An in-game promotion information card point delivery method, device and equipment

Abstract

The application discloses an in-game promotion information card point putting method, device and equipment. The method comprises the following steps: generating a plurality of putting time nodes based on a reference game node in a target game script and a promotion information material feature in a to-be-put material library, inserting promotion information materials matched with the putting time nodes into the game script, putting the target game script after the promotion information materials are inserted into a target reinforcement learning environment configured with a plurality of user intelligent body models for iterative adjustment, and outputting a putting strategy when an iteration stop condition is met; since the target game script after the promotion information materials are inserted is put into the target reinforcement learning environment, the user intelligent body models in the reinforcement learning environment simulate the real interactive behavior of users in the game to the promotion information, the dynamic optimization of the card point putting strategy is realized, the game rhythm is avoided from being interfered with while the full exposure of the promotion information is ensured, and the advertisement putting accuracy and the user experience are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of in-game information delivery technology, and in particular to a method, apparatus, and equipment for delivering promotional information at designated points within a game. Background Technology

[0002] With the development of digital gaming businesses such as mobile games, web games, and cloud games, in-game promotional information delivery has become a crucial component of game operation and monetization. Existing game products typically need to embed incentive-based, interactive, or native promotional information into various game nodes, such as level transitions, task completions, character resurrection, scene transitions, and settlement displays, without deviating from the main game flow, to increase the exposure, viewership, and conversion rates of promotional information. Meanwhile, the placement, timing, and content of promotional information directly affect users' perception of the game's pace, their immersion in the game environment, and their overall acceptance of the game experience. Therefore, how to effectively deliver promotional information during game script execution has become a key issue in the field of game promotion technology.

[0003] In existing technologies, the delivery of in-game promotional information typically involves either manually pre-configuring fixed points or configuring rule templates based on historical statistical results. Specifically, one approach involves operations personnel or the delivery system setting the insertion position of promotional information according to preset nodes in the target game script, matching promotional materials from the material library to the corresponding fixed points. Another approach involves offline adjustments to existing delivery rules based on historical exposure data, click data, skip data, or user tag data to create a new delivery configuration before re-deploying it online. While some solutions may incorporate user segmentation results, material classification results, or node popularity results for optimization, they generally still revolve around static rules, fixed nodes, or single-round statistical feedback, relying primarily on manual experience, preset rules, or simple historical data feedback mechanisms for adjustment.

[0004] However, existing technologies still have significant shortcomings in practical applications. First, they typically struggle to uniformly represent user experience constraints, game rhythm constraints, and game scene constraints at different time points based on the comprehensive game characteristics of each time point in the target game script, making it difficult to accurately generate delivery time points that are compatible with the game process. Second, existing technologies usually handle the selection of promotional information checkpoints and the matching of promotional materials separately, lacking a linkage optimization mechanism for both, which easily leads to insufficient matching between checkpoints and materials. Third, existing technologies have low utilization of user behavior feedback during gameplay, such as watching, skipping, interrupting, or resuming promotional information, making it difficult to build a dynamic iterative environment that can simulate real user interaction processes. This results in lagging updates to delivery strategies and an inability to continuously adapt to behavioral differences corresponding to different user profiles. Therefore, how to achieve dynamic optimization and accurate delivery of in-game promotional information checkpoints without affecting the user's gaming experience has become a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] The main objective of this invention is to provide a method, apparatus, and device for placing promotional information cards within a game, aiming to solve the technical problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for placing promotional information cards in games, the method comprising the following steps: Multiple delivery time nodes are generated based on the baseline game nodes in the target game script and the promotional information features of each promotional information material in the material library to be delivered. The baseline game nodes are used to characterize the comprehensive game features of each time node in the target game script and the key constraints for delivering promotional information at each time node. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints. Insert promotional materials that match each delivery time point into the target game script; The target game script, after inserting promotional materials, is deployed to the target reinforcement learning environment to iteratively adjust the deployment time points of the target game script and the promotional materials matched at each deployment time point. The target reinforcement learning environment is configured with multiple user agent models, which are constructed based on the user profile features of the platform to be deployed. The user agent models are configured to interact with the promotional materials at each deployment time point during the execution of the target game script. In response to the current iteration stopping condition being met, the current delivery time node of the target game script and the promotional information materials matching each delivery time node are output as the delivery strategy. The iteration stopping condition is that the current iteration number reaches a preset number threshold or the promotional information interaction behavior of the user intelligent agent model meets the preset interaction condition.

[0007] Furthermore, to achieve the above objectives, the present invention also proposes an in-game promotional information checkpoint delivery device, the in-game promotional information checkpoint delivery device comprising: The time node generation module is used to generate multiple delivery time nodes based on the baseline game node in the target game script and the promotion information features of each promotion information material in the material library to be delivered. The baseline game node is used to characterize the comprehensive game features of each time node in the target game script and the key constraints for delivering promotion information at each time node. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints. The material insertion module is used to insert promotional information materials that match each delivery time point into the target game script; The iterative adjustment module is used to deploy the target game script after inserting promotional information materials to the target reinforcement learning environment, so as to iteratively adjust the deployment time nodes of the target game script and the promotional information materials matched at each deployment time node. The target reinforcement learning environment is configured with multiple user intelligent agent models. The user intelligent agent models are constructed based on the user profile features of the platform to be deployed. The user intelligent agent models are configured to interact with the promotional information materials at each deployment time node during the operation of the target game script. The strategy output module is used to output the current deployment time node of the target game script and the promotional information materials matched with each deployment time node as a deployment strategy in response to the current iteration stopping condition. The iteration stopping condition is that the current iteration number reaches a preset number threshold or the promotional information interaction behavior of the user intelligent agent model meets the preset interaction condition.

[0008] Furthermore, to achieve the above objectives, this application also proposes an in-game promotional information card placement device, the device comprising: a memory, a processor, and an in-game promotional information card placement program stored in the memory, the processor being used to run the in-game promotional information card placement program, and the computer program being configured to implement the steps of the in-game promotional information card placement method as described above.

[0009] In addition, to achieve the above objectives, this application also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the in-game promotional information card placement method described above.

[0010] This invention deeply couples multidimensional game internal constraints with a reinforcement learning mechanism based on user profile simulation to construct a closed-loop automated delivery node optimization mechanism. This effectively uncovers the deep-seated matching logic between complex game process state transitions and multimodal promotional content features, significantly improving the accuracy of promotional information delivery and user interaction efficiency. Simultaneously, by utilizing intelligent agent models in a simulation computing environment for low-cost, high-concurrency iterative trial and error, it significantly reduces the reliance on real online traffic and the high trial-and-error costs of traditional manual testing. While maintaining a good gaming experience and a balance with the core game rhythm, it achieves a comprehensive improvement in the intelligence level and efficiency of the overall promotional information delivery strategy. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the structure of the in-game promotional information checkpoint delivery device in the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the first embodiment of the in-game promotional information card placement method of the present invention; Figure 3 This is a flowchart illustrating the second embodiment of the in-game promotional information card placement method of the present invention; Figure 4 This is a flowchart illustrating the third embodiment of the in-game promotional information card placement method of the present invention; Figure 5 This is a flowchart illustrating the fourth embodiment of the in-game promotional information card placement method of the present invention; Figure 6 This is a structural block diagram of the first embodiment of the in-game promotional information card placement device of the present invention.

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0015] Reference Figure 1 , Figure 1 This is a schematic diagram of the in-game promotional information checkpoint placement device structure in the hardware operating environment involved in the embodiments of the present invention.

[0016] like Figure 1 As shown, the promotional information distribution device within the game may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to establish communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0017] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the in-game promotional information card placement device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0018] like Figure 1 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a program for distributing promotional information cards within the game.

[0019] exist Figure 1 In the in-game promotional information card placement device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the in-game promotional information card placement device of the present invention can be set in the in-game promotional information card placement device, and the in-game promotional information card placement device calls the in-game promotional information card placement program stored in the memory 1005 through the processor 1001, and executes the in-game promotional information card placement method provided in the embodiment of the present invention.

[0020] This invention provides a method for placing promotional information cards within a game, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the in-game promotional information card placement method of the present invention.

[0021] In this embodiment, the method for placing promotional information cards in the game includes the following steps: Step S10: Generate multiple delivery time nodes based on the baseline game nodes in the target game script and the promotional information features of each promotional material in the material library to be delivered.

