A digital marketing content interaction generation method suitable for multi-scene delivery of a large model

Abstract

The application discloses a digital marketing content interaction generation method suitable for multi-scene delivery of large models, and relates to the technical field of interaction generation.The application can accurately depict the behavior transition characteristics of users in different content paths by setting an envelope boundary set modeling mechanism, effectively identifies high-value interaction nodes, and avoids the blindness and redundancy of content display.At the same time, a double-channel feedback mechanism divides the real-time behavior of users into high-speed feedback and verification feedback, realizes the collaborative linkage of high-frequency response and long-term optimization, and balances the immediacy of content scheduling and the stability of strategy.Combined with a dynamic exposure frequency matrix and a sorting and rearrangement strategy, the content delivery has structure self-adaptation and multi-scene generalization capability, and the personalized accuracy and user response efficiency of digital marketing driven by large models are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of interactive generation technology, and in particular to a method for interactively generating digital marketing content that is adapted to multi-scenario deployment of large models. Background Technology

[0002] Currently, digital marketing systems are evolving from simple content outreach to intelligent content adaptation that integrates user behavior perception and response scheduling to enhance user engagement depth and marketing conversion rates. Against this backdrop, a key direction for the evolution of digital marketing technology has emerged: how to integrate content generation and scheduling strategies based on multi-scenario user historical behavior data, real-time feedback signals, and content exposure paths to build an adaptive intelligent content delivery system. However, existing technologies are mostly limited to coarse-grained methods such as behavior clustering, periodic prediction, and tag generation, lacking dynamic boundary modeling and structural optimization mechanisms for the content delivery chain, making it difficult to meet the complex requirements of "multi-scenario—multi-feedback source—multi-node structural scheduling."

[0003] Prior art document CN119359343A discloses a digital marketing method and system based on the metaverse, which provides users with more timely and personalized content recommendations through behavioral clustering and cyclical trend prediction. While this solution has some innovation in user behavior feature modeling and feedback response, content generation mainly relies on dimensions such as the interaction frequency and dwell time of the metaverse platform, lacking modeling and optimization of the content display link itself. In addition, its response mechanism is a single-channel feedback path, failing to form a dual-path iterative mechanism of high-speed feedback and verification feedback. In high-frequency and ever-changing interaction scenarios, it is difficult to quickly adjust the delivery order and display parameters of content nodes, resulting in insufficient dynamic adaptation capabilities.

[0004] The prior art document CN111630550B proposes generating interactive messages with asynchronous media content and realizing media content rendering across multiple devices by constructing session data items. However, this solution focuses more on the media content identification and asynchronous loading processing of messages. Its main goal is to enhance cross-device rendering consistency. It does not involve how to dynamically adjust the content path and optimize the generation parameters based on real-time user feedback. Its content scheduling mechanism is passively and statically triggered, lacking linkage management of interactive behavior and display sequence, and cannot meet the closed-loop requirements of the digital marketing field for "real-time interaction - structural adjustment - model iteration". Summary of the Invention

[0005] In view of the problems existing in content delivery technology when dealing with complex interaction paths and multi-source feedback, this invention is proposed.

[0006] Therefore, the problem to be solved by this invention is how to achieve dynamic linkage scheduling between content generator parameters and path nodes.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] Firstly, this invention provides a digital marketing content interactive generation method adapted to multi-scenario deployment of large models. The method includes: forming an envelope boundary set of the content delivery path based on historical user interaction behavior data in the target scenario, combined with content display layout and time window information; dividing real-time user interaction behavior feedback into a high-speed feedback channel and a verification feedback channel, respectively forming scheduling instruction codes and parameter iteration instruction codes, and injecting them into the content generator; dynamically adjusting the display order and exposure frequency of each path node in the content delivery path according to the envelope boundary set and the scheduling instruction codes of the high-speed feedback channel; periodically updating the envelope boundary set and adjusting the content generator parameters through historical feedback data collected by the verification feedback channel to maintain the long-term stability and adaptability of the path structure.

