AI intelligent advertisement automatic putting method and system

Abstract

The invention relates to the technical field of advertisement putting, in particular to an AI intelligent automatic advertisement putting method and system, and the method comprises the steps: collecting the behavior characteristics of a current use session of a user, and generating a scene fingerprint representing the current use state; matching corresponding advertisement types from pre-established advertisement demand portraits based on the scene fingerprints, and determining an advertisement type set allowed to participate in putting; for the advertisements in the advertisement type set, acquiring corresponding conversion process information, and based on comparison between the continuous interaction capability of the currently used session and the conversion process information, screening out the advertisements with conversion completion conditions in the currently used session; the interaction type and the occurrence sequence of the put advertisements are recorded in the use session to form a session putting track, and the session evolution state is continuously generated in combination with the user behavior to control the putting participation range of the advertisements; after advertisement putting is completed, user feedback information is collected and stored according to the use state identification.

Description

Technical Field

[0001] This invention relates to the field of advertising delivery technology, and in particular to an AI-powered intelligent automatic advertising delivery method and system. Background Technology

[0002] With the widespread adoption of mobile internet and smart devices, advertising delivery has gradually evolved from traditional manual configuration to algorithm-based automated delivery. Existing automated advertising delivery methods typically use users' historical browsing behavior, interests, or long-term user profiles to filter, sort, and deliver ads. Some solutions also incorporate users' geographic location or device information to target ads by region or audience, thus achieving a degree of automation and scalability in advertising delivery.

[0003] In practical applications, users' usage status on the same terminal has obvious dynamic changes. For example, users may be commuting and browsing the application briefly, or they may be staying in a stable state and using the application for a long time. Existing advertising methods mainly rely on static user profiles or coarse-grained targeting information, which makes it difficult to accurately depict the real-time usage status of the user's current session. This makes it impossible for advertising decisions to differentiate between different usage statuses, thus affecting the adaptability of advertising. Summary of the Invention

[0004] To address the above shortcomings, this invention provides an AI-powered intelligent automatic advertising delivery method and system, aiming to improve the problem that advertising delivery decisions cannot differentiate between different usage states, thus affecting the adaptability of advertising delivery.

[0005] In a first aspect, the present invention provides the following technical solution: an AI-powered intelligent advertising automatic delivery method, comprising the following steps: S1. Collect the behavioral characteristics of the user's current session and generate a scenario fingerprint of the current usage state; S2. Based on the scenario fingerprint, match the corresponding ad type from the pre-established ad demand profile to determine the set of ad types allowed to participate in the current ad request processing; S3. For the advertisements in the set of advertisement types, obtain the corresponding conversion process information, compare the continuous interaction capability of the current usage session with the conversion process information, and filter out the advertisements that have the conditions to complete the conversion in the current usage session. S4. Within the current usage session, the interaction types and order of the ads that have been delivered to the filtered ad records are used to form a session delivery trajectory. The session delivery trajectory is then matched with a predefined conversion guidance sequence to determine the range of ad types for the next ad delivery within the ad type set. S5. Based on the session delivery trajectory and the continuity of user behavior in the current session, generate a session evolution state, and within the range of ad types, only allow ads that meet the delivery unlocking conditions corresponding to the session evolution state to participate in delivery; S6. After the advertisement is delivered, collect user feedback information on the delivered advertisement, record the usage status identifier corresponding to the feedback, and store the advertisement feedback information according to the usage status identifier.

[0006] By adopting the above technical solution, the collection and analysis of user's current usage session behavior characteristics are realized, and the advertising delivery process is controlled in stages based on the generated scene fingerprint. This enables the advertising delivery decision-making process to be combined with the user's current usage session status, thereby enabling the advertising delivery decision to be differentiated for different usage states. This improves the problem of insufficient ad delivery adaptability caused by the difficulty in characterizing the user's real-time usage status in existing advertising delivery methods.

[0007] Preferably, in S1, the collection of behavioral characteristics of the user's current usage session includes: After a user session begins, information on the user terminal’s location changes, session duration, and page dwell and page switching behavior is collected. The collected behavioral information is processed into a time series to form a set of session behavior features that reflect the continuity and stability of user operations within a session.

[0008] Preferably, in S1, generating the scenario fingerprint of the current usage state includes: The set of conversation behavior features is constructed into time series features and then input into the scene recognition model. The scene recognition model is built on a neural network for processing temporal features, the neural network including at least one of a recurrent neural network, a long short-term memory network, or a sequence modeling network based on an attention mechanism; The neural network is used to learn the sequence features of the session behavior features, and the output is a state vector representing the current state of the user's session. Based on the state vector, the usage scenario identifier corresponding to the user's current usage session is determined by the state classification algorithm, and the usage scenario identifier is used as the scenario fingerprint of the current usage state.

[0009] Preferably, in S2, matching the corresponding ad type from the pre-established ad demand profile includes: Obtain the advertising demand profile corresponding to the candidate advertisement. The advertising demand profile is used to describe the demand attributes of the advertisement in terms of interaction complexity, continuous operation requirements and exposure dependence characteristics. The scene fingerprint and the advertising demand profile are matched to determine the degree to which the scene fingerprint meets the advertising demand profile. Based on the adaptability analysis results, advertisements that meet the preset adaptability conditions are classified into the advertisement type corresponding to the current usage state, thereby determining the set of advertisement types that are allowed to participate in the current advertisement request processing.

[0010] Preferably, in S3, obtaining the corresponding conversion process information includes: For each candidate ad, a conversion process description corresponding to its conversion behavior is pre-established. The conversion process description is used to represent the sequence of operation steps required to complete the conversion of the ad. Associate corresponding sequential operation conditions with at least some of the operation steps, wherein the sequential operation conditions are used to describe the minimum sequential operation requirements required to complete the corresponding operation steps; The sequence of operation steps and the corresponding continuous operation conditions are used as the conversion process information for the advertisement.

