Content AI generation system and method based on large screen interaction

Abstract

The invention discloses a content AI generation system and generation method based on large screen interaction, relates to the technical field of AI generation, and provides personalized contents for multiple users and realizes fair and reasonable scheduling of screen domain resources by utilizing multi-mode perception technologies such as computational vision, voice recognition and gesture capture and combining an online knowledge base and a generation model. Different from a traditional fixed playing or simple polling mode, the method particularly emphasizes that people and screens are paired by adopting a mature combination optimization algorithm and a simplified judgment rule under the condition of multi-person concurrency, so that the uncertainty caused by complex weight scoring is avoided, and the interpretability and the implementability of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of AI generation technology, and more specifically, to a content AI generation system and method based on large-screen interaction. Background Technology

[0002] With the popularization of large-screen intelligence, various smart screens are widely deployed in public spaces and commercial scenarios. They not only play information but are also gradually taking on multiple interactive functions such as voice interaction, gesture recognition, QR code interaction, and user screen projection. However, most traditional large-screen systems adopt fixed playback or simple polling scheduling modes: first-come, first-served or simple polling cannot take into account user priority, queue length, and screen domain status. This results in some users staying for a long time or having low interaction efficiency. Generally, they play content randomly on the large screen according to fixed templates, which cannot be combined with users' behavior, interests, or intentions. The content push matching degree is low, and the long-term experience of the audience is degraded. There is no mechanism to restrict certain templates or users from frequently occupying high-quality screen domains, affecting the fair use of others. Existing systems cannot adjust content or recommendation strategies based on audience feedback, resulting in homogenized display and inability to adapt.

[0003] To address the above issues, some studies have attempted to apply multi-armed slot machine models to interactive recommendations, using reward signals to optimize content selection. However, these methods typically rely on complex combinations of feature weights, which are difficult to interpret and have high implementation costs. Based on this, this invention proposes a simplified and practical perception, matching, decision-making, and generation process. It employs two mature combined algorithms, including the Hungarian algorithm and the stable marriage algorithm, as well as a decision rule that uses only a few key parameters, maximizing matching efficiency and user satisfaction while satisfying fairness constraints.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a content AI generation system and generation method based on large-screen interaction to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A content AI generation method based on large-screen interaction, including the steps; Step S1: In each decision cycle, generate an eligibility matrix based on hard constraints and construct a user preference table and a screen domain preference table. Calculate the congestion load ratio and preference readiness and make a choice between the Hungarian algorithm and the stable marriage algorithm. Step S2: When the Hungarian algorithm is selected, construct the cost matrix according to the lexicographical cost of waiting priority distance and iteratively solve for the minimum total cost matching; when the stable marriage algorithm is selected, propose according to the preference list and accept until there is no obstruction; output a stable matching set. Step S3: For a stable matching set, select a single-user template from the template library according to template type and scene elements. For a set of users with the same template on the same screen, make three judgments based on screen margin and distance difference: forced merging, conditional merging, and no merging. If merging is required, switch to a dual-user template; otherwise, keep the single-user template. Step S4: Generate content for single-user templates and dual-user templates respectively, and design a rollback and redistribution mechanism for dual-user templates. When the resource width changes or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

[0007] In a preferred embodiment, step S1 includes the following: The sensing devices include cameras, depth sensors, microphones, gesture recognition devices, and barcode scanners. The information collected includes the user's location and posture, as well as the waiting time. Obtain the anonymous ID of each user u, the distance d(u,r) to each screen domain r, and the orientation angle. ; Record the cumulative waiting time of user u since the last time they obtained the screen domain service. ; The hard constraint is that the distance d(u,r) between the user and the screen domain ≤ a preset maximum distance, and the orientation angle... If the value of Elig(u,r) is ≤ the maximum value of the preset orientation angle, then the qualification matrix value Elig(u,r) is 1; otherwise, it is 0. Generate a user preference table and a user preference list. For each user u, collect the set of all screen domains that satisfy Elig(u,r)=1. Then the user's screen domain combination (u,r) takes priority. The user screen domain combination (u,r) is compared for distance, and the distances d(u,r) are sorted from smallest to largest to obtain the user preference list pu, in which r that does not meet the qualifications is excluded; For each screen domain r, sort all users who satisfy Elig(u,r)=1 in ascending order of distance d(u,r) to obtain the screen domain preference list. ; Calculate preference readiness The percentage of users who generated a complete sort for each user is recorded as follows: The percentage of screen domains that generate a complete sort for each screen domain is statistically analyzed and denoted as . ; like ,but ,otherwise ; Calculate the congestion load ratio kappa, which is the number of users in the current period divided by the sum of the capacities of all screen domains; according to Compared with the Kappa selection matching algorithm, In the case of a stable marriage, the algorithm for stable marriages is chosen; otherwise, the Hungarian algorithm is chosen.

