding to the abnormal cause with the highest impact after ranking, includes: Based on game type, channel characteristics, and current user game lifecycle stage, the influence weights of key business indicators are assigned. These key business indicators include user acquisition cost, paid conversion rate, user retention rate, and user game lifetime value. Based on the direct impact and duration of the aforementioned anomalies on the key business indicators, a comprehensive impact score is calculated on the accuracy of the channel partner's value assessment. If the game is in the new game launch phase, the weight of the impact of the abnormal cause on the user acquisition cost and the paid conversion rate is increased, and the comprehensive impact score is calculated based on the increased impact weight. If the game is in a mature operation stage, the impact weight of the abnormal cause on the user retention rate and the user's game lifetime value is increased, and the comprehensive impact score is calculated based on the increased impact weight. If the cause of the anomaly involves a specific channel, the influence weight of the key business indicators of the specific channel is adjusted based on the historical performance data of the specific channel, and the comprehensive influence score is calculated based on the increased influence weight. When the comprehensive influence score reaches a second preset threshold, the corresponding warning level and handling strategy are selected.

[0013] Optionally, when the comprehensive impact score reaches a second preset threshold, selecting the corresponding early warning level and processing strategy includes: Based on the game type, channel characteristics, and current user game lifecycle stage, the second preset threshold is adjusted, and the corresponding warning level and processing strategy are selected based on the adjusted second preset threshold. If the game is in the new game launch phase, the second preset threshold is lowered to the third preset threshold, and the comprehensive impact score is determined based on the third preset threshold, and the corresponding warning level and handling strategy are selected. If the game is in a mature operation stage, the second preset threshold is adjusted to the fourth preset threshold based on the user retention rate and the historical fluctuation range of the current user's game life cycle stage, and the comprehensive impact score is determined based on the fourth preset threshold to select the corresponding warning level and handling strategy. If the cause of the anomaly involves a specific channel, the second preset threshold is adjusted to the fifth preset threshold based on the historical data performance of the specific channel, and the comprehensive impact score is determined based on the fifth preset threshold to select the corresponding warning level and handling strategy.

[0014] Secondly, this application provides a game distribution channel optimization system, the system comprising: The data acquisition module is used to acquire raw user behavior information and raw event data from the channel, attach a system receiving timestamp to the raw event data and temporarily store it in the storage database; The status determination module is used to extract key user behavior events from the raw event data. If the preset final confirmation data is not received, the status of the raw event data is determined to be pending confirmation. The asynchronous scanning module is used to asynchronously scan the storage database and identify delayed key events in the raw event data, and extract the user identifier and original reporting timestamp from the key events; The correction and optimization module is used to trace and associate the key user behavior events in a preset time period adjacent to the occurrence time of the key event based on the user identifier and the original reporting timestamp, and to correct the status and attribution channel of the key event based on the key user behavior events.

[0015] Thirdly, a computer-readable storage medium is provided having a program stored thereon, which, when executed by a processor, implements the game distribution channel optimization method described in the first aspect.

[0016] Compared to related technologies, the game distribution channel optimization method and system provided in this application have at least the following technical effects: By receiving raw event data from channels and attaching a system receiving timestamp for temporary storage, the problem of data loss or mislabeling caused by data transmission delays and inconsistent formats in emerging channels is effectively solved. Furthermore, key user behavior events are judged as pending confirmation when final confirmation data is not received, and the temporary data is scanned asynchronously to identify delayed key events. Based on this, according to the user identifier and the original reporting timestamp, the user's behavior data before and after the event occurred is traced, thereby correcting the status and channel attribution of key events.

[0017] Through the aforementioned technical solution, this application can fundamentally solve the problem of channel quality assessment bias caused by data distortion in existing technologies. Specifically, by effectively processing delayed data and correcting the status of key events, it avoids the misclassification of a large number of real-world payment and retention behaviors as "not-occurring," thereby correcting the system's severe underestimation of the true value of users. It also allows the channel quality assessment system to update parameters based on more accurate and complete data, avoiding a vicious cycle of reinforcing negative judgments after "learning" from distorted data. This ensures that the budget allocation decision-making mechanism is based on real user value and channel performance, preventing the erroneous reduction of budgets for high-potential channels or even the cessation of advertising, thus effectively improving game release efficiency and maximizing marketing investment returns.

[0018] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a method for optimizing game distribution channels according to an exemplary embodiment.

[0020] Figure 2 This is a flowchart illustrating step S40 according to an exemplary embodiment.

[0021] Figure 3 This is a flowchart illustrating step S403 according to an exemplary embodiment.

[0022] Figure 4 This is a block diagram illustrating a game distribution channel optimization system according to an exemplary embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0024] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0025] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0026] Example 1

[0027] This invention provides a method for optimizing game distribution channels. Figure 1 This is a flowchart illustrating a game distribution channel optimization method according to an exemplary embodiment. Figure 1 As shown, the method includes: Step S10: Obtain the raw event data from the channel provider, append a system receiving timestamp to the raw event data, and temporarily store it in the storage database; in this example, its implementation mainly includes: 1. Continuously monitor data interfaces from various channels. Once raw event data is received, the system automatically generates a system reception timestamp accurate to the millisecond level and binds it to the raw event data. Subsequently, this raw event data with system reception timestamps is written to a high-speed, scalable temporary storage database, such as using Apache Kafka as a message queue for initial temporary storage, or directly written to the temporary area of ​​a distributed file system (such as HDFS).

[0028] 2. The data receiving module adopts a batch processing method, which pulls data from the channel API at preset time intervals (e.g., every 5 minutes). After the data is pulled, the system adds a unified system receiving timestamp (i.e. the time when batch processing started) to each piece of data in the batch, and then temporarily stores the data of the entire batch in the memory cache or temporary file.

[0029] Step S20: Extract key user behavior events from the raw event data. If no preset final confirmation data is received, the raw event data status is determined to be pending confirmation. In this example, its implementation mainly includes: 1. The system maintains a list of critical events, defining which events are considered critical user behavior events. When the event processing module identifies these critical events from the temporary data, it checks whether corresponding final confirmation data exists. If no final confirmation data is received within a preset time window (e.g., 30 minutes), the critical event is marked as "pending confirmation" and stored in a dedicated pending confirmation event table.

[0030] 2. The system sets a timer for each critical event. When a critical event is received, the timer starts. If no final confirmation data is received before the timer expires, the status of the critical event is automatically updated to "Pending Confirmation".

[0031] Step S30: Asynchronously scan the storage database and identify delayed key events in the raw event data, extracting user identifiers and original reporting timestamps from the key events. In this example, this is implemented using streaming processing technology, such as Apache Flink or Spark Streaming. When new raw event data flows into the temporary storage area, the streaming system compares its original reporting timestamp with the system's receiving timestamp in real time. If a significant delay is found, the tracing process is immediately triggered. During tracing, the system queries the user's historical behavior database to obtain the user's behavior sequence before and after the event occurrence time, and recalculates the attribution channel using a preset attribution model.

[0032] Step S40: Based on the user identifier and the original reporting timestamp, trace and associate key user behavior events in a preset time period adjacent to the occurrence time of the key event, and correct the status and attribution channel of the key event based on the key user behavior events.

