An intelligent media delivery and multi-dimensional integration linkage system based on an internet of things

Abstract

The application discloses a kind of wisdom media delivery and multi-dimensional integral linkage system based on Internet of Things, comprising: three-dimensional context generation module, user behavior data, equipment state data and environmental context information are collected, and three-dimensional context tensor is constructed and expanded to generate characteristic vector;Multi-target integral fusion module, viewing integral, interactive integral and propagation integral are weighted and fused to generate multi-target integral return value;Content path dependence modeling module, adjust content path dependence structure matrix to represent sequential correlation;Improved LinUCB score module, based on context characteristic vector, multi-target integral return value and content path dependence structure matrix, combined with integral volatility, integral drift rate and path coupling to score and sort content;Content push module pushes the highest score content to Internet of Things equipment, and integral feedback collection and update module realizes parameter dynamic update.The application improves the intelligent level of wisdom media delivery and integral linkage.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to an IoT-based smart media delivery and multi-dimensional points linkage system. Background Technology

[0002] Currently, the development of the Internet of Things and smart media platforms has driven the widespread application of content distribution and intelligent recommendation. Most existing smart media delivery systems are based on the collection of basic user attributes or single behaviors, combined with traditional recommendation methods such as collaborative filtering and tag matching to push content. However, they are difficult to dynamically reflect the multidimensional needs and interest changes of users in real-world scenarios, and their comprehensive utilization of information such as device status and environmental changes is also relatively limited, making it difficult to achieve accurate recommendations for different terminals and real-time situations.

[0003] Currently, most point-based incentive mechanisms use fixed rules to award static points to user behaviors such as browsing, interacting, or sharing. They lack refined modeling of the actual impact of personalized user behavior and content, and the point weights are difficult to adjust adaptively, resulting in insufficient flexibility and effectiveness of the incentive mechanism. At the same time, existing systems have limited ability to identify abnormal point distribution and device cheating, and are easily affected by abnormal operations such as point manipulation.

[0004] On the other hand, traditional content push priority calculation mainly relies on static features or single historical feedback, failing to effectively integrate multi-dimensional context and the sequential dependencies between content. Even when some platforms adopt algorithms such as LinUCB, they primarily focus on static input and single-objective optimization, lacking in-depth modeling of structural factors such as integral fluctuations and path coupling. Existing technologies struggle to support efficient content push and integral linkage in complex scenarios, failing to meet the actual needs of large-scale IoT smart media platforms.

[0005] Therefore, how to provide a smart media delivery and multi-dimensional points linkage system based on the Internet of Things is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a smart media delivery and multi-dimensional points linkage system based on the Internet of Things (IoT). This invention integrates multi-dimensional context awareness, points feedback fusion, content path dependency modeling, and an improved LinUCB scoring method. It describes in detail the system process for implementing content delivery priority ranking, dynamic points reward allocation, and user behavior feedback updates in an IoT scenario. It has the advantages of comprehensive information collection, accurate content push, and intelligent points linkage.

[0007] According to an embodiment of the present invention, a smart media delivery and multi-dimensional points linkage system based on the Internet of Things includes: The 3D context generation module is used to periodically collect user behavior data, device status data, and environmental context information, construct a 3D context tensor, expand the tensor, and generate context feature vectors. The multi-objective integral fusion module is used to collect viewing points, interaction points, and dissemination points associated with the delivered content. It uses non-negative weighting coefficients to fuse the three types of integrals, generate a multi-objective integral reward value, and associate it with the context feature vector. The content path dependency modeling module is used to adjust the content path dependency structure matrix based on historical delivery sequences and user feedback results, representing the sequential relationship of delivered content; An improved LinUCB scoring module is used to calculate the confidence upper bound score of the delivered content by utilizing context feature vectors, multi-objective integral reward values, and content path dependency structure matrices, based on the LinUCB algorithm which introduces integral volatility, integral drift rate, path coupling term, L1 norm sparse regularization term, and perturbation penalty term. The content push module is used to sort the confidence upper bound scores of all delivered content, select the content with the highest score to push to IoT devices, and record the content identifier, push time, and device identifier. The points feedback collection and update module is used to collect user points feedback on the target content, update the three-dimensional context tensor, multi-objective points reward value, content path dependency structure matrix and improved LinUCB algorithm parameters, and use the update results in the content push and feedback collection process.