[0011] Preferably, in S3, the comparison between the continuous interaction capability and conversion process information based on the current usage session includes: Based on the user's page dwell behavior, operation intervals, and operation interruptions in the current usage session, a continuous interaction capability description is generated to characterize the user's ability to perform continuous operations within the usage session. Following the order of the operation steps in the conversion process information, the descriptions of continuous interaction capabilities are matched with the continuous operation conditions of the corresponding operation steps in sequence. An ad is deemed to have the conditions to complete a conversion in the current usage session only if the description of continuous interaction capabilities can cover the continuous operation conditions corresponding to the operation steps.

[0012] Preferably, in S4, the formation of the session delivery trajectory includes: Within the user's current session, the interaction types of the ads that have been delivered are recorded, and the interaction types are used to characterize how the user interacts with the ads; The interaction types are sequentially processed according to the order in which the advertisements are delivered in the usage session, forming a session interaction sequence that reflects the process of advertisement delivery within the usage session. The conversation interaction sequence is used as the conversation delivery trajectory.

[0013] Preferably, in S5, generating the session evolution state based on the session delivery trajectory and the continuity of user behavior in the current usage session includes: Based on the session delivery trajectory, obtain the interaction types and delivery order information of the ads that have been delivered in the current session; Within a usage session, user page dwell time, continuous operation duration, and operation interruption are collected to characterize the continuity of user behavior within the usage session. By jointly analyzing the session delivery trajectory and behavioral continuity, the current session stage is determined, and the determined session stage is taken as the session evolution state.

[0014] Preferably, in S5, the release unlocking conditions include: Pre-set corresponding launch unlocking conditions for ads of different ad types or conversion process complexity levels. The launch unlocking conditions are used to limit the session evolution state that the ad needs to meet to participate in the launch. After generating the session evolution state, the session evolution state is matched with the delivery unlocking conditions to determine whether the current session meets the delivery unlocking conditions of the corresponding advertisement. An ad is allowed to participate in the delivery process within the current session only if the session evolution state meets the corresponding ad delivery unlocking conditions.

[0015] Secondly, the present invention provides the following technical solution: an AI-powered intelligent advertising automatic delivery system, the system comprising the following modules: The conversation behavior analysis module is used to collect the behavioral characteristics of the user's current usage session and generate a scenario fingerprint that represents the current usage state; The ad type matching module is used to match the corresponding ad type from the pre-established ad demand profile based on the scenario fingerprint, and determine the set of ad types that are allowed to participate in the current ad request processing. The conversion completion determination module is used to obtain the conversion process information corresponding to the advertisements in the set of advertisement types, and compare the conversion process information with the continuous interaction capability of the current usage session to filter out advertisements that have the conditions to complete the conversion in the current usage session. The session delivery trajectory management module is used to record the interaction types and occurrence order of the delivered advertisements within the current usage session, forming a session delivery trajectory, and matching the session delivery trajectory with a predefined conversion guidance sequence to determine the range of advertisement types corresponding to the next advertisement delivery within the set of advertisement types; The delivery unlocking control module is used to generate a session evolution state based on the session delivery trajectory and the continuity of user behavior in the current usage session, and within the range of the ad types, only ads that meet the delivery unlocking conditions corresponding to the session evolution state are allowed to participate in delivery; The feedback information management module is used to collect user feedback information on the advertised ads after the ad delivery is completed, record the usage status identifier corresponding to the feedback, and store the ad feedback information according to the usage status identifier.

[0016] The present invention has the following beneficial effects: 1. In the practical application scenarios of mobile terminal advertising, users may be in different usage states such as commuting, short-term browsing, or stable stay. By generating scene fingerprints based on usage session behavior characteristics and using these scene fingerprints as input for advertising placement decisions, the advertising placement process can incorporate a structured description of the user's current usage state, thereby differentiating the advertising placement logic between commuting or short-term usage scenarios and stable stay or long-term usage scenarios.

[0017] 2. In this invention, during commutes or in fragmented usage scenarios, user operations are easily interrupted, making it difficult to complete the advertising conversion process involving multiple steps. By introducing conversion process information and combining it with the continuous interaction capabilities generated within the usage session, the system can determine whether an advertisement has the conditions to complete the conversion in the current usage session. This allows the advertising screening process to consider the objective constraints of the continuity of operations in the current session, thereby preventing advertisements that cannot complete the conversion in the current usage scenario from entering the delivery process.

[0018] 3. In this invention, in actual usage scenarios where users stay on the same application for a long time, advertising conversion usually requires multiple exposures to gradually guide the conversion, rather than a single exposure. By recording the interaction types and their order of occurrence of the ads within the same usage session to form a session delivery trajectory, and matching the session delivery trajectory with a predefined conversion guidance sequence, the advertising delivery process can maintain logical consistency between the delivery order and the guidance path at the session level. Attached Figure Description

[0019] Figure 1 This is a flowchart of the AI-powered intelligent advertising automatic delivery method proposed in this invention; Figure 2 This is an architecture diagram of the AI-powered intelligent advertising automatic delivery system proposed in this invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: In the first embodiment of the present invention, the present invention provides an AI-powered intelligent advertising automatic delivery method and system, such as... Figure 1 As shown, it includes the following steps: S1. Collect the behavioral characteristics of the user's current session and generate a scenario fingerprint of the current usage state; Furthermore, in S1, the behavioral characteristics of the user's current session are collected, including: After a user session begins, information on the user terminal’s location changes, session duration, and page dwell and page switching behavior is collected. The collected behavioral information is processed into a time series to form a set of session behavior features that reflect the continuity and stability of user operations within a session.