[0033] In the technical solutions of the above embodiments, the key terms and implementation environments involved are described as follows: "Raw event data" refers to the unprocessed, raw user behavior data received from various game distribution channels, such as user clicks, browsing, downloads, registrations, and payment records. This data usually includes information such as user identifiers, event types, and event timestamps.

[0034] "Receive timestamp" refers to the time stamp automatically attached by the game distribution channel optimization system when it receives raw event data, used to record the time when the data arrives at the system.

[0035] "Key user behavior events" refer to user behaviors that are of great significance to the optimization of game distribution channels, such as first-time payment, registration, download completion, and completion of specific tasks. These events are usually directly related to the conversion effect and user value of the channel.

[0036] "Final confirmation data" refers to data sent by a channel or third-party system to confirm that a key user behavior event has actually occurred and been completed. For example, for a payment transaction, final confirmation data might be a callback notification from the payment platform.

[0037] "User identifier" refers to the identity information used to uniquely identify a user, which can be a device ID, user account ID, anonymous ID, etc.

[0038] "Original reporting timestamp" refers to the actual time when the event occurred, as recorded in the original event data.

[0039] The implementation environment of this application embodiment is typically a distributed data processing system, including a data receiving / acquisition module, a data storage module, a status determination module, an asynchronous scanning module, a correction and optimization module, etc. These modules work together to implement the above-mentioned optimization method.

[0040] The channel optimization method in this application first receives raw event data from various channels and adds a system reception timestamp to each data entry, temporarily storing it to ensure that all data is recorded and to provide objective time evidence of data arrival in the system. Then, it asynchronously scans the temporarily stored raw event data, makes a preliminary judgment on key user behavior events in the received raw event data, and determines the status as pending confirmation until final confirmation data is received. This avoids the error of immediately marking events as "not occurred" due to short-term data delays, providing a buffer for subsequent corrections. Finally, when such delayed key events are identified, the method actively traces the user's behavior data before and after the event occurrence time based on the event's user identifier and original reporting timestamp, reconstructing the user's true behavior path and context. This corrects the previously incorrectly judged status of key events and reassesses their correct channel affiliation, effectively addressing the data transmission delay and inconsistency issues of emerging channels, thereby ensuring the accuracy of channel affiliation and assessment.

[0041] In one example, suppose a user clicks on an advertisement on an emerging channel and completes a game download, but due to channel API delays, the download completion event is only reported several hours later. Traditional systems might mark this as "not occurred" because they didn't receive the event within a preset window. However, the method in this embodiment receives this delayed event and, based on its original reporting timestamp and user identifier, traces the user's behavior before and after the download. For example, it might discover that the user clicked on an advertisement before the download and launched the game after the download. Through this traced behavioral data, the actual occurrence of the download event can be confirmed, and it can be attributed to the correct advertising channel, thereby correcting the status and channel attribution of the critical event.

[0042] The aforementioned mechanism of this application enables the game distribution channel optimization method to obtain more authentic and comprehensive user behavior data by identifying, tracing, and correcting delayed events. This provides a more accurate basis for channel quality assessment and budget allocation, ultimately achieving channel optimization goals. It overcomes the problems of poor data quality in emerging channels, as well as the loss and mislabeling of "silent" data caused by data delays, and avoids underestimation of channel value and budget allocation errors due to data distortion.

[0043] In one exemplary embodiment, due to the complexity of diverse sources of user behavior data, dispersed identity information, and data delays, simply tracing may not be sufficient to accurately and comprehensively construct the complete user behavior path and assign it to a specific channel. Figure 2 This is a flowchart illustrating step S40 according to an exemplary embodiment. (Refer to the attached document.) Figure 2 In this embodiment, step S40 includes: S401. Standardize the raw user behavior information from different channels, extract user identification clues and behavior time information, and generate a unique internal number for each standardized user behavior information. In this embodiment, the standardization process includes unifying, cleaning, and structuring the raw and heterogeneous user behavior data obtained from various channels (e.g., social media, app stores, advertising platforms, offline events, etc.). Specifically, this process includes data parsing, field mapping, data type conversion, and missing value imputation, etc., with the aim of eliminating data heterogeneity and ensuring that all behavior information is consistent and comparable in subsequent processing. During this process, all available identification clues (e.g., device ID, user ID, IP address, cookie ID) and time information (e.g., event occurrence time, reporting time, etc.) are accurately extracted, and a unique internal number is assigned to each standardized behavior information for subsequent tracking and association. S402. The standardized user behavior information is transformed into a user behavior relationship graph. This graph includes nodes representing user identity clues, channels, user behavior events, and user behavior content, as well as edges representing the relationships between these nodes. In this embodiment, user identity clues, channels, user behavior events, and user behavior content are abstracted as nodes in the user behavior relationship graph, and the interactions and temporal relationships between these nodes are represented by edges. For example, a user identity clue node can be connected to multiple user behavior event nodes through an "occurrence" relationship, a user behavior event node can be connected to a channel node through a "from" relationship, and a user behavior content node can be connected to a user behavior event node through a "involvement" relationship. The purpose is to provide an intuitive and efficient data model for representing and analyzing complex user behavior patterns across different channels and time points.

[0044] S403. In the user behavior relationship graph, multiple user identity clues that are close in time and have common characteristics are marked as identification nodes and connected by association edges to form a user identity chain. For example, if a device ID is associated with a specific login account in a short period of time, and the device ID and the login account show a high degree of consistency in terms of geographical location, behavioral preferences, etc., then association edges can be established between these user identity clue nodes. The purpose is to solve the problem of fragmented user identity on different devices, different channels, or different points in time, thereby constructing a complete, cross-channel identity view of the user.

[0045] S404. When a key event is identified, the system uses the key event as a starting point and traverses backwards in the user behavior relationship graph to identify user behavior paths that meet the conditions from the user's initial touchpoint to the final conversion. For example, if a "paid purchase" event is identified as a key event, the system will start from that event node and trace back along the reverse time direction through the edges in the graph to the touchpoint where the user first came into contact with the game or channel. The system will identify all preceding events such as ad clicks, application downloads, registrations, and browsing, as well as their corresponding channels. The purpose is to comprehensively capture all possible paths from the user's initial contact to the final conversion, providing a complete data foundation for subsequent attribution analysis.

[0046] S405. A comprehensive assessment is made of the type influence, temporal proximity, interaction depth, and identity clue certainty of user touchpoints in each channel within the user behavior path. A contribution score is calculated for each channel in the user behavior path, and the user's channel affiliation weight is assigned based on the contribution score. In this embodiment, type influence includes the nature of the channel (e.g., brand advertising, performance advertising, organic search, etc.); temporal proximity refers to the distance between the touchpoint's occurrence time and the key event's occurrence time; interaction depth refers to the degree of interaction between the user and the channel's content (e.g., clicks, browsing duration, video completion rate, etc.); identity clue certainty refers to the reliability of clues used to associate the user's identity. The overall combination aims to achieve a fairer and more accurate channel affiliation weight through multi-dimensional evaluation, avoiding the bias that may arise from a single affiliation model.

