Multi-site oriented performing content adaptive matching system

Abstract

The present application relates to the technical field of performing arts content adaptive matching, in particular to a multi-venue oriented performing arts content adaptive matching system, wherein a content reorganization module is used to receive a performing arts content combination scheme output by a matching recommendation engine, retrieve a corresponding performance form template from a form template library module according to a resulting performance form template type, and reorganize a selected content atomic unit set according to a time axis, a spatial position and an interaction logic in the retrieved performance form template to generate a complete performing arts script and a performance control instruction sequence adapted to a target venue. The present application automatically collects physical characteristic parameters of a target venue, commercial target parameters and multi-dimensional characteristic data of an audience group, and dynamically recommends an atomic unit combination and a performance form template most adapted to the target venue by using a double-tower network and a combinatorial optimizer.

Description

Technical Field

[0001] This invention relates to the field of adaptive matching technology for performing arts content, specifically an adaptive matching system for performing arts content across multiple venues. Background Technology

[0002] The planning and production of offline performances (such as stage plays, musicals, immersive games, and cultural tourism performances) have long relied on human experience. This process requires planners to design scripts, manage actors, and arrange lighting and sound for each specific venue (including traditional theaters, shopping mall atriums, scenic areas, restaurants, and bookstores) from scratch. Adapting the same IP into different versions often requires complete re-arranging, leading to long production cycles and high costs. Furthermore, existing performance control systems mostly operate by playing lights, sound effects, and cues sequentially according to pre-programmed timecodes, unable to dynamically adjust the plot in real-time based on audience reactions (such as online comments, donation activity, and heart rate changes). Additionally, current technology cannot automatically match venue spatial parameters (dimensions, column grid, and pedestrian flow) with the performance content. Therefore, existing technologies have significant shortcomings. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive matching system for performance content across multiple venues, in order to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: an adaptive matching system for performance content across multiple venues, comprising: The site information acquisition module is used to acquire the physical characteristic parameters and commercial target parameters of the target site and construct the site feature vector corresponding to the target site. The audience profiling analysis module is used to acquire multi-dimensional feature data of the audience group from the user database or real-time sensors, and output the audience feature vector after normalization and embedding encoding. The content atom library module is used to automatically break down the original performing arts content into multiple content atom units and generate a metadata vector corresponding to each content atom unit; The template library module stores several preset performance format templates. The parameters of each preset performance format template include total duration, segmented time axis (the segmented time axis consists of several continuous time intervals, and each time interval is assigned a time slot number, the number of time slots varies between different performance format templates), a list of compatible atomic types for each time slot, a spatial location requirement vector (the spatial location requirement vector consists of the occupancy ratios of the main stage, secondary stage, aisle, and bar area respectively), a list of technical resource requirements (including the number of lighting channels, the number of sound channels, and the number of projectors), and interactive node density (the interactive node density represents the average number of interactive nodes per preset time interval). Based on each template parameter, a corresponding template feature vector is generated. The matching recommendation engine is used to calculate the matching degree between each content atom unit and each performance format template based on the obtained venue feature vector and audience feature vector, and output the top performance content combination schemes with the highest matching degree. Each scheme includes the selected set of content atom units and the corresponding performance format template identifier. The content reorganization module is used to receive the performance content combination scheme output by the matching recommendation engine, retrieve the corresponding performance form template from the form template library module according to the obtained performance form template type, and reorganize the selected content atomic unit set according to the time axis, spatial position and interactive logic in the retrieved performance form template to generate a complete performance script and performance control instruction sequence adapted to the target venue. The performance execution module is used to parse and execute the obtained performance control instruction sequence; and to synchronously collect audience interaction feedback data during the performance, and package the collected data into timing feedback frames for output. The dynamic adjustment module is used to perform sentiment analysis and intent recognition on the obtained interactive feedback data, generate plot branch adjustment instructions or ending switching instructions and feed them back to the content reorganization module. After receiving the instructions, the content reorganization module modifies the content atomic units in the remaining timeline in real time and regenerates the performance script and performance control instruction sequence adapted to the target venue.

[0005] Furthermore, the physical characteristic parameters include venue type, space size, maximum audience capacity, and audience flow pattern; the commercial objective parameters include box office revenue weight, brand exposure weight, educational purpose weight, and audience dwell time target. Specifically, venue types include theaters, shopping malls, scenic spots, restaurants, bookstores, and street blocks; spatial dimensions include length, width, and height; maximum audience capacity indicates the upper limit of the total number of audience members that the target venue can accommodate; audience flow patterns include free flow, one-way flow, and gathering and staying patterns; commercial target parameters include box office revenue weight, brand exposure weight, educational purpose weight, and audience stay time target. The value range of the box office revenue weight, brand exposure weight, and educational purpose weight parameters is [0,1], and the larger the value, the higher the priority of the corresponding commercial target. In the process of constructing the site feature vector corresponding to the target site, the unique encoding vector of the corresponding site type is obtained within the physical feature parameters of the target site; the unique hot encoding vector of the corresponding audience flow pattern is obtained within the physical feature parameters of the target site; and the original text information is replaced with the corresponding unique hot encoding vector. Then, the physical feature parameters of the target site after replacing the text information and the commercial target parameters are concatenated to obtain the site feature vector corresponding to the target site.