[0022] Furthermore, in S1, generating the scenario fingerprint of the current usage state includes: The set of conversation behavior features is constructed into time series features and then input into the scene recognition model. The scene recognition model is built on a neural network for processing temporal features, including at least one of recurrent neural networks, long short-term memory networks, or sequence modeling networks based on attention mechanisms; The neural network is used to learn the sequence features of the session behavior features, and the output is a state vector representing the current state of the user's session. Based on the state vector, the usage scenario identifier corresponding to the user's current usage session is determined by the state classification algorithm, and the usage scenario identifier is used as the scenario fingerprint of the current usage state.

[0023] Specifically, after the system detects that a user has entered a new usage session through a mobile terminal, it initiates a behavioral feature collection process for that usage session to obtain basic behavioral data that characterizes the user's current usage state. During the duration of the usage session, the system collects multi-dimensional behavioral information related to the current usage session from the terminal side. The aforementioned behavioral information includes at least the changes in the terminal's location information during the usage session, the duration of the usage session, and the user's page dwell and page switching behaviors in the application interface. Among them, location information changes are used to reflect the spatial movement characteristics of the terminal within the usage session, usage session duration information is used to reflect the overall dwell time of the user in the current usage session, and page dwell behavior and page switching behavior are used to reflect the user's operation rhythm and interaction continuity within the application. The system uses the session start time as the time base, performs time alignment processing on the collected behavioral information, and organizes various types of behavioral information into a time series format according to time order, thereby constructing a session behavior feature set corresponding to the current session; the session behavior feature set can be represented as: ; in, Represents a set of conversational behavior features. Indicates the number of types of behavioral characteristics. Indicates the first Class behavioral characteristics in time The feature value corresponding to time step. This indicates the use of time indexes within the session; through the above method, the system obtains a set of time-seriesd session behavior features that reflect the continuity and stability of user operations within the session. When generating the scene fingerprint of the current usage state, the system constructs the session behavior feature set into a time-series feature input form and inputs the time-series features into the scene recognition model. The scene recognition model is built on a sequence neural network, which is used to perform temporal modeling on the input time-series features to characterize the changing relationship of session behavior features in the time dimension. In one possible implementation, the sequence neural network can adopt any of the following structures: recurrent neural network structure, long short-term memory network structure, or attention-based sequence modeling network structure. The specific network structure can be configured according to the actual application requirements. During operation, the sequence neural network processes the time-series features sequentially, and models the usage behavior patterns through recursive updates of its internal states. The state update process can be represented as follows: ; in, Represents the time sequence of neural networks The hidden state vector corresponding to time step 1. Indicates time The hidden state vector corresponding to time step 1. Indicates time The conversation behavior feature vector input at any given time. This represents a nonlinear mapping function determined by the structure of a sequence neural network; After processing the features of the entire time series, the system generates a state vector to represent the current usage session state based on the final or aggregated implicit state. The system performs state classification processing based on state vectors. This state classification process uses a state classification algorithm to determine the state vectors. The state classification process can be represented as: ; in, This indicates the identified use case category identifier. This represents a predefined index of use case categories. In the state vector Under the condition of belonging to the first The probability of a class's use case; The system will use the determined usage scenario category as the usage scenario identifier of the current usage session, and store the usage scenario identifier as the scenario fingerprint of the current usage state for subsequent ad delivery related processes.

[0024] S2. Based on the scenario fingerprint, match the corresponding ad type from the pre-established ad demand profile to determine the set of ad types allowed to participate in the current ad request processing; Furthermore, in S2, matching the corresponding ad type from the pre-established ad demand profile includes: Obtain the ad demand profile corresponding to the candidate ad. The ad demand profile is used to describe the demand attributes of the ad in terms of interaction complexity, continuous operation requirements and exposure dependence characteristics. The scene fingerprint and the advertising demand profile are matched to determine the degree to which the scene fingerprint meets the advertising demand profile. Based on the adaptability analysis results, advertisements that meet the preset adaptability conditions are classified into the advertisement type corresponding to the current usage state, thereby determining the set of advertisement types that are allowed to participate in the current advertisement request processing.

[0025] Specifically, after generating the scenario fingerprint for the current usage session, the system enters the ad type matching process to determine the set of ad types that are allowed to participate in the current ad request processing. The system first obtains the set of candidate ads corresponding to the current ad request, and then reads the corresponding ad demand profile for each candidate ad. The ad demand profile is pre-established during the ad creation or deployment configuration stage and is used to describe the demand attributes of the ad at the interaction and exposure levels. The aforementioned demand attributes include at least the interaction complexity attribute, the continuous operation requirement attribute, and the exposure dependency feature. Among them, the interaction complexity attribute describes the complexity of the interaction steps required for the user to complete the ad, the continuous operation requirement attribute describes the requirements for continuous operation time or continuous interaction behavior during the ad interaction process, and the exposure dependency feature describes the dependence of the ad on the number of exposures or the exposure sequence before generating a user response. The ad demand profile can be represented in the form of a multi-dimensional demand vector, which is used to uniformly characterize the differences of different ads in the dimension of demand attributes. When performing adaptability analysis, the system jointly analyzes the scenario fingerprint corresponding to the current usage session and the advertising demand profile. The scenario fingerprint, after generation, is also represented as a multi-dimensional feature vector to reflect the state characteristics of the current usage session in terms of stability, continuity, and interaction capacity. The scenario fingerprint can be represented as a vector. Advertising demand profiles can be represented as vectors. ,in: ; in, This represents the scenario fingerprint vector corresponding to the current usage session. This represents the ad demand profile vector corresponding to the candidate ad. Indicates the number of requirement attribute dimensions. Indicates the scene fingerprint in the first... State values ​​on each requirement attribute dimension This indicates that the advertising demand profile is in the first place. Demand values ​​on each demand attribute dimension; The system performs adaptation analysis based on scene fingerprint vectors and advertising demand profile vectors to determine whether the current usage session meets the conditions for ad delivery on the corresponding demand attribute dimension. Fit analysis is performed using a dimension-by-dimensional comparison method, and its fit determination can be expressed as: ; in, Indicates the first The adaptation results on each demand attribute dimension are recorded as follows: when the state value of the scene fingerprint on this dimension meets the demand value in the advertising demand profile, it is recorded as adapted; otherwise, it is recorded as unsuitable. The system further determines the overall adaptation status of candidate ads in the current usage session based on the adaptation results of each demand attribute dimension. The overall adaptation status can be calculated by summarizing the adaptation results of each dimension, and the determination method can be expressed as follows: ; in, This indicates the overall fit of the candidate ad in the current usage session, if and only if all requirement attribute dimensions meet the fit conditions. This indicates that the advertisement meets the preset adaptation conditions; The system will classify advertisements whose overall adaptation results meet the preset adaptation conditions into the advertisement type corresponding to the current usage status, and determine the set of advertisements classified into the same advertisement type as the set of advertisement types that are allowed to participate in the current advertisement request processing; In this way, the system can achieve dynamic matching of ad types based on the usage session scenario state without changing the original definition of ad requirements attributes, thus providing input basis for subsequent ad screening and delivery processes.