[0008] In a preferred embodiment, step S2 includes the following: Perform the operation when using the Hungarian algorithm; For each excluded user screen domain combination (u,r), construct a cost matrix C(u,r) and use lexicographical cost. Set the first key value to 0 if the condition is met; otherwise, set it to 1. Then normalize the distance d(u,r) to [0,1] and denote it as As a secondary key value, the formula is expressed as: Where I is the index function, 10 is a sufficiently large constant, and if Elig(u,r)=0, then C(u,r) is set to a maximum value to indicate that it is unavailable; Subtract the minimum value of each row and the minimum value of each column from the cost matrix, and find the fewest lines to cover all zero elements until the number of lines equals the order of the matrix. By selecting a position for each row and column from the zero element of the matrix, a set of matching users and screen domains is obtained. ; The stable marriage algorithm is used for this operation. Set all users and screen domains to an idle state and match the initial set. Empty; Unmatched users submit proposals to screen domains in order of their preference list. Each screen domain accepts the current best applicant and makes a temporary match, while rejecting other applications. Rejected users continue to the next round of proposals. Repeat the proposal and acceptance process until all users have completed the matching or there are no more proposals available for the screen domain; Match output stable matching set .

[0009] In a preferred embodiment, step S3 includes the following: Three types of judgments are made based on screen domain margin and distance difference: forced merging, conditional merging, and no merging. Match output stable matching set After generation, maintain a template library M, and select template m* for each user screen domain combination (u,r) in the matching set; For the same screen domain r, group by template type m and filter the matching user sets. ,like No need to merge; if Then, they are paired up in order of arrival and enter the merging process. For candidate matching user sets The user duo The template m and screen domain r are determined solely by two parameters to decide whether to merge the screen domain margin. And distance difference Δd, when A forced merge will be performed, merging these two users; when Conditions will be merged in certain cases; otherwise, they will not be merged, and each user template will be displayed separately.

[0010] In a preferred embodiment, step S4 includes the following: For content generation of single-user templates, each user screen domain combination (u,r) generates content based on the selected single-user template m*, and automatically adjusts UI elements such as font size, speech rate, and interactive button size, taking into account the user's distance from the screen domain and orientation. For content generation of dual-user templates, the template has two user groups on each side. Set up independent information slots and fill them with individual preferences and recommended content; The rollback condition for dual-user templates is that the screen domain margin is ≥1 or the change in user distance difference does not meet the merging condition. During rollback, the user group is separated, steps S1-S3 are re-executed, and independent content is generated.

[0011] The content AI generation system based on large-screen interaction includes: an algorithm selection module, a human-screen matching module, a merging and judgment module, and a rollback and redistribution module, with signal connections between each module; The algorithm selection module generates an eligibility matrix and constructs a user preference table and a screen domain preference table based on hard constraints in each decision cycle. It calculates the congestion load ratio and preference readiness and makes a selection between the Hungarian algorithm and the stable marriage algorithm.