[0047] In the technical solution of the above embodiments, firstly, standardized processing unifies heterogeneous data sources, facilitating subsequent analysis; secondly, the standardized information is transformed into a user behavior relationship graph, providing a structured way to represent and manage complex user interactions at different channels and time points, making the exploration of user behavior paths more visual; thirdly, association edges are created between user identity clues that are close in time and have common characteristics, connecting scattered user identities to form a complete user identity chain, thereby overcoming the problem of fragmented user identities and ensuring the integrity and consistency of user behavior data; thus, when a key event is identified, a reverse traversal is performed starting from the key event node, which can systematically explore all possible preceding events and channel touchpoints, identify a complete set of behavior paths, and avoid missing key touchpoints; finally, by comprehensively considering multiple dimensions such as the type influence, time proximity, interaction depth, and identity clue certainty of user touchpoints in the channel through contribution scoring, a contribution score is calculated for each channel and the channel weight is assigned, achieving a more refined and fair allocation of channel weights, significantly improving the accuracy and reliability of user channel attribution weights.

[0048] In one example, suppose a game publisher is promoting a new game, and user X comes into contact with and ultimately purchases in-game items through the following path: The user sees and clicks on a game ad on social media platform A (touchpoint 1, channel A, device ID 1, time T1); is redirected to the game's official website via the ad (touchpoint 2, channel website, device ID 1, time T1 + 5 minutes); a few days later, searches for and downloads the game in app store B (touchpoint 3, channel B, device ID 1, time T2); upon first login, logs in using a WeChat account (touchpoint 4, in-game channel, WeChat ID 1, time T2 + 10 minutes); a week later, logs in using the same WeChat account on another device C (touchpoint 5, in-game channel, device ID 2, WeChat ID 1, time T3); finally, makes an in-game purchase on device C (key event, time T3 + 30 minutes).

[0049] When performing the method processing of step S40 in the embodiments of this application: First, the raw event data from social media platform A, the game's official website, app store B, and within the game will be standardized to extract identification clues such as device ID1, device ID2, and WeChat ID1, as well as time information such as T1, T2, and T3, and generate a unique internal number for each event.

[0050] Next, this standardized behavioral information will be transformed into a user behavior relationship graph. For example, Device ID1, Device ID2, and WeChat ID1 will serve as user identity clue nodes, Channel A, Channel Official Website, and Channel B will serve as channel nodes, and clicks, browsing, downloads, logins, and paid purchases will serve as behavioral event nodes.

[0051] Then, since Device ID1 and WeChat ID1 are associated in the in-game login behavior at time T2+10 minutes, and Device ID2 and WeChat ID1 are associated in the in-game login behavior at time T3, association edges will be created between Device ID1 and WeChat ID1, and between Device ID2 and WeChat ID1, thereby connecting these scattered identity clues to form the user identity chain of X.

[0052] When the key event of "paid purchase" is identified, the user behavior relationship graph will be traversed backwards from this key event node. By tracing back along the edges of the graph, all the preceding events and channel touchpoints that may have led to this paid purchase will be explored and identified. For example, tracing back from the paid purchase to login on device C, in-game behavior on device C, download on device B, and ad click on device A, the complete set of behavior paths will be identified.

[0053] Finally, considering the type of influence of each channel touchpoint (e.g., the exposure influence of social media ads, the direct conversion influence of app stores), time proximity (e.g., the most recent login behavior before a paid purchase), interaction depth (e.g., in-game login and playtime), and the certainty of identity clues (e.g., WeChat ID is more certain than device ID), a contribution score is calculated for Channel A, Channel Website, Channel B, and in-game channels. For example, social media ads may receive a certain exposure contribution score, app stores may receive a higher conversion contribution score, while in-game channels may receive a higher interaction and retention contribution score. Ultimately, based on these contribution scores, corresponding channel weights are assigned to each channel, thereby achieving precise optimization of game distribution channels.

[0054] In an exemplary embodiment of this application, simply associating based on time proximity and common features may be inaccurate. For example, in scenarios with complex user behavior patterns, device sharing, or multiple account usage, the accuracy and reliability weight of identity association may be affected, resulting in an inaccurate or broken construction of the user identity chain. Figure 3 This is a flowchart illustrating step S403 according to an exemplary embodiment. (Refer to the attached document.) Figure 3 In this embodiment, step S403 includes: S4031. Evaluate the reliability weight of common features, where the reliability weight indicates the importance of common features in identity association. In this embodiment, the reliability weight is used to assign different weight values ​​to different common features based on their importance and stability in user identification. For example, features such as device ID, IP address, and browser fingerprint may have different reliability weights. Device ID and login account are generally considered to have higher reliability weights, while IP address or browser fingerprint may have relatively lower reliability weights because they are more likely to change or be shared. Furthermore, the reliability weight can be dynamically adjusted based on historical data analysis, expert experience, or machine learning models, with the aim of ensuring that more reliable features are used when performing identity association.

[0055] S4032. Based on the continuity pattern of user behavior events, determine the channel to which user identity clues belong, thereby analyzing the user's behavioral sequence across different channels within a short period of time. For example, if a user clicks on channel A and then browses on channel B, and these two actions are very close in time, it can be assumed that these two actions likely belong to the same user. The continuity pattern of user behavior can include consecutive clicks, browsing, or downloads on different channels within a short period of time. The purpose is to use the coherence of behavioral patterns to help determine the attribution relationship of scattered identity clues.

[0056] Continuous patterns include consecutive clicks, views, or downloads across different channels within a short period of time. If a single user logs in through different channels, the user's identity clue will be attributed to the channel of the login account. In this implementation, when a user logs in through different channels, the login account will be given priority as the identity clue, because the login account is usually the most distinctive identifier of the user and can uniquely identify the user.

[0057] If multiple users have a history of logging into different accounts on the same device, the user's identity clues will be attributed to the channel party of the active account at that time, based on the time when the user's behavior event occurred. For example, if device 100 is logged in and generates behavior by account A in the morning, and logged in and generates behavior by account B in the afternoon, the behavior clues for the morning will be attributed to account A, and the behavior clues for the afternoon will be attributed to account B. The purpose is to solve the identity confusion problem in the device sharing scenario and ensure that the behavior data can be accurately attributed to the actual user and channel.

[0058] In scenarios involving shared devices or accounts, the attribution of user identity clues is determined based on the type and frequency of user behavior events. Types include in-game purchases and ad clicks; for example, key behavioral events such as in-game purchases or ad clicks typically have high user intent and uniqueness. If, on a shared device, one account frequently makes purchases while another account only browses, the overlap in identity clues can be more accurately differentiated and processed based on the type and frequency of these behaviors, thus avoiding the erroneous association of different users' behaviors.

[0059] S4033. Based on the reliability weights of common features, the continuity pattern of user behavior events, and the attribution channel of user identity clues, a comprehensive association confidence score for user identity clues is calculated. When the association confidence score reaches a first preset threshold, it is marked as an identification node. Association edges are created based on these identification nodes to form a user identity chain. In this embodiment, the association confidence score comprehensively considers the reliability weights of common features, the matching degree of behavior continuity patterns, and the certainty of user login status. For example, the confidence score of an identity association can be calculated using a weighted summation method, where each weight reflects its importance in identity recognition. When the association confidence score reaches the preset threshold, association edges are created to connect scattered identity clues, forming a complete user identity chain.

[0060] Furthermore, the first preset threshold can be adjusted according to actual business needs and data characteristics to balance the accuracy of identity association.