[0008] Optionally, modules can be integrated using the following methods: S1. Periodically collect user behavior data, device status data and environmental context information, construct a three-dimensional context tensor, update the three-dimensional context tensor in real time and expand the tensor to generate the context feature vector of the content to be delivered. S2. Collect the viewing points, interaction points, and dissemination points associated with each piece of content, and use a non-negative weighting coefficient to fuse the three types of points to generate the corresponding multi-objective point reward value. S3. Establish a content path dependency structure matrix, and dynamically adjust the matrix parameters according to the historical delivery sequence and user feedback results to represent the content sequence relationship. S4. Using the improved LinUCB algorithm, the confidence upper bound score of the delivered content is calculated by utilizing the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S5. Sort the confidence upper bound scores of all delivered content, select the content with the highest score as the target delivered content, and push it to the designated IoT device; S6. Collect actual point feedback from users on the target content, and update the three-dimensional context tensor, multi-objective point reward value, content path dependency structure matrix and related parameters of the improved LinUCB algorithm. S7. Based on the updated 3D context tensor, multi-objective integral reward value, content path dependency structure matrix, and relevant parameters of the improved LinUCB algorithm, continue to execute the content push and feedback collection process.

[0009] Optionally, the improved LinUCB algorithm includes: The algorithm inputs the context feature vector obtained by expanding the three-dimensional context tensor tensor and integrates user behavior data, device status data and environmental context information into a feature representation. Introducing integral volatility and integral drift rate, the integral variance and integral mean change rate of multi-objective integral return values ​​are calculated respectively within a set time window; The content path dependency structure matrix is ​​used to describe the sequential relationship of the delivered content, and the path coupling term is calculated through the content path dependency structure matrix. The influence of the sequential relationship between content is included in the confidence upper bound score. In the regression model, an L1 norm sparse regularization term and a perturbation penalty term are set to handle abnormal perturbations, fake integral feedback, and sparse abnormal behavior in the context feature vector and multi-objective integral reward value. The integral volatility, integral drift rate, path coupling term, and regression model output are all included in the confidence upper bound score to complete the dynamic ranking of content delivery.

[0010] Optionally, S1 specifically includes: S11. Collect user behavior data, including content browsing behavior, content click behavior, content interaction behavior, and content dissemination behavior; S12. Collect device status data, including the unique identifier of the IoT terminal, the current operating status of the device, and the network connection status; S13. Collect environmental context information, including the time information when the content is delivered and related environmental parameters; S14. Map user behavior data, device status data and environmental context information to the user dimension, device dimension and time dimension of the three-dimensional context tensor respectively, and construct the three-dimensional context tensor. S15. Perform tensor expansion on the three-dimensional context tensor to convert the three-dimensional structure into a one-dimensional context feature vector. Each component represents the content feature relationship corresponding to different combinations of user, device and time information.

[0011] Optionally, S2 specifically includes: S21. Collect the viewing points, interaction points, and dissemination points associated with each piece of content. Viewing points represent user feedback on browsing the content, interaction points represent user interaction with the content, and dissemination points represent feedback when the content is forwarded. S22. Assign non-negative weighting coefficients to the viewing points, interaction points, and dissemination points respectively, in order to determine the weight of each type of point in the multi-objective point reward value; S23. Combine the viewing points, interaction points and dissemination points in a weighted manner to obtain a multi-objective point reward value. The multi-objective point reward value is the result of the weighted sum of the three types of points. S24. Establish a correspondence between the multi-objective integral reward value and the context feature vector for use in the content scoring and ranking process.

[0012] Optionally, S3 specifically includes: S31. Based on the historical delivery sequence of the delivered content, record the push order of each delivered content and user points feedback to form content sequence association information; S32. Match user points feedback data with content sequence association information, and extract the association relationship of each content delivery in different time sequences; S33. Based on historical delivery sequences and user points feedback data, construct a content path dependency structure matrix. Each element of the content path dependency structure matrix is ​​used to represent the sequential dependency relationship between one delivery content and another delivery content. The rows and columns of the matrix correspond to each delivery content. S34. Based on the latest user points feedback data, dynamically adjust the parameters in the content path dependency structure matrix so that the content path dependency structure matrix reflects the changes in the content order relationship. The content path dependency structure matrix is ​​used to characterize the content order relationship and is used for the calculation of the confidence upper bound score.

[0013] Optionally, S4 specifically includes: S41. Take the context feature vector, multi-objective integral reward value and content path dependency structure matrix as input, and set the confidence interval adjustment coefficient, path dependency adjustment coefficient, integral volatility adjustment coefficient and integral drift rate adjustment coefficient. S42. Initialize the model parameter vector and feature correlation matrix for each delivered content, and set the initial value of the content path dependency structure matrix. S43. Statistically analyze the multi-objective points return value feedback data for each content delivery, calculate the variance of the points feedback sequence within a set time window, and record it as points volatility. S44. Statistically analyze the multi-target points return value feedback data for each delivered content, and calculate the rate of change of the mean of the points feedback sequence within a set time window, which is recorded as the points drift rate. S45. Based on the content path dependency structure matrix, calculate the sequential coupling term, which is the weighted sum of the corresponding row of the content path dependency structure matrix and the content score value. S46. Calculate the confidence upper bound score based on the context feature vector, multi-objective integral reward value, content path dependency structure matrix, integral volatility, and integral drift rate. S47. In the process of calculating the confidence upper bound score, set an L1 norm sparse regularization term and a perturbation penalty term to process abnormal perturbations, fake integral feedback and sparse abnormal behavior in the context feature vector and multi-objective integral reward value, and use the confidence upper bound score value for the ranking of delivered content.

