Big data-based media promotion targeted accurate delivery system and method

Abstract

The application is suitable for the field of media technology, and provides a media promotion directional accurate delivery system and method based on big data. The system comprises a log collection module, an inertia graph construction module, an event identification module, a content library construction module and a strategy selection module. The method collects user terminal application switching logs in real time, constructs a cognitive inertia graph with application types as nodes and multiple weight directed edges for each user, identifies task switching gap events and judges corresponding attention residual depth, constructs a promotion content library with pre-labeled cognitive mode labels and establishes an expected delivery mapping, dynamically selects a delivery strategy according to the attention residual depth, and determines the final delivery content through multi-dimensional weighted sorting. The application can realize directional accurate delivery of media promotion based on the real-time cognitive state of the user, and effectively improve the delivery adaptation degree and conversion effect.

Description

Technical Field

[0001] This invention belongs to the field of media technology, and in particular relates to a targeted and precise media promotion system and method based on big data. Background Technology

[0002] The deep integration of mobile internet and big data technologies is driving the media promotion industry towards refinement, with information flow scenarios becoming the core reach channel for media promotion. Currently, the industry generally relies on user behavior data for targeted advertising, using the collection and analysis of users' online behavior to deliver promotional content in a targeted manner. The iterative upgrading of precision targeting technology has become a core direction for improving the conversion efficiency of media promotion. With the continuous enrichment of the mobile application ecosystem, users are switching between different types of applications more frequently, and user attention is exhibiting fragmented and dynamic characteristics. Traditional advertising models based on static profiles are gradually becoming unable to adapt to users' real-time changing attention states, creating a clear industry demand for new advertising technologies capable of capturing users' instantaneous cognitive states.

[0003] Existing targeted media advertising solutions typically collect user interaction data such as browsing and clicking, build a user interest tagging system based on historical behavior data, and use tag matching to filter promotional content against users. Finally, content is pushed into the news feed according to predetermined rules. Some optimization solutions incorporate data from dimensions such as user app usage frequency and active time periods to supplement and improve user profiles, thereby enhancing the match between content and user interests and achieving basic targeted advertising results.

[0004] Current technologies rely on static user tags and historical interest data for ad placement decisions, failing to adjust the placement logic based on the real-time cognitive state of users during application switching. They cannot identify and utilize the differences in user attention states at task switching nodes, and the content matching dimension is limited to the user interest level, making it difficult to adapt to the user's instantaneous cognitive acceptance mode. The placement strategy lacks a dynamic and differentiated adjustment mechanism, and the adaptability of promotional content to the user's real-time cognitive state is insufficient, resulting in a significant bottleneck in improving the overall placement accuracy and conversion effect. Summary of the Invention

[0005] The purpose of this invention is to provide a media promotion targeted and precise delivery system based on big data, which aims to solve the technical problems existing in the prior art as identified in the background section.

[0006] This invention is implemented as follows: a big data-based targeted and precise media promotion system, the system comprising: The log collection module is used to collect application switching logs from user terminals in real time. The application switching logs include switching time, source application ID, target application ID, source application type, and target application type. An inertial map construction module is used to construct and update a cognitive inertial map for each user based on the application switching logs. The cognitive inertial map includes directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights. The event recognition module is used to identify task switching gap events that meet predetermined conditions from the application switching log. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, the module queries the current user's cognitive inertia graph to obtain the weight of the directed edge from the source application type to the information flow type, and determines the attention residual depth corresponding to this switch based on the weight. The content library construction module is used to build a media promotion content library. It determines the corresponding expected attention retention mode based on the source application type, and establishes an expected delivery mapping with the cognitive mode tag based on the expected attention retention mode. The promotion content of the media promotion content library is pre-labeled with cognitive mode tags, which include visual, logical and interactive types. The strategy selection module is used to dynamically select a delivery strategy based on the attention residual depth corresponding to the task switching interval event when the event is detected. The delivery strategy includes strong follow delivery and weak transition delivery.

[0007] Another objective of this invention is to provide a method for targeted and precise media promotion based on big data, the method comprising: Real-time collection of application switching logs from user terminals, including switching time, source application ID, target application ID, source application type, and target application type; Based on the application switching logs, a cognitive inertia graph is constructed and updated for each user. The cognitive inertia graph contains directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights. Identify task switching gap events that meet predetermined conditions from the application switching logs. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, query the current user's cognitive inertia graph to obtain the weight of the directed edge from the source application type to the information flow type. Determine the attention residual depth corresponding to this switch based on the weight. A media promotion content library is constructed. Based on the source application type, the corresponding expected attention retention pattern is determined, and an expected delivery mapping is established between the expected attention retention pattern and the cognitive pattern tag. The promotion content in the constructed media promotion content library is pre-labeled with cognitive pattern tags, which include visual, logical, and interactive types. When the task switching interval event is detected, a delivery strategy is dynamically selected based on the attention residual depth corresponding to this switch. The delivery strategy includes strong follow delivery and weak transition delivery.

[0008] As a further aspect of the present invention, the real-time collection of application switching logs from user terminals specifically includes: The application switching event is obtained through the event listening interface of the terminal operating system. The source application ID that was exited and the target application ID that was entered when the switch occurred are extracted, and the switching time is recorded. The system calls a pre-established category mapping table to map the source application ID to the source application type and the target application ID to the target application type. The pre-established category mapping table is based on the application store category tags. The value range of the source application type and the target application type includes image, game, document, development, social and information flow categories. The switching time, source application ID, target application ID, source application type, and target application type are encapsulated into an application switching log and stored in the application switching log database.