[0026] S3. For ads in the ad type set, obtain the corresponding conversion process information, compare the continuous interaction capability of the current usage session with the conversion process information, and filter out ads that have the conditions to complete the conversion in the current usage session. Furthermore, in S3, obtaining the corresponding conversion process information includes: For each candidate ad, a conversion process description corresponding to its conversion behavior is pre-established. The conversion process description is used to indicate the sequence of operation steps required to complete the conversion of the ad. Associate corresponding sequential operation conditions with at least some of the operation steps. The sequential operation conditions are used to describe the minimum sequential operation requirements required to complete the corresponding operation steps. The sequence of operation steps and the corresponding continuous operation conditions are used as the conversion process information for the advertisement.

[0027] Furthermore, in S3, the comparison between the continuous interaction capabilities and conversion process information based on the current usage session includes: Based on the user's page dwell behavior, operation intervals, and operation interruptions in the current usage session, a continuous interaction capability description is generated to characterize the user's ability to perform continuous operations within the usage session. Following the order of the operation steps in the conversion process information, the descriptions of continuous interaction capabilities are matched with the continuous operation conditions of the corresponding operation steps in sequence. An ad is deemed to have the conditions to complete a conversion in the current usage session only if the description of continuous interaction capabilities can cover the continuous operation conditions corresponding to the operation steps.

[0028] Specifically, after determining the set of ad types, the system performs a conversion feasibility assessment on the candidate ads that have entered the ad type set to determine whether the candidate ads meet the basic conditions for conversion in the current usage session. The system first obtains the conversion process information corresponding to each candidate advertisement. This conversion process information is pre-established during the advertisement configuration phase and corresponds to the specific conversion behavior of the advertisement. The conversion process information describes the user operation path required to complete the advertisement conversion, including the sequence of operation steps the user needs to complete and the continuous operation conditions associated with some of these steps. The conversion process information is represented as an ordered set of steps, used to clarify the execution order of user operations during the advertisement conversion process. The conversion process information can be represented as: ; in, This indicates the conversion process information corresponding to the candidate ads. This indicates the number of steps required to complete an ad conversion. Indicates the first One operation step; The system associates corresponding continuous operation conditions with at least a portion of the operation steps in the set of operation steps. These conditions describe the minimum continuous operation requirements that must be met to complete the operation step. The continuous operation conditions are defined in the form of minimum continuous operation duration and can be expressed as: ; in, This represents the set of consecutive operation conditions corresponding to the transformation process information. Indicates completion of the first The minimum continuous operation time required for each operation step; when a certain operation step does not have a continuous operation condition set, the corresponding... Take a zero value; through the above method, the system constructs the conversion process information of candidate advertisements by combining the sequence of operation steps and the continuous operation conditions; When generating continuous interaction capabilities, the system quantifies the user's continuous operation capability within the current usage session based on behavioral data. The system collects page dwell time, operation intervals, and operation interruptions from the usage session and performs time-series analysis on these behaviors to identify the time interval during which the user can continuously perform interactive operations without interruption. Continuous interaction capability is represented by the length of continuous operable time. Continuous interaction capability can be represented as: ; in, Indicates the duration of continuous interaction within the current session. This indicates the length of the continuous operation time interval during which no operation interruption occurs in the usage session; After generating continuous interaction capabilities, the system will perform a process-level comparison between the continuous interaction capabilities and the continuous operation conditions in the conversion process information. The system calculates the margin relationship between the continuous operation conditions and continuous interaction capabilities corresponding to each operation step according to the sequence of operation steps in the conversion process information; for the first step... For each operation step, its continuous operation margin can be expressed as: ; in, Indicates continuous interaction capability relative to the first The system further summarizes and analyzes the continuous operation margins corresponding to all operation steps in the conversion process to identify bottleneck steps that restrict the execution of the current conversion process; the overall continuous operation margin of the conversion process can be expressed as: ; in, This indicates the minimum continuous operation margin of the conversion process in the current usage session. The system determines the feasibility of conversion based on the minimum continuous operation margin of the conversion process. When the minimum continuous operation margin meets the preset judgment conditions, it is determined that the candidate advertisement has the conditions to complete the conversion in the current usage session, and the advertisement is retained in the subsequent advertisement delivery process.

[0029] S4. Within the current usage session, the interaction types and order of the ads that have been delivered to the filtered ad records are used to form a session delivery trajectory. The session delivery trajectory is then matched with a predefined conversion guidance sequence to determine the range of ad types for the next ad delivery within the ad type set. Furthermore, in S4, the formation of a session delivery trajectory includes: Within the user's current session, the interaction types of the ads that have been delivered are recorded. The interaction types are used to characterize how the user interacts with the ads. The interaction types are sequentially processed according to the order in which the advertisements are delivered in the usage session, forming a session interaction sequence that reflects the process of advertisement delivery within the usage session. Use the conversation interaction sequence as the conversation delivery trajectory.