[0014] Optionally, S5 specifically includes: S51. Collect the confidence upper bound score value corresponding to each content delivery. The confidence upper bound score value is calculated by the improved LinUCB algorithm based on the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S52. Sort the confidence upper bound scores of all the delivered content, arrange them from high to low scores, and determine the ranking position of each delivered content. S53. Determine the content with the highest score as the target content. The target content is the content with the highest confidence upper bound score. S54. Push the target content to the designated IoT device, record the content identifier, push time, and IoT device identifier of the target content, and retain relevant parameter information for points feedback collection and parameter updates.

[0015] Optionally, S6 specifically includes: S61. Record user points feedback data related to the target content, including viewing points, interaction points, and dissemination points corresponding to content identifiers, push times, and IoT device identifiers; S62. Based on the user, device, and time information in the integral feedback data, fill the data into the corresponding three-dimensional context tensor. S63. Based on the integral feedback data, adjust the non-negative weighting coefficients of each item, calculate the new multi-objective integral reward value, and map the new reward value to the context feature vector; S64. Using integral feedback data, correct the element values ​​of the content path dependency structure matrix so that the content path dependency structure matrix represents the actual feedback relationship of the current content delivery order. S65. Apply integral feedback data to adjust the relevant parameters of the improved LinUCB algorithm. The relevant parameters include the model parameter vector, feature correlation matrix, integral volatility adjustment coefficient, integral drift rate adjustment coefficient, path dependence adjustment coefficient, confidence interval adjustment coefficient, L1 norm sparse regularization term parameter, and perturbation penalty term parameter.

[0016] The beneficial effects of this invention are: This invention, through the design of a three-dimensional context tensor, organically integrates user behavior, device status, and environmental information. This allows content delivery to adaptively adjust based on the actual scenario, the user's current state, and the specific terminal type, enabling compatibility with changes in user interests and multi-scenario collaboration, thus overcoming the limitations of the traditional single static data-driven model. Based on multi-objective integral fusion and a content path-dependent structure matrix, dynamic linkage is achieved between integral feedback and content order. The system can optimize integral allocation and push logic in real time according to different users and content types, and the priority ranking method is more closely aligned with actual interactive behavior.

[0017] An improved LinUCB scoring method is adopted, which fully considers integral volatility, integral drift rate, and sequential coupling between content. The confidence upper bound score is more sensitive to integral anomalies, behavioral fluctuations, and device interference. The system can promptly identify and correct unreasonable feedback, improving the overall stability and robustness of content delivery. The L1 norm sparse regularization term and perturbation penalty term mechanism in the algorithm have a certain tolerance for integral anomalies and forged data, which helps maintain the healthy operation of the integral incentive mechanism.

[0018] In summary, this invention not only improves the real-time and personalized nature of content distribution, but also makes the points-based linkage more scientific and transparent. The entire system has better adaptability and operational efficiency in large-scale IoT scenarios, providing users with a richer and more tailored content push experience, while also giving platforms more flexible space to adjust their business strategies. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an IoT-based smart media delivery and multi-dimensional points linkage system proposed in this invention; Figure 2 This is a schematic diagram of the three-dimensional context tensor generation process of a smart media delivery and multi-dimensional integral linkage system based on the Internet of Things proposed in this invention. Figure 3 This is an improved LinUCB scoring flowchart for an IoT-based smart media delivery and multi-dimensional scoring linkage system proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-3 A smart media delivery and multi-dimensional points linkage system based on the Internet of Things includes: The 3D context generation module is used to periodically collect user behavior data, device status data, and environmental context information, construct a 3D context tensor, expand the tensor, and generate context feature vectors. The multi-objective integral fusion module is used to collect viewing points, interaction points, and dissemination points associated with the delivered content. It uses non-negative weighting coefficients to fuse the three types of integrals, generate a multi-objective integral reward value, and associate it with the context feature vector. The content path dependency modeling module is used to adjust the content path dependency structure matrix based on historical delivery sequences and user feedback results, representing the sequential relationship of delivered content; An improved LinUCB scoring module is used to calculate the confidence upper bound score of the delivered content by utilizing context feature vectors, multi-objective integral reward values, and content path dependency structure matrices, based on the LinUCB algorithm which introduces integral volatility, integral drift rate, path coupling term, L1 norm sparse regularization term, and perturbation penalty term. The content push module is used to sort the confidence upper bound scores of all delivered content, select the content with the highest score to push to IoT devices, and record the content identifier, push time, and device identifier. The points feedback collection and update module is used to collect user points feedback on the target content, update the three-dimensional context tensor, multi-objective points reward value, content path dependency structure matrix and improved LinUCB algorithm parameters, and use the update results in the content push and feedback collection process.