[0009] As a further aspect of the present invention, the step of constructing and updating a cognitive inertia map for each user specifically includes: Initialize a directed weighted graph as a cognitive inertia graph, where the nodes are all application types defined in a preset category mapping table; Read all application switching logs of the current user from the application switching log database, and parse out the source application type and target application type for each log entry. For each pair of source application type and target application type, update the switching frequency of the directed edge from the source application type node to the target application type node in the cognitive inertia graph, and increment the switching frequency by 1; Based on the switching time in the application switching log, calculate the average switching interval of the most recent switching from the source application type to the target application type, and update the average switching interval to the average switching interval weight of the corresponding directed edge. Based on the updated switching frequency and average switching interval, the switching intensity is calculated by weighted summation of the reciprocals of the switching frequency and the average switching interval. The switching intensity is then updated to the switching intensity weight of the corresponding directed edge, thus completing the incremental update of the cognitive inertia map.

[0010] As a further aspect of the present invention, the step of determining the attention residual depth corresponding to this switch based on the weight specifically includes: Scan the application switching logs of the current user in the application switching log database, select log records where the target application type is information flow and the source application type is not information flow, mark the selected log records as task switching interval events, and each task switching interval event corresponds to one application switching log. For each marked task switching gap event, the source application type is parsed from the application switching log corresponding to the task switching gap event, the current user's cognitive inertia graph is queried, and the switching intensity weight of the directed edge from the source application type node to the information flow type node is read. The switching intensity weight is compared with the preset intensity threshold. If the switching intensity weight is greater than the preset intensity threshold, it is determined that the attention residual depth corresponding to the task switching gap event is higher than the preset intensity threshold; otherwise, it is determined that the attention residual depth is not higher than the preset intensity threshold.

[0011] As a further aspect of the present invention, the establishment of the expected projection mapping with the cognitive pattern label specifically includes: Receive uploaded media promotion content and assign cognitive mode tags to each piece of media promotion content. If it contains dynamic visual media, it is labeled as visual; if it contains text structured information components, it is labeled as logical; if it contains interactive controls, it is labeled as interactive. Each identified media promotion content item and its corresponding cognitive pattern tag will be associated and stored in the media promotion content library; Based on the preset correspondence rules between source application types and expected attention residual patterns, a corresponding expected attention residual pattern is determined for each source application type. If the source application type is an image-based or game-based application, the expected attention residual pattern is visual attention residual; if the source application type is a document-based or development-based application, the expected attention residual pattern is logical attention residual; if the source application type is a social application, the expected attention residual pattern is social attention residual. Each desired attention residual pattern is mapped to a cognitive pattern tag in the media promotion content library to establish a desired delivery mapping relationship, wherein visual attention residual maps to visual type, logical attention residual maps to logical type, and social attention residual maps to interactive type.

[0012] As a further aspect of the present invention, the dynamic selection of the delivery strategy specifically includes: When a task switching gap event is detected, the judgment result is obtained as to whether the attention residual depth corresponding to the task switching gap event is higher than a preset depth threshold. If the attention residual depth is higher than the preset depth threshold, then strong follow-up delivery is performed: the source application type is parsed from the application switching log corresponding to the task switching gap event, and according to the expected attention residual mode corresponding to the source application type, the cognitive mode tag with the expected delivery mapping relationship with the expected attention residual mode is queried in the expected delivery mapping relationship, the promotion content with the cognitive mode tag is screened from the media promotion content library, a first candidate set is generated, and the promotion content in the first candidate set is inserted into the content stream of the target application corresponding to the task switching gap event for priority delivery; If the attention residual depth is not higher than the preset depth threshold, weak transition delivery is performed: promotional content whose cognitive pattern tags and the expected attention residual patterns corresponding to the source application type do not have the expected delivery mapping relationship are selected from the media promotion content library, a second candidate set is generated, and the promotional content in the second candidate set is inserted into the content stream of the target application corresponding to the task switching gap event for priority delivery.

[0013] As a further embodiment of the present invention, after generating the first candidate set and the second candidate set, a candidate sorting step is also included: For each promotional content in the candidate set, the publication time and historical display count of the promotional content are extracted from the media promotional content library. The timeliness weight is calculated based on the time interval between the publication time and the current time, and the historical display count is used as historical conversion data. From the current user's cognitive inertia map, read the switching frequency of the source application type node corresponding to the task switching gap event, and use it as the basis for calculating the user's preference strength for the source application type; The promotional content is selected based on a weighted ranking system that considers timeliness, historical conversion data, and switching frequency. The highest-ranking content is then chosen as the final target for ad placement.

[0014] The beneficial effects of this invention are: This invention constructs a cognitive inertia map by collecting application switching logs from user terminals, transforming user application switching behavior into quantifiable cognitive features. This enables a structured characterization of user attention shift inertia. Based on this, it identifies task switching gap events and determines the corresponding attention residual depth, capturing the instantaneous cognitive state of users when entering an information flow scenario. This invention constructs a media promotion content library with cognitive pattern tags and establishes a desired delivery mapping relationship, opening up the matching link between user cognitive state and promotion content attributes. It combines dynamic switching of strong follow-up delivery and weak transition delivery strategies based on attention residual depth to adapt to users' content acceptance tendencies under different cognitive states. Simultaneously, it completes the final content selection through multi-dimensional weighted sorting of historical conversion data and user preference intensity based on timeliness. The entire solution uses the user's real-time cognitive state as the core for delivery decisions, breaking the limitations of traditional static tag delivery, achieving dynamic and refined delivery strategies, comprehensively improving the adaptability of media promotion content to user states, effectively improving delivery accuracy and content conversion effects. Furthermore, the entire mechanism operates based on native user behavior data, requiring no additional user intervention, balancing delivery efficiency and user experience. Attached Figure Description