[0030] Specifically, after filtering the ads that can be placed in the current usage session, the system enters the ad placement order control process based on session history to continuously manage the ad placement behavior within the same usage session. During the duration of a user's current session, the system records each advertisement that has been delivered and simultaneously collects the user's interactive behavior in response to the delivered advertisements, in order to characterize the delivery and interaction process of advertisements within the user session. After an advertisement is delivered, the system records the corresponding interaction type. The interaction type is used to characterize the user's interaction after receiving the advertisement. The interaction type can include behaviors such as clicking, browsing, jumping, and ignoring. Interaction types are predefined during the ad configuration phase and correspond to the ad's interaction design method, used to uniformly identify different ad interaction behaviors at the session level; Within a usage session, the system sequentially processes recorded interaction types according to the order in which ads are delivered, thus forming a session interaction sequence that reflects the ad delivery process within the usage session; the session interaction sequence can be represented as: ; in, This indicates the session interaction sequence corresponding to the currently used session. This indicates the number of ads that have been delivered within the current session. Indicates the first The interaction type identifier corresponding to each ad placement; The system maintains the session interaction sequence as the session delivery trajectory of the current session and updates it dynamically throughout the session. When making subsequent ad delivery decisions, the system obtains a predefined conversion guidance sequence to describe the order of ad interaction types from ad information presentation to achieving the target conversion. The conversion guidance sequence is pre-established during the ad strategy configuration phase to characterize the ad's guidance path within the session, and its form can also be represented as an ordered sequence. The conversion guidance sequence can be represented as: ; in, This indicates the conversion guide sequence. This indicates the number of interaction stages included in the conversion guidance path. Indicates the first The types of ad interactions corresponding to each onboarding phase; When determining the scope of the next ad delivery, the system performs sequence matching between the session delivery trajectory corresponding to the current user session and the conversion guidance sequence. Based on the matching position of the session delivery trajectory in the conversion guidance sequence, the system determines the completed guidance phase of the current user session. The system determines the current guidance progress by finding the longest prefix match length of the session interaction sequence in the conversion guidance sequence. The matching process can be represented as follows: ; in, This indicates the number of completed bootstrapping phases in the current session's conversion bootstrapping sequence; based on the matching results, the system will assign the current session to the next bootstrapping phase in the conversion bootstrapping sequence. The type of ad interaction corresponding to each stage is used to determine the range of ad types for the next ad campaign. Within the set of ad types, the system performs the next ad delivery selection process based on the determined range of ad types, thereby maintaining logical consistency between the ad delivery order and the conversion guidance path within the same user session.

[0031] S5. Generate a session evolution state based on the session delivery trajectory and the continuity of user behavior in the current session, and within the scope of ad types, only allow ads that meet the delivery unlocking conditions corresponding to the session evolution state to participate in delivery; Furthermore, in S5, the generation of session evolution states based on session delivery trajectories and the continuity of user behavior in the current session includes: Based on the session delivery trajectory, obtain the interaction types and delivery order information of the ads that have been delivered in the current session; Within a usage session, user page dwell time, continuous operation duration, and operation interruption are collected to characterize the continuity of user behavior within the usage session. By jointly analyzing the session delivery trajectory and behavioral continuity, the current session stage is determined, and the determined session stage is taken as the session evolution state.

[0032] Furthermore, in S5, the unlock conditions include: Pre-set corresponding launch unlock conditions for ads of different ad types or conversion process complexity levels. Launch unlock conditions are used to limit the session evolution state that the ad needs to meet to participate in the launch. After generating the session evolution state, the session evolution state is matched with the delivery unlocking conditions to determine whether the current session meets the delivery unlocking conditions of the corresponding advertisement. An ad is allowed to participate in the delivery process within the current session only if the session evolution state meets the corresponding ad delivery unlocking conditions.