[0022] This invention achieves multi-dimensional perception and fusion of user behavior, device status, and environmental information by constructing a three-dimensional context generation module, a multi-objective integral fusion module, a content path dependency modeling module, and an improved LinUCB scoring module. It provides a complete data link for content push and points allocation, enhances the real-time response and self-learning capabilities of the smart media platform, and meets the needs of personalized content distribution in complex IoT scenarios.

[0023] In this embodiment, the modules are interconnected using the following method: S1. Periodically collect user behavior data, device status data and environmental context information, construct a three-dimensional context tensor, update the three-dimensional context tensor in real time and expand the tensor to generate the context feature vector of the content to be delivered. S2. Collect the viewing points, interaction points, and dissemination points associated with each piece of content, and use a non-negative weighting coefficient to fuse the three types of points to generate the corresponding multi-objective point reward value. S3. Establish a content path dependency structure matrix, and dynamically adjust the matrix parameters according to the historical delivery sequence and user feedback results to represent the content sequence relationship. S4. Using the improved LinUCB algorithm, the confidence upper bound score of the delivered content is calculated by utilizing the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S5. Sort the confidence upper bound scores of all delivered content, select the content with the highest score as the target delivered content, and push it to the designated IoT device; S6. Collect actual points feedback from users on the target content, and update the three-dimensional context tensor, multi-objective integral reward value, content path dependency structure matrix and related parameters of the improved LinUCB algorithm. S7. Based on the updated 3D context tensor, multi-objective integral reward value, content path dependency structure matrix, and relevant parameters of the improved LinUCB algorithm, continue to execute the content push and feedback collection process.

[0024] This invention ensures the efficient flow and iteration of contextual features, points rewards, path structures, and scoring results through a series of interconnected processing steps. It strengthens information collaboration between modules and dynamic updates of model parameters, achieving an automatic closed loop for content push, points allocation, and feedback collection, thereby improving the system's linkage capabilities and overall efficiency.

[0025] In this embodiment, the improved LinUCB algorithm includes: The algorithm inputs the context feature vector obtained by expanding the three-dimensional context tensor tensor and integrates user behavior data, device status data and environmental context information into a feature representation. Introducing integral volatility and integral drift rate, the integral variance and integral mean change rate of multi-objective integral return values ​​are calculated separately within a set time window; The content path dependency structure matrix is ​​used to describe the sequential relationship of the delivered content, and the path coupling term is calculated through the content path dependency structure matrix. The influence of the sequential relationship between content is included in the confidence upper bound score. In the regression model, an L1 norm sparse regularization term and a perturbation penalty term are set to handle abnormal perturbations, fake integral feedback, and sparse abnormal behavior in the context feature vector and multi-objective integral reward value. The integral volatility, integral drift rate, path coupling term, and regression model output are all included in the confidence upper bound score to complete the dynamic ranking of content delivery.

[0026] This invention collects user behavior, device status, and environmental parameters, maps multi-source data into a unified three-dimensional context tensor, and transforms it into feature vectors through tensor expansion, effectively improving the expressive power of the data structure. This enables the system to comprehensively capture the characteristics of users across multiple terminals and scenarios, providing a solid foundation for subsequent content filtering and push.

[0027] In this embodiment, S1 specifically includes: S11. Collect user behavior data, including content browsing behavior, content click behavior, content interaction behavior, and content dissemination behavior; S12. Collect device status data, including the unique identifier of the IoT terminal, the current operating status of the device, and the network connection status; S13. Collect environmental context information, including the time information when the content is delivered and related environmental parameters; S14. Map user behavior data, device status data and environmental context information to the user dimension, device dimension and time dimension of the three-dimensional context tensor respectively, and construct the three-dimensional context tensor. S15. Perform tensor expansion on the three-dimensional context tensor to convert the three-dimensional structure into a one-dimensional context feature vector. Each component represents the content feature relationship corresponding to different combinations of user, device and time information.

[0028] This invention collects viewing points, interaction points, and dissemination points for content, and flexibly integrates these three types of points using non-negative weighting coefficients to generate multi-objective point reward values. This allows point incentives to be dynamically adjusted according to content characteristics and user behavior, optimizing the adaptability and diversity of content push and incentives.