[0015] Figure 1 A structural block diagram of a big data-based targeted and precise media promotion system provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a big data-based targeted and precise media promotion method provided in this embodiment of the invention; Figure 3 A flowchart for real-time collection of application switching logs of user terminals provided in an embodiment of the present invention; Figure 4 A flowchart for constructing and updating a cognitive inertia map for each user, provided as an embodiment of the present invention; Figure 5 This is a flowchart provided by an embodiment of the present invention for determining the attention residual depth corresponding to this switch based on the weight; Figure 6 A flowchart for establishing a desired projection mapping between the cognitive pattern label and the embodiment of the present invention; Figure 7 A flowchart illustrating the dynamic selection of delivery strategies provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] Figure 1The structural block diagram of the big data-based targeted media promotion system provided in this embodiment of the invention is as follows: Figure 1 As shown, the system includes: The log collection module 100 is used to collect application switching logs of user terminals in real time. The application switching logs include switching time, source application ID, target application ID, source application type and target application type. The inertial map construction module 200 is used to construct and update a cognitive inertial map for each user based on the application switching log. The cognitive inertial map includes directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights. The event recognition module 300 is used to identify task switching gap events that meet predetermined conditions from the application switching log. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, the current user's cognitive inertia map is queried to obtain the weight of the directed edge from the source application type to the information flow type, and the attention residual depth corresponding to this switch is determined according to the weight. The content library construction module 400 is used to construct a media promotion content library, determine the corresponding expected attention retention mode according to the source application type, and establish an expected delivery mapping with the cognitive mode tag based on the expected attention retention mode. The promotion content of the constructed media promotion content library is pre-labeled with cognitive mode tags, and the cognitive mode tags include visual, logical and interactive. The strategy selection module 500 is used to dynamically select a delivery strategy based on the attention residual depth corresponding to the task switching interval event when the event is detected. The delivery strategy includes strong follow delivery and weak transition delivery.

[0018] Figure 2 A flowchart of a big data-based targeted media promotion method provided in this embodiment of the invention is shown below. Figure 2 As shown, the method includes: S100, collect application switching logs from user terminals in real time. The application switching logs include switching time, source application ID, target application ID, source application type, and target application type. By leveraging the event listening interface of the terminal operating system to capture application process switching events, the source application identifier corresponding to the exiting application and the target application identifier corresponding to the entering application are extracted at the moment the switching action is triggered. The precise timestamp of the switching event is recorded synchronously. Then, a pre-built category mapping table is called to complete the mapping conversion from application identifier to application type. The mapping table is built based on the category tags of mainstream app stores, covering all preset categories such as image, game, document, development, social, and information flow. Finally, the switching time, source application identifier, target application identifier, source application type, and target application type are encapsulated into a structured application switching log and stored in a dedicated log database.

[0019] This step acquires native behavioral data from the terminal system level. The data collection process does not require the additional implantation of third-party tracking points, ensuring the authenticity and time accuracy of the behavioral data. The standardized type mapping system unifies the classification dimensions of different applications, providing a unified basis for comparing and calculating user behavior patterns across applications. The entire collection mechanism can cover the switching behavior of all terminal applications, fully reconstructing the user's attention transfer path between different applications, supporting the calculation of user cognitive features across the entire chain, and promoting the upgrade of the targeting dimension from single application granularity to application type cognitive granularity.

[0020] When users switch between different applications on mobile devices, the system can complete event capture, data parsing, and log storage within a very short time window. All switching behaviors are automatically included in the data collection scope without the need for users to actively submit information, thus completing the accumulation of full behavioral data while ensuring user experience.

[0021] S200, based on the application switching log, construct and update a cognitive inertia graph for each user, the cognitive inertia graph containing directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights; Based on application switching logs, a cognitive inertia graph is constructed and updated for each user. This step first initializes a directed weighted graph as the basic structure of the graph. The nodes in the graph correspond to all application types defined by a preset category mapping table. Then, the full switching logs of the corresponding user are read from the log database, and the source application type and target application type in each log are parsed one by one. For each pair of types identified, the switching frequency count of the corresponding directed edge is updated. At the same time, the average switching interval between the most recent switching behaviors between the corresponding types is calculated based on the timestamps in the logs, and this is used as the interval weight of the directed edge. Finally, the switching intensity is calculated by combining the updated switching frequency and the average switching interval through a weighted sum of the switching frequency and the reciprocal of the average switching interval. The switching intensity is used as the core intensity weight of the directed edge, completing the incremental update of the entire graph.

[0022] The Cognitive Inertia Map transforms users' fragmented switching behaviors into structured, graph-like cognitive features. It quantifies the degree of cognitive correlation between different application types from two dimensions: total behavior and time density. It fully depicts users' attention shifting inertia in different application scenarios. The map adopts an incremental update mechanism, which can continuously iterate as user behavior data accumulates, always keeping up with the latest user behavior habits and cognitive patterns. The entire map system transforms users' cognitive inertia, which cannot be directly observed, into quantifiable and callable numerical indicators, giving the deployment decisions based on users' cognitive states a feasible computational basis.

[0023] As users use different types of applications over a long period of time, the graph will update its weight values ​​in sync with each application switching behavior, continuously accumulating user behavioral characteristics, and finally forming a unique personal cognitive behavior profile, providing personalized judgment basis for matching advertising strategies.