[0022] It should be understood that the executing entity of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a terminal electronic device capable of performing the above functions. The following description uses a promotional information card placement device within a game (hereinafter referred to as the placement device) as an example to illustrate this embodiment and the subsequent embodiments.

[0023] It should be noted that the benchmark game node is used to characterize the comprehensive game features of each time point in the target game script and the key constraints for delivering promotional information at each time point. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints.

[0024] It should be noted that the target game script refers to a collection of program code or structured configuration files containing complete game logic flow, scene rendering sequence, and event triggering rules. A baseline game node can be a key time anchor point in the game process that triggers a specific interactive event, completes a specific level, or experiences a significant transition in the game's underlying state. A delivery time node refers to a specific timestamp or game event progress point determined on the game's timeline to trigger the rendering and display of promotional information.

[0025] It should be noted that the characteristics of promotional information can be a vectorized representation of data from dimensions such as the media format type of the promotional content, the content duration, the target audience profile preference tags, and the content sentiment.

[0026] Key constraints refer to a set of multi-dimensional rules that limit the timing of promotional information display to avoid triggering negative player feedback. User experience constraints may include rules limiting the frequency of promotional pop-ups or preventing promotional images from obscuring the core gameplay interaction. Game rhythm constraints may be conditional rules that prevent interrupting the player's flow during intense combat or high-frequency operation phases. Game scene constraints refer to matching rules that require the visual style and display logic of promotional information to be consistent with the current game's visual style and world setting.

[0027] In practical implementation, the delivery device can pre-parse and extract the runtime flow trajectory of the target game script, identify multiple potential benchmark game nodes, and simultaneously quantify the comprehensive game characteristics corresponding to these nodes. Combined with the promotional information features extracted from the library of materials to be delivered, the matching degree between various promotional materials and each benchmark game node is calculated in terms of context, time gaps, etc. Based on the calculated matching degree score, a multi-dimensional rule filtering mechanism is introduced to strictly eliminate matching pairs that violate user experience constraints, game rhythm constraints, and game scene constraints. Finally, under the premise of satisfying all constraints, multiple optimal delivery time nodes are selected and generated.

[0028] For example, in the static settlement interface after completing a game in an action game (which is a baseline game node and a low-frequency operation period that conforms to the game rhythm constraints), the delivery device will match a short and similarly styled equipment promotion video (combining the characteristics of promotional information with game scene constraints).

[0029] It is understandable that this embodiment deeply couples and filters the multidimensional features of promotional information with the node status of the game itself. Under the premise of strictly ensuring the continuity of the player's game and the underlying experience, it accurately locks the time window suitable for information exposure, thereby improving the rationality of the display of promotional content and the potential conversion probability from the source.

[0030] Step S20: Insert promotional information materials that match each delivery time point into the target game script.

[0031] It should be noted that the matching promotional materials can be multimedia data entities in the form of video clips, static image stickers, interactive mini-games, or virtual prop models with jump links.

[0032] In practice, the delivery device dynamically modifies the rendering sequence or event hook logic of the target game script according to the delivery time node, embedding the loading instructions of the corresponding promotional materials and UI display code segments into the corresponding positions of the original code. Through dynamic injection at the code level, the deep integration of external promotional content and the game's native business logic is achieved, ensuring that the promotional information can be accurately triggered and displayed at the predetermined key points.

[0033] In some embodiments, the delivery device can cache the corresponding promotional material data in local memory in advance by asynchronous loading or pre-downloading. When the game engine runs in real time and advances to the corresponding delivery time node, the matching promotional information material is seamlessly rendered and superimposed on the current game screen by calling the system's graphics rendering interface in the form of floating layers, scene embedded textures, or dynamic transition animations.

[0034] Step S30: Deploy the target game script with inserted promotional materials to the target reinforcement learning environment to iteratively adjust the deployment time of the target game script and the promotional materials matched at each deployment time.

[0035] It should be noted that the target reinforcement learning environment is configured with multiple user agent models. The user agent models are constructed based on the user profile features of the platform to be deployed. The user agent models are configured to interact with the promotional information materials at each deployment time point during the execution of the target game script.

[0036] It should be noted that the target reinforcement learning environment is pre-configured with state space definitions, action space definitions, and reward functions for high-fidelity simulation of game operation and evaluation of user behavior feedback. The user agent model can be built based on deep neural networks, capable of simulating the decision-making mechanisms of real players in the game, such as field of view movement, key operations, and attention shifting. User profile features can be aggregated statistical data from dimensions such as the age distribution of real users on the platform to be advertised, preferred game genres, historical ad click habits, and willingness to pay.

[0037] In its implementation, the delivery device can deploy the entire target game script, after code injection and material embedding, into the target reinforcement learning environment for execution. During the virtual environment's runtime, multiple user agent models with different user profiles are generated. These agents take over game control and simulate real player actions. When the game progress triggers a preset delivery time point and displays promotional materials, the system captures real-time interaction feedback data from the user agent models (such as simulated gaze duration, click-to-jump actions, ignore or active close actions). Based on these real-time interactions, the target reinforcement learning environment calculates the expected return (reward value) of the current delivery strategy using a preset reward function. It then updates the underlying policy network weights using reinforcement learning algorithms such as policy gradients or value functions, dynamically updating and iteratively adjusting the delivery time point and material matching relationships in subsequent rounds to find delivery combinations with higher reward values.

[0038] It is understandable that this embodiment innovatively introduces an intelligent agent model driven by real business data for sandbox simulation interaction. It can automatically test and optimize parameters on a large scale for complex point-of-sale deployment schemes without consuming real online traffic resources. This effectively uncovers the nonlinear hidden correlation between user behavior preference characteristics and the timing of promotional information presentation, and greatly improves the reliability of the deployment strategy and the efficiency of model iteration.

[0039] Step S40: In response to the current iteration stopping condition being met, output the current delivery time node of the target game script and the promotional information materials matching each delivery time node as the delivery strategy.

[0040] It should be noted that the iteration stops when the current iteration count reaches a preset threshold or when the user agent model's promotional information interaction behavior meets preset interaction conditions. The delivery strategy can be a structured configuration data package that has been optimized through reinforcement learning and includes one or more optimal delivery timestamps and corresponding multimedia material hash value mappings.

[0041] The preset threshold for the number of training rounds can be a pre-configured maximum number of training rounds to control computational resource consumption and time costs. Preset interaction conditions refer to evaluation metrics reflecting a stable and high level of positive feedback from the agent to promotional information, such as simulating click-through rates remaining above a preset expected baseline for multiple consecutive periods.

[0042] In its implementation, after each iteration of parameter updates in the target reinforcement learning environment, the delivery device initiates a global monitoring program for status verification. The delivery device determines whether the cumulative number of simulated iterations has reached a preset threshold, or simultaneously evaluates whether the user agent model's positive interaction with the current version of the promotional information has consistently met preset interaction conditions (meaning the optimization model has converged or found a locally optimal delivery solution). Once either of these conditions is met, the delivery device immediately halts the reinforcement learning environment's training process and extracts all delivery time nodes and associated promotional materials determined in the latest target game script. The delivery device cleans and packages these matching relationships, outputting a delivery strategy file that can be directly called by the production environment, thus ending the entire optimization process. This prevents the unlimited and ineffective waste of computing resources and ensures that the final extracted delivery strategy has the expected optimal or near-optimal promotional benefits.

[0043] This embodiment deeply couples multi-dimensional game internal constraints with a reinforcement learning mechanism based on user profile simulation to construct a closed-loop automated delivery node optimization mechanism. This effectively uncovers the deep-seated matching logic between complex game process state transitions and multi-modal promotional content features, significantly improving the accuracy of promotional information delivery and user interaction efficiency. Simultaneously, by utilizing the low-cost, high-concurrency iterative trial and error of the intelligent agent model in a simulation computing environment, it greatly reduces the dependence on real online traffic and the high trial and error costs of traditional manual testing. While maintaining a good gaming experience and a balance with the core game rhythm, it achieves a comprehensive improvement in the intelligence level and efficiency of the overall promotional information delivery strategy.

[0044] refer to Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the in-game promotional information card placement method of the present invention.

[0045] Based on the first embodiment described above, in this embodiment, step S10 further includes: Step S101: Perform game feature analysis on each time node in the target game script to obtain the comprehensive game features of each time node.