[0033] Specifically, after completing the formation of the session delivery trajectory and obtaining the range of ad types corresponding to the next ad delivery, the system enters the session evolution state generation and delivery unlocking control process to conduct phased access control on the ads that can participate in the delivery within the current session. The system first obtains the interaction types and their order of advertising in the current session based on the session delivery trajectory, and uses the order of advertising as one of the behavioral evidences of the in-session guidance progress; in one possible implementation, the session delivery trajectory is maintained in the form of a session interaction sequence, which contains interaction type identifiers arranged in the order of delivery, and is used to characterize the cumulative process of advertising interactions in the current session. The system synchronously collects user page dwell behavior, continuous operation duration, and operation interruption events within the current usage session to form a set of continuity features characterizing the continuity of user behavior within the usage session. Among them, page dwell behavior reflects the user's continuous dwell time on a single page, continuous operation duration reflects the duration of continuous interaction without significant operation interruptions, and operation interruption events reflect the break in the interaction chain within the usage session. The system performs time alignment and windowing processing on the above continuity feature set to form a continuity metric that can be used for joint analysis at the session scale. In one possible implementation, the system segments the user session based on a preset time window, and within each time window, it counts the page dwell time, operation interval, and number of interruption events to form a window-level continuity vector. The system further calculates a session-level continuity metric based on the window-level continuity vector to support session phase determination. The session-level continuity metric can be expressed as: ; in, This represents a session-level continuity metric. This indicates the cumulative duration during which the continuous operation criteria are met within the current session. This indicates the duration of the current session from its start to the current moment. The continuous operation determination criteria can be configured based on the operation interval threshold or the interruption event threshold. The operation interval threshold is used to limit the maximum allowable interval between two adjacent operations, and the interruption event threshold is used to limit the number of interruptions allowed within the statistical window. After obtaining session delivery trajectory information and session-level continuity measurement, the system performs joint analysis of session delivery trajectory and behavior continuity to determine the session stage of the current usage session and takes that session stage as the session evolution state. The determination of a session phase can employ a rule-based phase discrimination algorithm. This algorithm uses pre-defined phase division rules to jointly map the guidance progress corresponding to the session delivery trajectory with the usage stability corresponding to the session-level continuity metric. The session phase determination can be expressed as: ; in, This indicates the session stage identifier corresponding to the session evolution state. This indicates the number of ads served or the number of interaction types recorded within the current session. This represents a session-level continuity metric. This represents the phase partitioning mapping function, which is determined by a pre-configured set of phase rules. The set of phase rules can include quantity range constraints and continuity range constraints set for different session phases. Quantity range constraints are used to limit the cumulative progress range of the session delivery trajectory, and continuity range constraints are used to limit the value range of the session-level continuity metric. After generating the session evolution state, the system matches the delivery unlock conditions to determine whether different ads meet the admission requirements for participating in the current usage session. The conditions for unlocking ad placement are set in advance during the ad configuration stage, and corresponding session stage requirements are established for different ad types or conversion process complexity levels to limit the session evolution state that the ad must meet to participate in ad placement. In one possible implementation, the conditions for unlocking the delivery are stored in the form of a mapping table or a set of rules, which associates the ad type identifier or complexity level identifier with the set of session stages that are allowed to be delivered. When performing unlock matching, the system reads the ad type identifier or conversion process complexity level identifier corresponding to the candidate ad within the ad type range, and obtains the corresponding unlock stage requirements accordingly. The unlock stage requirements can be represented as the minimum allowed session stage threshold, or as the set of allowed session stages. When using the minimum session stage threshold for judgment, unlock matching can be represented as follows: ; in, Indicates candidate ads The unlocking result, This indicates the session stage identifier corresponding to the current session's session evolution state. Indicates candidate ads The corresponding minimum session stage threshold; The system allows a candidate ad to participate in the delivery process within the current usage session only if the unlocking result of the candidate ad meets the preset conditions, and excludes candidate ads that do not meet the unlocking conditions from the delivery candidate set of the current usage session. After completing the delivery unlock control, the system outputs the candidate ads that meet the unlock conditions as the candidate set allowed to participate in delivery within the current usage session, and performs subsequent delivery selection processing within the range of ad types to maintain logical consistency between the session evolution state and the ad admission conditions.

[0034] S6. After the advertisement is delivered, collect user feedback information on the delivered advertisement, record the usage status identifier corresponding to the feedback, and store the advertisement feedback information according to the usage status identifier.

[0035] Specifically, after the system completes an ad delivery and outputs the ad content to the user terminal, the system enters the ad feedback collection and status association storage process to record the user's interactive feedback on the delivered ads in a structured manner and establish a correspondence between the feedback and the usage status. The system generates a unique event identifier for each ad impression during ad delivery and associates the event identifier with the ad identifier, delivery time information, and current session identifier so that the source of feedback can be traced based on the event identifier in the subsequent feedback collection stage. Event identifiers can be generated by the system according to preset encoding rules. The encoded content can include session identifier fragments, timestamp fragments, and ad identifier fragments. The specific encoding method can be configured according to the project implementation needs. The system listens for feedback trigger events related to the advertisement on the user terminal side or the delivery link side, and collects the corresponding feedback information after detecting the feedback trigger event. The feedback information may include one or more of the following interactive events: click, pause, swipe, close, jump, form submission, order placement, etc. The set of feedback information types can be predefined on the advertising platform or application side to ensure the comparability of feedback data for different advertisements and different terminal environments. When collecting feedback information, the system binds the feedback information with the event identifier to form a feedback record that can be stored and processed later. Simultaneously, the system obtains the usage status identifier corresponding to the time the feedback occurs and writes this identifier into the status field of the feedback record. The usage status identifier is generated by the usage session analysis process and maintained within the usage session. It characterizes the current usage session status when feedback occurs. The usage status identifier can be a scene category identifier corresponding to a scene fingerprint, or a session stage identifier corresponding to a session evolution state. The specific identifier format can be selected based on the system configuration. The system uses the above method to distinguish the source of the same advertisement feedback under different usage states, so as to avoid the loss of context information in the feedback record during the storage stage; The system writes feedback records containing usage status identifiers into the feedback storage unit and categorizes and stores the feedback records according to the usage status identifiers. In one possible implementation, the feedback storage unit includes multiple logical partitions, each corresponding to a usage status identifier value or a range of usage status identifier values. The system selects the corresponding logical partition for writing based on the usage status identifier in the feedback record. In another possible implementation, the feedback storage unit uses the same storage table structure for writing, and uses the usage status identifier field as the partition key or index key in the storage table to support subsequent retrieval and aggregation processing based on the usage status identifier. To ensure consistent storage and retrieval of feedback records, the system can represent feedback records as structured data objects. These structured data objects should include at least a delivery event identifier, an advertisement identifier, a feedback type identifier, a feedback occurrence time, and a usage status identifier. In one possible implementation, a feedback record can be represented as: ; in, This represents a feedback record. Indicates the event identifier. Indicates advertising logo, Indicates the feedback type identifier. Indicates the time when the feedback occurred. This indicates the usage status identifier corresponding to when the feedback was generated; the system uses the usage status identifier when storing feedback records. As a classification dimension, the feedback record is written to satisfy... The corresponding storage area; After completing the categorized storage, the system uses the feedback records as a usable input data source for subsequent ad delivery processing. Subsequent processing can read the corresponding set of feedback records according to the usage status identifier when needed, to support statistical analysis, rule updates, or model training related to usage status. When the system reads the feedback set according to the usage status identifier, it can be represented as follows: ; in, Indicates the use of status flags. A set of feedback records, where each element in the set satisfies the condition that the status identifier field equals [the specified value]. Feedback records; Through the above methods, the system can collect advertising feedback information, associate status identifiers, and store them according to status identifiers, providing a foundation of feedback data that can be accessed by status dimension for subsequent delivery processing.