[0009] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario deployment described in this invention, the formation of the envelope boundary set includes: extracting the number of clicks, dwell time, and jump frequency of each path node from historical user interaction behavior data associated with the target scenario, and generating a behavior focus factor matrix according to time windows; combining the content display layout information under the target scenario and the behavior focus factor matrix, generating a path node-level behavior response sequence within a continuous time window, and calculating the behavior response slope sequence between adjacent path nodes; delineating the boundary transition zone of multiple heterogeneous content sets based on the abrupt change segments in the behavior response slope sequence, and extracting the envelope boundary set containing the core behavior jump position.

[0010] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario deployment described in this invention, the determination of the mutation segment includes: calculating the variance of the slope difference within a sliding time window for the behavior response slope sequence, and marking behavior response slope sequence segments whose slope difference variance exceeds the mutation threshold as a candidate mutation segment set; performing consistency verification on the slope change direction of each behavior response slope sequence segment in the candidate mutation segment set, filtering out oscillating slope rebound segments, and retaining behavior transition segments with consistent direction and exceeding the mutation threshold; and aggregating the behavior transition segments into a mutation segment set according to the path node position.

[0011] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario deployment described in this invention, the step of dividing real-time user interaction behavior feedback into high-speed feedback channels and verification feedback channels includes: constructing a behavior intensity index threshold group by combining the behavior focus factor matrix and the historical response intensity of each path node in the envelope boundary set, and identifying the response priority interval of real-time feedback; the behavior intensity index threshold group includes click frequency threshold, dwell time surge ratio, and jump reverse ratio; for the user's real-time interaction behavior feedback data stream, the feedback is divided into high-speed feedback channels or verification feedback channels based on whether the corresponding path node is at the core behavior transition position in the boundary set and whether it meets the behavior intensity index threshold group.

[0012] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario deployment described in this invention, the determination of whether the corresponding path node is at the core behavior transition position in the boundary set includes: extracting the context identifier information of the interaction behavior from the real-time interaction behavior feedback data stream, constructing a feedback behavior structural feature tuple, wherein the structural features include the display position identifier, the page loading timestamp, and the user device window range; constructing a feedback matching judgment condition set based on the layout number, display time interval, and content exposure position range of each path node in the envelope boundary set, wherein the judgment condition is: the path node display position number is the same, the page loading timestamp is within the effective display time window, and the user device window range intersects with the display area range; if there are multiple path nodes that meet the conditions, the path node with the time window closest to the page loading timestamp is selected as the matching path node; the determination of whether the feedback diversion judgment threshold set is met includes: comparing the behavior intensity index group of the real-time interaction behavior feedback data stream with the behavior intensity index threshold group, if any behavior intensity index is higher than the corresponding threshold, and the obtained matching path node is at the core behavior transition position in the boundary set, it is assigned to the high-speed feedback channel; the rest are assigned to the verification channel.

[0013] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario delivery as described in this invention, the dynamic adjustment of the display order and exposure frequency of each path node in the content delivery path includes: extracting the core behavior transition node group located within the envelope boundary set in the current path node set based on the scheduling instruction encoding in the high-speed feedback channel, and constructing a display structure priority value; calculating the instantaneous priority score of each path node by combining the historical response intensity of the path node and the difference in the behavior response threshold in the scheduling instruction; locally rearranging the display order of each path node in the current content delivery path according to the instantaneous priority score, while preserving the order continuity of the behavior transition nodes; and dynamically adjusting the exposure frequency matrix by allocating exposure frequency gain or suppression factor to each node in the path node set after the display order is updated, based on the user dwell time change trend and scheduling feedback density.

[0014] As a preferred embodiment of the digital marketing content interaction generation method for adapting to large models and multi-scenario deployment described in this invention, the instant priority score is calculated by weighting the click frequency increase rate, dwell time change rate, and jump direction offset rate corresponding to each path node to form an instant behavior response index group. The weighted values ​​of the instant behavior response index group are then embedded into the basic priority weights of the previous time window to form the instant priority score for the current time window. The dynamic adjustment of the exposure frequency matrix includes: constructing a dwell time change trend sequence based on user dwell time records of the path node set within multiple consecutive time windows, and classifying it into different dwell trend categories according to the trend slope, including a continuous growth segment, a stable segment, and a declining segment; statistically analyzing the number of times each path node is scheduled per unit time in the high-speed feedback channel to determine the scheduling feedback density index of the path node, and constructing a scheduling feedback density matrix; determining the exposure frequency of each path node according to the dwell trend category and scheduling feedback density value; selecting the corresponding gain or suppression factor from the set exposure frequency adjustment coefficient group to adjust the exposure frequency matrix and form an updated version.