[0046] It should be noted that comprehensive game features refer to a set of data that quantitatively describes the operating environment and operating logic of a target game script at a specific moment from multiple dimensions, including game rhythm features and game scene features.

[0047] Game rhythm characteristics can be abstract metrics that measure the density of player input responses and screen refresh rate within a specific time period in a target game script. Game scene characteristics can be a set of attributes that characterize the style of virtual buildings, the distribution of lighting and shadow parameters, and the types of virtual items within the currently rendered screen of the target game script.

[0048] In its implementation, the delivery device acquires the continuous code execution sequence corresponding to the running process of the target game script. The delivery device uses a built-in feature analysis algorithm to analyze each time node in the target game script, which is divided into preset time steps. It extracts game rhythm features such as the frequency of virtual key presses and the rate of interface switching within each time node. At the same time, the delivery device identifies game scene features such as the environmental layout and virtual weather status in the rendered screen within each time node. The delivery device performs data splicing and multimodal data fusion processing on the extracted game rhythm features and game scene features to accurately obtain comprehensive game features covering multiple dimensions of each time node.

[0049] It is understandable that this embodiment deconstructs the running state of the target game script in a fine-grained manner, comprehensively quantifies the underlying logical performance of the game process from the dual perspectives of pace and scene layout, and effectively mines the potential interactive context information in the time sequence.

[0050] Step S102: Construct a benchmark game node database based on the comprehensive game features of each time point. The benchmark game node database stores multiple benchmark game nodes.

[0051] It should be noted that the benchmark game node database refers to a set of key time anchor point features that have been screened, verified, and meet the preset game smoothness requirements and are persistently stored in a structured form.

[0052] In the specific implementation, the delivery device evaluates the value of each time node containing comprehensive game features according to preset screening rules, removes time nodes that indicate a high-frequency and intense confrontation state from the comprehensive game features, and marks the remaining time nodes that meet the preset smooth state threshold (e.g., preset smoothness 0.8) as benchmark game nodes. The selected benchmark game nodes, together with their corresponding comprehensive game features, are structured and encoded, and the encoded data structure is written in batches into a pre-configured non-relational storage space, thereby constructing a benchmark game node database that can support high-concurrency retrieval.

[0053] It is understandable that this embodiment significantly improves the efficiency of historical data retrieval and feature matching for game nodes by transforming dynamic and fragmented feature data into a structured and persistent standard data set, reduces the computing power consumption of the online real-time comparison process, and ensures the stability and rapid response capability of subsequent node calling processes.

[0054] Step S103: Perform feature analysis on each promotional material in the material library to obtain the promotional information features of each promotional material.

[0055] It should be noted that the material library to be deployed includes multiple promotional materials to be deployed, including incentive-based promotional materials, interactive promotional materials, and native promotional materials.

[0056] It should be noted that promotional materials for incentive behaviors can be specific advertising packages that distribute virtual game tokens or resurrection items to users after guiding them to watch multimedia resources.

[0057] Interactive promotional materials can be mini-client trial programs that allow users to perform actions such as swiping, clicking, or simple dragging.

[0058] Native promotional materials can be scene patches, virtual billboards, or brand co-branded item display components whose visual style is completely integrated into the current game world view.

[0059] Promotional information features refer to data feature vectors used to quantitatively characterize the color distribution, interaction categories, content duration, and audience preference tags of multimedia content.

[0060] In practice, the delivery device retrieves a pre-built library of materials to be delivered from an external source and traverses the various types of promotional materials stored in the library. By performing feature analysis on promotional materials for incentivized behaviors, interactive behaviors, and native promotional materials, it extracts multi-dimensional attributes such as color distribution, duration, interaction method, and resource reward type for each type of promotional material. These multi-dimensional attributes are then mapped into standardized numerical vectors to obtain the promotional features of each promotional material.

[0061] It is understandable that this embodiment achieves standardized expression of promotional content for different business forms by uniformly reducing the dimensionality and extracting features from massive multimodal promotional resources, and effectively extracts the inherent attribute logic of promotional materials.

[0062] Step S104: Generate multiple delivery time nodes based on the promotion information features and the benchmark game node.

[0063] In the specific implementation, the delivery device calls the pre-built feature similarity calculation model, performs inner product operation on the obtained promotion information features and the comprehensive game features of each benchmark game node in the benchmark game node database, evaluates the semantic alignment and scene fit between the promotion information features and the comprehensive game features, filters out matching combinations with inner product calculation scores greater than a preset score line (e.g., the preset score is 80 points), extracts the timestamp information corresponding to the benchmark game nodes in the high-scoring matching combinations, and establishes the timestamp information as multiple delivery time nodes suitable for content display.

[0064] It is understandable that this embodiment performs in-depth logical correlation calculations between static material attributes and dynamic game progress, and accurately locates the best exposure time that matches the tone of the material without disrupting the game experience in a data-driven manner, which greatly improves the scientificity and rationality of the overall content distribution strategy.

[0065] This embodiment unifies the analysis of game rhythm protection, scene visual coordination, and material interaction attributes. It maximizes the fit between initial nodes and materials while ensuring that the core game experience is not disturbed. This provides a high-quality solution space for subsequent reinforcement learning iteration and optimization, significantly reduces the cost of strategy trial and error and the investment of operational manpower, and improves the intelligent decision-making level of promotional information delivery in the game and the long-term stability of the game.

[0066] refer to Figure 4 , Figure 4 This is a flowchart illustrating the third embodiment of the in-game promotional information card placement method of the present invention.

[0067] Based on the above embodiments, in this embodiment, after step S30, the method further includes: Step S31: Monitor the promotional information interaction behavior between the user agent models and the promotional materials at each delivery time point during the operation of the target game script in the target reinforcement learning environment.

[0068] It should be noted that promotional information interaction behavior can be a simulated operation command reflecting the data exchange between the user's intelligent agent model and the displayed multimedia content in a virtual environment. User behavior monitoring results refer to the structured analysis results generated after aggregating and statistically analyzing all page jumps, dwell times, and closing operations triggered by the virtual model over multiple periods.

[0069] In practice, the delivery device calls the interface within the target reinforcement learning environment. As the target game script progresses along the timeline to each delivery time node and displays promotional materials, it captures in real time the action trajectory data of each user intelligent agent model with specific profile tags, such as simulated clicks, simulated pauses, and simulated closing actions, in response to the displayed content.

[0070] For example, data records of simulated dwell time (e.g., 3 seconds) of intelligent agent models in strategy games when encountering equipment display pages can be captured and then aggregated to form interaction behavior logs.

[0071] Furthermore, in order to deeply explore the true preference patterns of multi-dimensional audiences under different exposure points, thereby improving the accuracy of the targeting strategy, step S31 above may include: Step S311: Collect node trigger logs, material playback logs, and user operation logs of the target game script in the target reinforcement learning environment to obtain the interaction logs.

[0072] It should be noted that node trigger logs are data files that record system events and timestamps generated when the game script runs to the preset ad delivery time. Material playback logs can record process data such as the rendering progress, playback duration, and audio output status of promotional materials in the virtual interface. User operation logs are behavioral data that record the types and frequency of interaction commands input by the user's intelligent agent model during game execution. Interaction logs are standardized data sets that integrate node trigger records, material playback status, and intelligent agent operation commands.

[0073] In the specific implementation, the delivery device captures the timestamp trigger records, ad creative rendering frame rate data, and click and swipe command streams input by the agent in real time during the game script execution by calling the data tracking interface built into the reinforcement learning environment. The timestamp trigger records are aligned with the timeline and marked as node trigger logs. The ad creative rendering frame rate data and audio decoding status are merged and recorded as creative playback logs. The click and swipe command streams are classified and archived as user operation logs according to operation type. Then, the log aggregation engine is used to concatenate the fields and unify the format of the node trigger logs, creative playback logs, and user operation logs, and finally outputs the structured interaction logs.

[0074] For example, during the copy settlement phase, if the agent clicks to skip a reward advertisement, the frame rate data of the advertisement playing for only two seconds is recorded simultaneously to generate a complete log entry.

[0075] Step S312: Generate the node interaction sequence corresponding to each deployment time node based on the interaction log and the identifier of each user agent.