[0029] In this embodiment, S2 specifically includes: S21. Collect the viewing points, interaction points, and dissemination points associated with each piece of content. Viewing points represent user feedback on browsing the content, interaction points represent user interaction with the content, and dissemination points represent feedback when the content is forwarded. S22. Assign non-negative weighting coefficients to the viewing points, interaction points, and dissemination points respectively, in order to determine the weight of each type of point in the multi-objective point reward value; S23. Combine the viewing points, interaction points and dissemination points in a weighted manner to obtain a multi-objective point reward value. The multi-objective point reward value is the result of the weighted sum of the three types of points. S24. Establish a correspondence between the multi-objective integral reward value and the context feature vector for use in the content scoring and ranking process.

[0030] This invention establishes a content path dependency structure matrix, adjusts matrix parameters in real time based on historical delivery and user feedback, effectively models the sequential relationship between content, enhances the system's adaptability to changes in user interests and content synergy, and provides structured data support for push priority determination.

[0031] In this embodiment, S3 specifically includes: S31. Based on the historical delivery sequence of the delivered content, record the push order of each delivered content and user points feedback to form content sequence association information; S32. Match user points feedback data with content sequence association information, and extract the association relationship of each content delivery in different time sequences; S33. Based on historical delivery sequences and user points feedback data, construct a content path dependency structure matrix. Each element of the content path dependency structure matrix is ​​used to represent the sequential dependency relationship between one delivery content and another delivery content. The rows and columns of the matrix correspond to each delivery content. S34. Based on the latest user points feedback data, dynamically adjust the parameters in the content path dependency structure matrix so that the content path dependency structure matrix reflects the changes in the content order relationship. The content path dependency structure matrix is ​​used to characterize the content order relationship and is used for the calculation of the confidence upper bound score.

[0032] This invention utilizes an improved LinUCB scoring algorithm that integrates contextual features, integral rewards, and path dependencies, combined with integral volatility, integral drift rate, and regularization penalties, to achieve a scientific scoring of content priority. This helps improve the accuracy and robustness of the recommendation system and prevents scoring fraud from affecting the ranking results.

[0033] In this embodiment, S4 specifically includes: S41. Take the context feature vector, multi-objective integral reward value and content path dependency structure matrix as input, and set the confidence interval adjustment coefficient, path dependency adjustment coefficient, integral volatility adjustment coefficient and integral drift rate adjustment coefficient. S42. Initialize the model parameter vector and feature correlation matrix for each delivered content, and set the initial value of the content path dependency structure matrix. The model parameter vector and feature correlation matrix are used for the linear regression part of the subsequent confidence upper bound score value and the confidence interval calculation part, respectively. S43. Statistically analyze the multi-target points return value feedback data for each content delivery, calculate the variance of the points feedback sequence within a set time window, and denot it as points volatility σ. Points volatility is used to reflect the numerical fluctuation of points feedback and is a dynamic adjustment item. S44. Statistically analyze the multi-target points return value feedback data for each delivered content, calculate the rate of change of the mean of the points feedback sequence within a set time window, and denot it as the points drift rate δ. The points drift rate is used to reflect the trend of the mean of the points feedback changing over time. S45. Based on the content path dependency structure matrix, calculate the sequential coupling term. The sequential coupling term is the weighted sum of the corresponding row of the content path dependency structure matrix and the content score value. This term comes from the requirement of path structure modeling to express the sequential association between content. S46. Substitute the context feature vector, multi-objective integral reward value, content path dependency structure matrix, integral volatility, and integral drift rate into the confidence upper bound score calculation formula. The formula derivation is derived from the confidence interval idea of ​​the LinUCB algorithm and the linear combination extension of the scoring structure. The score calculation formula is as follows: ; Where s is the upper confidence bound score of the delivered content, x is the context feature vector, θ is the model parameter vector, α is the confidence interval adjustment coefficient, A is the feature correlation matrix, γ is the path dependence adjustment coefficient, Pij is the element in the i-th row and j-th column of the content path dependence structure matrix, sj is the score of the j-th content, β1 is the integral volatility adjustment coefficient, β2 is the integral drift rate adjustment coefficient, σ is the integral volatility, and δ is the integral drift rate. Based on LinUCB, this formula adds a path coupling term and a dynamic penalty term to achieve joint modeling of the content sequence association and integral feedback dynamics. S47. In the process of calculating the confidence upper bound score, an L1 norm sparse regularization term and a perturbation penalty term are set to process abnormal perturbations, fake integral feedback and sparse abnormal behavior in the context feature vector and multi-objective integral reward value. The above regularization term and penalty term are used to enhance the sparsity of the model parameters and the ability to resist perturbations. Finally, the confidence upper bound score value is used for the ranking of the delivered content.

[0034] This invention sorts the confidence upper bound scores and pushes the highest-scoring content to IoT devices. The system also records the push information to ensure smooth data links for subsequent feedback collection and parameter correction, forming an integrated push and feedback process to improve the targeting and effectiveness of the delivery.