[0012] The human-screen matching module, when selecting the Hungarian algorithm, constructs a cost matrix based on the lexicographical cost order of waiting priority distance and iteratively solves for the minimum total cost matching; when selecting the stable marriage algorithm, it proposes according to the preference list and accepts until there are no obstructions; the matching output is a stable matching set. The merge determination module selects a single-user template from the template library based on template type and scene elements for a stable matching set. For user sets with the same template on the same screen, it makes three determinations based on screen margin and distance difference: forced merge, conditional merge, and no merge. When merging is required, it switches to a dual-user template; otherwise, it keeps the single-user template. The rollback and redistribution module generates content for single-user and dual-user templates respectively and designs a rollback and redistribution mechanism for dual-user templates. When the resource width increases or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

[0013] The technical effects and advantages of this invention's content AI generation method based on large-screen interaction: By introducing a qualification matrix based on hard constraints such as distance and orientation, and a dual-keyword sorting mechanism, combined with an intelligent switching strategy of the Hungarian algorithm and the stable marriage algorithm, the system can quickly and reasonably complete the matching of users and screen domains in multi-user concurrent scenarios, effectively avoiding the long-term occupation of resources by a few users and significantly improving the fairness of screen domain resource utilization. The system adopts simplified decision rules and a few key parameters such as waiting time, distance difference, and screen domain reserve for matching and merging, which reduces the complexity and uncertainty brought by traditional multi-feature weight scoring, making the system decision-making process clear and controllable, and facilitating actual deployment and operation and maintenance. By intelligently selecting single-user or dual-user templates from the template library based on user context information such as location, waiting time, and behavioral intent, and calling a multimodal generation model to generate personalized content in real time, the system automatically adjusts display elements based on viewing distance and orientation to improve the matching degree between content and user needs. Through dual judgment of screen space availability and user distance difference, the system dynamically triggers the merging and display of user content, effectively alleviating the queuing pressure when screen space resources are tight. At the same time, a rollback mechanism is designed to automatically separate and merge content when resources are abundant or user status changes, ensuring system flexibility and user experience continuity. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the content AI generation method based on large-screen interaction according to the present invention.

[0015] Figure 2 This is a schematic diagram of the content AI generation system module based on large-screen interaction according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 Please see Figure 1 As shown, this invention discloses a content AI generation method based on large-screen interaction, including the following steps: Step S1: In each decision cycle, generate an eligibility matrix based on hard constraints and construct a user preference table and a screen domain preference table. Calculate the congestion load ratio and preference readiness and make a choice between the Hungarian algorithm and the stable marriage algorithm. Step S2: When the Hungarian algorithm is selected, construct the cost matrix according to the lexicographical cost of waiting priority distance and iteratively solve for the minimum total cost matching; when the stable marriage algorithm is selected, propose according to the preference list and accept until there is no obstruction; output a stable matching set. Step S3: For a stable matching set, select a single-user template from the template library according to template type and scene elements. For a set of users with the same template on the same screen, make three judgments based on screen margin and distance difference: forced merging, conditional merging, and no merging. If merging is required, switch to a dual-user template; otherwise, keep the single-user template. Step S4: Generate content for single-user templates and dual-user templates respectively, and design a rollback and redistribution mechanism for dual-user templates. When the resource width changes or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

[0018] In step S1, at each decision cycle, an eligibility matrix is ​​generated based on hard constraints, and a user preference table and a screen domain preference table are constructed. The congestion load ratio and preference readiness are calculated, and a choice is made between the Hungarian algorithm and the stable marriage algorithm. Specific details include: Within the decision-making cycle t, information including location and posture information and waiting time is collected using cameras, depth sensors, microphones, gesture recognition, and barcode scanning devices; Position and orientation information: Obtain the anonymous ID of each user u, the distance d(u,r) to each screen domain r, and the orientation angle. Distance is measured in meters using a camera and depth sensor, and orientation angle is output by a pose analysis algorithm, with a value ranging from 0° to 180°. Waiting time records the cumulative waiting time of user u since the last time they obtained the domain service. If this is the user's first appearance, then ; The above features are combined into a user context vector. It includes a distance list, orientation, wait time, and other features.