[0076] It should be noted that the node interaction sequence can be an ordered data chain arranged in chronological order, reflecting the material display process and the agent's operation response at a specific delivery time node.

[0077] In the specific implementation, the delivery device extracts the identifiers of each user agent by parsing the device source address field in the interaction log. Based on the user agent identifiers, the interaction logs are grouped and isolated. For each group of logs, the timestamp of the corresponding delivery time node is extracted as the starting anchor point of the sequence. The playback progress data of the material after the time node is concatenated and arranged with the agent operation instructions in chronological order. A sliding window algorithm is used to remove redundant records that exceed the preset observation period. Finally, a node interaction sequence corresponding to each delivery time node is generated. For example, the video advertisement playback record triggered in the third minute and the closing instruction in the fifth second of the log identified as agent A are arranged in sequence to construct an independent interaction sequence.

[0078] It is understood that this embodiment transforms discrete multi-source logs into structured behavioral sequences with clear causal relationships through a mechanism of identifier grouping and time sequence concatenation. This effectively eliminates data cross-interference caused by the parallel operation of multiple agents, establishes a clear behavior tracking path to support subsequent state label extraction, and significantly improves the accuracy of sequence generation and data processing efficiency.

[0079] Step S313: Based on the node interaction sequence, determine the complete display marker, active close marker, passive interruption marker, and interruption recovery marker corresponding to each delivery time node.

[0080] It should be noted that the complete display mark refers to the status indicator that the promotional material has been played without interference until the end at the specified time.

[0081] The active shutdown flag can be a status indicator that represents the user's intelligent agent model manually triggering a skip or exit operation during the playback of the material.

[0082] Passive interruption flags are status indicators that represent promotional materials being forcibly paused due to game logic conflicts or system resource scheduling.

[0083] The interruption resumption flag is a status indicator that indicates that promotional materials that have been forcibly paused will resume playback after the interference factors are eliminated.

[0084] In the specific implementation, the delivery device performs state matching by traversing the playback progress field and operation instruction field of the material in the node interaction sequence. When the playback progress reaches the preset complete threshold and no termination instruction is detected, a complete display mark is generated; when the playback progress does not reach the threshold and a clear exit instruction is detected by the agent, an active close mark is generated; when a high-priority plot event is triggered by the game script or the system resource usage is overloaded, causing playback to pause, a passive interruption mark is generated; and when the playback progress is detected to be reactivated and continue until the end after the passive interruption mark is generated, an interruption recovery mark is generated. The four types of marks are bound and mapped to the corresponding delivery time nodes.

[0085] For example, in a battle scene node, if the recording of an ad being paused for three seconds due to an agent releasing a skill is detected, the passive interruption and interruption recovery status can be marked simultaneously.

[0086] Step S314: Construct a set of node behavior features based on the complete display marker, active shutdown marker, passive interruption marker, and interruption recovery marker corresponding to each delivery time node.

[0087] In its implementation, the delivery device calculates basic behavioral weights by statistically analyzing the frequency and distribution ratio of four types of markers at each delivery time point. The cumulative duration of fully displayed markers is converted into a positive experience index, the trigger interval for actively closing markers is converted into a rejection strength index, and the number of alternations between passively interrupted and interrupted recovery markers is converted into an environmental interference index. A feature encoding matrix is ​​used to vectorize and concatenate the positive experience index, rejection strength index, and environmental interference index, followed by normalization to eliminate dimensional differences, ultimately outputting a structured set of node behavioral features. Specifically, the positive experience index represents the user's agent model's acceptance and viewing intention of promotional materials; the rejection strength index represents the user's agent model's aversion or avoidance tendency towards promotional materials; and the environmental interference index represents the obstructive impact of game logic or system status on the promotional information display process.

[0088] For example, the high-frequency active shutdown records of a certain plot node can be transformed into a high repulsion intensity vector and merged with other indices to generate a feature subset of that node.

[0089] Step S315: Based on the node behavior feature set, determine the promotion information interaction behavior between each user intelligent agent model and the promotion information materials at each delivery time node.

[0090] In its implementation, the delivery device reads multi-dimensional vector data from the node behavior feature set by calling a behavior classifier. It then inputs the positive experience index, rejection strength index, and environmental interference index into a pre-defined decision tree model for threshold comparison and path optimization. Based on the model output, the interaction patterns of each user agent model at the corresponding delivery time node are categorized into deep viewing, quick skipping, environmental blocking, or fluctuating adaptation. The categorization results are associated and encapsulated with the original node identifier and material identifier, ultimately outputting clearly defined promotional information interaction behaviors. For example, a combination of a high positive experience index and a low environmental interference index is identified as deep viewing behavior and recorded in the interaction behavior profile.

[0091] It is understandable that this embodiment incorporates subjective skipping intention, objective environmental interference, and interruption recovery capability into a unified analysis framework. Under the premise of ensuring that the parallel data of multiple agents does not cross-contaminate, it maximizes the semantic richness and computational availability of behavioral features, and provides high-confidence feedback signals for subsequent deployment strategy iteration and optimization.

[0092] Step S32: Adjust the delivery time node based on the user behavior monitoring results, and re-match the adjusted delivery time node with each promotional information material in the material library to be delivered.

[0093] It should be noted that the adjusted delivery time points refer to those established after offset calculations or reassessment, and are used to replace the original inefficient exposure delivery opportunities.

[0094] In practice, the delivery device analyzes the collected user behavior monitoring results to identify inefficient delivery time nodes where the frequency of triggering simulated close actions is higher than the preset jump frequency threshold. Combining the context features within the target game script, the timestamps of the inefficient nodes are shifted to the game's calmer periods. After generating new adjusted delivery time nodes, the feature matching algorithm is called again to select static images, texts, or short videos from the material library to be delivered that are more suitable for the current new node's tags. For example, video materials that were originally delivered during intense team battles are adjusted to static banner ads on the game's settlement screen.

[0095] It is understandable that this embodiment uses multi-dimensional monitoring feedback data as a guiding signal to dynamically correct the original suboptimal or conflicting exposure timings, and simultaneously completes the adaptive rotation of content materials and secondary semantic alignment, effectively eliminating redundant interference links that are prone to negative experience, and greatly enhancing the dynamic adaptability of the push solution.

[0096] Furthermore, in order to maximize the natural integration and exposure effectiveness of advertising content while ensuring that the core gaming experience is not disturbed, step S32 above may include: Step S321: Calculate the set of node evaluation parameters corresponding to each deployment time node based on user behavior monitoring results and key constraints.

[0097] It should be noted that key constraints can be a set of rules that restrict the placement of promotional information to ensure the normal operation of the game. The node evaluation parameter set refers to a data set that summarizes multi-dimensional quantitative indicators such as experience interference, content appeal, commercial value, and constraint penalties at each deployment time point.

[0098] In practical implementation, the deployment device reads skip frequency, dwell time, and interaction conversion rate data from user behavior monitoring results. Combining these with key constraints such as user experience thresholds, game pace safety zones, and scene compatibility standards, a weighted scoring model is used. Skip frequency is converted into an experience interference index, dwell time into a content attraction index, and interaction conversion rate into a commercial value index. A constraint penalty function is used to apply negative weights to nodes that exceed the pace safety zone or violate scene compatibility standards. The experience interference index, content attraction index, commercial value index, and constraint penalty weights are then vector-concatenated and normalized to output a set of node evaluation parameters. For example, nodes whose experience interference index spikes due to frequent skipping during the climax of a battle are marked as having a low-scoring parameter combination.

[0099] Step S322: Determine the node adjustment priority corresponding to each deployment time node based on the node evaluation parameter set, and obtain a candidate adjustment node set.

[0100] It should be noted that node adjustment priority refers to the urgency level of each delivery time node under the current strategy, indicating the need for location or content optimization. The candidate adjustment node set can be a group of time nodes identified as requiring migration or replacement operations after being filtered by urgency scoring.

[0101] In its implementation, the deployment device analyzes the experience interference index and commercial value index within the node evaluation parameter set to call a priority ranking algorithm to calculate the overall optimization urgency score for each node. Nodes with an overall optimization urgency score higher than a preset adjustment threshold are marked as high-priority optimization targets, while those with a score lower than the threshold are marked as low-priority retention targets. The high-priority optimization targets are then sorted in descending order of their overall optimization urgency scores, and the top-ranked target nodes are selected. The timestamps of these target nodes are packaged with their original parameter configurations to generate a candidate adjustment node set. For example, nodes with moderate scores but significant pacing conflicts during plot transitions are prioritized for inclusion in the candidate queue for potential relocation.