[0024] S300, identify task switching gap events that meet predetermined conditions from the application switching log. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, query the current user's cognitive inertia graph, obtain the weight of the directed edge from the source application type to the information flow type, and determine the attention residual depth corresponding to this switch based on the weight. This step identifies task switching gap events from application switching logs and determines the corresponding attention residual depth. It scans all switching logs for the corresponding user in the log database, filters out log records where the target application type is news feed and the source application type is not news feed, and marks these as task switching gap events. Each event corresponds to one switching log. For each identified gap event, the corresponding source application type is extracted, the current user's cognitive inertia graph is queried, and the switching intensity weight corresponding to the directed edge from the source application type to the news feed type is read. The switching intensity weight is compared with a preset intensity threshold. If the weight is higher than the threshold, the attention residual depth corresponding to this switch is determined to be higher than the preset level; otherwise, it is determined to be lower than the preset level.

[0025] This step anchors the key node when a user enters the information feed scenario, taking the moment when the user switches from other applications to the information feed as the time window for ad delivery. By using the switching intensity index in the cognitive inertia graph to indirectly reflect the degree of residual previous cognitive state of the user, the cognitive state can be judged in real time without additional subjective feedback from the user. The entire recognition mechanism can complete the state determination at the moment the switching event occurs, accurately capturing the key gap of user attention shift, and providing a dual judgment basis of timing and state for differentiated ad delivery.

[0026] During daily use of the terminal, whenever a user switches from a non-information feed application to an information feed application, the system will automatically trigger an event recognition and status determination process. This process assesses the user's attention status during the loading of information feed content, allowing sufficient decision-making time for content delivery.

[0027] S400, construct a media promotion content library, determine the corresponding expected attention retention mode according to the source application type, and establish an expected delivery mapping with the cognitive mode tag based on the expected attention retention mode. The promotion content in the constructed media promotion content library is pre-labeled with cognitive mode tags, and the cognitive mode tags include visual, logical and interactive. The process involves building a media promotion content library and establishing a target audience mapping relationship. This step receives uploaded media promotion content of various types and assigns a corresponding cognitive pattern tag to each piece of content. Content containing dynamic visual media corresponds to a visual tag, content containing structured text information components corresponds to a logical tag, and content containing interactive controls corresponds to an interactive tag. All content is then associated with its corresponding tag and stored in the content library. Subsequently, based on preset matching rules, a target attention residual pattern is matched for each type of source application: image-based and game-based applications correspond to visual attention residual, document-based and development-based applications correspond to logical attention residual, and social applications correspond to social attention residual. Then, a mapping is established between each attention residual pattern and the cognitive pattern tags in the content library: visual attention residual corresponds to a visual tag, logical attention residual corresponds to a logical tag, and social attention residual corresponds to an interactive tag, forming a complete target audience mapping system.

[0028] This step completes the standardized labeling of content-side cognitive attributes and the establishment of matching rules for user-side cognitive states. It establishes a clear matching link between user cognitive states and promotional content, ensuring that each campaign under different cognitive states has a corresponding content pool to support it. The pre-fixed mapping rules simplify the content selection logic during the campaign process and improve the execution efficiency of campaign decisions. At the same time, the clearly categorized content tag system also provides clear content selection boundaries for differentiated campaign strategies, ensuring that content under different campaign strategies can match the corresponding cognitive attributes.

[0029] All promotional materials can be attribute-labeled during the data entry stage. Content with different presentation formats is automatically categorized into corresponding tags. With the help of preset mapping rules, it can directly adapt to the needs of different scenarios without the need to perform content attribute judgment on a temporary basis before each campaign.

[0030] S500, when the task switching interval event is detected, the delivery strategy is dynamically selected based on the attention residual depth corresponding to this switch. The delivery strategy includes strong follow delivery and weak transition delivery.

[0031] When a task switching gap event is detected, a delivery strategy is dynamically selected based on the depth of attention persistence. This step involves obtaining the depth of attention persistence for the event when it is triggered. If the depth of attention persistence is higher than a preset threshold, strong follow-up delivery is executed. The source application type corresponding to the event is extracted, and the corresponding expected attention persistence mode is matched. The corresponding cognitive mode tag is queried through the expected delivery mapping relationship. Promotional content with the corresponding tag is selected from the content library to form a first candidate set, and the content in the candidate set is inserted into the target application's content stream for priority delivery. If the depth of attention persistence is not higher than the preset threshold, weak transition delivery is executed. Promotional content with no mapping relationship to the corresponding expected attention persistence mode is selected from the content library to form a second candidate set, which is also inserted into the target application's content stream for priority delivery.

[0032] After the candidate set is generated, sorting optimization is required. The publication time and historical display frequency of each item are extracted. The timeliness weight is calculated by the interval between the publication time and the current time. The historical display frequency is used as historical conversion data. At the same time, the switching frequency of the corresponding source application type node is read from the cognitive inertia graph as the basis for calculating the strength of user preference. Finally, the timeliness weight, historical conversion data and switching frequency are combined to complete the weighted sorting, and the content with the highest ranking is selected as the final target for delivery.

[0033] This step executes a differentiated delivery strategy based on users' real-time attention state, deeply matching users' current cognitive acceptance tendencies. When attention is high, content with similar cognitive patterns is matched to lower the cognitive threshold and increase content acceptance. When attention is low, differentiated content is introduced to break cognitive inertia, enhancing content freshness and user attention capture. The three-dimensional weighted ranking mechanism takes into account both the freshness and historical performance of the content and users' personalized preferences, ensuring that the final delivered content meets multiple requirements of timeliness, conversion effect, and personalized matching. The entire dynamic delivery mechanism is driven entirely by real-time user behavior data, upgrading the delivery strategy from uniform push to personalized strategies for each user, comprehensively improving the accuracy and conversion efficiency of media promotion.