[0015] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the digital marketing content interactive generation method for adapting to large models and multi-scenario deployment as described in the first aspect of the present invention are implemented.

[0016] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the digital marketing content interactive generation method for adapting to large models and multi-scenario deployment as described in the first aspect of the present invention are implemented.

[0017] The beneficial effects of this invention are as follows: By setting an envelope boundary set modeling mechanism, this invention can accurately characterize the behavioral transition features of users in different content paths, effectively identify high-value interaction nodes, and avoid the blindness and redundancy of content display. Simultaneously, the dual-channel feedback mechanism divides real-time user behavior into high-speed feedback and verification feedback, achieving synergistic linkage between high-frequency response and long-term optimization, balancing the immediacy of content scheduling with the stability of the strategy. Furthermore, combined with a dynamic exposure frequency matrix and ranking reshuffling strategy, this enables content delivery to possess structural adaptability and multi-scenario generalization capabilities, significantly improving the personalized accuracy and user response efficiency of digital marketing driven by a large model. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a digital marketing content interaction generation method adapted for multi-scenario deployment of large models. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] As mentioned in the background section, how to integrate content generation and scheduling strategies based on multi-scenario user historical behavior data, real-time feedback signals, and content exposure paths to build an adaptive intelligent content delivery system has become a key direction in the evolution of digital marketing technology. However, most existing technologies are limited to coarse-grained methods such as behavior clustering, periodic prediction, and tag generation, lacking dynamic boundary modeling and structural optimization mechanisms for the content delivery chain, making it difficult to meet the complex requirements of "multi-scenario—multi-feedback source—multi-node structural scheduling".

[0024] Figure 1 A flowchart illustrating a digital marketing content interaction generation method adapted for multi-scenario deployment of large models according to an embodiment of the present invention. Figure 1 As shown, the digital marketing content interaction generation method adapted to multi-scenario deployment of large models includes:

[0025] S1: Based on historical user interaction data in the target scenario, combined with content display layout and time window information, form a set of envelope boundaries for the content delivery path.

[0026] The target business scenarios include, but are not limited to, product display pages, recommendation modules, search results blocks, and information flow content areas.

[0027] Historical user interaction data specifically includes user click data, page dwell time data, and node jump link logs recorded during the content display cycle. Each type of behavior data must correspond one-to-one with a content path node. Path nodes refer to the functional units or content carriers that users can interact with in the display structure, such as product cards, navigation blocks, recommended items, or jump buttons.

[0028] In this embodiment of the invention, the formation of the envelope boundary set includes:

[0029] A. Extract the number of clicks, dwell time, and jump frequency of each path node from the historical user interaction behavior data associated with the target scenario, and generate a behavior focus factor matrix according to the time window to represent the distribution of behavior response intensity of each path node in different time periods.

[0030] Specifically, a rolling time window structure is set up, which is defined as a set of time slices of equal width with a granularity of minutes or hours, used to characterize the evolution of behavioral response over time.

[0031] Within each time window, the normalized values ​​of click count, dwell time, and jump frequency are input into the behavior focus factor generation unit to generate a three-dimensional behavior response tensor. By slicing this three-dimensional behavior response tensor along the path node direction and expanding it along the time window dimension, a behavior focus factor matrix can be formed. This matrix uses path nodes as row labels and time window indices as column labels, with each cell representing the behavior response intensity of the current node within the corresponding time window. The behavior response intensity can be expressed using a weighted fusion method.

[0032] B. Combining the content display layout information and behavior focus factor matrix in the target scenario, generate a path node-level behavior response sequence within a continuous time window, and calculate the behavior response slope sequence between adjacent path nodes.

[0033] Specifically, content display layout information refers to the structural hierarchy and node order of the target page or content flow, including the relative display position, hierarchical relationship, and content type of path nodes. This structure can be represented as an ordered linked list of path nodes, used to clarify the physical or logical adjacency relationship between adjacent path nodes.