[0102] It is understandable that this embodiment achieves differentiated hierarchical management of massive deployment nodes through urgency scoring calculation and threshold screening mechanism, effectively avoiding system computing power overload and strategy oscillation caused by synchronous adjustment of all nodes, establishing an optimization queue focusing on core issues to guide subsequent resource allocation, and significantly improving the computational efficiency and strategy convergence speed of iterative adjustments.

[0103] Step S323: Perform node location migration processing based on the candidate adjustment node set and the benchmark game node database to obtain the adjusted deployment time node.

[0104] In its implementation, the delivery device iterates through the target node identifiers within the candidate adjustment node set to retrieve alternative time windows from the baseline game node database that have a higher similarity to the original game scene features of the target node than a preset matching benchmark. It then calculates the player idle rate and interface rendering load rate within these alternative time windows, selecting the optimal migration position where the player idle rate is greater than a preset idle threshold and the interface rendering load rate is less than a preset load threshold. The trigger timestamp of the target node is then shifted to the optimal migration position, and the associated logical trigger configuration is updated. A timeline conflict check is performed to eliminate overlapping intervals, and finally, the adjusted delivery time node is output. For example, a reward advertisement originally scheduled to trigger at the moment of skill release is moved to a three-second idle window after the skill cooldown ends to generate a new node sequence.

[0105] Step S324: Perform context matching based on the adjusted delivery time node and the promotion information features of each promotional material to obtain the material matching result.

[0106] In practical implementation, the delivery device can construct a node context feature vector by extracting scene color parameters, plot emotion tags, and background music rhythm patterns corresponding to the adjusted delivery time nodes. It also reads the visual main color tone, narrative style attributes, and audio beat data from the promotional information features of each promotional material to construct a material context feature vector. The semantic fit score between the node context feature vector and the material context feature vector is calculated. Material node pairings with semantic fit scores higher than a preset alignment threshold are selected. The identifier mapping relationship and fit score of the pairings are structured and recorded, and the material matching results are output. For example, a heartwarming plot node can be bound to a warm-toned native promotional material to generate a high-fit matching record.

[0107] Step S325: Based on the material matching results and the material rotation constraints of the material library to be deployed, perform material reselection processing on the adjusted deployment time node, and rematch the adjusted deployment time node with each promotional information material in the material library to be deployed.

[0108] It should be noted that the creative rotation constraint refers to the operational rules that limit the frequency of display, total exposure, and budget allocation of similar promotional information in order to maintain a healthy advertising ecosystem.

[0109] In practice, the delivery device reads high-fit matching records from the creative matching results to verify the daily exposure limit, cooldown period of similar creatives, and budget consumption rate in the creative rotation constraints of the creative library to be delivered. Invalid matching records that violate the daily exposure limit or are within the cooldown period of similar creatives are eliminated. The remaining valid matching records are then comprehensively weighted and sorted according to budget consumption rate and fit score. The top-ranked promotional creative is selected as the target replacement object, and the resource reference address of the target replacement object is written into the loading configuration table of the adjusted delivery time node. This completes the creative reselection process and establishes a rematching relationship between the adjusted delivery time node and each promotional creative in the creative library to be delivered. For example, interactive playable ads exceeding the exposure limit can be replaced with rewarded video ads that meet the cooldown period and have sufficient budget to update the final delivery configuration.

[0110] Step S33: Remove the promotional materials from the target game script and insert promotional materials that match the adjusted delivery time nodes.

[0111] In practice, the delivery device sends a resource release command to the rendering engine of the target game script to accurately locate and unload the memory occupied by the old version of the promotional information material corresponding to each previous delivery time node, and release the original loading hook logic. Then, at the adjusted delivery time node position at the code level, the rendering component of the latest matched promotional information material is dynamically injected. For example, after unloading the original incentive video component that occupied the main interface, a smaller interactive trial play widget code segment is embedded at the new settlement node, thereby completing the replacement process of the entire material environment.

[0112] Step S34: Re-deploy the target game script with inserted promotional information materials to the target reinforcement learning environment, and return to the step of monitoring the interaction behavior between each user agent model and the promotional information materials at each deployment time point during the operation of the target game script in the target reinforcement learning environment, so as to iteratively simulate the interaction process between the user and the target game script.

[0113] In the specific implementation, the deployment device redeploys the updated target game script and starts it in the target reinforcement learning environment. It uses the built-in state machine control logic to trigger a new round of simulation operation cycle. During the operation, it continuously calls probes to track the interactive feedback of each user agent model to the newly embedded material. It repeatedly executes the data collection and strategy fine-tuning process until it determines that the current cycle has reached the preset cycle threshold (50 times or 100 times) and actively stops the entire simulation process.

[0114] This embodiment transforms a static, single-deployment plan into a dynamic lifecycle management process with self-correction and optimization capabilities. This not only effectively eliminates subjective biases caused by manual experience-based settings but also continuously uncovers the optimal exposure window hidden deep within complex game logic, thereby greatly improving the accuracy of the overall checkpoint strategy and scene compatibility.

[0115] refer to Figure 5 , Figure 5 This is a flowchart illustrating the fourth embodiment of the in-game promotional information card placement method of the present invention.

[0116] Based on the above embodiments, in this embodiment, step S30 further includes: Step S301: Based on the historical game behavior data of users in the target game script and the historical interaction behavior data of users with promotional materials in the platform to be deployed, perform user stratification and clustering to obtain multiple user profile categories.

[0117] In practical implementation, the ad delivery device reads historical game behavior data stored on the platform's server, such as player login frequency, level dwell time, and item consumption rate. This allows for the simultaneous extraction of historical interaction data, including ad click-through rate, video completion rate, and skip operation trigger interval. The multi-dimensional behavioral data undergoes dimensionality reduction to eliminate feature redundancy (e.g., using principal component analysis). The reduced feature vectors are then input into a Gaussian mixture model for probability density estimation and iterative optimization. Based on the Bayesian information criterion, the optimal number of clusters is determined, grouping players with highly similar behavioral patterns into the same group. The final output includes multiple user profile categories, such as hardcore paying users, casual users, and ad-sensitive users. For example, players who are online for more than three hours daily and have an ad skip rate of less than 10% are categorized as a high-tolerance paying user group.

[0118] It is understandable that this embodiment effectively overcomes the strategy generalization problem caused by traditional extensive user tagging by finely segmenting the player group, establishes a highly cohesive and loosely coupled profile classification system to support the subsequent construction of differentiated intelligent agents, and significantly improves the accuracy of user feature extraction.

[0119] Step S302: Construct the corresponding user intelligent agent model based on each user profile category.

[0120] It should be noted that the state vector of the user intelligent agent model includes game state features, promotional information fatigue features, interaction preference features, and progress interruption tolerance features.

[0121] Game state characteristics refer to a set of environmental parameters reflecting the current game progress, scene complexity, and operational intensity. Promotional information fatigue characteristics are quantitative indicators that characterize user aversion or decreased attention due to frequent advertising exposure. Interaction preference characteristics refer to a set of parameters recording the weights of user preferences for different material types, reward mechanisms, and display formats. Progress interruption tolerance characteristics refer to a psychological threshold indicator that measures the user's acceptance of game interruptions by external content without risk of churn.

[0122] In the specific implementation, the device extracts the current game running progress, scene complexity index and operation density parameters to splice them to generate game state features. It statistically analyzes the ad exposure frequency and skip rate decay curves in the historical period to fit and generate promotional information fatigue features. It analyzes the click weight of material type and reward redemption tendency to encode and generate interaction preference features. It calculates the operation delay threshold of key plot nodes and the retention probability after forced interruption to quantify and generate progress interruption tolerance features. After dimensional alignment and tensor splicing of the four types of features, they are injected into the state input layer of the algorithm entity to complete the parameterized construction of the user intelligent agent model.

[0123] Furthermore, in order to accurately construct the user intelligent agent model and thus improve the accuracy of the interaction simulation, step S302 above may include: Step S3021: Obtain the historical game behavior sequence and historical promotion information behavior sequence corresponding to each user profile category.