[0036] Example 2: In practical mobile terminal advertising application scenarios, the usage status of the same user varies significantly across different usage sessions. For example, during commutes or short-term app browsing, user behavior exhibits fragmented characteristics, while during stable or prolonged app use, user behavior shows strong continuity. Existing advertising systems typically make delivery decisions based on user history profiles or single ad request information, making it difficult to dynamically control the advertising delivery process by considering the user's current usage session status. This results in a lack of differentiation in ad delivery type and order across different usage states. To address these issues, this invention provides an AI-powered intelligent automatic ad delivery system, the structure of which is as follows: Figure 2 As shown. The specific implementation process of this system is as follows: The conversation behavior analysis module is used to collect the behavioral characteristics of the user's current usage session and generate a scenario fingerprint that represents the current usage state; The ad type matching module is used to match the corresponding ad type from the pre-established ad demand profile based on the scenario fingerprint, and determine the set of ad types that are allowed to participate in the current ad request processing. The conversion completion determination module is used to obtain the conversion process information corresponding to the advertisements in the set of advertisement types, and compare the continuous interaction capability of the current usage session with the conversion process information to filter out advertisements that have the conditions to complete the conversion in the current usage session. The Session Delivery Tracking Management module is used to record the interaction types and order of the ads delivered in the current session, forming a session delivery track. It then matches the session delivery track with a predefined conversion guidance sequence to determine the range of ad types for the next ad delivery within the ad type set. The delivery unlock control module is used to generate a session evolution state based on the session delivery trajectory and the continuity of user behavior in the current session, and within the scope of ad types, only ads that meet the delivery unlock conditions corresponding to the session evolution state are allowed to participate in delivery; The feedback information management module is used to collect user feedback on the ads after the ads have been delivered, record the usage status identifier corresponding to the feedback, and store the ad feedback information according to the usage status identifier.

[0037] Specifically, the AI-powered intelligent ad delivery system is deployed on an ad delivery server or cloud computing platform and establishes a communication connection with user terminals and the ad delivery platform. This connection is used to uniformly schedule and control the ad delivery process when an ad request occurs. When a user enters the application through a mobile terminal and triggers an ad request, the system first activates the session behavior analysis module to identify the current usage session and enter the session-level processing flow.

[0038] The session behavior analysis module collects behavioral feature information related to the current session from the user terminal during the session. The behavioral feature information includes, but is not limited to, changes in the terminal's location within the session, session duration, and page dwell and page switching behaviors.

[0039] The conversation behavior analysis module analyzes the usage conversation based on the collected behavioral characteristics, generates a scenario fingerprint to represent the current usage state, and outputs the scenario fingerprint to the subsequent advertising delivery processing module.

[0040] After obtaining the scene fingerprint, the ad type matching module reads the pre-established ad demand profile and performs type matching processing on the candidate ads based on the adaptation relationship between the scene fingerprint and the ad demand profile.

[0041] The ad type matching module determines the set of ad types that are allowed to participate in the current ad request processing based on the matching results, and passes the set of ad types to the conversion feasibility judgment module as the input range for subsequent filtering.

[0042] Upon receiving the set of ad types, the conversion feasibility assessment module retrieves the corresponding conversion process information for each ad within the set and, combined with the user's continuous interaction capabilities within the current session, determines whether the ad meets the basic conditions for conversion. Ads that do not meet the conversion completion conditions in the current session are removed from the candidate list; ads that meet the conditions are retained and passed to the session delivery trajectory management module.

[0043] The session delivery trajectory management module records the ads delivered within a user session, including the interaction type corresponding to the ad and the order in which the ads were delivered. Based on these records, the module generates a session delivery trajectory and matches it with a predefined conversion guidance sequence to determine the current user session's position in the conversion guidance path. This allows the module to determine the range of ad types for the next ad delivery within the ad type set.

[0044] After receiving the range of ad types, the ad delivery unlocking control module generates a session evolution state based on the session delivery trajectory and the continuity of user behavior in the current session. The ad delivery unlocking control module further matches the session evolution state with pre-set ad delivery unlocking conditions, allowing only ads that meet the unlocking conditions corresponding to the current session evolution state to participate in the delivery process within the current session; other ads do not participate in the delivery process within the current session.

[0045] After ad delivery is completed, the feedback management module collects user feedback on the delivered ads and records the corresponding usage status identifier at the time of feedback. The feedback management module categorizes and stores ad feedback information according to usage status identifiers, so that the feedback information can be retrieved and processed according to usage status during subsequent ad delivery processing.

[0046] Through the coordinated operation of the above modules, the system can uniformly control the type, order, and timing of ad delivery within the same user session, thereby completing the automatic ad delivery process based on the user session state.

[0047] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-powered intelligent method for automatic ad delivery, characterized in that, Includes the following steps: S1. Collect the behavioral characteristics of the user's current session and generate a scenario fingerprint of the current usage state; S2. Based on the scenario fingerprint, match the corresponding ad type from the pre-established ad demand profile to determine the set of ad types allowed to participate in the current ad request processing; S3. For the advertisements in the set of advertisement types, obtain the corresponding conversion process information, compare the continuous interaction capability of the current usage session with the conversion process information, and filter out the advertisements that have the conditions to complete the conversion in the current usage session. S4. Within the current usage session, the interaction types and order of the ads that have been delivered to the filtered ad records are used to form a session delivery trajectory. The session delivery trajectory is then matched with a predefined conversion guidance sequence to determine the range of ad types for the next ad delivery within the ad type set. S5. Based on the session delivery trajectory and the continuity of user behavior in the current session, generate a session evolution state, and within the range of ad types, only allow ads that meet the delivery unlocking conditions corresponding to the session evolution state to participate in delivery; S6. After the advertisement is delivered, collect user feedback information on the delivered advertisement, record the usage status identifier corresponding to the feedback, and store the advertisement feedback information according to the usage status identifier.

2. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S1, the collection of behavioral characteristics of the user's current session includes: After a user session begins, information on the user terminal’s location changes, session duration, and page dwell and page switching behavior is collected. The collected behavioral information is processed into a time series to form a set of session behavior features that reflect the continuity and stability of user operations within a session.

3. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S1, generating the scene fingerprint of the current usage state includes: The set of conversation behavior features is constructed into time series features and then input into the scene recognition model. The scene recognition model is built on a neural network for processing temporal features, the neural network including at least one of a recurrent neural network, a long short-term memory network, or a sequence modeling network based on an attention mechanism; The neural network is used to learn the sequence features of the session behavior features, and the output is a state vector representing the current state of the user's session. Based on the state vector, the usage scenario identifier corresponding to the user's current usage session is determined by the state classification algorithm, and the usage scenario identifier is used as the scenario fingerprint of the current usage state.

4. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S2, matching the corresponding ad type from the pre-established ad demand profile includes: Obtain the advertising demand profile corresponding to the candidate advertisement. The advertising demand profile is used to describe the demand attributes of the advertisement in terms of interaction complexity, continuous operation requirements and exposure dependence characteristics. The scene fingerprint and the advertising demand profile are matched to determine the degree to which the scene fingerprint meets the advertising demand profile. Based on the adaptability analysis results, advertisements that meet the preset adaptability conditions are classified into the advertisement type corresponding to the current usage state, thereby determining the set of advertisement types that are allowed to participate in the current advertisement request processing.

5. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S3, obtaining the corresponding conversion process information includes: For each candidate ad, a conversion process description corresponding to its conversion behavior is pre-established. The conversion process description is used to represent the sequence of operation steps required to complete the conversion of the ad. Associate corresponding sequential operation conditions with at least some of the operation steps, wherein the sequential operation conditions are used to describe the minimum sequential operation requirements required to complete the corresponding operation steps; The sequence of operation steps and the corresponding continuous operation conditions are used as the conversion process information for the advertisement.

6. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S3, the comparison between the continuous interaction capability and conversion process information based on the current usage session includes: Based on the user's page dwell behavior, operation intervals, and operation interruptions in the current usage session, a continuous interaction capability description is generated to characterize the user's ability to perform continuous operations within the usage session. Following the order of the operation steps in the conversion process information, the descriptions of continuous interaction capabilities are matched with the continuous operation conditions of the corresponding operation steps in sequence. An ad is deemed to have the conditions to complete a conversion in the current usage session only if the description of continuous interaction capabilities can cover the continuous operation conditions corresponding to the operation steps.

7. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S4, the formation of the session delivery trajectory includes: Within the user's current session, the interaction types of the ads that have been delivered are recorded, and the interaction types are used to characterize how the user interacts with the ads; The interaction types are sequentially processed according to the order in which the advertisements are delivered in the usage session, forming a session interaction sequence that reflects the process of advertisement delivery within the usage session. The conversation interaction sequence is used as the conversation delivery trajectory.

8. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S5, generating the session evolution state based on the session delivery trajectory and the continuity of user behavior in the current session includes: Based on the session delivery trajectory, obtain the interaction types and delivery order information of the ads that have been delivered in the current session; Within a usage session, user page dwell time, continuous operation duration, and operation interruption are collected to characterize the continuity of user behavior within the usage session. By jointly analyzing the session delivery trajectory and behavioral continuity, the current session stage is determined, and the determined session stage is taken as the session evolution state.

9. The AI-powered intelligent advertising automatic delivery method according to claim 1, characterized in that, In S5, the release unlocking conditions include: Pre-set corresponding launch unlocking conditions for ads of different ad types or conversion process complexity levels. The launch unlocking conditions are used to limit the session evolution state that the ad needs to meet to participate in the launch. After generating the session evolution state, the session evolution state is matched with the delivery unlocking conditions to determine whether the current session meets the delivery unlocking conditions of the corresponding advertisement. An ad is allowed to participate in the delivery process within the current session only if the session evolution state meets the corresponding ad delivery unlocking conditions.

10. An AI-powered intelligent automatic advertising delivery system, characterized in that: The AI-powered intelligent advertising automatic delivery method according to any one of claims 1-9, the system comprising the following modules: The conversation behavior analysis module is used to collect the behavioral characteristics of the user's current usage session and generate a scenario fingerprint that represents the current usage state; The ad type matching module is used to match the corresponding ad type from the pre-established ad demand profile based on the scenario fingerprint, and determine the set of ad types that are allowed to participate in the current ad request processing. The conversion completion determination module is used to obtain the conversion process information corresponding to the advertisements in the set of advertisement types, and compare the conversion process information with the continuous interaction capability of the current usage session to filter out advertisements that have the conditions to complete the conversion in the current usage session. The session delivery trajectory management module is used to record the interaction types and occurrence order of the delivered advertisements within the current usage session, forming a session delivery trajectory, and matching the session delivery trajectory with a predefined conversion guidance sequence to determine the range of advertisement types corresponding to the next advertisement delivery within the set of advertisement types; The delivery unlocking control module is used to generate a session evolution state based on the session delivery trajectory and the continuity of user behavior in the current usage session, and within the range of the ad types, only ads that meet the delivery unlocking conditions corresponding to the session evolution state are allowed to participate in delivery; The feedback information management module is used to collect user feedback information on the advertised ads after the ad delivery is completed, record the usage status identifier corresponding to the feedback, and store the ad feedback information according to the usage status identifier.

Images