[0034] When users switch to the news feed application at different times and in different scenarios, the system will automatically match the corresponding delivery strategy and content. Each delivery is tailored to the user's current cognitive state. Over the long term, the matching degree between the delivery content and the user's state can be continuously optimized, improving the reach of promotional content without interfering with the user's normal use.

[0035] like Figure 3 As shown, the real-time collection of application switching logs from user terminals specifically includes: S110 obtains application switching events through the event listening interface of the terminal operating system, extracts the source application ID that exits and the target application ID that enters when the switching occurs, and records the switching time. S120, invoke a pre-established preset category mapping table to map the source application ID to the source application type and the target application ID to the target application type. The preset category mapping table is established based on the application store category tags. The value range of the source application type and the target application type includes image type, game type, document type, development type, social type and information flow type. S130 encapsulates the switching time, source application ID, target application ID, source application type, and target application type into an application switching log and stores it in the application switching log database.

[0036] like Figure 4 As shown, the construction and updating of a cognitive inertia map for each user specifically includes: S210, initialize a directed weighted graph as a cognitive inertia graph, in which the nodes are all application types defined in the preset category mapping table; Users' app switching behavior is not random. The patterns of switching essentially reflect users' cognitive habits, attention shift patterns, and content preferences. However, a single switching behavior is easily affected by accidental factors and cannot stably depict users' long-term behavioral characteristics and implicit cognitive inertia.

[0037] The specific functions of cognitive inertial mapping include: 1. Structure user's scattered application switching behavior data and quantify the cognitive association strength between different application types in the form of a directed weighted graph to form a calculable and incrementally updatable user cognitive behavior profile; 2. To provide a quantitative data foundation for subsequent judgment of attention residual depth during task switching, transforming the user's cognitive state, which cannot be directly observed, into an objective indicator that can be calculated through behavioral data; 3. It connects multiple system modules such as log collection, event recognition, and strategy selection, serving as the underlying cognitive feature carrier of the entire precision targeting system to ensure the consistency and accuracy of the targeting logic.

[0038] S220, Read all application switching logs of the current user from the application switching log database, and parse out the source application type and target application type for each log entry; S230, for each pair of source application type and target application type, update the switching frequency of the directed edge from the source application type node to the target application type node in the cognitive inertia graph, and increment the switching frequency by 1; S240, calculate the average switching interval of the most recent multiple switchings from the source application type to the target application type based on the switching time in the application switching log, and update the average switching interval to the average switching interval weight of the corresponding directed edge. ; in: The average switching interval weight from the source application type to the target application type is the average switching interval attribute value of the corresponding directed edge in the cognitive inertia graph. This represents the total number of recent switching actions by the selected user, from the source application type to the target application type. The first in ascending order of time The switching timestamp corresponding to this type of switching behavior. For the first The switching timestamp corresponding to this type of switching behavior. The time interval between two consecutive switches of the same type.

[0039] Switching frequency alone can only measure the total number of switching behaviors, but cannot reflect the intensity of switching behaviors over time. The time interval is a key dimension that reflects the strength of user behavior inertia. For the same number of switching, there is an essential difference in the strength of cognitive inertia corresponding to short-term intensive switching and long-term scattered switching.

[0040] Calculating the average switching interval can supplement the quantification of users' switching behavior characteristics from the perspective of time density. The shorter the average switching interval, the more frequently and consistently users switch between two types of applications, indicating a stronger cognitive association and a stronger inertia in attention shifting. This metric can effectively distinguish between "high switching frequency accumulated over a long period of time" and "high switching frequency occurring in a short period of time," avoiding the evaluation bias caused by using switching frequency alone. It provides time dimension support for subsequent calculations of switching intensity, making the characterization of cognitive inertia more closely reflect the user's actual behavioral state.

[0041] S250: Based on the updated switching frequency and average switching interval, calculate the switching intensity by weighted summation of the reciprocal of the switching frequency and the average switching interval, update the switching intensity to the switching intensity weight of the corresponding directed edge, and complete the incremental update of the cognitive inertia map. ; in: The switching intensity weight from the source application type to the target application type is the switching intensity attribute value of the corresponding directed edge in the cognitive inertia graph. The updated switching frequency is the total number of times the user has switched from the source application type to the target application type. The average switching interval weight for the corresponding directed edge; The preset weighting coefficients for switching frequency dimensions, These are preset weighting coefficients for the inverse dimension of the average handover interval. Both are positive preset constants used to adjust the contribution ratio of the two indicators in the handover intensity calculation.

[0042] Switching frequency and average switching interval reflect the characteristics of switching behavior from two independent dimensions: "total quantity" and "time density". A single indicator cannot fully measure the cognitive association strength between the two types of applications: looking only at switching frequency, it is impossible to distinguish between short-term intensive switching and long-term sporadic switching; looking only at switching interval, it is impossible to reflect the long-term cumulative effect of behavior.

[0043] By using a weighted sum, information from both dimensions can be integrated to more comprehensively and accurately quantify the strength of a user's cognitive inertia when switching between two types of applications. Switching frequency reflects the long-term accumulation of this switching pattern; higher frequency indicates a more stable behavioral pattern. The reciprocal of the average switching interval reflects the temporal density of the switching; shorter intervals result in a larger reciprocal, indicating stronger recent behavioral inertia. The weighting coefficients allow for flexible adjustment of the weights of the two dimensions to adapt to different business scenarios. The final output switching strength more reliably reflects the strength of a user's cognitive inertia, providing accurate quantitative evidence for subsequent assessment of attention retention depth.