[0124] It should be noted that historical game behavior sequences refer to sequences arranged chronologically that record the game operation trajectories and progress of players within a specific user profile category. Historical promotional information behavior sequences can be sequences arranged chronologically that record the interactive responses and state changes of players within a specific user profile category when they encounter advertising materials.

[0125] Step S3022: Extract game state preference parameters corresponding to each user profile category based on the historical game behavior sequence.

[0126] It should be noted that the game state preference parameter refers to a quantitative indicator that represents the degree of preference of players of a specific user profile category for different game scene types and rhythm intensities.

[0127] In practice, the deployment device can identify the player's preference distribution for high-intensity combat scenarios or gentle exploration scenarios by analyzing the peak duration of level dwell time and operation frequency in the historical game behavior sequence. It uses time series decomposition algorithm to remove random fluctuation components to extract stable behavior cycles. The scenario preference distribution and stable behavior cycles are weighted and fused to quantify and generate game state preference parameters. For example, the record of hardcore paying players frequently challenging high-difficulty dungeons at night is converted into high combat intensity preference weights and entered into the parameter configuration table.

[0128] It is understood that this embodiment transforms discrete game operation records into structured parameters that characterize the player's scene adaptability through temporal decomposition and tendency quantization mechanisms. This effectively eliminates the interference of short-term accidental operations on long-term habit judgments, establishes an accurate state preference mapping model to guide the agent's environmental perception initialization, and significantly enhances the stability of parameter extraction and the accuracy of behavioral representation.

[0129] Step S3023: Extract the promotion information fatigue parameter and interaction preference parameter corresponding to each user profile category based on the historical promotion information behavior sequence.

[0130] It should be noted that the promotional information fatigue parameter refers to a quantitative threshold that characterizes the decreased attention or aversion tendency of players in a specific user profile category due to frequent exposure to advertisements. The interaction preference parameter can be a parameter that characterizes the weight of players in a specific user profile category's preference for different types of advertising creatives, reward mechanisms, and display formats.

[0131] In practical implementation, the delivery device can fit the marginal effect model of the player's attention decreasing with the number of impressions by traversing the ad exposure frequency and skip rate decay curve in the historical promotion information behavior sequence. The slope intercept of the marginal effect model is transformed into the promotion information fatigue parameter. The click conversion rate and reward redemption tendency of different material types in the sequence are statistically analyzed to construct a preference weight matrix. The preference weight matrix is ​​normalized and encoded to generate interaction preference parameters. For example, the data of the skip rate of the ad sensitive group rising sharply after watching the incentive video three times in a row is fitted as a high fatigue sensitivity coefficient.

[0132] Step S3024: Extract the progress interruption tolerance parameter corresponding to each user profile category based on the historical game behavior sequence and the historical promotion information behavior sequence.

[0133] In practice, the delivery device can calculate the player's psychological tolerance boundary for game interruption by cross-comparing the operation delay records of key plot nodes in the historical game behavior sequence with the retention probability after forced pop-up triggering in the historical promotion information behavior sequence. The survival analysis model is used to evaluate the user churn risk curve under different interruption durations, and the safe interval threshold of the risk curve is converted into a progress interruption tolerance parameter.

[0134] Step S3025: Construct a corresponding user agent model based on the game state preference parameters, the promotion information fatigue parameters, the interaction preference parameters, and the progress interruption tolerance parameters.

[0135] In the specific implementation, the delivery device can map the game state preference parameters to the weights of the game state feature input layer nodes by initializing the multilayer perceptron network architecture, encode the promotion information fatigue parameter into the bias term of the promotion information fatigue feature hidden layer, transform the interaction preference parameter into the activation threshold of the interaction preference feature output layer, configure the progress interruption tolerance parameter into the global constraint mask of the progress interruption tolerance feature, and use tensor splicing technology to perform dimensional alignment and structural fusion of the four types of features to complete the network parameter injection and topology solidification to generate the user intelligent agent model.

[0136] Step S303: Construct a target reinforcement learning environment based on the delivery time node, promotion information characteristics, and key constraints.

[0137] It should be noted that the action space of the target reinforcement learning environment includes node retention actions, node migration actions, and material replacement actions.

[0138] Node retention refers to the control instruction that maintains the matching relationship between the current delivery time node and the creative material. Node migration refers to the control instruction that shifts the timestamp of the delivery time node to an adjacent time window that meets the key constraints. Creative replacement refers to the control instruction that unbinds the promotional information from the current node and rematches alternative creative materials.

[0139] In its implementation, the delivery device initializes the virtual sandbox by loading the game script configuration file marked with delivery time nodes. It reads the duration, visual style, and reward mechanism parameters of the materials bound to each node to construct an environment observation matrix. It compiles the user experience protection threshold, rhythm safety interval, and scene compatibility rules into an environment penalty function, defines the set of operation instructions that the agent is allowed to execute, encapsulates the instruction to maintain the current node position and material binding as a node retention action, encapsulates the instruction to adjust the node trigger timestamp to the adjacent safety window as a node migration action, and encapsulates the instruction to unbind the current material and rematch the alternative library resources as a material replacement action. The observation matrix, penalty function, and operation instruction set are systematically integrated to generate the target reinforcement learning environment.

[0140] Step S304: Configure the collaboration mechanism and message mechanism based on the target reinforcement learning environment.

[0141] It should be noted that the collaborative mechanism is used to configure policy networks and value networks for each user agent model, and the messaging mechanism is used to enable user agent models corresponding to different user profile categories to pass node feedback information in the same iteration round.

[0142] In its implementation, the delivery device instantiates an independent multilayer perceptron as a policy network for each user agent model by calling the deep learning framework interface to output the action probability distribution. Simultaneously, it constructs a critic network with shared parameters as a value network to evaluate the expected global state benefit. It establishes a cross-agent communication bus and defines a standardized data packet format. After each agent completes its current round of actions, it extracts core indicators such as node interference index, material matching score, and reward decay rate. These core indicators are packaged into node feedback information and broadcast to the receiving ports of other agents corresponding to different profile categories via the communication bus. This allows each agent to integrate external feedback in the next decision step to correct its own policy gradient, thus completing the underlying deployment of the collaboration and messaging mechanisms.

[0143] Furthermore, in order to maximize the global policy coordination efficiency and reward signal accuracy while ensuring the flexibility of independent decision-making among multiple agents, step S304 above may include: Step S3041: Determine the first reward parameter based on the behavior of watching the promotional information in its entirety, skipping the promotional information, and closing the promotional information.

[0144] It should be noted that "watching the entire promotional message" refers to the user's intelligent agent model not triggering any interruption or exit commands during the ad creative display until playback ends naturally. "Skipping the promotional message" refers to the user's intelligent agent model actively triggering fast-forward or skip commands before the ad creative playback progress reaches a preset point. "Closing the promotional message" refers to the user's intelligent agent model directly calling the interface exit API to terminate playback during the ad creative display.

[0145] In the specific implementation, the delivery device extracts the complete viewing time record of the advertising material from triggering to natural end, the timestamp distribution of the skip command triggered during playback, and the operation frequency of directly clicking the close button. The complete viewing time is mapped as a positive base score, the case where the skip command is triggered earlier than the preset progress threshold is mapped as a negative decay score, and the direct close operation is mapped as a strong negative penalty score. Through weighted summation and normalization, the first reward parameter is output.

[0146] Step S3042: Determine the second reward parameter based on node interruption behavior, node recovery behavior, and repeated exposure behavior.

[0147] It should be noted that node interruption behavior refers to the forced pause of promotional materials due to high-priority game events or system resource contention. Node resumption behavior can refer to the automatic continuation of the remaining playback progress of promotional materials after the external interference factors are eliminated. Repeated exposure behavior refers to the behavior of the same promotional material being triggered and displayed to the same user agent model multiple times within a limited time period.

[0148] It should be noted that the first reward parameter refers to a numerical indicator that quantifies the user's agent model's subjective willingness to accept and interact with the promotional information content. The second reward parameter refers to a numerical indicator that quantifies the degree of interference of the game's operating environment with ad display and the healthy level of exposure frequency.