[0034] During the operation, firstly, for each path node, the sequence of behavioral response values ​​within a continuous time window is extracted along the time window dimension to form a single-node behavioral response time series; then, in order to extract the response change trend between spatially adjacent nodes, path node pairs are obtained according to the layout order, the behavioral response difference under each time window is calculated and divided by the relative displacement or structural jump number between path nodes to generate a behavioral response slope sequence.

[0035] C. Based on the abrupt change segments in the behavioral response slope sequence, delineate the boundary transition zones of multiple heterogeneous content sets and extract the envelope boundary set containing the core behavioral transition locations.

[0036] The determination of abrupt change segments includes: calculating the variance of the slope difference within a sliding time window for the behavioral response slope sequence (i.e., calculating the difference and variance between adjacent slope values ​​within a continuous time window); if the slope variance of a certain node pair exceeds the abrupt change threshold in multiple time windows, then the behavioral response slope sequence segment is marked as a candidate abrupt change segment set; the consistency of the slope change direction of each behavioral response slope sequence segment in the candidate abrupt change segment set is checked, and oscillating slope rebound segments (repeated fluctuations in direction, sudden increases followed by declines, etc.) are screened out, while behavioral transition segments with consistent direction and exceeding the abrupt change threshold are retained; the behavioral transition segments are aggregated into a set of abrupt change segments according to the path node position.

[0037] Finally, all behavioral transition segments are aggregated according to their corresponding path node positions to form a set of mutation segments. Boundary transition zones are generated based on the node groups on both sides of the mutation segments, and integrated into a set of envelope boundaries for the content path. The set of envelope boundaries can be understood as the boundaries of segments in the content display path where user behavior changes drastically.

[0038] Furthermore, based on the set of mutation segments, node intervals constituting boundary transitions are extracted from the continuous path node groups before and after the mutation segments. Each boundary transition zone consists of several path nodes adjacent to the mutation segments, used to express the user response transition area where the content structure gradually switches from one behavioral state to another.

[0039] Within the boundary transition zone, the set of path nodes located at the peak of the behavioral response change rate is further identified. Combining these path nodes with characteristic indicators such as context jump frequency, response surge locations, and abrupt behavioral change points, the core behavioral transition locations are identified. Specifically, a path node can be identified as a core behavioral transition location if it simultaneously meets two or more of the following three conditions: the path node's jump frequency is significantly higher than adjacent path nodes (high behavioral path divergence); the total response volume of the path node (clicks / swipes, etc.) is significantly higher than adjacent path nodes (abnormal behavioral trigger intensity); and the path node's dwell time change rate is inversely proportional to and significantly different from that of the preceding and following path nodes (abrupt behavioral pattern change).

[0040] S2: Divide real-time user interaction feedback into a high-speed feedback channel and a verification feedback channel, which are then used to form real-time scheduling instruction codes and parameter iteration instruction codes, respectively, and injected into the content generator.

[0041] S2.1: Divide real-time user interaction feedback into a high-speed feedback channel and a verification feedback channel.

[0042] S2.1.1: Combining the historical response intensity of each path node in the behavior focus factor matrix and the envelope boundary set, construct a behavior intensity index threshold group to identify the response priority range of real-time feedback; the behavior intensity index threshold group includes click frequency threshold, dwell time surge ratio and jump reverse ratio.

[0043] Ideally, the statistical characteristic values ​​of the behavioral intensity distribution exhibited by each path node in the behavioral focus factor matrix are calculated separately. Specifically, this includes: the maximum time window value and mean difference of click frequency; the window burst ratio of dwell time (i.e., the relative growth rate of dwell time in adjacent time windows); and the reverse jump ratio (i.e., the frequency ratio of users jumping back after entering the path node). These three behavioral intensity indicators are then compared through a local mean-fluctuation window (for example, the mean of the behavioral indicator within the window can be calculated for each path node, plus the product of the threshold coefficient and the standard deviation), forming a set of behavioral intensity indicator thresholds for real-time feedback judgment.