[0044] like Figure 5 As shown, determining the attention residual depth corresponding to this switch based on the weight specifically includes: S310, Scan the application switching logs of the current user in the application switching log database, select log records where the target application type is information flow and the source application type is not information flow, mark the selected log records as task switching interval events, and each task switching interval event corresponds to one application switching log. S320: For each marked task switching gap event, parse the source application type from the application switching log corresponding to the task switching gap event, query the current user's cognitive inertia graph, and read the switching intensity weight of the directed edge from the source application type node to the information flow type node. S330: Compare the switching intensity weight with the preset intensity threshold. If the switching intensity weight is greater than the preset intensity threshold, it is determined that the attention residual depth corresponding to the task switching gap event is higher than the preset intensity threshold; otherwise, it is determined that the attention residual depth is not higher than the preset intensity threshold.

[0045] The theory of attentional residue in cognitive psychology states that when a person switches from one task to another, some cognitive resources remain in the preceding task, forming attentional residue. Furthermore, the more frequent and coherent the switching behavior, the higher the cognitive activation level of the preceding task, and the greater the amount of attentional residue. When a user switches from a non-feedback app to a feed app, the higher the intensity of the switch and the more stable this switching pattern, the deeper the attentional residue from the preceding app.

[0046] This step establishes a quantitative mapping relationship between "observable user behavior data" and "indirectly observable user cognitive states." It indirectly infers the user's real-time attention persistence depth through objective switching intensity indicators, enabling cognitive state judgment without additional subjective user feedback. Based on this, the system can dynamically adjust its delivery strategy according to the user's real-time cognitive state, breaking away from the traditional uniform delivery model and achieving refined and precise delivery based on user cognitive states, thereby improving the acceptance and conversion rate of promotional content.

[0047] like Figure 6 As shown, establishing the expected projection mapping with the cognitive pattern label specifically includes: S410 receives uploaded media promotion content and assigns cognitive mode tags to each piece of media promotion content. If it contains dynamic visual media, it is labeled as visual; if it contains text structured information components, it is labeled as logical; and if it contains interactive controls, it is labeled as interactive. S420 will associate and store each identified media promotion content and its corresponding cognitive pattern tag in the media promotion content library. S430, based on the preset correspondence rules between source application type and expected attention residual mode, determines the corresponding expected attention residual mode for each source application type. If the source application type is image type or game type, the expected attention residual mode is visual attention residual; if the source application type is document type or development type, the expected attention residual mode is logical attention residual; if the source application type is social type, the expected attention residual mode is social attention residual. S440, map each desired attention residual pattern to the cognitive pattern tag in the media promotion content library to establish a desired delivery mapping relationship, wherein visual attention residual maps to visual type, logical attention residual maps to logical type, and social attention residual maps to interactive type.

[0048] Different types of source applications trigger different attention persistence patterns (e.g., visual attention persistence for image / game applications and logical attention persistence for document / development applications), and promotional content also corresponds to different cognitive pattern tags. The two need to establish a matching relationship to achieve the adaptation of content to the user's current cognitive state.

[0049] Establishing a target delivery mapping enables precise matching between user cognitive states and promotional content, ensuring that the delivered content aligns with the user's current attention patterns, reducing cognitive load, and improving content acceptance and conversion rates. This step aims to pre-define matching rules, simplifying content filtering logic during delivery. Upon detecting a switching event, the system can quickly filter content with corresponding tags directly through the mapping relationship, improving system response speed and operational efficiency. This step also provides clear content filtering criteria for both strong follow-up and weak transition strategies: strong follow-up selects content with mapping matches, while weak transition selects content with non-matching mappings, ensuring the feasibility of both strategies.

[0050] like Figure 7 As shown, the dynamically selected delivery strategy specifically includes: S510, when a task switching gap event is detected, obtain the judgment result of whether the attention residual depth corresponding to the task switching gap event is higher than a preset depth threshold; S520, if the attention residual depth is higher than the preset depth threshold, then strong follow-up delivery is performed: the source application type is parsed from the application switching log corresponding to the task switching gap event, and according to the expected attention residual mode corresponding to the source application type, the cognitive mode tag with the expected delivery mapping relationship with the expected attention residual mode is queried in the expected delivery mapping relationship, the promotion content with the cognitive mode tag is selected from the media promotion content library, a first candidate set is generated, and the promotion content in the first candidate set is inserted into the content stream of the target application corresponding to the task switching gap event for priority delivery; S530, if the attention residual depth is not higher than the preset depth threshold, then perform weak transition delivery: select promotional content from the media promotion content library whose cognitive mode tags and the expected attention residual modes corresponding to the source application type do not have the expected delivery mapping relationship, generate a second candidate set, and insert the promotional content in the second candidate set into the content stream of the target application corresponding to the task switching gap event for priority delivery.

[0051] Users exhibit significant differences in their acceptance of promotional content based on their level of attention retention: when attention retention is deep, users are still in the cognitive mode of the previous application, and have a lower cognitive threshold and higher acceptance of similar content; when attention retention is shallow, users have basically left the previous cognitive state, and similar content is more likely to cause aesthetic fatigue, while differentiated content is more attractive.

[0052] A differentiated dynamic delivery strategy is implemented based on users' real-time cognitive states, balancing delivery effectiveness and user experience. For scenarios with high attention retention, a strong follow-up delivery approach is used to align with the user's current cognitive pattern and maximize content conversion efficiency. For scenarios with low attention retention, a weak transition delivery approach breaks cognitive inertia, avoids user aversion caused by content homogenization, and simultaneously expands the user's attention span, enhancing the content's freshness. The combination of these two strategies achieves superior delivery results in different scenarios.