[0061] In the technical solutions of the above embodiments, by evaluating the reliability weight of common features, more stable and identifiable identity clues can be prioritized, avoiding noise introduced by low-reliability features. Simultaneously, the introduction of continuous patterns in user behavior allows for the capture of instantaneous behavioral associations across channels, enabling the inference of user identity through the coherence of behavioral trajectories even in the absence of explicit identifiers. Using login accounts as identity clues leverages the most definitive user identity identifier, significantly improving the certainty of association. For complex scenarios involving shared devices or accounts, by distinguishing the active time points of different accounts on the same device and analyzing the type and frequency of behavioral events, this application can effectively separate and attribute the behaviors of different users, avoiding identity confusion. Finally, by comprehensively calculating the association confidence score and setting a first preset threshold, the reliability of identity associations is quantitatively evaluated, ensuring that only associations with sufficient confidence are adopted, thereby constructing a more accurate, stable, and complete user identity chain, significantly improving the accuracy and robustness of user identity chain construction in game distribution channels. Therefore, even in scenarios with complex user behavior, shared devices, or multiple accounts, a more complete and accurate user identity chain can be formed, providing a more reliable data foundation for subsequent user behavior path tracing, key event status correction, and channel attribution, ultimately improving the accuracy and optimization effect of channel attribution.

[0062] In one example, suppose user Y clicks on an ad on their phone to access the download page for game A, but does not download it immediately; subsequently, on the same phone, the user sees an advertisement for game A on another social media channel and clicks to access the official website of game A; a few minutes later, the user downloads game A from an app store on the same phone and logs in to the game using their social media account.

[0063] In this scenario, raw event data from different channels is first received, and a timestamp is added to each data point. When constructing the user behavior relationship graph, multiple identity clues are identified: such as the device ID for clicking on ads, the device ID and IP address for browsing social media, the device ID for downloading from the app store, and the social media account that is ultimately logged in.

[0064] According to the scheme of this application, the reliability weight of common features is first evaluated: the reliability weight of the login account is the highest, followed by the device ID, and the IP address is the lowest.

[0065] Secondly, based on the continuous pattern of user behavior: ad clicks, social media browsing, and app store downloads occur consecutively within a short period of time, and the device IDs are consistent, indicating a highly continuous behavioral pattern.

[0066] Secondly, the login account was used as an identity clue: since the user ultimately logged into the game using a social media account, this login account was prioritized as the most important identity clue.

[0067] Finally, the overall association confidence score is calculated, taking into account the certainty of the logged-in account (high weight), the consistency of the device ID (medium weight), and the matching degree of short-term continuous behavioral patterns (high matching degree). For example, if the certainty score of the logged-in account is 0.9, the consistency score of the device ID is 0.7, and the matching degree of behavioral continuity is 0.8, with their respective weights of 0.5, 0.3, and 0.2, then the overall association confidence score might be 0.9*0.5 + 0.7*0.3 + 0.8*0.2 = 0.45 + 0.21 + 0.16 = 0.82.

[0068] When the association confidence score (e.g., 0.82) reaches a preset threshold (e.g., 0.75), an association edge will be created, successfully linking these scattered device IDs, IP addresses, and social media account clues to form a complete user identity chain.

[0069] Through the above process, the embodiments of this application can accurately identify that this is the behavior of the same user on different channels and at different times, and attribute it to the correct user identity, thereby providing accurate data for subsequent channel attribution weights and user behavior analysis.

[0070] In an exemplary embodiment, if the calculation method of the association confidence score fails to adequately adapt to various dynamic factors such as game type, channel characteristics, user behavior stage, and marketing activities, it may lead to a decrease in the accuracy of identity association, thereby affecting the integrity of the user behavior path and the accuracy of the channel attribution. In this embodiment, step S4033 includes: If the channel is newly launched or has limited data, the weight of the channel to which user identity leads belong will be increased, and identity will be associated with short-term, high-frequency behavioral patterns to calculate the association confidence score. In the early stages of operation, when data accumulation is insufficient or a channel is just beginning, due to a lack of sufficient historical user behavior data to establish complex behavioral patterns, strong identity identifiers such as user login accounts will be prioritized. Furthermore, high-frequency behaviors such as consecutive clicks, browsing, or downloads by users across different channels within a short period usually strongly point to the same user; therefore, the matching degree of these behavioral patterns will be given higher weight to compensate for the uncertainty caused by data sparsity.

[0071] If the channel is a mature channel, the reliability weights of common features are dynamically adjusted based on historical data analysis results. This is combined with long-term behavioral patterns and user lifecycle stages to calculate an association confidence score. For mature channels that have accumulated a large amount of user behavior data, in-depth analysis of historical data can more accurately assess the reliability of common features such as device ID and IP address in identity association. For example, if historical data shows that a device ID is frequently associated with multiple different accounts within a specific time period, the reliability weight of that device ID may be reduced. Furthermore, long-term user behavioral patterns (e.g., fixed gaming time, payment habits) and their lifecycle stage (e.g., paying user, highly active user) are considered for more refined identity association.

[0072] During game version updates or major marketing events, the weighting of the continuity pattern of user behavior events is temporarily adjusted, and the association confidence score is calculated. When a game is updated or a large-scale marketing campaign is launched, user behavior patterns may temporarily deviate from the norm. For example, users may frequently switch channels, download multiple versions, or engage in extensive interactions within a short period to experience new features or participate in activities. In this case, the weighting of the degree of matching the continuity pattern of behavior will be temporarily increased to avoid misclassifying these event-driven behaviors, which still belong to the same user, as behaviors of different users, thereby ensuring the integrity of the identity chain.

[0073] The system monitors fluctuations in the association confidence score and the breakage rate of user identity chains. An alert is triggered when abnormal fluctuations or an increased breakage rate occur. In this embodiment, the overall distribution of the calculated association confidence score (e.g., mean, variance) and the stability of the established user identity chains (e.g., average chain length, breakage frequency) are continuously monitored. Once significant abnormal fluctuations in these indicators are detected (e.g., a sudden and sharp drop or rise in scores) or an abnormally high breakage rate in the identity chain is detected, the system will immediately trigger an alert mechanism, prompting operations or technical personnel to intervene and investigate potential data quality issues, malicious behavior, or model failure.

[0074] In summary, the channel optimization method of this application adaptively adjusts the evaluation function and parameters of the association confidence score based on various dynamic factors such as game type, channel characteristics, user behavior stage, and marketing activities. This overcomes the limitations of traditional static calculation methods in complex and ever-changing scenarios, enabling more accurate capture of user identity association characteristics in different contexts. For example, it prioritizes the certainty of login status in new channels, utilizes rich historical data for in-depth analysis in mature channels, and flexibly adapts to changes in user behavior patterns during special events. Furthermore, by monitoring the association confidence score and the stability of the identity chain in real time, this application can promptly detect and respond to potential anomalies, thereby ensuring the accuracy and integrity of the user identity chain.

[0075] In one example, suppose a newly launched casual game is being promoted through multiple emerging advertising channels. Since these channels are newly launched, the amount of data is relatively small. In this case, based on the characteristic of "newly launched channels or channels with limited data," the weight of the certainty of user login status in the association confidence score is dynamically increased. For example, if a user enters the game through an ad click within a short period and immediately logs in using their social media account, even if their subsequent behavioral data is not yet rich, this login behavior will be highly trusted and associated with previous ad clicks, forming a preliminary user identity chain.