[0006] The multidimensional feature data includes age distribution (in this invention, the age distribution is divided into five intervals: 0-17 years old, 18-25 years old, 26-35 years old, 36-50 years old, and over 50 years old, with the total number of people in each interval being 1), consumption preference tags (in this invention, the consumption preference tags are in the form of unique hot codes, and different consumption preference tags correspond to different unique hot codes), interaction willingness scores (with values ​​ranging from [0,1]), and the historical viewing behavior sequence of the most recent preset performances. The historical viewing behavior sequence of each performance consists of the unique hot code and score of the performance type corresponding to the performance performance. The metadata vector includes: unique hot encoding of the content atomic unit type, minimum number of viewers required, interaction intensity level (value range includes 0, 1 and 2), commercial conversion potential score (value range is [0,1], the higher the value, the stronger the commercial monetization ability of the unit), duration, number of actors required, emotional tag vector (the emotional tags of this invention include six basic emotions: happiness, sadness, anger, fear, surprise and neutral), and replaceability tag (value is 0 or 1, a value of 1 means that the content atomic unit supports dynamic replacement, a value of 0 means that the content unit is not replaceable); the content atomic unit types include five performance forms: dialogue, song, dance, interactive game and famous scene reproduction.

[0007] Furthermore, the metadata vector of each content atom unit in the content atom library module is also bound to a real-time replacement cost parameter, which is used to characterize the cost required to replace the original content atom unit with the corresponding content atom unit during dynamic adjustment; the logical connection verification submodule prioritizes the content atom unit with the lowest corresponding replacement cost when calling the alternative units.

[0008] Furthermore, the matching recommendation engine includes a venue-audience joint encoder, a content-template dual-tower network, a similarity calculation unit, and a combinatorial optimizer; The venue-audience joint encoder is used to concatenate the venue feature vector and the audience feature vector to generate a venue-audience joint query vector. Specifically, a two-layer fully connected neural network combined with the ReLU activation function is used to perform deep feature fusion and dimensionality mapping, generating a 128-dimensional venue-audience joint query vector. The calculation formulas involved are as follows:

[0009] Where q represents the venue-audience joint query vector generated by the mapping, with a dimension of 128; ReLU() represents the non-linear activation function; This represents the site feature vector corresponding to the target site. Represents the audience feature vector; Representing vectors with vector The splicing result; This represents the weight matrix of the first fully connected network, with a dimension of 256×(dv+da), where dv represents the embedding dimension of the site feature vector corresponding to the target site, and da represents the embedding dimension of the audience feature vector. This represents the bias vector of the first fully connected layer, with a dimension of 256; This represents the weight matrix of the second fully connected network, with dimensions of 128×256; This represents the bias vector of the second fully connected network, with a dimension of 128; , , and All of these are parameters of a trainable model; The content-template dual-tower network is used to map the metadata vector of each content atom unit and each template feature vector to an embedding space of the same dimension as the venue-audience joint query vector, respectively, to obtain the atom embedding feature vector and the template embedding feature vector. The calculation formulas involved are as follows:

[0010]

[0011] in, This represents the atomic embedding feature vector corresponding to the i-th content atomic unit; This represents the metadata vector of the i-th content atom unit; This represents the template embedding feature vector corresponding to the j-th performance format template; Let represent the template feature vector of the j-th performance form template; tanh represents the hyperbolic tangent activation function; Represents the characteristic projection matrix of atomic units. The dimension is Multiply the dimension by 128; The projection matrix represents the feature of the performance format template. The dimension is Multiply the dimension by 128; This represents the mapping layer bias vector corresponding to the content atomic unit; This represents the mapping layer bias vector corresponding to the template; , , and All of these are parameters of a trainable model.

[0012] The similarity calculation unit is used to calculate the dot product similarity between the obtained venue-audience joint query vector and each atom embedded feature vector, as well as the compatibility score between each content atom unit and each performance format template; the calculation formulas involved are as follows:

[0013]

[0014] in, The dot product similarity between the obtained venue-audience joint query vector q and the atomic embedding feature vector corresponding to the i-th atomic unit is expressed as follows: Let q be the transpose of the obtained venue-audience joint query vector; This represents the compatibility score between the i-th content atom unit and the j-th performance format template. This represents the Sigmoid activation function.

[0015] The combined optimizer is used to pair and combine a set of content atomic units and a performance format template that maximizes the overall compatibility score based on integer linear programming. The decision variables configured in the combined optimizer include x, which represents whether a content atomic unit is selected; y, which represents whether a performance format template is selected; and z, which represents whether a content atomic unit is assigned to the time slot corresponding to the template. The optimization objective is to maximize the overall compatibility score. The constraints of the combined optimizer include: the total duration of the selected content atomic units does not exceed the template duration; the total interaction intensity does not exceed the template's allowed upper limit; and the number of actors does not exceed the venue's upper limit. The combined optimizer outputs a performance content combination scheme, which includes a selected sequence of content atomic units, a selected performance format template, and a time slot allocation mapping table. The integer linear programming in the combinatorial optimizer is solved using the branch and bound method or the bundle search algorithm. When the solution time exceeds a preset upper limit, the current optimal feasible solution is returned.