[0053] like Figure 7 As shown, after generating the first candidate set and the second candidate set, a candidate sorting step is also included: S540, for each promotional content in the candidate set, extract the publication time and historical display count of the promotional content from the media promotional content library, calculate the timeliness weight based on the time interval between the publication time and the current time, and use the historical display count as historical conversion data; The timeliness weight is calculated using the exponential decay model commonly used in recommendation systems, which aligns with the value decay pattern of media content: ; in: The timeliness weight of a single promotional content item, with a value range of [value range missing]. The weight of content is 1 at the moment of publication, and gradually decreases over time; This is the preset time-sensitivity decay coefficient, which is a positive constant. The larger the value, the faster the time interval decays the weight. This is the time interval between the current system time and the time when the promotional content was published, usually in hours or days.

[0054] In the design logic of this system, the number of times the promoted content has been displayed in history is a direct proxy indicator of its historical performance: the more times a piece of content can be displayed in history, the more rounds of validation it has undergone, and the more stable audience acceptance and performance base it has.

[0055] The purpose of using historical impression counts as historical conversion data is to serve as a quantitative evaluation dimension of content quality and historical performance. When ranking candidate content, priority is given to content that has been validated by the market and has a more stable historical performance, thereby reducing the risk of launching new or low-quality content and ensuring the stability and lower limit of the overall campaign performance. At the same time, historical impression data is easy to collect and has low computational costs, which can meet the performance requirements of real-time ranking.

[0056] S550, read the switching frequency of the source application type node corresponding to the task switching interval event from the current user's cognitive inertia map, and use it as the basis for calculating the user's preference strength for the source application type; The higher the frequency of switching source application types, the more frequently users use this type of application, the more time and attention they invest, and the higher their acceptance of the cognitive patterns corresponding to this type of application. Essentially, this reflects the strength of users' preference for this type of application and its corresponding content.

[0057] Using switching frequency as a measure of preference strength allows for the integration of users' long-term behavioral preferences into the content ranking logic. This ensures that the ranking results not only consider the attributes of the content itself but also align with the user's current personalized behavioral characteristics. The stronger a user's preference for the source application type, the higher the weight of content of that type. Ultimately, the selected content will better match the user's long-term interests, further enhancing the personalization and accuracy of the targeting.

[0058] S560 uses a weighted sorting system based on timeliness, historical conversion data, and switching frequency to select the highest-ranking promotional content as the final delivery target. ; in: The score is the overall ranking score of a single promotional content. The higher the score, the higher the ranking in the candidate set, and the more likely it is to be selected as the final target for delivery. The timeliness of the promoted content will be given a weight. This refers to the historical number of times the promotional content has been displayed, i.e., the historical conversion data defined in the plan; This represents the cumulative switching frequency of the source application type node corresponding to this task switching interval event in the current user cognitive inertia graph. , , These are preset weighting coefficients corresponding to timeliness weight, historical conversion data, and switching frequency, all of which are positive constants used to balance the influence of the three dimensions in the ranking.

[0059] The three metrics correspond to the three core dimensions of content ranking. However, a single metric can only evaluate the value of content placement from a single dimension, which has obvious limitations. Timeliness weighting reflects the freshness of content and ensures its immediate appeal, but it cannot reflect the historical impact of content and user preferences. Historical conversion data reflects the historical performance of content delivery, ensuring the basic effectiveness of delivery, but it cannot take into account both content freshness and user personalization. Switching frequency reflects users' personalized preferences and improves the adaptability of the campaign, but it cannot guarantee the quality and timeliness of the content itself.

[0060] The combination of these three factors can take into account the three core dimensions of content freshness, historical performance, and user personalization preferences. By balancing the influence of each dimension through weighting, the final ranking results are more scientific and comprehensive. This ensures both the timeliness and conversion effect of the promoted content and aligns with users' personal behavior habits, ultimately achieving an overall improvement in the accuracy of ad targeting and conversion efficiency.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A media promotion targeted and precise delivery system based on big data, characterized in that: The system includes: The log collection module is used to collect application switching logs from user terminals in real time. The application switching logs include switching time, source application ID, target application ID, source application type, and target application type. An inertial map construction module is used to construct and update a cognitive inertial map for each user based on the application switching logs. The cognitive inertial map includes directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights. The event recognition module is used to identify task switching gap events that meet predetermined conditions from the application switching log. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, the module queries the current user's cognitive inertia graph to obtain the weight of the directed edge from the source application type to the information flow type, and determines the attention residual depth corresponding to this switch based on the weight. The content library construction module is used to build a media promotion content library. It determines the corresponding expected attention retention mode based on the source application type, and establishes an expected delivery mapping with the cognitive mode tag based on the expected attention retention mode. The promotion content of the media promotion content library is pre-labeled with cognitive mode tags, which include visual, logical and interactive types. The strategy selection module is used to dynamically select a delivery strategy based on the attention residual depth corresponding to the task switching interval event when the event is detected. The delivery strategy includes strong follow delivery and weak transition delivery.

2. A method for targeted and precise media promotion based on big data, characterized in that: The method includes: Real-time collection of application switching logs from user terminals, including switching time, source application ID, target application ID, source application type, and target application type; Based on the application switching logs, a cognitive inertia graph is constructed and updated for each user. The cognitive inertia graph contains directed edges with different application types as nodes and the switching intensity, switching frequency and average switching interval between application types as weights. Identify task switching gap events that meet predetermined conditions from the application switching logs. The predetermined conditions are that the target application type is information flow type and the source application type is not information flow type. For the identified task switching gap events, query the current user's cognitive inertia graph to obtain the weight of the directed edge from the source application type to the information flow type. Determine the attention residual depth corresponding to this switch based on the weight. A media promotion content library is constructed. Based on the source application type, the corresponding expected attention retention pattern is determined, and an expected delivery mapping is established between the expected attention retention pattern and the cognitive pattern tag. The promotion content in the constructed media promotion content library is pre-labeled with cognitive pattern tags, which include visual, logical, and interactive types. When the task switching interval event is detected, a delivery strategy is dynamically selected based on the attention residual depth corresponding to this switch. The delivery strategy includes strong follow delivery and weak transition delivery.