[0076] As the game operates successfully, some channels gradually develop into "mature channels," accumulating a large amount of user behavior data. At this point, based on the characteristics of these "mature channels," historical data analysis results are used to dynamically adjust the reliability weights of common features. For example, analysis reveals that in mature channels, device IDs have a very high correlation with user identity, while IP addresses have relatively low reliability (possibly due to shared networks). Therefore, the weight of device IDs among common features is increased, and combined with users' long-term gaming habits, payment records, and other "long-term behavioral patterns," as well as their "user lifecycle stage" (e.g., whether they are high-value paying users), identity association is constructed, thereby building a more stable and comprehensive user identity chain.

[0077] Furthermore, when a game undergoes a "version update" or launches a "major marketing event" (such as an anniversary celebration), users may frequently check event details, download update packages, or participate in community discussions across different channels within a short period. These behaviors may exhibit a high "behavioral continuity pattern." To avoid misclassifying these event-driven behaviors, which still belong to the same user, as behaviors of different users, it is necessary to temporarily adjust the weight of the matching degree of the behavioral continuity pattern in the association confidence score to ensure the accurate maintenance of the integrity of the user identity chain even during special periods.

[0078] Finally, the system continuously monitors the distribution of association confidence scores and the stability of user identity chains. For example, if the system detects a sudden, abnormally low association confidence score for a channel, or an increased "breakage rate" in the user identity chain (i.e., many previously stable user identity chains suddenly break down), this may indicate data transmission problems, fraudulent activity, or a malfunctioning system model within that channel. In such cases, the system will immediately trigger an alert, notifying relevant personnel to investigate and address the issue, thereby mitigating losses and maintaining the accuracy of channel attribution.

[0079] An exemplary embodiment monitors fluctuations in the association confidence score and the breakage rate of the user identity chain. When abnormal fluctuations occur or the breakage rate of the user identity chain rises to a second preset threshold, an early warning is triggered, including: Collect and analyze abnormal characteristic information of abnormal fluctuations or increased identity chain breakage rate, including abnormal time point, abnormal duration and abnormal impact range.

[0080] Extract the original event data transmission logs from the abnormal feature information and analyze the data transmission quality of the original event data transmission logs to obtain data transmission quality diagnosis results. The original event data transmission logs record the entire process of event data being sent from the channel end to the system receiving end. By checking the logs, it can be determined whether there are data transmission quality problems such as packet loss, transmission delay, and data format errors. For example, if a large amount of event data is found to be dropped or delayed during transmission, it may indicate that the anomaly is caused by data transmission problems.

[0081] In user behavior relationship diagrams, user behavior patterns are analyzed during abnormal periods to diagnose whether these patterns represent genuine changes in user behavior and obtain diagnostic results. This includes observing user behavior sequences such as clicks, browsing, downloads, registrations, and payments to determine whether these patterns conform to normal user behavior patterns or whether there are abnormal or non-genuine changes in user behavior. For example, if a large number of users complete registration within a short period without any subsequent interaction, this may indicate an abnormal behavior pattern, thus leading to a diagnostic result for the user behavior pattern.

[0082] Based on historical user data and user profile information from the channel provider, we can identify whether users exhibit characteristics of suspected malicious traffic manipulation, thus obtaining a diagnostic result for malicious traffic manipulation. Malicious traffic manipulation typically manifests as abnormal user acquisition costs, artificially high paid conversion rates, extremely low user retention rates, or a large number of user activities originating from specific IP addresses or devices within a short period. By comparing historical data and user profiles, we can determine whether the current anomaly matches these characteristics of malicious behavior, thereby obtaining a diagnostic result for malicious traffic manipulation.

[0083] Based on the diagnostic results of data transmission quality, user behavior patterns, and malicious traffic manipulation, an anomaly diagnostic report is generated. The alert level and handling strategy are then selected based on this report. The anomaly diagnostic report includes the type and severity of the anomaly to indicate its possible causes. The report displays possible causes of the anomaly, such as data transmission failure, changes in genuine user behavior patterns, or malicious traffic manipulation. Based on the anomaly diagnostic report, the alert level (e.g., escalating from a general alert to an emergency alert) and subsequent handling strategies (e.g., contacting the channel to investigate the problem, initiating countermeasures, or adjusting the attribution model parameters) are adjusted accordingly.

[0084] In the technical solution of the above embodiments, by deeply analyzing abnormal feature information, data transmission quality, user behavior patterns and potential malicious traffic-boosting behavior, the underlying causes of anomalies can be identified and an anomaly diagnosis report can be generated. Ultimately, accurate diagnosis of channel anomalies and corresponding preset processing can be achieved, thereby adjusting the warning level and subsequent processing strategies in a targeted manner. This improves the efficiency and accuracy of anomaly handling, ensures the stability and reliability of the game distribution channel optimization method, and avoids resource waste and losses caused by misjudgment or untimely processing.

[0085] Based on these diagnostic results, the system will generate an anomaly diagnostic report, such as indicating that "data transmission delays caused the loss of some events, and there are a small number of suspected fraudulent activities." According to this report, the system can adjust the alert level (e.g., from a normal alert to an emergency alert) and match the corresponding handling strategy (e.g., immediately contact the channel to investigate the data transmission problem and activate the countermeasure mechanism to filter suspected fraudulent users).

[0086] An exemplary embodiment states that since anomaly diagnostic reports may indicate multiple causes of anomalies, and these causes may vary in type, severity, and priority of their impact on business, simple adjustments may not achieve an effective response, and could even lead to wasted resources or unresolved issues in a timely manner. The selection of alert levels and handling strategies based on anomaly diagnostic reports specifically includes: Based on the anomaly diagnosis report, the system matches the corresponding alert level from a pre-defined alert level matrix and the corresponding handling strategy from a pre-defined handling strategy library. Anomaly causes can include data transmission failures, abnormal user behavior patterns, and malicious traffic manipulation, while severity can be quantified based on the anomaly's impact on key business metrics (such as user acquisition cost and paid conversion rate). The pre-defined alert level matrix can be a multi-dimensional table, with dimensions including anomaly cause type and severity. Each cell corresponds to a specific alert level (e.g., information level, warning level, severe level, urgent level). The pre-defined handling strategy library contains standardized operating procedures for different alert levels and anomaly causes, such as data link checks, user behavior analysis model adjustments, and countermeasure rule updates. If the anomaly diagnosis report identifies multiple causes of anomalies with different priorities, the causes are ranked according to a preset priority rule, and the warning level and handling strategy corresponding to the anomaly with the highest priority after ranking are selected. In this embodiment, the preset priority rule can be defined based on factors such as the potential impact range, urgency, and difficulty of resolving the anomaly. For example, data transmission interruption may be given a higher priority than slight fluctuations in user behavior patterns, because the former may lead to complete data loss, which has a more direct and severe impact on business. It is necessary to solve the most critical and impactful problem first, ensuring that limited resources are allocated to the anomalies that require the most attention.

[0087] If the anomaly diagnosis report points out multiple anomalies with the same priority, the anomalies will be ranked according to their impact on the accuracy of channel value assessment. The warning level and handling strategy corresponding to the anomaly with the greatest impact after ranking will be selected. For example, if there are two anomalies with the same priority, one is a slight data delay and the other is a broken user identity chain that may lead to incorrect attribution, the latter will have a greater impact on the accuracy of channel value assessment and will therefore be handled first.