[0016] Specifically, use Characterizing whether the i-th atomic unit is selected in the final performance scheme and The value is 0 or 1, using Indicates whether the j-th performance format template is selected and The value is 0 or 1, using The time slot representing whether the i-th atomic unit is assigned to the j-th template and The value is 0 or 1; the optimization objective function involved is as follows:

[0017] in, λ represents the optimization objective; λ represents the preset weight balance coefficient.

[0018] The constraints of the combined optimizer include: ① The total duration of the selected content atomic units does not exceed the template duration; ② The total interaction intensity does not exceed the template's allowed upper limit; ③ The number of actors does not exceed the venue's upper limit; ④ Only one fixed performance format template is used for a single performance; ⑤ All selected atomic units are uniquely assigned to their corresponding time slots; ⑥ Content atomic units can only be assigned to the already selected performance format templates.

[0019] The combined optimizer solves the constructed optimization objective function and outputs the performance combination scheme that maximizes the optimization objective.

[0020] Furthermore, the content reorganization module includes: a timeline mapping submodule, a spatial layout mapping submodule, a logical connection verification submodule, and an instruction generation unit; The timeline mapping submodule is used to sort the selected content atomic units according to the allocated time slots based on the time slot division of the selected performance format template, and generate a timeline event list. The spatial layout mapping submodule is used to receive a plan view of the target site, map the required spatial position of each content atom unit to actual physical coordinates, and generate a space-event binding table; the specific mapping process involves the following functions:

[0021] in, This represents the result of the optimal performance coordinate mapping function for the corresponding type of content atomic unit; p represents the set of candidate coordinate points for similar performances within the venue; p represents the candidate point set. Any coordinate vector in the vector; The reference coordinates of the geometric center of the candidate point set; This indicates that the optimal coordinate point P corresponding to the minimum Euclidean distance is obtained; the optimal performance coordinate mapping function result of the spatial layout mapping submodule determines the optimal performance point of each content atom unit, and finally generates the space-event binding table; The logic connection verification submodule is used to detect the plot coherence between adjacent content atomic units, and constructs a plot logic knowledge graph using the sentiment tags and key events of the content atomic units as nodes and causal or temporal relationships as edges. For adjacent content atomic units on the timeline, the cosine distance of the corresponding sentiment tag vector and the shortest path length in the knowledge graph are calculated. If either the cosine distance of the corresponding sentiment tag vector or the shortest path length in the knowledge graph exceeds a preset threshold, the adjacent content atomic units are determined to have a logical break, and a replaceable alternative unit is called from the content atomic library for insertion or replacement. The calculation formula for the cosine distance of the corresponding sentiment tag vector is as follows:

[0022] in, This indicates that the i-th adjacent content atom unit is related to the i-th content atom unit. The cosine distance between the sentiment tag vectors corresponding to each content atom unit; This represents the sentiment tag vector corresponding to the i-th content atom unit; Indicates the first The sentiment tag vector corresponding to each content atom unit; express The modulus length; express The length of the module.

[0023] The instruction generation unit is used to generate a sequence of performance control instructions based on the timeline event list and the space-event binding table; each instruction in the performance control instruction sequence includes a timestamp, device identifier, instruction type, and parameter value.

[0024] Furthermore, the interactive feedback data includes reward amounts, bullet screen text, voting selections, virtual product purchase data, wearable device heart rate data, and facial expression features; The dynamic adjustment module includes: an emotion computing unit, an intent recognition unit, a branch decision unit, and a conflict detection unit; The emotion computing unit is used to perform fusion analysis on bullet screen text, heart rate data, and facial expressions, and output a real-time emotion intensity value; specifically, the calculation formula for calculating the real-time emotion intensity value of the audience at time t is as follows:

[0025] in, This represents the intensity of the audience's emotions at time t; This is a pre-trained BERT sentiment analysis function; The text displayed during the live chat at time t; This represents the heart rate data collected from the wearable device at time t; This indicates the length of the preset normal heart rate fluctuation range; This represents the preset baseline heart rate value for the audience; Let t represent the facial expression feature vector at time t. Each element in the facial expression feature vector is the probability of determining the preset emotion label corresponding to the corresponding element position. The preset emotion labels corresponding to different element positions in the facial expression feature vector are different. This represents the maximum probability value corresponding to each element position in the facial expression feature vector; , and These represent different preset weighting coefficients.

[0026] The intent recognition unit is used to fuse data on reward amounts, voting choices, and virtual product purchases to output an audience intent vector; The branch decision unit is used to select a plot branch based on the obtained real-time emotional intensity value, the audience intention vector and the current plot node, through a preset decision tree. The conflict detection unit is used to maintain the current state (position, emotion, task progress) of each character in the completed performance and check whether the requirements of the selected branch on the character's state are compatible with the current state; if compatible, it outputs an adjustment instruction; otherwise, it requires the branch decision unit to reselect the plot branch; the adjustment instruction includes the content atomic unit to be replaced, the new content atomic unit, the timeline offset, and the ending switch flag; the conflict detection unit also records the list of available props and the sequence of completed actions for each character; when the selected branch requires the use of a certain prop but the state machine shows that the prop is unavailable, the conflict detection unit returns a conflict flag; After receiving the adjustment instructions, the content reorganization module dynamically modifies the execution order of content atomic units, replaces or inserts new content atomic units, changes the ending node, and calls the instruction generation unit again to output the updated performance control instruction sequence and send it to the performance execution module within the remaining performance timeline.