[0035] In this embodiment, S5 specifically includes: S51. Collect the confidence upper bound score value corresponding to each content delivery. The confidence upper bound score value is calculated by the improved LinUCB algorithm based on the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S52. Sort the confidence upper bound scores of all the delivered content, arrange them from high to low scores, and determine the ranking position of each delivered content. S53. Determine the content with the highest score as the target content. The target content is the content with the highest confidence upper bound score. S54. Push the target content to the designated IoT device, record the content identifier, push time, and IoT device identifier of the target content, and retain relevant parameter information for points feedback collection and parameter updates.

[0036] This invention utilizes an integral feedback collection and update mechanism to synchronously adjust the three-dimensional context tensor, multi-objective integral reward value, and path dependency matrix using user feedback data. This enables continuous optimization of model parameters and push strategies, enhancing the platform's ability to perceive and adapt to actual user behavior.

[0037] In this embodiment, S6 specifically includes: S61. Record user points feedback data related to the target content, including viewing points, interaction points, and dissemination points corresponding to content identifiers, push times, and IoT device identifiers; S62. Based on the user, device, and time information in the integral feedback data, fill the data into the corresponding three-dimensional context tensor. S63. Based on the integral feedback data, adjust the non-negative weighting coefficients of each item, calculate the new multi-objective integral reward value, and map the new reward value to the context feature vector; S64. Using integral feedback data, correct the element values ​​of the content path dependency structure matrix so that the content path dependency structure matrix represents the actual feedback relationship of the current content delivery order. S65. Apply integral feedback data to adjust the relevant parameters of the improved LinUCB algorithm. The relevant parameters include the model parameter vector, feature correlation matrix, integral volatility adjustment coefficient, integral drift rate adjustment coefficient, path dependence adjustment coefficient, confidence interval adjustment coefficient, L1 norm sparse regularization term parameter, and perturbation penalty term parameter.

[0038] This invention refines the dynamic update path of various parameters in the improved LinUCB algorithm. By driving the synchronous adjustment of model parameters, regularization coefficients, and path coefficients through integral feedback, it ensures that the system can continuously improve the effectiveness and intelligence of content scoring and push strategies under environmental changes.

[0039] Example 1: To verify the feasibility of this invention in practice, it was applied to a smart media operation scenario in a large commercial complex. This complex integrates multiple traffic entry points, including shopping malls, restaurants, subway entrances, and bus stops, and is equipped with various IoT devices such as indoor and outdoor LED screens, electronic wayfinding systems, smart terminals in brand stores, and mobile advertising screens. Previously, the operations team relied on scheduled content rotation and a fixed points allocation mechanism, which made it difficult to dynamically identify user interest hotspots. Content recommendations had a low match with the actual scenario, resulting in stagnant user click and interaction conversion rates. Simultaneously, due to insufficient intelligence in points management, user participation was low, and point-farming behavior was difficult to detect in a timely manner, leading to increased marketing costs and wasted content exposure resources.

[0040] After the system is launched, the 3D context generation module automatically collects user behavior data from different areas and terminals within the business district. Combining this data with device operating status and environmental parameters for the day (such as customer density, temperature, and promotional activities), it dynamically generates a 3D context tensor and expands it into a context feature vector in real time. When each piece of content is viewed, commented on, or shared by a user, the system categorizes and calculates viewing points, interaction points, and dissemination points, then uses dynamically adjusted weighting coefficients to fuse them, generating a comprehensive multi-objective reward value. The content path dependency modeling module continuously analyzes the user's browsing and interaction sequences within a certain time period, automatically adjusting the sequential correlation strength between content items to help identify trends in user interest shifts.

[0041] In the content recommendation decision-making stage, the improved LinUCB scoring module calculates the upper confidence bound score based on the contextual feature vector, multi-objective integral reward value, and content path dependency structure matrix of each piece of content, taking into account factors such as integral volatility, integral drift rate, path coupling term, and regularization parameters. The system automatically selects the content with the highest score and prioritizes its push to the user's current IoT terminal. User-generated integral feedback is collected in real time, and the system adjusts the three-dimensional context tensor, integral reward value, path dependency matrix, and scoring parameters accordingly, achieving a push-feedback-adaptive closed-loop iteration.

[0042] During the actual operation period, the platform monitored the entire process of core indicators such as content clicks, completed views, total interactive points, abnormal point identification rate, and user activation. Compared with the original system, the implementation results showed that the content click conversion rate increased from 2.1% to 4.0%, the average content completion rate increased from 50% to 66%, and the average daily total interactive points increased from 93,500 to 140,700. Point cheating and point manipulation were automatically identified and blocked through point volatility and drift rate analysis, and the abnormal point identification rate increased to 8.5%. The average response latency for content push was reduced from 2.7 seconds to 1.6 seconds, and the number of dormant users activated increased significantly. Specific comparative data is shown in Table 1: Table 1 Performance Comparison of Smart Media Delivery Systems

[0043] Operational data also shows that the flexibility of brands in configuring their own points incentive strategies has increased during the promotion period. During a new product launch event for a certain brand, the platform achieved a 34% higher conversion rate than usual through dynamic adjustments to points weights and real-time optimization of the path dependency matrix. During peak hours and themed promotions, the system can accurately identify user groups with popular interests, automatically switch content priorities, and adaptively adjust the push rhythm, ensuring exposure opportunities for high-quality content while avoiding redundant waste of resources.