[0088] In the technical solutions of the above embodiments, by conducting multi-dimensional evaluation and prioritization of the causes of anomalies, the system can adopt the most appropriate early warning level and handling strategy, thereby improving the efficiency and accuracy of anomaly response. Especially when facing complex and ever-changing anomaly scenarios, such as when multiple anomalies occur simultaneously or when the impact of anomalies is difficult to directly determine, intelligent decision-making can be made based on the potential impact of anomalies on the accuracy of channel value assessment, ensuring that the assessment of key business indicators is not interfered with, and effectively maintaining the authenticity and impartiality of channel data attribution.

[0089] An exemplary embodiment suggests that relying solely on static or empirical evaluation methods may fail to adequately consider complex factors such as game type, channel characteristics, and user lifecycle stages, leading to inaccurate assessments of anomaly impacts and consequently affecting the timeliness and accuracy of channel optimization decisions. This model rankes various anomaly causes based on their impact on the accuracy of channel value assessment, and selects the alert level and handling strategy corresponding to the anomaly cause with the highest ranking impact, including: Based on game type, channel characteristics, and the current user lifecycle stage, key business metrics (KCMs) are weighted. These KCMs include user acquisition cost, conversion rate, user retention rate, and user lifetime value. Weighting these KCMs involves differentiating their importance based on different game types (e.g., casual games, hardcore RPGs), channel characteristics (e.g., user acquisition channels, brand channels, community channels), and user lifecycle stage (e.g., new users, active users, churned users). For example, in the initial launch phase of a new game, user acquisition cost and first-day conversion rate may be given higher weights, as quickly acquiring users and validating the business model are the primary goals. Conversely, for games in the mature operational phase, user retention rate and lifetime value may be more critical to ensure long-term profitability.

[0090] Based on the direct impact and duration of the anomaly on key business indicators, a comprehensive impact score on the accuracy of channel value assessment is calculated. This score objectively reflects the overall potential impact of the anomaly on the accuracy of channel value assessment. If the game is in the new game launch phase, the weight of the impact of the cause of the anomaly on user acquisition cost and paid conversion rate will be increased, and the comprehensive impact score will be calculated based on the increased impact weight. Since the core objective at this stage is to quickly acquire users and verify market feedback, the impact of the cause of the anomaly on user acquisition cost and first-day paid conversion rate is considered more critical.

[0091] If the game is in a mature operational phase, the weighting of anomalies on user retention rate and user lifetime value will be increased, and the overall impact score will be calculated based on the increased weighting. At this stage, the game's focus shifts to user retention and long-term value extraction, and the weighting of anomalies on user retention rate and lifetime value will be increased. For example, if an anomaly leads to an abnormally high churn rate or a decrease in user lifetime value, even if user acquisition costs do not fluctuate significantly, the overall impact score will increase due to the significant influence of the high-weighted indicators.

[0092] If the anomaly involves a specific channel, the impact weights of key business metrics for that channel are adjusted based on its historical performance data. A comprehensive impact score is then calculated based on these increased weights. When the comprehensive impact score reaches a second preset threshold, the corresponding warning level and handling strategy are selected. If the specific channel is a newly partnered channel or a channel with historically poor performance, the impact weights of its relevant business metrics need to be finely adjusted by combining its historical performance data (e.g., average user acquisition cost, conversion rate, retention rate, etc.). For example, a channel with a historically low conversion rate might be given a higher impact weight even with a slight conversion rate anomaly to avoid potentially greater losses.

[0093] Ultimately, when the calculated comprehensive impact score reaches the second preset threshold, the system will automatically select the corresponding warning level (e.g., general warning, moderate warning, emergency warning) and handling strategy (e.g., data verification, channel suspension, manual intervention, etc.) based on the severity indicated by the score. Furthermore, it is understood that the second preset threshold can be dynamically adjusted based on actual operational experience and risk tolerance, thereby ensuring more accurate assessment of the causes of anomalies in different business scenarios, and thus selecting the warning level and handling strategy that best suits the current business objectives.

[0094] In one exemplary embodiment, a fixed preset threshold may not adequately adapt to the complexity and dynamic changes of different game types, channel characteristics, and user lifecycle stages. If the threshold is set improperly, the system may be overly sensitive or insensitive to anomalies, thereby affecting the accuracy of warnings and the timeliness and effectiveness of handling strategies. When the comprehensive impact score reaches a second preset threshold, the corresponding warning level and handling strategy are selected, including: Based on the game type, channel characteristics, and the current user's game lifecycle stage, adjust the second preset threshold, and select the corresponding warning level and handling strategy according to the adjusted second preset threshold. For example, different types of games (such as casual games and hardcore RPGs) may have different tolerances for channel anomalies; the data characteristics and risk patterns of different channels (such as user acquisition channels and brand channels) also differ; and the sensitivity of users to abnormal events should also be treated differently depending on their lifecycle stage (such as new users, active users, and churned users). If the game is in the new game launch phase, the second preset threshold is lowered to the third preset threshold. The overall impact score is then determined based on the third preset threshold, and the corresponding warning level and handling strategy are selected. During this phase, tolerance for channel anomalies is lower, requiring earlier and more sensitive identification of potential problems to avoid early user churn and resource waste. Therefore, the preset threshold is appropriately lowered so that even with a relatively low overall impact score, warnings and actions can still be triggered, ensuring close monitoring of the new game's channel health.

[0095] If the game is in a mature operational phase, the second preset threshold is adjusted to the fourth preset threshold based on the historical fluctuation range of user retention rate and the current user's game lifecycle stage. The overall impact score is then determined based on the fourth preset threshold, and the corresponding warning level and handling strategy are selected. At this stage, the adjustment of the second preset threshold will be combined with the historical fluctuation range of user retention rate and lifetime value. Specifically, the normal fluctuation range of these key indicators over a period of time is analyzed, and a threshold that better reflects the current operational status is set accordingly. For example, if historical data shows that fluctuations in user retention rate within a certain range are normal, then only when the overall impact score causes the predicted retention rate to exceed this range will it be judged as an anomaly requiring a warning, thus avoiding overreaction.

[0096] If the cause of the anomaly involves a specific channel, the second preset threshold is adjusted to the fifth preset threshold based on the historical data performance of that channel. The comprehensive impact score is then determined based on the fifth preset threshold, and the corresponding warning level and handling strategy are selected. Since each channel has its unique operational characteristics and risk patterns, for example, some channels may be more prone to fraudulent traffic activities, while others may have inherent delays in data transmission. By combining the historical performance of a specific channel, the second preset threshold can be adjusted in a targeted manner, making the anomaly judgment for that channel more accurate and avoiding a "one-size-fits-all" judgment approach.

[0097] In one example, suppose a game company launches a new casual game and operates multiple user acquisition channels simultaneously. In the early stages of the game's launch, due to high user acquisition costs and insufficient user behavior data accumulation, the system will lower the second preset threshold. For instance, if a warning is typically triggered only when the overall impact score reaches 0.7, the second preset threshold might be adjusted to a third preset threshold of 0.5 during the new game launch phase. This means that even with a relatively low overall impact score, potential channel anomalies will be more sensitively identified, such as a sudden spike in user acquisition costs or a significantly lower-than-expected first-day paid conversion rate, thus triggering a warning in a timely manner. This allows the operations team to quickly intervene, adjust the advertising strategy, or investigate technical issues, preventing early user churn.