[0027] Furthermore, the dynamic adjustment module also includes a safety monitoring unit, which monitors the abnormal proportion of crowd density or audience heart rate in real time. When the monitored abnormal proportion is greater than or equal to a preset safety threshold, a safety adjustment command is generated to switch the plot to a preset safety mode.

[0028] Furthermore, the system also includes a self-learning update module, used to collect performance data after each performance as a reward signal, and update the dual-tower network parameters every preset time interval based on the obtained reward signal. The performance data includes attendance rate, audience satisfaction, and commercial conversion revenue. During the update of the dual-tower network parameters, the reward signal is a weighted sum of the attendance rate, audience satisfaction, and commercial conversion revenue in the corresponding performance data, and the weight coefficients corresponding to each parameter item in the performance data are preset. The relevant model parameters are updated using a mini-batch gradient descent method, and a loss function is constructed. , This represents the actual compatibility score of the sample.

[0029] Compared with the prior art, the beneficial effects achieved by the present invention are: (1) This invention decomposes the performance content into content atomic units with metadata and separates them from the performance format template, so that the same IP work can be quickly adapted to different venue types without having to start from scratch, which can significantly shorten the production cycle to a certain extent. (2) This invention automatically collects the physical characteristic parameters, commercial target parameters and multi-dimensional characteristic data of the target venue and the audience, and uses a dual-tower network and combinatorial optimizer to dynamically recommend the most suitable atomic unit combination and performance form template for the target venue; (3) This invention can effectively enhance the audience’s sense of participation by collecting audience feedback such as bullet comments, rewards, votes, heart rate, and expressions in real time, and after emotion calculation and intent recognition, combined with the conflict detection mechanism to intelligently obtain subsequent plot branches or endings. Attached Figure Description

[0030] 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 schematic diagram of the structure of the adaptive matching system for performance content across multiple venues according to the present invention. Detailed Implementation

[0031] 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.

[0032] Please see Figure 1 This embodiment provides an adaptive matching system for performance content across multiple venues, including: The venue information acquisition module is used to acquire the physical characteristic parameters and commercial target parameters of the target venue and construct the venue feature vector corresponding to the target venue. The physical characteristic parameters include venue type, space size, maximum audience capacity, and audience flow pattern. The commercial target parameters include box office revenue weight, brand exposure weight, educational purpose weight, and audience dwell time target. Specifically, venue type includes theater, shopping mall, scenic spot, restaurant, bookstore, and street; space size includes length, width, and height; maximum audience capacity represents the upper limit of the total number of audience members that the target venue can accommodate; audience flow pattern includes free flow mode, one-way flow mode, and gathering and dwelling mode; the commercial target parameters include box office revenue weight, brand exposure weight, educational purpose weight, and audience dwell time target. The value range of the box office revenue weight, brand exposure weight, and educational purpose weight parameters is [0,1], and the larger the value, the higher the priority of the corresponding commercial target. In the process of constructing the site feature vector corresponding to the target site, the unique encoding vector of the corresponding site type is obtained within the physical feature parameters of the target site; the unique hot encoding vector of the corresponding audience flow pattern is obtained within the physical feature parameters of the target site; and the original text information is replaced with the corresponding unique hot encoding vector. Then, the physical feature parameters of the target site after replacing the text information and the commercial target parameters are concatenated to obtain the site feature vector corresponding to the target site.

[0033] The audience profile analysis module is used to acquire multi-dimensional feature data of the audience group from the user database or real-time sensors, and output the audience feature vector after normalization and embedding encoding. The multi-dimensional feature data includes age distribution (in this invention, the age distribution is divided into five intervals: 0-17 years old, 18-25 years old, 26-35 years old, 36-50 years old, and over 50 years old, with the total number of people in each interval being 1), consumption preference tags (in this invention, the consumption preference tags are in the form of unique hot codes, and different consumption preference tags correspond to different unique hot codes), interaction willingness scores (with values ​​ranging from [0,1]), and the historical viewing behavior sequence of the most recent preset performances. The historical viewing behavior sequence of each performance consists of the unique hot code and score of the performance type corresponding to the performance session. The content atom library module is used to automatically decompose the original performing arts content into multiple content atom units and generate a metadata vector corresponding to each content atom unit. The metadata vector includes: the unique hot code of the content atom unit type, the minimum number of viewers required, the interaction intensity level (the value range includes 0, 1 and 2), the commercial conversion potential score (the value range is [0,1], the higher the value, the stronger the commercial monetization ability of the unit), the duration, the number of actors required, the emotional tag vector (the emotional tags of this invention include six basic emotions: happiness, sadness, anger, fear, surprise and neutral), and the replaceability tag (the value is 0 or 1, the value of 1 means that the content atom unit supports dynamic replacement, and the value of 0 means that the content unit is not replaceable). The content atom unit types include five types of performing arts: dialogue, song, dance, interactive game and famous scene reproduction. In the content atom library module, the metadata vector of each content atom unit is also bound to a real-time replacement cost parameter, which is used to characterize the cost required to replace the original content atom unit with the corresponding content atom unit during dynamic adjustment; when calling the alternative units, the logical connection verification submodule prioritizes the content atom unit with the lowest corresponding replacement cost.