[0019] For each user screen domain combination (u,r), calculate the eligibility matrix Elig(u,r) based on hard constraints: If and only if both conditions are met When, Elig(u,r)=1, where, 4 meters can be taken. 30° is acceptable; If any condition is not met, Elig(u,r) = 0, and the user screen domain combination (u,r) is marked as unmatchable. The two key parameters d(u,r) are then compared with... Save it in association with the user's screen domain combination (u,r); Generate a user preference table. For each user u, collect the set of all screen domains that satisfy Elig(u,r)=1 and sort them according to the following rules; The first key question is whether the waiting time has reached the threshold. For example, 10 seconds, if If so, then the user's screen domain combination (u,r) takes priority; The second key is to compare the distance between user screen domain combinations (u,r) with the same first key, and sort the distances d(u,r) from smallest to largest; This yields the user preference list. Among them, r that does not meet the eligibility criteria is excluded; Generate a screen domain preference table. For each screen domain r, sort all users who satisfy Elig(u,r)=1 according to the following rules. The first key question is whether the waiting time has reached the threshold. ; Second key: Sort the distances d(u,r) from smallest to largest to obtain the screen domain preference list pr; calculate the preference readiness. The percentage of users who can generate a complete sort for each user is denoted as . The percentage of screen domains that can generate a complete sort for each screen domain is denoted as . ; like ,but ,otherwise ; Calculate the congestion load ratio (kappa), where kappa is the number of users in the current period divided by the sum of the capacities of all screen domains. For example, 1.2 indicates a crowded state.

[0020] according to Compared to the Kappa selection matching algorithm, it is crowded and has complete preferences, i.e. and The stable marriage algorithm is chosen, which achieves unhindered pair matching through proposal and acceptance iterations. In other cases, a cost matrix C(u,r) is constructed, and the Hungarian algorithm is used to solve for the minimum total cost matching. The Hungarian algorithm is a polynomial-time algorithm that can solve assignment problems.

[0021] In step S2, when the Hungarian algorithm is selected, a cost matrix is ​​constructed based on the lexicographical cost order following the waiting priority distance, and the minimum total cost matching is iteratively solved; when the stable marriage algorithm is selected, proposals are made according to the preference list, and acceptance continues until there are no obstruction pairs; the matching output is a stable matching set, which specifically includes: If the Hungarian algorithm is selected in step one, then perform the following operations; For each excluded user screen domain combination (u,r), construct a cost matrix C(u,r) and use lexicographical cost: if If the first key is 0, then set its value to 0; otherwise, set it to 1. Then normalize the distance d(u,r) to [0,1] and denote it as As a secondary key value, the formula is expressed as: Where I is the index function, 10 is a sufficiently large constant to make the waiting priority higher than the distance, and if Elig(u,r)=0, then C(u,r) is set to a maximum value to indicate unavailability; Perform row and column preprocessing, that is, subtract the minimum value of each row and the minimum value of each column from the cost matrix, so that the matrix contains at least one zero element; Perform the zero-element masking operation, find the fewest lines to cover all zero elements. If the number of lines is equal to the matrix order, then the current set of zero elements constitutes the optimal match; otherwise, proceed to the next adjustment step. Adjustments and iterations are performed. The minimum value is found among the uncovered elements. The minimum value is subtracted from the uncovered elements and added to the minimum value for the elements covered by the two lines. The operation of covering zero elements is repeated until the number of lines equals the order of the matrix. By selecting a position for each row and column from the zero element of the matrix, a set of matching users and screen domains is obtained. And the corresponding minimum total cost; If the stable marriage algorithm is selected in step one, then the following operations are performed; Set all users and screen domains to an idle state and match the initial set. Empty; The process of proposing and accepting proposals is as follows: Unmatched users submit proposals to screen domains in order of their preference list. Each screen domain accepts the current best applicant and makes a temporary match, while rejecting other applications. Rejected users continue to the next round of proposals. Repeat the proposal and accept operation until all users have completed the matching or there is no more proposal in the screen domain, ensuring that there is no pair of u and r that prefer each other but are not together. Match output stable matching set .