[0044] S2.1.2: For the user's real-time interactive behavior feedback data stream, based on whether the corresponding path node is at the core behavior transition position in the boundary set and whether it meets the feedback diversion judgment threshold group, the feedback is divided into a high-speed feedback channel or a verification feedback channel.

[0045] S2.1.2.1: Extract contextual identifier information of interactive behaviors from the real-time interactive behavior feedback data stream, and construct feedback behavior structural feature tuples, including display position identifiers (such as content area numbers or layout block IDs), page load timestamps, and user device viewport ranges. Each feedback behavior is bound to a structural feature tuple for subsequent path node location. The construction of feedback structural feature tuples adopts a client-side SDK or Web layer collection mechanism. During the collection phase, de-identification encoding and compression are performed, and the data is reported to the intermediate layer cache pool of the content scheduling engine through a pipeline to achieve low-latency feedback access and accurate structural tracking capabilities.

[0046] S2.1.2.2: Based on the layout number, display time interval, and content exposure position range of each path node in the envelope boundary set, construct a set of feedback matching judgment conditions; the judgment conditions are: the path node display position number is the same, the page load timestamp is within the valid display time window, and the user device window range and the display area range intersect; if there are multiple path nodes that meet the conditions, the path node with the time window closest to the page load timestamp is selected as the matching path node; the marking results will serve as the input basis for feedback channel splitting and instruction encoding.

[0047] S2.1.2.3: If the obtained matching path node is located at the core behavior transition position in the boundary set, and the behavior intensity index group of the real-time interactive behavior feedback data stream is compared with the behavior intensity index threshold group, if any behavior intensity index is higher than the corresponding threshold, it is assigned to the high-speed feedback channel; the rest of the feedback is assigned to the verification channel.

[0048] It can be seen that the present invention adopts a dual determination mechanism of structural position + behavior intensity to realize structure-behavior coupled classification of the feedback channel, which has stronger stability and generalization ability than the existing behavior-driven classification method.

[0049] The high-speed feedback channel receives immediate feedback such as high-frequency clicks and abnormal redirects; the verification feedback channel receives low-speed feedback such as timeouts and periodic revisits.

[0050] S2.2: For the data in the high-speed feedback channel, combined with the historical response direction trend of the path nodes, a scheduling instruction code is generated. The code includes the path node number, feedback timestamp, behavior change direction (calculated by the behavior difference sign in the sliding window), and scheduling trigger priority weight (which can be calculated based on the degree of exceeding the threshold of the intensity index, but this invention does not limit it to a single value).

[0051] This instruction is directly injected into the content scheduling part, driving the content exposure order, content replacement priority, and interaction structure adjustment strategy to ensure that sudden user behavior can trigger interface response in an instant.

[0052] S2.3: For the data in the verification feedback channel, accumulate and generate the content generator parameter offset vector within a set period, and combine it with the evolution stability of the boundary segment to which the path node belongs to form the parameter iteration instruction code; wherein, the evolution stability is calculated by the cross-window volatility of each path node.

[0053] This encoding is used for long-term structural evolution or content template fine-tuning to ensure that the generation system has a slow-heat response capability to low-frequency structural behavior changes, avoiding overfitting or noise scheduling.

[0054] S3: Based on the envelope boundary set and the scheduling instruction code of the high-speed feedback channel, dynamically adjust the display order and exposure frequency of each path node in the content delivery path.

[0055] S3.1: Based on the scheduling instruction encoding in the high-speed feedback channel, extract the core behavior transition node group located within the envelope boundary set in the current path node set, and construct the display structure priority value.

[0056] Whether a path node is within the envelope boundary set is determined by two conditions being met simultaneously: First, the structural positioning information of the path node (e.g., layout block number, display position number) must be completely consistent with the marking information of the boundary set; second, the display cycle of the path node must intersect with the behavioral transition cycle of the boundary set. Nodes that meet these two conditions are defined as core behavioral transition path nodes, and these nodes are identified as key response areas for sudden changes in user behavior.

[0057] After extracting the core behavior transition path node group, a dynamic weight priority list of the path nodes is constructed by combining the frequency of their appearance in the high-speed feedback channel, their feedback weight value, and their behavior response direction. A time weight decay mechanism is adopted, in which the path nodes extracted in the feedback with the more recent encoding time have higher weights, so as to maintain the system's sensitivity and response speed to changes in current user preferences.