[0034] The template library module stores several preset performance format templates. The parameters of each preset performance format template include total duration, segmented time axis (the segmented time axis consists of several continuous time intervals, and each time interval is assigned a time slot number, the number of time slots varies between different performance format templates), a list of compatible atomic types for each time slot, a spatial location requirement vector (the spatial location requirement vector consists of the occupancy ratios of the main stage, secondary stage, aisle, and bar area respectively), a list of technical resource requirements (including the number of lighting channels, the number of sound channels, and the number of projectors), and interactive node density (the interactive node density represents the average number of interactive nodes per preset time interval). Based on each template parameter, a corresponding template feature vector is generated. The matching recommendation engine is used to calculate the matching degree between each content atom unit and each performance format template based on the obtained venue feature vector and audience feature vector, and output the top performance content combination schemes with the highest matching degree. Each scheme includes the selected set of content atom units and the corresponding performance format template identifier. The matching recommendation engine includes a venue-audience joint encoder, a content-template dual-tower network, a similarity calculation unit, and a combinatorial optimizer. The venue-audience joint encoder is used to concatenate the venue feature vector and the audience feature vector to generate a venue-audience joint query vector. Specifically, a two-layer fully connected neural network combined with the ReLU activation function is used to perform deep feature fusion and dimensionality mapping, generating a 128-dimensional venue-audience joint query vector. The calculation formulas involved are as follows:

[0035] Where q represents the venue-audience joint query vector generated by the mapping, with a dimension of 128; ReLU() represents the non-linear activation function; This represents the site feature vector corresponding to the target site. Represents the audience feature vector; Representing vectors with vector The splicing result; This represents the weight matrix of the first fully connected network, with a dimension of 256×(dv+da), where dv represents the embedding dimension of the site feature vector corresponding to the target site, and da represents the embedding dimension of the audience feature vector. This represents the bias vector of the first fully connected layer, with a dimension of 256; This represents the weight matrix of the second fully connected network, with dimensions of 128×256; This represents the bias vector of the second fully connected network, with a dimension of 128; , , and All of these are parameters of a trainable model; The content-template dual-tower network is used to map the metadata vector of each content atom unit and each template feature vector to an embedding space of the same dimension as the venue-audience joint query vector, respectively, to obtain the atom embedding feature vector and the template embedding feature vector. The content-template dual-tower network is used to map the metadata vector of each content atom unit and each template feature vector to an embedding space of the same dimension as the venue-audience joint query vector, respectively, to obtain the atom embedding feature vector and the template embedding feature vector. The calculation formulas involved are as follows:

[0036]

[0037] in, This represents the atomic embedding feature vector corresponding to the i-th content atomic unit; This represents the metadata vector of the i-th content atom unit; This represents the template embedding feature vector corresponding to the j-th performance format template; Let represent the template feature vector of the j-th performance form template; tanh represents the hyperbolic tangent activation function; Represents the characteristic projection matrix of atomic units. The dimension is Multiply the dimension by 128; The projection matrix represents the feature of the performance format template. The dimension is Multiply the dimension by 128; This represents the mapping layer bias vector corresponding to the content atomic unit; This represents the mapping layer bias vector corresponding to the template; , , and All of these are parameters of a trainable model.

[0038] The similarity calculation unit is used to calculate the dot product similarity between the obtained venue-audience joint query vector and each atom embedded feature vector, as well as the compatibility score between each content atom unit and each performance format template. The calculation formulas involved are as follows:

[0039]

[0040] in, The dot product similarity between the obtained venue-audience joint query vector q and the atomic embedding feature vector corresponding to the i-th atomic unit is expressed as follows: Let q be the transpose of the obtained venue-audience joint query vector; This represents the compatibility score between the i-th content atom unit and the j-th performance format template. This represents the Sigmoid activation function.

[0041] The combined optimizer is used to pair and combine a set of content atomic units and a performance format template that maximizes the overall compatibility score based on integer linear programming. The decision variables configured in the combined optimizer include x, which represents whether a content atomic unit is selected; y, which represents whether a performance format template is selected; and z, which represents whether a content atomic unit is assigned to the time slot corresponding to the template. The optimization objective is to maximize the overall compatibility score. The constraints of the combined optimizer include: the total duration of the selected content atomic units does not exceed the template duration; the total interaction intensity does not exceed the template's allowed upper limit; and the number of actors does not exceed the venue's upper limit. The combined optimizer outputs a performance content combination scheme, which includes a selected sequence of content atomic units, a selected performance format template, and a time slot allocation mapping table. The integer linear programming problem in the combined optimizer is solved using either branch and bound or bundle search algorithms. When the solution time exceeds a preset upper limit, the current optimal feasible solution is returned. Specifically, using... Characterizing whether the i-th atomic unit is selected in the final performance scheme and The value is 0 or 1, using Indicates whether the j-th performance format template is selected and The value is 0 or 1, using The time slot representing whether the i-th atomic unit is assigned to the j-th template and The value is 0 or 1; the optimization objective function involved is as follows:

[0042] in, λ represents the optimization objective; λ represents the preset weight balance coefficient. In this embodiment, the value of λ is 0.7 by default, which is used to balance the weight ratio of audience adaptation similarity and template compatibility in the overall score.