[0022] In step S3, for a stable matching set, a single-user template is selected from the template library according to template type and scene elements. For user sets with the same template on the same screen, three types of judgments are made based on screen margin and distance difference: forced merging, conditional merging, and no merging. When merging is required, the template is switched to a dual-user template; otherwise, the single-user template is maintained. Specific details include: Match output stable matching set After generation, maintain a template library M. For each user screen domain combination (u, r) in the matching set, select a single-user version template m according to the following rules. Pre-label each template in the template library with a set of scene element tags, including the location and theme domain, available interaction forms such as voice scanning, gesture display, network and authorization requirements, and whether personalized slots are allowed. For any user and screen domain combination, extract its scene element set based on the current perception and operating status. Define a consistency index between the template and the scene, with a value of one if and only if the essential element set of the template is completely covered by the scene element set; otherwise, the value is zero. The first decision step is to filter the candidate set with a consistency of one in the template library, denoted as the first candidate set. If the first candidate set is empty, select the template with the fewest missing elements that can be replaced by the template's built-in degradation switch, such as degrading gestures to button clicks and network-based material retrieval to local general materials, to form a backup candidate set. Then, continue to the second step to select the most suitable template m* from the template library.

[0023] After obtaining the single-user template, the content is not generated immediately, but instead a merge determination is initiated.

[0024] Perform merged candidate identification and filtering. For the same screen domain r, group by template type m, and filter matching user sets. That is, given the screen domain r and template type m, This indicates that all elements matching the given screen domain r have been identified. , and the user set to which it is assigned a template of type m; like No need to merge; if Then, they are paired up in order of arrival and enter the merging process. For candidate matching user sets The user duo The template m and screen domain r are determined solely by two parameters to decide whether to merge, including screen domain margin and distance difference; Screen domain margin is the remaining number of card bits in screen domain r. This equals the screen domain's capacity minus the number of currently occupied slots. Where: the screen domain's capacity is the total number of slots that screen domain r can simultaneously support in the current time slot; the number of currently occupied slots is the number of slots that have been allocated or are being displayed in the current time slot within the screen domain. If there are no free slots, then... It is necessary to save space by merging; Distance difference is represented as a pair of users The difference in distance between them is expressed as ; Δd is relatively small, the viewing distance between the two people is close, and after merging, a unified font size can be used and the viewing angle is consistent; The merge decision logic includes forced merge, conditional merge, and no merge. If so, a forced merge will be performed, merging these two users; If the condition is met, the conditions will be merged; otherwise, they will not be merged, and each user template will be displayed separately. If merging is not required, generate content using a single-user template for each (u,r). If merging is required, use a dual-user template, filling in the content slots for both users, sharing common information while retaining their unique fields.

[0025] In step S4, content is generated for both single-user and dual-user templates, and a rollback and reallocation mechanism is designed for dual-user templates. If resource width changes or distance differences cause the merging conditions to no longer be met, a rollback is triggered in the next cycle, and the respective content is regenerated. The specific content includes: For content generation of single-user templates, each user screen domain combination (u,r) generates content based on the selected single-user template m*. The preset editable slots in the template are filled with corresponding information based on the user's anonymous ID, waiting time, behavioral intent and preferences. Combined with the online knowledge base, text or multimodal generation models are called to generate personalized text, image or voice content for the user. When generating content, the system considers the user's distance from the screen and their orientation, automatically adjusting UI elements such as font size, speech rate, and interactive button size to suit the user's viewing conditions. After content generation, it is pushed to the screen and displayed according to a predetermined time window. During the display, the system continues to monitor user behavior feedback for subsequent adjustments.

[0026] For content generation of dual-user templates, when it is determined that two users need to be merged based on screen space and distance difference, the dual-user version of the template is selected, and the common area displays general information related to the screen space or scene; the left and right or top and bottom sides of the template represent the two user groups respectively. Set up independent information slots, fill them with individual preferences and recommended content, and keep them displayed simultaneously without obscuring each other; The font size and layout of interactive elements are designed uniformly based on the average distance between two people. When users have similar preferences, the personalized recommendations can be merged to make the theme seen by both people more consistent. After the content is generated, it is pushed to the same screen area for joint display. A rollback and redistribution mechanism for dual-user templates was designed. The rollback condition is that the merging of dual-user templates is mainly to solve scenarios with limited screen space or users in close proximity. When screen space resources become available again... Or a pair of users When the distance difference changes significantly, i.e. the merging condition is no longer met, a rollback should be triggered; The rollback process involves grouping the two users in the next decision cycle t. Separate from the double-person merge queue and re-execute the matching from step S1 to step S3; Once separated, the user duo The generated content will become invalid. Instead, a suitable template will be selected for each user, and the generation model will be called again to generate personalized content based on the user's current status before pushing it to the corresponding screen domain.