[0098] Once the game enters its mature operational phase, user retention rate and lifetime value become more important metrics. At this stage, historical data analysis reveals that user retention rate typically fluctuates within a specific range. If the overall impact score indicates that user retention rate will fall below the lower limit of this historical fluctuation range, an alert is triggered based on an adjusted fourth preset threshold (e.g., it might revert to 0.7 or be dynamically adjusted according to the historical fluctuation range). For example, if a channel experiences a sudden and abnormal drop in user retention rate, causing the overall impact score to reach the adjusted threshold, an alert will be issued immediately, and the operations team will be advised to check the user quality or content delivery strategy for that channel.

[0099] Furthermore, if historical data for a specific channel (e.g., an emerging short video advertising channel) shows significant fluctuations in its user payment conversion rate, the preset threshold will be fine-tuned when assessing the channel's anomalies, taking into account its historical performance. For example, even if the overall impact score for this channel is slightly higher than the threshold for other established channels, the highest level alert may not be triggered immediately. Instead, a more in-depth data analysis will be conducted to distinguish between normal fluctuations and genuine anomalies.

[0100] Secondly, this application provides a game distribution channel optimization system, the system comprising: The data acquisition module is used to acquire raw user behavior information and raw event data from the channel, attach a system receiving timestamp to the raw event data and temporarily store it in the storage database; The status determination module is used to extract key user behavior events from the raw event data. If the preset final confirmation data is not received, the status of the raw event data is determined to be pending confirmation. The asynchronous scanning module is used to asynchronously scan the storage database and identify delayed critical events in the raw event data, and extract the user identifier and the original reporting timestamp from the critical events; The correction and optimization module is used to trace and associate key user behavior events within a preset time period adjacent to the occurrence time of the key event based on the user identifier and the original reporting timestamp, and to correct the status and attribution channel of the key event based on the key user behavior events.

[0101] In summary, the game distribution channel optimization method provided by this invention effectively solves the problem of data loss or incorrect labeling caused by data transmission delays and inconsistent formats in emerging channels by receiving raw event data from the channel and attaching a system receiving timestamp for temporary storage. Furthermore, it determines key user behavior events as pending confirmation when final confirmation data has not been received, asynchronously scans the temporary storage data to identify delayed key events, and on this basis, traces the user's behavior data before and after the event occurrence time based on the user identifier and the original reporting timestamp, thereby correcting the status and channel attribution of key events.

[0102] Through the aforementioned technical solution, this application can fundamentally solve the problem of channel quality assessment bias caused by data distortion in existing technologies. Specifically, by effectively processing delayed data and correcting the status of key events, it avoids the misclassification of a large number of real-world payment and retention behaviors as "not-occurring," thereby correcting the system's severe underestimation of the true value of users. It also allows the channel quality assessment system to update parameters based on more accurate and complete data, avoiding a vicious cycle of reinforcing negative judgments after "learning" from distorted data. This ensures that the budget allocation decision-making mechanism is based on real user value and channel performance, preventing the erroneous reduction of budgets for high-potential channels or even the cessation of advertising, thus effectively improving game release efficiency and maximizing marketing investment returns.

[0103] Example 2

[0104] Embodiment 2 of the present invention provides a game distribution channel optimization system. Figure 4 This is a block diagram illustrating a game distribution channel optimization system according to an exemplary embodiment. (Refer to...) Figure 4 The system includes: Data acquisition module 01 is used to acquire raw user behavior information and raw event data from the channel, attach a system receiving timestamp to the raw event data and temporarily store it in the storage database; The status determination module 02 is used to extract key user behavior events from the raw event data. If the preset final confirmation data is not received, the status of the raw event data is determined to be pending confirmation. Asynchronous scanning module 03 is used to asynchronously scan the storage database and identify delayed key events in the raw event data, and extract user identifiers and original reporting timestamps from the key events; The optimization module 04 is used to trace and associate key user behavior events in a preset time period adjacent to the occurrence time of the key event based on the user identifier and the original reporting timestamp, and to correct the status and attribution channel of the key event based on the key user behavior events.

[0105] The game distribution channel optimization system provided in Embodiment 2 of this invention effectively solves the problem of data loss or mislabeling caused by data transmission delays and inconsistent formats in emerging channels by receiving raw event data from channels and attaching a system receiving timestamp for temporary storage. Furthermore, it determines key user behavior events as pending confirmation when final confirmation data is not received, asynchronously scans the temporary data to identify delayed key events, and, based on user identifiers and original reporting timestamps, traces user behavior data before and after the event occurrence time, thereby correcting the status and channel attribution of key events. This ensures that the budget allocation decision-making mechanism is based on real user value and channel performance, avoiding erroneous reduction of budgets for high-potential channels or even cessation of advertising, thus effectively improving game distribution efficiency and maximizing marketing investment return.

[0106] Example 3

[0107] Embodiment 3 of the present invention provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, it implements the game distribution channel optimization method provided in Embodiment 1.

[0108] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0109] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, is used to cause the terminal device to perform the steps of implementing the game distribution channel optimization method in Embodiment 1.

[0110] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A game distribution channel optimization method, characterized by, The method comprises: acquiring original event data from a channel party, attaching a system receiving timestamp to the original event data and temporarily storing the original event data to a storage database; extracting a key user behavior event in the original event data, and determining a state of the original event data as to be confirmed when a preset final confirmation data is not received; asynchronously scanning the storage database and identifying a key event delayed in arriving in the original event data, extracting a user identifier and an original reporting timestamp in the key event; tracing and associating the key user behavior event in a preset time period adjacent to a time of occurrence of the key event according to the user identifier and the original reporting timestamp, and correcting a state and a belonging channel of the key event according to the key user behavior event.

2. The optimization method of claim 1, wherein, The tracing and associating the key user behavior event in the preset time period adjacent to the time of occurrence of the key event according to the user identifier and the original reporting timestamp comprises: standardizing the original user behavior information from different channel parties, extracting a user identification clue and behavior time information, and generating a unique internal number for each standardized user behavior information; transforming the standardized user behavior information into a user behavior relationship graph, wherein the user behavior relationship graph comprises nodes representing a user identity clue, a channel, a user behavior event and user behavior content, and associated edges representing relationships between the nodes; in the user behavior relationship graph, marking a plurality of user identity clues close in time and having common characteristics as identification nodes and connecting the identification nodes through the associated edges to form a user identity chain; when the key event is identified, traversing the user behavior relationship graph in reverse from the key event as a starting point to identify a user behavior path from a user initial touch point to a final conversion that meets the conditions; comprehensively judging a type influence, a time proximity, an interaction depth and an identity clue certainty of a user touch point in each channel party in the user behavior path, calculating a contribution score of each channel party in the user behavior path, and distributing a belonging channel weight of the user according to the contribution score.