[0044] More importantly, the platform provides operators with real-time displays of the effects of various content pushes, points trends, and user activity distribution through a data dashboard, assisting the operations team in continuously optimizing content strategies and incentive rules. Practical application has proven that this system significantly improves the accuracy of smart media content pushes and the scientific nature of points-based incentives, achieving the goals of dynamic user interest identification and integrated management of content delivery and points in commercial scenarios. It effectively solves the problems of insufficient accuracy, incentive failure, and difficulty in controlling anomalies in traditional smart media delivery systems.

[0045] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart media delivery and multi-dimensional points linkage system based on the Internet of Things, characterized in that, include: The 3D context generation module is used to periodically collect user behavior data, device status data, and environmental context information, construct a 3D context tensor, expand the tensor, and generate context feature vectors. The multi-objective integral fusion module is used to collect viewing points, interaction points, and dissemination points associated with the delivered content. It uses non-negative weighting coefficients to fuse the three types of integrals, generate a multi-objective integral reward value, and associate it with the context feature vector. The content path dependency modeling module is used to adjust the content path dependency structure matrix based on historical delivery sequences and user feedback results, representing the sequential relationship of delivered content; An improved LinUCB scoring module is used to calculate the confidence upper bound score of the delivered content by utilizing context feature vectors, multi-objective integral reward values, and content path dependency structure matrices, based on the LinUCB algorithm which introduces integral volatility, integral drift rate, path coupling term, L1 norm sparse regularization term, and perturbation penalty term. The content push module is used to sort the confidence upper bound scores of all delivered content, select the content with the highest score to push to IoT devices, and record the content identifier, push time, and device identifier. The points feedback collection and update module is used to collect user points feedback on the target content, update the three-dimensional context tensor, multi-objective points reward value, content path dependency structure matrix and improved LinUCB algorithm parameters, and use the update results in the content push and feedback collection process.

2. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 1, characterized in that, The modules are connected in the following way: S1. Periodically collect user behavior data, device status data and environmental context information, construct a three-dimensional context tensor, update the three-dimensional context tensor in real time and expand the tensor to generate the context feature vector of the content to be delivered. S2. Collect the viewing points, interaction points, and dissemination points associated with each piece of content, and use a non-negative weighting coefficient to fuse the three types of points to generate the corresponding multi-objective point reward value. S3. Establish a content path dependency structure matrix, and dynamically adjust the matrix parameters according to the historical delivery sequence and user feedback results to represent the content sequence relationship. S4. Using the improved LinUCB algorithm, the confidence upper bound score of the delivered content is calculated by utilizing the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S5. Sort the confidence upper bound scores of all delivered content, select the content with the highest score as the target delivered content, and push it to the designated IoT device; S6. Collect actual point feedback from users on the target content, and update the three-dimensional context tensor, multi-objective point reward value, content path dependency structure matrix and related parameters of the improved LinUCB algorithm. S7. Based on the updated 3D context tensor, multi-objective integral reward value, content path dependency structure matrix, and relevant parameters of the improved LinUCB algorithm, continue to execute the content push and feedback collection process.

3. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, The improved LinUCB algorithm includes: The algorithm inputs the context feature vector obtained by expanding the three-dimensional context tensor tensor and integrates user behavior data, device status data and environmental context information into a feature representation. Introducing integral volatility and integral drift rate, the integral variance and integral mean change rate of multi-objective integral return values ​​are calculated respectively within a set time window; The content path dependency structure matrix is ​​used to describe the sequential relationship of the delivered content, and the path coupling term is calculated through the content path dependency structure matrix. The influence of the sequential relationship between content is included in the confidence upper bound score. In the regression model, an L1 norm sparse regularization term and a perturbation penalty term are set to handle abnormal perturbations, fake integral feedback, and sparse abnormal behavior in the context feature vector and multi-objective integral reward value. The integral volatility, integral drift rate, path coupling term, and regression model output are all included in the confidence upper bound score to complete the dynamic ranking of content delivery.

4. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S1 specifically includes: S11. Collect user behavior data, including content browsing behavior, content click behavior, content interaction behavior, and content dissemination behavior; S12. Collect device status data, including the unique identifier of the IoT terminal, the current operating status of the device, and the network connection status; S13. Collect environmental context information, including the time information when the content is delivered and related environmental parameters; S14. Map user behavior data, device status data and environmental context information to the user dimension, device dimension and time dimension of the three-dimensional context tensor respectively, and construct the three-dimensional context tensor. S15. Perform tensor expansion on the three-dimensional context tensor to convert the three-dimensional structure into a one-dimensional context feature vector. Each component represents the content feature relationship corresponding to different combinations of user, device and time information.

5. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S2 specifically includes: S21. Collect the viewing points, interaction points, and dissemination points associated with each piece of content. Viewing points represent user feedback on browsing the content, interaction points represent user interaction with the content, and dissemination points represent feedback when the content is forwarded. S22. Assign non-negative weighting coefficients to the viewing points, interaction points, and dissemination points respectively, in order to determine the weight of each type of point in the multi-objective point reward value; S23. Combine the viewing points, interaction points and dissemination points in a weighted manner to obtain a multi-objective point reward value. The multi-objective point reward value is the result of the weighted sum of the three types of points. S24. Establish a correspondence between the multi-objective integral reward value and the context feature vector for use in the content scoring and ranking process.

6. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S3 specifically includes: S31. Based on the historical delivery sequence of the delivered content, record the push order of each delivered content and user points feedback to form content sequence association information; S32. Match user points feedback data with content sequence association information, and extract the association relationship of each content delivery in different time sequences; S33. Based on historical delivery sequences and user points feedback data, construct a content path dependency structure matrix. Each element of the content path dependency structure matrix is ​​used to represent the sequential dependency relationship between one delivery content and another delivery content. The rows and columns of the matrix correspond to each delivery content. S34. Based on the latest user points feedback data, dynamically adjust the parameters in the content path dependency structure matrix so that the content path dependency structure matrix reflects the changes in the content order relationship. The content path dependency structure matrix is ​​used to characterize the content order relationship and is used for the calculation of the confidence upper bound score.

7. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S4 specifically includes: S41. Take the context feature vector, multi-objective integral reward value and content path dependency structure matrix as input, and set the confidence interval adjustment coefficient, path dependency adjustment coefficient, integral volatility adjustment coefficient and integral drift rate adjustment coefficient. S42. Initialize the model parameter vector and feature correlation matrix for each delivered content, and set the initial value of the content path dependency structure matrix. S43. Statistically analyze the multi-objective points return value feedback data for each content delivery, calculate the variance of the points feedback sequence within a set time window, and record it as points volatility. S44. Statistically analyze the multi-target points return value feedback data for each delivered content, and calculate the rate of change of the mean of the points feedback sequence within a set time window, which is recorded as the points drift rate. S45. Based on the content path dependency structure matrix, calculate the sequential coupling term, which is the weighted sum of the corresponding row of the content path dependency structure matrix and the content score value. S46. Calculate the confidence upper bound score based on the context feature vector, multi-objective integral reward value, content path dependency structure matrix, integral volatility, and integral drift rate. S47. In the process of calculating the confidence upper bound score, set an L1 norm sparse regularization term and a perturbation penalty term to process abnormal perturbations, fake integral feedback and sparse abnormal behavior in the context feature vector and multi-objective integral reward value, and use the confidence upper bound score value for the ranking of delivered content.

8. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S5 specifically includes: S51. Collect the confidence upper bound score value corresponding to each content delivery. The confidence upper bound score value is calculated by the improved LinUCB algorithm based on the context feature vector, multi-objective integral reward value and content path dependency structure matrix. S52. Sort the confidence upper bound scores of all the delivered content, arrange them from high to low scores, and determine the ranking position of each delivered content. S53. Determine the content with the highest score as the target content. The target content is the content with the highest confidence upper bound score. S54. Push the target content to the designated IoT device, record the content identifier, push time, and IoT device identifier of the target content, and retain relevant parameter information for points feedback collection and parameter updates.

9. The IoT-based smart media delivery and multi-dimensional points linkage system according to claim 2, characterized in that, S6 specifically includes: S61. Record user points feedback data related to the target content, including viewing points, interaction points, and dissemination points corresponding to content identifiers, push times, and IoT device identifiers; S62. Based on the user, device, and time information in the integral feedback data, fill the data into the corresponding three-dimensional context tensor. S63. Based on the integral feedback data, adjust the non-negative weighting coefficients of each item, calculate the new multi-objective integral reward value, and map the new reward value to the context feature vector; S64. Using integral feedback data, correct the element values ​​of the content path dependency structure matrix so that the content path dependency structure matrix represents the actual feedback relationship of the current content delivery order. S65. Apply integral feedback data to adjust the relevant parameters of the improved LinUCB algorithm. The relevant parameters include the model parameter vector, feature correlation matrix, integral volatility adjustment coefficient, integral drift rate adjustment coefficient, path dependence adjustment coefficient, confidence interval adjustment coefficient, L1 norm sparse regularization term parameter, and perturbation penalty term parameter.

Images