[0027] Please see Figure 2 As shown, the present invention discloses a content AI generation system based on large-screen interaction, including: an algorithm selection module, a human-screen matching module, a merging and determination module, and a rollback and redistribution module, with signal connections between the modules; The algorithm selection module generates an eligibility matrix and constructs a user preference table and a screen domain preference table based on hard constraints in each decision cycle. It calculates the congestion load ratio and preference readiness and makes a selection between the Hungarian algorithm and the stable marriage algorithm.

[0028] The human-screen matching module, when selecting the Hungarian algorithm, constructs a cost matrix based on the lexicographical cost order of waiting priority distance and iteratively solves for the minimum total cost matching; when selecting the stable marriage algorithm, it proposes according to the preference list and accepts until there are no obstructions; the matching output is a stable matching set. The merge determination module selects a single-user template from the template library based on template type and scene elements for a stable matching set. For user sets with the same template on the same screen, it makes three determinations based on screen margin and distance difference: forced merge, conditional merge, and no merge. When merging is required, it switches to a dual-user template; otherwise, it keeps the single-user template. The rollback and redistribution module generates content for single-user and dual-user templates respectively and designs a rollback and redistribution mechanism for dual-user templates. When the resource width increases or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

[0029] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0030] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0031] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0032] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0034] In conclusion, 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A content AI generation method based on large-screen interaction, characterized in that, Includes steps; Step S1: In each decision cycle, generate an eligibility matrix based on hard constraints and construct a user preference table and a screen domain preference table. Calculate the congestion load ratio and preference readiness and make a choice between the Hungarian algorithm and the stable marriage algorithm. Step S2: When the Hungarian algorithm is selected, construct the cost matrix according to the lexicographical cost of waiting priority distance and iteratively solve for the minimum total cost matching; when the stable marriage algorithm is selected, propose according to the preference list and accept until there is no obstruction; output a stable matching set. Step S3: For a stable matching set, select a single-user template from the template library according to template type and scene elements. For a set of users with the same template on the same screen, make three judgments based on screen margin and distance difference: forced merging, conditional merging, and no merging. If merging is required, switch to a dual-user template; otherwise, keep the single-user template. Step S4: Generate content for single-user templates and dual-user templates respectively, and design a rollback and redistribution mechanism for dual-user templates. When the resource width changes or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

2. The content AI generation method based on large-screen interaction according to claim 1, characterized in that, The sensing devices include cameras, depth sensors, microphones, gesture recognition devices, and barcode scanners. The information collected includes the user's location and posture, as well as the waiting time. Obtain the anonymous ID of each user u, the distance d(u,r) to each screen domain r, and the orientation angle. ; Record the cumulative waiting time of user u since the last time they obtained the screen domain service. .

3. The content AI generation method based on large-screen interaction according to claim 2, characterized in that, The hard constraint is that the distance d(u,r) between the user and the screen domain ≤ a preset maximum distance, and the orientation angle... If the value of Elig(u,r) is ≤ the maximum value of the preset orientation angle, then the qualification matrix value Elig(u,r) is 1; otherwise, it is 0.

4. The content AI generation method based on large-screen interaction according to claim 2, characterized in that, Generate a user preference table and a user preference list. For each user u, collect the set of all screen domains that satisfy Elig(u,r)=1. Then the user's screen domain combination (u,r) takes priority. The user screen domain combination (u,r) is compared for distance, and the distances d(u,r) are sorted from smallest to largest to obtain the user preference list pu, in which r that does not meet the qualifications is excluded; For each screen domain r, sort all users who satisfy Elig(u,r)=1 in ascending order of distance d(u,r) to obtain the screen domain preference list. .

5. The content AI generation method based on large-screen interaction according to claim 2, characterized in that, Calculate preference readiness The percentage of users who generated a complete sort for each user is recorded as follows: The percentage of screen domains that generate a complete sort for each screen domain is statistically analyzed and denoted as . ; like and ,but ,otherwise ; Calculate the congestion load ratio kappa, which is the number of users in the current period divided by the sum of the capacities of all screen domains; according to Compared with the Kappa selection matching algorithm, and In the case of a stable marriage, the algorithm for stable marriages is chosen; otherwise, the Hungarian algorithm is chosen.