3. The optimization method of claim 2, wherein, The marking a plurality of user identity clues close in time and having common characteristics as identification nodes and connecting the identification nodes through the associated edges to form a user identity chain in the user behavior relationship graph comprises: evaluating a reliability weight of the common characteristics, wherein the reliability weight is used to indicate an importance of the common characteristics in identity association; judging a belonging channel of the user identity clue according to a continuity mode of the user behavior event, the continuity mode comprising continuous clicking, browsing or downloading behaviors in different channels within a short time, wherein if there is a login behavior of a single user in different channel parties, the user identity clue is attributed to a channel party of a login account; if there are historical records of a plurality of users logging in different accounts on the same device, the user identity clue is attributed to a channel party of an active account at a time point of occurrence of the user behavior event. If there is a shared device or account scenario, the attribution channel of the user identity clue is comprehensively judged according to the type and frequency of the user behavior event, wherein the type includes in-game payment behavior or advertisement click behavior; Based on the reliability weight of the common feature, the continuity mode of the user behavior event and the attribution channel of the user identity clue, the association confidence score of the user identity clue is comprehensively calculated, and the recognition node is marked when the association confidence score reaches a first preset threshold, and the association edge string is created based on the recognition node to form the user identity chain.

4. The optimization method of claim 3, wherein, The comprehensive calculation of the association confidence score of the user identity clue based on the reliability weight of the common feature, the continuity mode of the user behavior event and the attribution channel of the user identity clue includes: If the channel is a newly online channel or a channel with less data, the weight of the attribution channel of the user identity clue is increased, and identity association is performed in combination with a short-time high-frequency behavior mode to calculate the association confidence score; If the channel is a mature channel, the weight of the reliability weight of the common feature is dynamically adjusted according to the historical data analysis result, and identity association is performed in combination with a long-term behavior mode and a user life cycle stage to calculate the association confidence score; If it is during a game version update or a major market activity, the weight of the continuity mode of the user behavior event is temporarily adjusted to calculate the association confidence score; The fluctuation of the association confidence score and the breakage rate of the user identity chain are monitored, and an early warning is triggered when abnormal fluctuation or the breakage rate of the user identity chain increases.

5. The optimization method of claim 4, wherein, The monitoring of the fluctuation of the association confidence score and the breakage rate of the user identity chain, and the triggering of the early warning when the abnormal fluctuation or the breakage rate of the user identity chain increases to a second preset threshold, includes: Abnormal feature information of the abnormal fluctuation or the breakage rate of the identity chain is collected and analyzed, wherein the abnormal feature information includes an abnormal time point, an abnormal duration and an abnormal influence range; The original event data transmission log in the abnormal feature information is extracted and the data transmission quality of the original event data transmission log is analyzed to obtain a data transmission quality diagnosis result; In the user behavior relationship graph, the user behavior mode in the abnormal duration is analyzed to diagnose whether the user behavior mode is a real user behavior mode change, and a user behavior mode diagnosis result is obtained; According to the historical user data and user portrait information of the channel party, whether the user has characteristics of suspected malicious traffic brushing behavior is identified to obtain a malicious traffic brushing behavior diagnosis result; According to the data transmission quality diagnosis result, the user behavior mode diagnosis result and the malicious traffic brushing behavior diagnosis result, an abnormal diagnosis report is generated, and an early warning level and a processing strategy are selected according to the abnormal diagnosis report, wherein the abnormal diagnosis report includes an abnormal reason type and an abnormal severity, and is used to indicate the possible reason of the abnormality.

6. The optimization method of claim 5, wherein, The selection of the early warning level and the processing strategy according to the abnormal diagnosis report includes: According to the abnormal diagnosis report, a corresponding warning level is matched from a preset warning level matrix, and a corresponding processing strategy is matched from a preset processing strategy library, wherein If the abnormal diagnosis report indicates multiple abnormal causes but with different priorities, multiple abnormal causes are sorted according to a preset priority rule, and a warning level and a processing strategy corresponding to an abnormal cause with the highest priority after sorting are selected; If the abnormal diagnosis report indicates multiple abnormal causes with the same priority, multiple abnormal causes are sorted according to the influence degree of the abnormal causes on the accuracy of the channel party value evaluation, and a warning level and a processing strategy corresponding to an abnormal cause with the largest influence degree after sorting are selected.

7. The optimization method of claim 6, wherein, The sorting of multiple abnormal causes according to the influence degree of the abnormal causes on the accuracy of the channel party value evaluation, and the selection of a warning level and a processing strategy corresponding to an abnormal cause with the largest influence degree after sorting, include: According to the game type, the channel characteristics and the current user game life cycle stage, the influence weight of the key business indicators is allocated, wherein the key business indicators include user acquisition cost, payment conversion rate, user retention rate and user game life cycle value; Based on the direct influence and duration of the abnormal cause on the key business indicators, a comprehensive influence score of the abnormal cause on the accuracy of the channel party value evaluation is calculated, wherein If the game is in the new game online stage, the influence weight of the abnormal cause on the user acquisition cost and the payment conversion rate is increased, and the comprehensive influence score is calculated according to the increased influence weight; If the game is in the mature operation stage, the influence weight of the abnormal cause on the user retention rate and the user game life cycle value is increased, and the comprehensive influence score is calculated according to the increased influence weight; If the abnormal cause involves a specific channel, based on the historical performance data of the specific channel, the influence weight of the key business indicators of the specific channel is adjusted, and the comprehensive influence score is calculated according to the increased influence weight, and when the comprehensive influence score reaches a second preset threshold, a corresponding warning level and processing strategy are selected.

8. The optimization method of claim 7, wherein, The selection of a corresponding warning level and a processing strategy when the comprehensive influence score reaches a second preset threshold includes: According to the game type, the channel characteristics and the current user game life cycle stage, the second preset threshold is adjusted, and a corresponding warning level and processing strategy are selected according to the adjusted second preset threshold, wherein If the game is in the new game online stage, the second preset threshold is reduced to a third preset threshold, and the comprehensive influence score is determined according to the third preset threshold, and a corresponding warning level and processing strategy are selected; If the game is in the mature operation stage, the second preset threshold is adjusted to a fourth preset threshold according to the historical fluctuation range of the user retention rate and the current user game life cycle stage, and the comprehensive influence score is determined according to the fourth preset threshold, and a corresponding warning level and processing strategy are selected; If the abnormal cause involves a specific channel, the second preset threshold is adjusted to a fifth preset threshold in combination with historical data performance of the specific channel, and the comprehensive influence score is judged according to the fifth preset threshold, and a corresponding early warning level and processing strategy are selected.

9. A game distribution channel optimization system, characterized by, The system comprises: a data acquisition module for acquiring original user behavior information and original event data from a channel party, attaching a system receiving time stamp to the original event data and temporarily storing the original event data in a storage database; a state determination module for extracting a key user behavior event in the original event data, and determining the state of the original event data as to be confirmed when no preset final confirmation data is received; an asynchronous scanning module for asynchronously scanning the storage database and identifying a delayed-arrival key event in the original event data, extracting a user identifier and an original reporting time stamp in the key event; a correction and optimization module for tracing and associating the key user behavior event in a preset time period adjacent to the occurrence time of the key event according to the user identifier and the original reporting time stamp, and correcting the state and attribution channel of the key event according to the key user behavior event. 10.A computer readable storage medium having a program stored thereon, wherein the program, when executed by a processor, implements the game distribution channel optimization method of claims 1-8.

Images