[0149] In its implementation, the delivery device captures records of ad playback pauses caused by game core logic or system resource scheduling, timestamps of automatic resumption after pauses, and exposure frequencies of the same material being called multiple times within a preset time window. The pause records are calculated as environmental interference deductions based on the interruption duration and the criticality of the scene. The resume records are calculated as experience compensation bonuses based on the smoothness of the resume. Records with repeated exposure frequencies exceeding the fatigue threshold are calculated as marginally decreasing penalty points. A non-linear activation function is used for fusion compression to eliminate dimensional differences, and a second reward parameter is output.

[0150] Step S3043: Construct a reward evaluation model based on the first reward parameter and the second reward parameter.

[0151] In its implementation, the delivery device initializes a multi-layer feedforward neural network architecture, connecting the first reward parameter to the subjective intention input layer node and the second reward parameter to the objective environment input layer node. It sets the number of hidden layer neurons and configures a hyperbolic tangent activation function to capture the non-linear coupling relationship between subjective and objective parameters. It uses a historical interaction sample library to perform backpropagation training to optimize the network weight matrix. It introduces an attention mechanism module to dynamically adjust the contribution ratio of subjective and objective parameters in different game scenarios. After completing the network topology solidification and parameter convergence verification, it outputs a reward evaluation model with adaptive weighting capabilities.

[0152] Step S3044: Configure policy network, value network and message routing table for each user agent model based on the reward evaluation model and each user profile category.

[0153] In the specific implementation, the delivery device instantiates an independent multilayer perceptron as a policy network for the corresponding intelligent agent by traversing the core behavioral tags of each user profile category to output the action probability distribution. Simultaneously, it constructs a critic architecture with shared underlying features as a value network to evaluate the expected global state benefit. The output interface of the reward evaluation model is connected to the loss calculation module of the value network via a data link. Based on the interaction similarity matrix between profile categories, a node communication topology is generated. A target receiving address and data forwarding rules are assigned to each intelligent agent to construct a message routing table. After completing the network weight initialization and routing table entry loading, a differentiated intelligent agent architecture is established.

[0154] Step S3045: Configure the coordination mechanism and the messaging mechanism based on the policy network, the value network and the message routing table.

[0155] In its implementation, the delivery device sets the value network as the global revenue evaluation center to receive state-action pairs from each agent and output a unified benchmark signal. It sets the policy network as a local decision-making execution unit to independently update action distribution parameters based on the benchmark signal. It activates the communication ports in the message routing table and establishes an asynchronous data bus. It sets a fixed communication period so that each agent packages the hidden layer state vector and local reward gradient according to the routing table rules and broadcasts them to the associated nodes. After completing the underlying binding of the gradient synchronization protocol and the state sharing interface, it officially enables the collaboration mechanism and the message mechanism.

[0156] Step S305: Deploy the target game script with inserted promotional information materials to the target reinforcement learning environment, and control each user agent model to output iterative adjustment results based on the state vector and the node feedback information, so as to iteratively adjust the deployment time nodes of the target game script and the promotional information materials matched at each deployment time node.

[0157] In its implementation, the delivery device loads the target game script with embedded promotional materials by starting a virtual sandbox. This triggers each user agent model to read the current game state features, promotional information fatigue features, interaction preference features, and progress interruption tolerance features to form a real-time state vector. Simultaneously, it receives node feedback information transmitted via the communication bus and performs feature fusion. The fused feature tensor is then input into the policy network for forward propagation calculation. The probability distribution of each action space dimension is output, and the target execution command is sampled to determine the target execution command. Based on the target execution command, the node timestamp configuration is updated or the material reference address is replaced. The environmental reward value is calculated and backpropagation is used to update the network weights. The process of state reading, command output, and parameter update is executed cyclically to generate iterative adjustment results.

[0158] This embodiment effectively breaks through the technical limitations of traditional methods that rely on single user tags or fixed scheduling rules. It incorporates differentiated group psychological characteristics, game rhythm protection constraints, and multi-agent collaborative decision-making into a unified computing framework. While ensuring that the core game experience is not disturbed, it maximizes the accurate reach and natural integration of advertising content, provides high-confidence strategy output for game operation, significantly reduces the cost of manual trial and error and system computing power, and improves the long-term health and adaptive evolution capabilities of the game advertising ecosystem.

[0159] Furthermore, this invention also proposes a computer-readable storage medium storing a game promotion information checkpoint delivery program. When the game promotion information checkpoint delivery program is executed by a processor, it implements the steps of the game promotion information checkpoint delivery method described above.

[0160] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0161] The aforementioned computer-readable storage medium may be included in the promotional information card distribution device within the game; or it may exist independently and not be installed in the promotional information card distribution device within the game.

[0162] Furthermore, this invention also proposes a computer program product, including a promotional information card placement program within a game. When the promotional information card placement program within the game is executed by a processor, it implements the steps of the promotional information card placement method within the game as described above.

[0163] The specific implementation of the computer program product of the present invention is basically the same as the various embodiments of the above-mentioned method for placing promotional information cards in the game, and will not be repeated here.

[0164] Reference Figure 6 , Figure 6 This is a structural block diagram of the first embodiment of the in-game promotional information card placement device of the present invention.

[0165] like Figure 6 As shown, the in-game promotional information display device proposed in this embodiment of the invention includes: The time node generation module 10 is used to generate multiple delivery time nodes based on the baseline game node in the target game script and the promotion information features of each promotion information material in the material library to be delivered. The baseline game node is used to characterize the comprehensive game features of each time node in the target game script and the key constraints for delivering promotion information at each time node. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints. The material insertion module 20 is used to insert promotional information materials that match each delivery time node into the target game script; The iterative adjustment module 30 is used to deploy the target game script after inserting promotional information materials to the target reinforcement learning environment, so as to iteratively adjust the deployment time nodes of the target game script and the promotional information materials matched at each deployment time node. The target reinforcement learning environment is configured with multiple user intelligent agent models. The user intelligent agent models are constructed based on the user profile features of the platform to be deployed. The user intelligent agent models are configured to interact with the promotional information materials at each deployment time node during the operation of the target game script. The strategy output module 40 is used to output the current deployment time node of the target game script and the promotional information materials matched with each deployment time node as a deployment strategy in response to the current iteration stopping condition. The iteration stopping condition is that the current iteration number reaches a preset number threshold or the promotional information interaction behavior of the user intelligent agent model meets the preset interaction condition.

[0166] This embodiment deeply couples multi-dimensional game internal constraints with a reinforcement learning mechanism based on user profile simulation to construct a closed-loop automated delivery node optimization mechanism. This effectively uncovers the deep-seated matching logic between complex game process state transitions and multi-modal promotional content features, significantly improving the accuracy of promotional information delivery and user interaction efficiency. Simultaneously, by utilizing the low-cost, high-concurrency iterative trial and error of the intelligent agent model in a simulation computing environment, it greatly reduces the dependence on real online traffic and the high trial and error costs of traditional manual testing. While maintaining a good gaming experience and a balance with the core game rhythm, it achieves a comprehensive improvement in the intelligence level and efficiency of the overall promotional information delivery strategy.

[0167] The in-game promotional information checkpoint delivery device provided in this application, employing the in-game promotional information checkpoint delivery method described in the above embodiments, can solve the technical problem of in-game promotional information checkpoint delivery. Compared with the prior art, the beneficial effects of the in-game promotional information checkpoint delivery device provided in this application are the same as those of the in-game promotional information checkpoint delivery method described in the above embodiments, and other technical features in the in-game promotional information checkpoint delivery device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0168] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solution of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any limitations on this. The above preset thresholds and parameters are merely exemplary descriptions for ease of understanding, and can be dynamically configured according to business scenarios in actual applications, and do not constitute a limitation of the present invention.

[0169] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0170] In addition, for technical details not described in detail in this embodiment, please refer to the in-game promotional information card placement method provided in any embodiment of the present invention, which will not be repeated here.

[0171] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0172] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0174] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for placing promotional information cards within a game, characterized in that, The methods for placing promotional information at designated points within the game include: Multiple delivery time nodes are generated based on the baseline game nodes in the target game script and the promotional information features of each promotional information material in the material library to be delivered. The baseline game nodes are used to characterize the comprehensive game features of each time node in the target game script and the key constraints for delivering promotional information at each time node. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints. Insert promotional materials that match each delivery time point into the target game script; The target game script, after inserting promotional materials, is deployed to the target reinforcement learning environment to iteratively adjust the deployment time points of the target game script and the promotional materials matched at each deployment time point. The target reinforcement learning environment is configured with multiple user agent models, which are constructed based on the user profile features of the platform to be deployed. The user agent models are configured to interact with the promotional materials at each deployment time point during the execution of the target game script. In response to the current iteration stopping condition being met, the current delivery time node of the target game script and the promotional information materials matching each delivery time node are output as the delivery strategy. The iteration stopping condition is that the current iteration number reaches a preset number threshold or the promotional information interaction behavior of the user intelligent agent model meets the preset interaction condition.