[0058] S3.2: Calculate the instantaneous priority score of each path node by combining the historical response strength of the path node and the difference in the behavioral response threshold in the scheduling instructions.

[0059] In this embodiment of the invention, the instant priority score is calculated by weighting the click frequency increase rate, dwell time change rate and jump direction offset rate corresponding to each path node to form an instant behavior response index group. The weighted values ​​of the instant behavior response index group are then embedded in the basic priority weight of the previous time window to form the instant priority score of the current time window.

[0060] Among them, the click frequency increase rate is defined as the ratio of the click frequency of the current time window to the click frequency of the previous time window; the dwell time change rate is defined as the ratio of the total dwell time of the current time window to the historical baseline dwell time; the jump direction deviation rate is calculated by the ratio change of positive jump and reverse jump, and measures the trend of jump behavior direction deviation.

[0061] S3.3: Based on the real-time priority score, the display order of each path node in the current content delivery path is partially rearranged to preserve the sorting continuity of behavior transition nodes.

[0062] Before sorting, all path nodes are divided into two groups: the first group is the core behavior transition node group, and the second group is the regular node group. The reordering operation only performs priority sorting within the regular node group; for the core behavior transition node group, the current order will be maintained to preserve the structural coherence of the user behavior habit chain and the logical consistency of the content understanding path.

[0063] During the local reordering process, the top N% of nodes by priority score in the regular node group are first extracted and moved to the beginning of the path as preliminary auxiliary content. The remaining path nodes are arranged in descending order of score, and the content type relevance distance between path nodes is added to ensure that the smoothness of the content is not disrupted. Here, N is a constant.

[0064] S3.4: For the set of path nodes after the display order is updated, combine the trend of user dwell time changes and scheduling feedback density, allocate exposure frequency gain or suppression factor according to the node, and dynamically adjust the exposure frequency matrix.

[0065] The dynamic adjustment of the exposure frequency matrix includes: constructing a sequence of dwell time change trends based on the user dwell time records of the path node set in multiple consecutive time windows, and classifying them into different dwell trend categories according to the trend slope (fitted by the least squares method), including a continuous growth segment, a stable segment, and a decline segment.

[0066] The number of times each path node is scheduled per unit time in the high-speed feedback channel is counted to determine the scheduling feedback density index of the path node, and a scheduling feedback density matrix is ​​constructed.

[0067] Using a two-dimensional mapping method, the exposure frequency of each path node is determined by the classification of dwell trends and the scheduling feedback density value. From the set exposure frequency adjustment coefficient group, the corresponding gain or suppression factor is selected to adjust the exposure frequency matrix, forming an updated version. For example, if a path node belongs to a continuously increasing segment and the feedback density is higher than the upper threshold, a strong gain is selected; if it belongs to a decreasing segment and the feedback density is lower than the lower threshold, a strong suppression is selected.

[0068] Finally, the exposure frequency matrix is ​​updated by multiplying the current exposure frequency value of each path node by the selected factor.

[0069] S4: By verifying historical feedback data collected through the feedback channel, the envelope boundary set is updated periodically, and the content generator parameters are adjusted to maintain the long-term stability and adaptability of the path structure.

[0070] The verification feedback channel primarily collects low-speed evolutionary behavior feedback data, including but not limited to: records of excessively long dwell times, periodic page access trajectories, and delayed redirection path markers. Using fixed time periods (e.g., daily or every few hours) as the scheduling unit, feedback records for all path nodes in the verification channel within the previous period are collected in batches and their structures are analyzed.

[0071] To improve the adaptability of the envelope boundary set, the validation feedback data needs to be organized and summarized within a set time period (e.g., every 24 hours). The organization process includes recording the occurrence time, type (e.g., re-click, repeated jump), and display context information (e.g., display area, content category) of the feedback behavior based on the path node identifier corresponding to each feedback. After accumulating data over a certain period, a structured feedback dataset that can be used to update the boundary determination criteria is formed.