3. The method according to claim 2, characterized in that, The real-time collection of application switching logs from user terminals specifically includes: The application switching event is obtained through the event listening interface of the terminal operating system. The source application ID that was exited and the target application ID that was entered when the switch occurred are extracted, and the switching time is recorded. The system calls a pre-established category mapping table to map the source application ID to the source application type and the target application ID to the target application type. The pre-established category mapping table is based on the application store category tags. The value range of the source application type and the target application type includes image, game, document, development, social and information flow categories. The switching time, source application ID, target application ID, source application type, and target application type are encapsulated into an application switching log and stored in the application switching log database.

4. The method according to claim 3, characterized in that, The process of constructing and updating a cognitive inertia map for each user specifically includes: Initialize a directed weighted graph as a cognitive inertia graph, where the nodes are all application types defined in a preset category mapping table; Read all application switching logs of the current user from the application switching log database, and parse out the source application type and target application type for each log entry. For each pair of source application type and target application type, update the switching frequency of the directed edge from the source application type node to the target application type node in the cognitive inertia graph, and increment the switching frequency by 1; Based on the switching time in the application switching log, calculate the average switching interval of the most recent switching from the source application type to the target application type, and update the average switching interval to the average switching interval weight of the corresponding directed edge. Based on the updated switching frequency and average switching interval, the switching intensity is calculated by weighted summation of the reciprocals of the switching frequency and the average switching interval. The switching intensity is then updated to the switching intensity weight of the corresponding directed edge, thus completing the incremental update of the cognitive inertia map.

5. The method according to claim 4, characterized in that, The step of determining the attention residual depth corresponding to this switch based on the weight specifically includes: Scan the application switching logs of the current user in the application switching log database, select log records where the target application type is information flow and the source application type is not information flow, mark the selected log records as task switching interval events, and each task switching interval event corresponds to one application switching log. For each marked task switching gap event, the source application type is parsed from the application switching log corresponding to the task switching gap event, the current user's cognitive inertia graph is queried, and the switching intensity weight of the directed edge from the source application type node to the information flow type node is read. The switching intensity weight is compared with the preset intensity threshold. If the switching intensity weight is greater than the preset intensity threshold, it is determined that the attention residual depth corresponding to the task switching gap event is higher than the preset intensity threshold; otherwise, it is determined that the attention residual depth is not higher than the preset intensity threshold.

6. The method according to claim 5, characterized in that, The establishment of the expected projection mapping with the cognitive pattern label specifically includes: Receive uploaded media promotion content and assign cognitive mode tags to each piece of media promotion content. If it contains dynamic visual media, it is labeled as visual; if it contains text structured information components, it is labeled as logical; if it contains interactive controls, it is labeled as interactive. Each identified media promotion content item and its corresponding cognitive pattern tag will be associated and stored in the media promotion content library; Based on the preset correspondence rules between source application types and expected attention residual patterns, a corresponding expected attention residual pattern is determined for each source application type. If the source application type is an image-based or game-based application, the expected attention residual pattern is visual attention residual; if the source application type is a document-based or development-based application, the expected attention residual pattern is logical attention residual; if the source application type is a social application, the expected attention residual pattern is social attention residual. Each desired attention residual pattern is mapped to a cognitive pattern tag in the media promotion content library to establish a desired delivery mapping relationship, wherein visual attention residual maps to visual type, logical attention residual maps to logical type, and social attention residual maps to interactive type.

7. The method according to claim 6, characterized in that, The dynamically selected delivery strategy specifically includes: When a task switching gap event is detected, the judgment result is obtained as to whether the attention residual depth corresponding to the task switching gap event is higher than a preset depth threshold. If the attention residual depth is higher than the preset depth threshold, then strong follow-up delivery is performed: the source application type is parsed from the application switching log corresponding to the task switching gap event, and according to the expected attention residual mode corresponding to the source application type, the cognitive mode tag with the expected delivery mapping relationship with the expected attention residual mode is queried in the expected delivery mapping relationship, the promotion content with the cognitive mode tag is screened from the media promotion content library, a first candidate set is generated, and the promotion content in the first candidate set is inserted into the content stream of the target application corresponding to the task switching gap event for priority delivery; If the attention residual depth is not higher than the preset depth threshold, weak transition delivery is performed: promotional content whose cognitive pattern tags and the expected attention residual patterns corresponding to the source application type do not have the expected delivery mapping relationship are selected from the media promotion content library, a second candidate set is generated, and the promotional content in the second candidate set is inserted into the content stream of the target application corresponding to the task switching gap event for priority delivery.

8. The method according to claim 7, characterized in that, After generating the first candidate set and the second candidate set, the process also includes a candidate sorting step: For each promotional content in the candidate set, the publication time and historical display count of the promotional content are extracted from the media promotional content library. The timeliness weight is calculated based on the time interval between the publication time and the current time, and the historical display count is used as historical conversion data. From the current user's cognitive inertia map, read the switching frequency of the source application type node corresponding to the task switching gap event, and use it as the basis for calculating the user's preference strength for the source application type; The promotional content is selected based on a weighted ranking system that considers timeliness, historical conversion data, and switching frequency. The highest-ranking content is then chosen as the final target for ad placement.

Images