2. The method for placing promotional information cards within a game as described in claim 1, characterized in that, The process generates multiple delivery time nodes based on the baseline game nodes in the target game script and the promotional information features of each promotional material in the material library to be delivered, including: Game feature analysis is performed on each time point in the target game script to obtain comprehensive game features for each time point, including game rhythm features and game scene features; A benchmark game node database is constructed based on the comprehensive game features at each time point, and the benchmark game node database stores multiple benchmark game nodes; Feature analysis is performed on each promotional material in the material library to be deployed to obtain the promotional information features of each promotional material. The material library to be deployed includes multiple promotional materials to be deployed, and the promotional materials include incentive behavior promotional materials, interactive behavior promotional materials, and native promotional materials. Multiple delivery time nodes are generated based on the promotional information features and the benchmark game node.

3. The method for placing promotional information cards within a game as described in claim 1, characterized in that, After the target game script containing the inserted promotional materials is deployed to the target reinforcement learning environment, the process further includes: Monitor the promotional information interaction behavior between each user agent model and the promotional materials at each delivery time point during the operation of the target game script in the target reinforcement learning environment; The delivery time points are adjusted based on user behavior monitoring results, and the adjusted delivery time points are re-matched with each promotional material in the material library to be delivered. Remove promotional materials from the target game script and insert promotional materials that match the adjusted delivery time points; The target game script, after inserting promotional materials, is re-deployed into the target reinforcement learning environment. Then, the process of monitoring the interaction between the user agent models and the promotional materials at each deployment time point during the operation of the target game script in the target reinforcement learning environment is returned to iteratively simulate the interaction process between the user and the target game script.

4. The method for placing promotional information cards within a game as described in claim 3, characterized in that, The monitoring of the interaction behavior between the user agent models and promotional materials at each delivery time point during the operation of the target game script in the target reinforcement learning environment includes: Collect node trigger logs, material playback logs, and user operation logs of the target game script in the target reinforcement learning environment to obtain interaction logs; Generate node interaction sequences corresponding to each deployment time point based on interaction logs and the identifiers of each user's intelligent agent. Based on the node interaction sequence, determine the complete display marker, active close marker, passive interruption marker, and interruption recovery marker corresponding to each delivery time node; A set of node behavior features is constructed based on the complete display markers, active shutdown markers, passive interruption markers, and interruption recovery markers corresponding to each delivery time node; Based on the set of node behavior features, the promotional information interaction behavior between each user agent model and the promotional materials at each delivery time node is determined.

5. The method for placing promotional information cards within a game as described in claim 3, characterized in that, The step of adjusting the delivery time based on user behavior monitoring results and re-matching the adjusted delivery time with each promotional material in the material library to be delivered includes: Based on user behavior monitoring results and key constraints, calculate the set of node evaluation parameters corresponding to each deployment time node; Based on the set of node evaluation parameters, the node adjustment priority corresponding to each deployment time node is determined, and a set of candidate adjustment nodes is obtained. Based on the candidate adjustment node set and the benchmark game node database, node location migration processing is performed to obtain the adjusted release time node; Context matching is performed based on the adjusted delivery time points and the promotional information features of each promotional material to obtain the material matching results; Based on the material matching results and the material rotation constraints of the material library to be deployed, the adjusted deployment time nodes are reselected, and the adjusted deployment time nodes are rematched with each promotional information material in the material library to be deployed.

6. The method for placing promotional information cards within a game as described in claim 1, characterized in that, The step of deploying the target game script, after inserting promotional materials, to the target reinforcement learning environment, and iteratively adjusting the deployment time points of the target game script and the promotional materials matched at each deployment time point, includes: Based on the historical game behavior data of users in the target game script and the historical interaction data of users with promotional materials in the platform to be advertised, user stratification and clustering are performed to obtain multiple user profile categories. Based on each user profile category, a corresponding user intelligent agent model is constructed. The state vector of the user intelligent agent model includes game state features, promotional information fatigue features, interaction preference features, and progress interruption tolerance features. A target reinforcement learning environment is constructed based on the delivery time node, promotion information characteristics and key constraints. The action space of the target reinforcement learning environment includes node retention action, node migration action and material replacement action. Based on the target reinforcement learning environment configuration, a collaborative mechanism and a messaging mechanism are configured. The collaborative mechanism is used to configure a policy network and a value network for each user agent model. The messaging mechanism is used to enable user agent models corresponding to different user profile categories to pass node feedback information in the same iteration round. The target game script, after inserting promotional materials, is deployed to the target reinforcement learning environment. Each user agent model is controlled to output iterative adjustment results based on the state vector and the node feedback information, so as to iteratively adjust the deployment time nodes of the target game script and the promotional materials matched at each deployment time node.

7. The method for placing promotional information cards within a game as described in claim 6, characterized in that, The construction of corresponding user intelligent agent models based on each user profile category includes: Obtain the historical game behavior sequence and historical promotional information behavior sequence corresponding to each user profile category; Based on the historical game behavior sequence, game state preference parameters corresponding to each user profile category are extracted; Based on the historical promotional information behavior sequence, extract the promotional information fatigue parameter and interaction preference parameter corresponding to each user profile category; Based on the historical game behavior sequence and the historical promotion information behavior sequence, extract the progress interruption tolerance parameter corresponding to each user profile category; A corresponding user agent model is constructed based on the game state preference parameters, the promotion information fatigue parameters, the interaction preference parameters, and the progress interruption tolerance parameters.

8. The method for placing promotional information cards in a game as described in claim 6, characterized in that, The configuration of the collaboration mechanism and messaging mechanism based on the target reinforcement learning environment includes: The first reward parameter is determined based on the behaviors of watching the entire promotional information, skipping the promotional information, and closing the promotional information. The second reward parameter is determined based on node interruption behavior, node recovery behavior, and repeated exposure behavior. A reward evaluation model is constructed based on the first reward parameter and the second reward parameter; Based on the reward evaluation model and each user profile category, configure the policy network, value network and message routing table for each user agent model; The coordination mechanism and message mechanism are configured based on the policy network, the value network, and the message routing table.

9. A device for distributing promotional information cards within a game, characterized in that, The device employs the in-game promotional information checkpoint delivery method according to any one of claims 1 to 8, and the in-game promotional information checkpoint delivery device comprises: The time node generation module is used to generate multiple delivery time nodes based on the baseline game node in the target game script and the promotion information features of each promotion information material in the material library to be delivered. The baseline game node is used to characterize the comprehensive game features of each time node in the target game script and the key constraints for delivering promotion information at each time node. The key constraints include user experience constraints, game rhythm constraints, and game scene constraints. The material insertion module is used to insert promotional information materials that match each delivery time point into the target game script; The iterative adjustment module is used to deploy the target game script after inserting promotional information materials to the target reinforcement learning environment, so as to iteratively adjust the deployment time nodes of the target game script and the promotional information materials matched at each deployment time node. The target reinforcement learning environment is configured with multiple user intelligent agent models. The user intelligent agent models are constructed based on the user profile features of the platform to be deployed. The user intelligent agent models are configured to interact with the promotional information materials at each deployment time node during the operation of the target game script. The strategy output module is used to output the current deployment time node of the target game script and the promotional information materials matched with each deployment time node as a deployment strategy in response to the current iteration stopping condition. The iteration stopping condition is that the current iteration number reaches a preset number threshold or the promotional information interaction behavior of the user intelligent agent model meets the preset interaction condition.

10. A device for placing promotional information cards within a game, characterized in that, The in-game promotional information checkpoint delivery device includes: a memory, a processor, and an in-game promotional information checkpoint delivery program stored on the memory. The processor is used to run the in-game promotional information checkpoint delivery program, and the in-game promotional information checkpoint delivery program is configured to implement the in-game promotional information checkpoint delivery method as described in any one of claims 1 to 8.

Images