[0072] By comparing and organizing the feedback dataset with the location distribution and response behavior records of nodes in the existing envelope boundary set, it is determined whether the boundary set structure needs to be adjusted. The determination mechanism is as follows: if, in the verification feedback data, there are frequent reverse migrations of user behavior between path nodes on both sides of the boundary set (such as moving from a node with less exposure to a core node), or if non-core path nodes show a continuous trend of increasing response over multiple periods, then the path node is included in the boundary transition area, and the boundary position is appropriately expanded.

[0073] The update process does not rely on real-time data, but rather slowly corrects the boundary range based on medium- to long-term user behavior deviation trends, thereby avoiding boundary distortion caused by short-term fluctuations and improving the stability and inclusiveness of the content path in the long term.

[0074] After the envelope boundary set is updated, the relevant generation parameters in the content generator are adjusted synchronously, mainly including: the content display style weight of path nodes; the content category adaptation coefficient of the displayed content; and the behavioral trend weight in the user preference feature vector.

[0075] The adjustments are based on the long-term response changes of each node before and after the boundary update. If the proportion of a certain content category in the updated boundary nodes increases significantly, the generation parameters of that category will receive higher priority or display intensity in subsequent generation tasks; conversely, the frequency will be suppressed.

[0076] In addition, based on the high-frequency combinations of content tags or layout positions contained in the feedback, the generator's layout configuration template is updated to improve the consistency between subsequent content delivery and user path structure.

[0077] This invention employs a dual strategy of gradual boundary changes and parameter fine-tuning, enabling the path structure to adapt to long-term behavioral evolution while responding quickly to high-speed feedback. Compared to methods relying solely on real-time data, this invention effectively reduces user experience interference caused by content drift and sudden display changes, maintaining the coherence, stability, and preference matching accuracy of the recommended path.

[0078] This embodiment also provides a computer device suitable for the digital marketing content interactive generation method adapted to multi-scenario deployment of large models, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the digital marketing content interactive generation method adapted to multi-scenario deployment of large models as proposed in the above embodiment.

[0079] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0080] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the digital marketing content interactive generation method for adapting to large models and delivering multiple scenarios as proposed in the above embodiments.

[0081] In summary, this invention, by establishing an envelope boundary set modeling mechanism, can accurately characterize the behavioral transitions of users across different content paths, effectively identify high-value interaction nodes, and avoid the blindness and redundancy of content display. Simultaneously, the dual-channel feedback mechanism divides real-time user behavior into high-speed feedback and verification feedback, achieving synergistic linkage between high-frequency response and long-term optimization, balancing the immediacy of content scheduling with the stability of the strategy. Furthermore, combined with a dynamic exposure frequency matrix and ranking reshuffling strategy, content delivery possesses structural adaptability and multi-scenario generalization capabilities, significantly improving the personalized accuracy and user response efficiency of digital marketing driven by a large model.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A digital marketing content interaction generation method for adapting multi-scene delivery of a large model, characterized by: include: Based on historical user interaction data in the target scenario, combined with content display layout and time window information, an envelope boundary set of content delivery paths is formed. The real-time user interaction feedback is divided into a high-speed feedback channel and a verification feedback channel, which are then used to form scheduling instruction codes and parameter iteration instruction codes, respectively, and injected into the content generator. Based on the envelope boundary set and the scheduling instruction code of the high-speed feedback channel, the display order and exposure frequency of each path node in the content delivery path are dynamically adjusted. By verifying historical feedback data collected through the feedback channel, the envelope boundary set is updated regularly, and the content generator parameters are adjusted to maintain the long-term stability and adaptability of the path structure. The formation of the envelope boundary set includes: extracting the number of clicks, dwell time, and jump frequency of each path node from the historical user interaction behavior data associated with the target scene, and generating a behavior focus factor matrix according to time windows; combining the content display layout information under the target scene and the behavior focus factor matrix, generating a path node-level behavior response sequence within a continuous time window, and calculating the behavior response slope sequence between adjacent path nodes; based on the abrupt change segments in the behavior response slope sequence, delineating the boundary transition zones of multiple heterogeneous content sets, and extracting the envelope boundary set containing the core behavior jump positions; The step of dividing real-time user interaction feedback into high-speed feedback channels and verification feedback channels includes: constructing a set of behavior intensity index thresholds by combining the behavior focus factor matrix and the historical response intensity of each path node in the envelope boundary set, and identifying the response priority range of real-time feedback; the set of behavior intensity index thresholds includes click frequency threshold, dwell time surge ratio, and jump reverse ratio; for the user's real-time interaction feedback data stream, the feedback is divided into high-speed feedback channels or verification feedback channels based on whether the corresponding path node is at the core behavior transition position in the boundary set and whether it meets the set of behavior intensity index thresholds.