[0043] The constraints of the combined optimizer include: ① The total duration of the selected content atomic units does not exceed the template duration; ② The total interaction intensity does not exceed the template's allowed upper limit; ③ The number of actors does not exceed the venue's upper limit; ④ Only one fixed performance format template is used for a single performance; ⑤ All selected atomic units are uniquely assigned to their corresponding time slots; ⑥ Content atomic units can only be assigned to the already selected performance format templates.

[0044] The combined optimizer solves the constructed optimization objective function and outputs the performance combination scheme that maximizes the optimization objective.

[0045] The content reorganization module is used to receive the performance content combination scheme output by the matching recommendation engine, retrieve the corresponding performance form template from the form template library module according to the obtained performance form template type, and reorganize the selected content atomic unit set according to the time axis, spatial position and interactive logic in the retrieved performance form template to generate a complete performance script and performance control instruction sequence adapted to the target venue. The content reorganization module includes: a timeline mapping submodule, a spatial layout mapping submodule, a logical connection verification submodule, and an instruction generation unit; The timeline mapping submodule is used to sort the selected content atomic units according to the allocated time slots based on the time slot division of the selected performance format template, and generate a timeline event list. The spatial layout mapping submodule is used to receive a plan view of the target site, map the required spatial position of each content atom unit to actual physical coordinates, and generate a space-event binding table; the specific mapping process involves the following functions:

[0046] in, This represents the result of the optimal performance coordinate mapping function for the corresponding type of content atomic unit; p represents the set of candidate coordinate points for similar performances within the venue; p represents the candidate point set. Any coordinate vector in the vector; The reference coordinates of the geometric center of the candidate point set; This indicates that the optimal coordinate point P corresponding to the minimum Euclidean distance is obtained; the optimal performance coordinate mapping function result of the spatial layout mapping submodule determines the optimal performance point of each content atom unit, and finally generates the space-event binding table; The logic connection verification submodule is used to detect the plot coherence between adjacent content atomic units, and constructs a plot logic knowledge graph using the sentiment tags and key events of the content atomic units as nodes and causal or temporal relationships as edges. For adjacent content atomic units on the timeline, the cosine distance of the corresponding sentiment tag vectors and the shortest path length in the knowledge graph are calculated. If either exceeds a preset threshold, the adjacent content atomic units are determined to have a logical break, and a replacement alternative unit is called from the content atomic library for insertion or replacement. The calculation formula for the cosine distance of the corresponding sentiment tag vectors is as follows:

[0047] in, This indicates that the i-th adjacent content atom unit is related to the i-th content atom unit. The cosine distance between the sentiment tag vectors corresponding to each content atom unit; This represents the sentiment tag vector corresponding to the i-th content atom unit; Indicates the first The sentiment tag vector corresponding to each content atom unit; express The modulus length; express The modulus length; when If the shortest path distance in the knowledge graph is greater than 2, a logical break in the plot is determined. Then, candidate content atomic units are automatically retrieved from the content atomic library. The cosine distance between the i-th content atomic unit and the corresponding sentiment tag vectors of the candidate content atomic units is calculated, along with the distance between the i-th content atomic unit and the corresponding sentiment tag vectors of the candidate content atomic units. The cosine distance between the sentiment tag vectors corresponding to each content atom unit and the candidate content atom units is selected. The candidate content atom units are then compared with the sentiment tag vectors corresponding to the i-th content atom unit and the i-th content atom unit. The cell whose sum of the corresponding cosine distances between the content atomic cells is the smallest is used for insertion or replacement repair.

[0048] The instruction generation unit is used to generate a sequence of performance control instructions based on the timeline event list and the space-event binding table; each instruction in the performance control instruction sequence includes a timestamp, device identifier, instruction type, and parameter value.

[0049] The performance execution module is used to parse and execute the obtained performance control instruction sequence; and to synchronously collect audience interaction feedback data during the performance, and package the collected data into a time-series feedback frame for output; the interaction feedback data includes reward amount, bullet screen text, voting selection, virtual product purchase data, wearable device heart rate data, and facial expression features; The dynamic adjustment module includes: an emotion computing unit, an intent recognition unit, a branch decision unit, and a conflict detection unit; The emotion computing unit is used to perform fusion analysis on bullet screen text, heart rate data, and facial expressions, and output a real-time emotion intensity value; specifically, the calculation formula for calculating the real-time emotion intensity value of the audience at time t is as follows:

[0050] in, This represents the intensity of the audience's emotions at time t; This is a pre-trained BERT sentiment analysis function, and the output of the BERT sentiment analysis function takes values ​​in the range of [-1, 1]. The text displayed during the live chat at time t; This represents the heart rate data collected from the wearable device at time t; This indicates the length of the preset normal heart rate fluctuation range; This represents the preset baseline heart rate value for the audience; Let t represent the facial expression feature vector at time t. Each element in the facial expression feature vector is the probability of determining the preset emotion label corresponding to the corresponding element position. The preset emotion labels corresponding to different element positions in the facial expression feature vector are different. In this embodiment, the preset emotion labels include happiness, sadness, anger, fear, surprise, and neutrality. This represents the maximum probability value corresponding to each element position in the facial expression feature vector; , and These represent different preset weighting coefficients, and in this embodiment, the values ​​of the three are 0.5, 0.3, and 0.2, respectively.