6. The content AI generation method based on large-screen interaction according to claim 1, characterized in that, Perform the operation when using the Hungarian algorithm; For each excluded user screen domain combination (u,r), construct a cost matrix C(u,r) and use lexicographical cost. Set the first key value to 0 if the condition is met; otherwise, set it to 1. Then normalize the distance d(u,r) to [0,1] and denote it as As a secondary key value, the formula is expressed as: Where I is the index function, 10 is a sufficiently large constant, and if Elig(u,r)=0, then C(u,r) is set to a maximum value to indicate that it is unavailable; Subtract the minimum value of each row and the minimum value of each column from the cost matrix, and find the fewest lines to cover all zero elements until the number of lines equals the order of the matrix. Select a position for each row and column from the zero element of the matrix to obtain the matching set St of user and screen domain.

7. The content AI generation method based on large-screen interaction according to claim 6, characterized in that, The stable marriage algorithm is used for this operation. Set all users and screen domains to an idle state and match the initial set. Empty; Unmatched users submit proposals to screen domains in order of their preference list. Each screen domain accepts the current best applicant and makes a temporary match, while rejecting other applications. Rejected users continue to the next round of proposals. Repeat the proposal and acceptance process until all users have completed the matching or there are no more proposals available for the screen domain; Match output stable matching set .

8. The content AI generation method based on large-screen interaction according to claim 1, characterized in that, Three types of judgments are made based on screen domain margin and distance difference: forced merging, conditional merging, and no merging. After the stable matching set St is generated, a template library M is maintained, and a template m* is selected for each user screen domain combination (u,r) in the matching set; For the same screen domain r, group by template type m and filter the matching user sets. ,like No need to merge; if Then, they are paired up in order of arrival and enter the merging process. For candidate matching user sets The user duo The template m and screen domain r are determined solely by two parameters to decide whether to merge the screen domain margin. And distance difference Δd, when A forced merge will be performed, merging these two users; when Conditions will be merged in certain cases; otherwise, they will not be merged, and each user template will be displayed separately.

9. The content AI generation method based on large-screen interaction according to claim 1, characterized in that, For content generation of single-user templates, each user screen domain combination (u,r) generates content based on the selected single-user template m*, and automatically adjusts UI elements such as font size, speech rate, and interactive button size, taking into account the user's distance from the screen domain and orientation. For content generation of dual-user templates, the template has two user groups on each side. Set up independent information slots and fill them with individual preferences and recommended content; The rollback condition for dual-user templates is that the screen domain margin is ≥1 or the change in user distance difference does not meet the merging condition. During rollback, the user group is separated, steps S1-S3 are re-executed, and independent content is generated.

10. A content AI generation system based on large-screen interaction, used to implement the content AI generation method based on large-screen interaction as described in any one of claims 1-9, characterized in that, The algorithm selection module generates an eligibility matrix and constructs a user preference table and a screen domain preference table based on hard constraints in each decision cycle. It calculates the congestion load ratio and preference readiness and makes a selection between the Hungarian algorithm and the stable marriage algorithm. The human-screen matching module, when selecting the Hungarian algorithm, constructs a cost matrix based on the lexicographical cost order of waiting priority distance and iteratively solves for the minimum total cost matching; when selecting the stable marriage algorithm, it proposes according to the preference list and accepts until there are no obstructions; the matching output is a stable matching set. The merge determination module selects a single-user template from the template library based on template type and scene elements for a stable matching set. For user sets with the same template on the same screen, it makes three determinations based on screen margin and distance difference: forced merge, conditional merge, and no merge. When merging is required, it switches to a dual-user template; otherwise, it keeps the single-user template. The rollback and redistribution module generates content for single-user and dual-user templates respectively and designs a rollback and redistribution mechanism for dual-user templates. When the resource width increases or the distance difference changes so that the merging conditions are no longer met, a rollback is triggered in the next cycle to regenerate the respective content.

Images