2. The method of claim 1, wherein the method further comprises: The determination of the mutation segment includes: The variance of the slope difference within a sliding time window is calculated for the behavioral response slope sequence, and behavioral response slope sequence segments whose slope difference variance exceeds the mutation threshold are marked as a set of candidate mutation segments; The consistency of the slope change direction before and after each behavioral response slope sequence segment in the candidate mutation segment set is checked, and oscillating slope rebound segments are screened out, while behavioral transition segments with consistent direction and exceeding the mutation threshold are retained. The behavioral transition segments are aggregated into a set of mutation segments according to the path node positions.

3. The method of claim 2, wherein the method further comprises: The determination of whether the corresponding path node is at the core behavior transition position in the boundary set includes: Extract contextual identifier information of interactive behavior from real-time interactive behavior feedback data stream, and construct feedback behavior structural feature tuples. The structural features include display position identifier, page load timestamp, and user device viewport range. Based on the layout number, display time interval, and content exposure location range of each path node in the envelope boundary set, a set of feedback matching judgment conditions is constructed, wherein the judgment conditions are: The path nodes have the same display position number, the page load timestamp is within the valid display time window, and the user device's viewport area overlaps with the display area area; If multiple path nodes satisfy the condition, the path node whose time window is closest to the page load timestamp is selected as the matching path node. The determination of whether the behavior intensity index threshold group is met includes: The behavior intensity index group of the real-time interactive behavior feedback data stream is compared with the behavior intensity index threshold group. If any behavior intensity index is higher than the corresponding threshold and the obtained matching path node is at the core behavior transition position in the boundary set, it is assigned to the high-speed feedback channel; the rest are assigned to the verification channel.

4. The digital marketing content interactive generation method for adapting to large models and multi-scenario deployment as described in claim 1, characterized in that: The dynamic adjustment of the display order and exposure frequency of each path node in the content delivery path includes: Based on the scheduling instruction encoding in the high-speed feedback channel, the core behavior transition node group located within the envelope boundary set in the current path node set is extracted, and the display structure priority value is constructed. The instantaneous priority score of each path node is calculated by combining the historical response intensity of the path nodes and the differences in the behavioral response thresholds in the scheduling instructions. Based on the real-time priority score, the display order of each path node in the current content delivery path is partially rearranged to preserve the sorting continuity of the behavior transition nodes. For the set of path nodes after the display order is updated, the exposure frequency matrix is ​​dynamically adjusted by allocating exposure frequency gain or suppression factor to nodes based on the trend of user dwell time changes and scheduling feedback density.

5. The method of claim 4, wherein the method further comprises: The instant priority score is calculated by weighting the click frequency increase rate, dwell time change rate and jump direction offset rate of each path node to form an instant behavior response index group. The weighted values ​​of the instant behavior response index group are then embedded in the basic priority weight of the previous time window to form the instant priority score of the current time window. The dynamically adjusted exposure frequency matrix includes: Based on the user dwell time records of the path node set within multiple consecutive time windows, a dwell time change trend sequence is constructed, and it is divided into different dwell trend categories according to the trend slope, including a continuous growth segment, a stable segment, and a decline segment. The number of times each path node is scheduled per unit time in the high-speed feedback channel is counted to determine the scheduling feedback density index of the path node and construct the scheduling feedback density matrix. The exposure frequency of each path node is determined by the classification of dwell trends and the scheduling feedback density value. In the set exposure frequency adjustment coefficient group, the corresponding gain or suppression factor is selected to adjust the exposure frequency matrix and form an updated version.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the digital marketing content interactive generation method for adapting large models and multi-scenario deployment as described in any one of claims 1 to 5.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the digital marketing content interactive generation method for adapting large models and multi-scenario deployment as described in any one of claims 1 to 5.

Images