[0051] The intent recognition unit is used to fuse data on reward amounts, voting choices, and virtual product purchases to output an audience intent vector; In this embodiment, the audience intent vector is obtained by taking as input data the amount of the reward, the voting selection, and the purchase of virtual goods, and is processed through a pre-trained machine learning model. The calculation method involved is as follows:

[0052] in, This represents the audience intention vector at time t; This represents the normalized activation function; This represents the reward amount at time t; This represents the voting result at time t; This represents the number of virtual goods purchased at time t; This represents a trainable weight matrix of size 3×3; This represents a trainable three-dimensional bias vector.

[0053] The branch decision unit is used to select a plot branch based on the obtained real-time emotional intensity value, the audience intention vector and the current plot node, through a preset decision tree. The conflict detection unit is used to maintain the current state (position, emotion, task progress) of each character in the completed performance and check whether the requirements of the selected branch on the character's state are compatible with the current state; if compatible, it outputs an adjustment instruction; otherwise, it requires the branch decision unit to reselect the plot branch; the adjustment instruction includes the content atomic unit to be replaced, the new content atomic unit, the timeline offset, and the ending switch flag; the conflict detection unit also records the list of available props and the sequence of completed actions for each character; when the selected branch requires the use of a certain prop but the state machine shows that the prop is unavailable, the conflict detection unit returns a conflict flag; After receiving the adjustment instructions, the content reorganization module dynamically modifies the execution order of content atomic units, replaces or inserts new content atomic units, changes the ending node, and calls the instruction generation unit again to output the updated performance control instruction sequence and send it to the performance execution module within the remaining performance timeline.

[0054] The dynamic adjustment module also includes a safety monitoring unit, which monitors the abnormal percentage of crowd density or audience heart rate in real time. When the monitored abnormal percentage is greater than or equal to a preset safety threshold, a safety adjustment command is generated to switch the plot to a preset safety mode.

[0055] The system also includes a self-learning update module, used to collect performance data after each performance as a reward signal, and update the dual-tower network parameters every preset time interval based on the obtained reward signal. The performance data includes attendance rate, audience satisfaction, and commercial conversion revenue. During the update of the dual-tower network parameters, the reward signal is a weighted sum of the attendance rate, audience satisfaction, and commercial conversion revenue in the corresponding performance data, and the weight coefficients corresponding to each parameter item in the performance data are preset. The relevant model parameters are updated using a mini-batch gradient descent method, and a loss function is constructed. , This represents the actual compatibility score of the sample.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0057] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive matching system for performance content across multiple venues, characterized in that: include: The site information acquisition module is used to acquire the physical characteristic parameters and commercial target parameters of the target site and construct the site feature vector corresponding to the target site. The audience profiling analysis module is used to acquire multi-dimensional feature data of the audience group from the user database or real-time sensors, and output the audience feature vector after normalization and embedding encoding. The content atom library module is used to automatically break down the original performing arts content into multiple content atom units and generate a metadata vector corresponding to each content atom unit; The template library module stores several preset performance templates. The parameters of each preset performance template include total duration, segmented time axis, list of atom types adapted to each time slot, spatial location requirement vector, list of technical resource requirements, and interactive node density. Based on the parameters of each template, a corresponding template feature vector is generated. The matching recommendation engine is used to calculate the matching degree between each content atom unit and each performance format template based on the obtained venue feature vector and audience feature vector, and output the top performance content combination schemes with the highest matching degree. Each scheme includes the selected set of content atom units and the corresponding performance format template identifier. The content reorganization module is used to receive the performance content combination scheme output by the matching recommendation engine, retrieve the corresponding performance form template from the form template library module according to the obtained performance form template type, and reorganize the selected content atomic unit set according to the time axis, spatial position and interactive logic in the retrieved performance form template to generate a complete performance script and performance control instruction sequence adapted to the target venue. The performance execution module is used to parse and execute the obtained performance control instruction sequence; During the performance, the system simultaneously collects audience interaction feedback data and packages the collected data into time-series feedback frames for output. The dynamic adjustment module is used to perform sentiment analysis and intent recognition on the obtained interactive feedback data, generate plot branch adjustment instructions or ending switching instructions and feed them back to the content reorganization module. After receiving the instructions, the content reorganization module modifies the content atomic units in the remaining timeline in real time and regenerates the performance script and performance control instruction sequence adapted to the target venue.

2. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The physical characteristic parameters include venue type, space size, maximum audience capacity, and audience flow pattern; the commercial objective parameters include box office revenue weight, brand exposure weight, educational purpose weight, and audience dwell time target. The multidimensional feature data includes age distribution, consumption preference tags, interaction willingness scores, and historical performance viewing behavior sequences; The metadata vector includes: unit type, minimum number of viewers required, interaction intensity level, commercial conversion potential score, duration, number of actors required, emotional tag vector, and replaceable tags.

3. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The metadata vector of each content atom unit in the content atom library module is also bound to a real-time replacement cost parameter, which is used to characterize the cost required to replace the original content atom unit with the corresponding content atom unit during dynamic adjustment; the logical connection verification submodule prioritizes the content atom unit with the lowest corresponding replacement cost when calling the alternative units.

4. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The matching recommendation engine includes a venue-audience joint encoder, a content-template dual-tower network, a similarity calculation unit, and a combinatorial optimizer. The venue-audience joint encoder is used to concatenate the venue feature vector and the audience feature vector to generate a venue-audience joint query vector. The content-template dual-tower network is used to map the metadata vector of each content atom unit and each template feature vector to an embedding space of the same dimension as the venue-audience joint query vector, respectively, to obtain the atom embedding feature vector and the template embedding feature vector. The similarity calculation unit is used to calculate the dot product similarity between the obtained venue-audience joint query vector and each atom embedded feature vector, as well as the compatibility score between each content atom unit and each performance format template. The combined optimizer is used to pair and combine a set of content atomic units and a performance format template that maximizes the overall compatibility score based on integer linear programming search. The decision variables configured in the combined optimizer include x, which represents whether a content atomic unit is selected; y, which represents whether a performance format template is selected; and z, which represents whether a content atomic unit is assigned to the time slot corresponding to the template. The optimization goal is to maximize the overall compatibility score; The constraints of the combinatorial optimizer include: the total duration of the selected content atomic units does not exceed the template duration, the total interaction intensity does not exceed the template's allowed upper limit, and the number of actors does not exceed the venue's upper limit; the combinatorial optimizer outputs a performance content combination scheme, which includes the selected content atomic unit sequence, the selected performance format template, and a time slot allocation mapping table; The integer linear programming in the combinatorial optimizer is solved using the branch and bound method or the bundle search algorithm. When the solution time exceeds a preset upper limit, the current optimal feasible solution is returned.

5. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The content reorganization module includes: a timeline mapping submodule, a spatial layout mapping submodule, a logical connection verification submodule, and an instruction generation unit; The timeline mapping submodule is used to sort the selected content atomic units according to the allocated time slots based on the time slot division of the selected performance format template, and generate a timeline event list. The spatial layout mapping submodule is used to receive a plan view of the target site, map the required spatial position of each content atom unit to the actual physical coordinates, and generate a space-event binding table. The logic connection verification submodule is used to detect the plot coherence between adjacent content atomic units, and to construct a plot logic knowledge graph with the emotional tags and key events of the content atomic units as nodes and causal or temporal relationships as edges. For adjacent content atomic units on the time axis, the cosine distance of the corresponding emotional tag vector and the shortest path length in the knowledge graph are calculated. If either exceeds a preset threshold, the adjacent content atomic units are determined to be logically broken, and alternative units are called from the content atomic library for insertion or replacement. The instruction generation unit is used to generate a sequence of performance control instructions based on the timeline event list and the space-event binding table; each instruction in the performance control instruction sequence includes a timestamp, device identifier, instruction type, and parameter value.

6. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The interactive feedback data includes reward amount, bullet screen text, voting selection, virtual product purchase data, wearable device heart rate data, and facial expression features; The dynamic adjustment module includes: an emotion computing unit, an intent recognition unit, a branch decision unit, and a conflict detection unit; The emotion computing unit is used to fuse and analyze bullet screen text, heart rate data, and facial expressions to output real-time emotion intensity values; The intent recognition unit is used to fuse data on reward amounts, voting choices, and virtual product purchases to output an audience intent vector. The branch decision unit is used to select a plot branch based on the obtained real-time emotional intensity value, the audience intention vector and the current plot node, through a preset decision tree. The conflict detection unit is used to maintain the current state of each character in the completed performance and check whether the requirements of the selected branch for the character's state are compatible with the current state; if compatible, it outputs an adjustment instruction; otherwise, it requires the branch decision unit to reselect the plot branch; the adjustment instruction includes the content atomic unit to be replaced, the new content atomic unit, the timeline offset, and the ending switching flag; the conflict detection unit also records the list of available props and the sequence of completed actions for each character; when the selected branch requires the use of a certain prop but the state machine shows that the prop is unavailable, the conflict detection unit returns a conflict flag; After receiving the adjustment instructions, the content reorganization module dynamically modifies the execution order of content atomic units, replaces or inserts new content atomic units, changes the ending node, and calls the instruction generation unit again to output the updated performance control instruction sequence and send it to the performance execution module within the remaining performance timeline.

7. The adaptive matching system for performance content across multiple venues according to claim 1, characterized in that: The dynamic adjustment module also includes a safety monitoring unit, which monitors the abnormal percentage of crowd density or audience heart rate in real time. When the monitored abnormal percentage is greater than or equal to a preset safety threshold, a safety adjustment command is generated to switch the plot to a preset safety mode.

8. The adaptive matching system for performance content across multiple venues according to claim 4, characterized in that: The system also includes a self-learning update module, which collects performance data after each performance as a reward signal and updates the dual-tower network parameters every preset time interval based on the obtained reward signal. The performance data includes attendance rate, audience satisfaction, and commercial conversion amount.

Images