Film and television membership management system based on big data

Abstract

This invention discloses a film and television membership management system based on big data, including a data acquisition module, a membership lifecycle segmentation module, a membership lifecycle conversion module, and a film and television membership intelligent management module. This invention relates to the field of data processing technology, specifically to a film and television membership management system based on big data. This solution innovatively combines membership lifecycle segmentation with membership lifecycle conversion, enabling real-time dynamic adjustment of management and conversion strategies based on film and television membership data, thus improving the real-time adaptability of film and television membership management. It proposes an improved clustering algorithm using granular sphere quadratic optimization, a two-factor density calculation method, and a dynamic allocation strategy, improving the accuracy and adaptability of lifecycle segmentation. Furthermore, it employs a group feedback learning strategy and a logarithmic decreasing strategy to improve the particle optimization algorithm, enhancing the accuracy and adaptability of the model output results, promoting precise optimization of film and television membership conversion strategies, and realizing intelligent and personalized management of film and television members.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a film and television membership management system based on big data. Background Technology

[0002] The big data-based film and television membership management system is a system that uses big data technology and artificial intelligence algorithms to comprehensively manage the members of film and television platforms. By collecting and analyzing film and television membership data, it accurately segments member groups and achieves intelligent management, helping the platform improve member participation and satisfaction, thereby effectively increasing user retention and loyalty.

[0003] However, traditional film and television membership management systems suffer from technical problems such as the inability to dynamically manage members' needs in real time, leading to low management efficiency, high churn rates, and insufficient member activity. Existing clustering algorithms for segmenting film and television members also suffer from technical problems such as the inability to accurately segment members with differentiated behaviors and the lack of consideration for the global correlation of member groups, resulting in low accuracy in segmenting the lifecycle stages of film and television members. Furthermore, existing hyperparameter settings for member lifecycle transformation models are unreasonable, and the reliance on linear inertia weight adjustment and insufficient inter-particle information interaction during hyperparameter optimization make individuals prone to getting trapped in local optima, thus affecting the accuracy of model output results. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a big data-based film and television membership management system. Addressing the technical problems of traditional film and television membership management systems, such as the inability to dynamically manage members' needs in real time, leading to low management efficiency, high churn rates, and insufficient member activity, this solution innovatively combines member lifecycle segmentation with member lifecycle conversion. First, an improved clustering algorithm is used to accurately segment the member lifecycle stages, and then the conversion strategy is dynamically adjusted based on these stages. This allows for real-time adjustments to management and conversion strategies based on member data, improving the system's real-time adaptability, enhancing the real-time and dynamic nature of member conversion, effectively increasing member activity and loyalty, and achieving efficient and intelligent management of film and television members. Furthermore, addressing the technical problems of existing clustering algorithms for film and television membership segmentation, such as the inability to accurately segment members with differentiated behaviors and the lack of consideration for the global correlation of member groups, resulting in low accuracy in segmenting the member lifecycle stages, this solution innovatively proposes a granular sphere secondary optimization, a two-factor density calculation method, and a dynamic allocation strategy. An improved clustering algorithm enhances the accuracy of member lifecycle stage boundaries, ensuring accurate segmentation of members with differentiated behaviors. This addresses the limitation of traditional methods to local information, strengthens consideration of global and local correlations, and effectively improves the accuracy and adaptability of lifecycle segmentation, thus achieving intelligent management of film and television memberships. Addressing the technical issues of existing member lifecycle conversion models having unreasonable hyperparameter settings, relying on linear inertia weight adjustments and insufficient inter-particle information interaction during hyperparameter optimization, leading to individuals easily getting trapped in local optima and thus affecting the accuracy of model output results, this solution innovatively adopts a swarm feedback learning strategy in the particle swarm optimization algorithm. This increases the dimension of inter-particle information interaction, preventing particles from excessively clustering near local optima. A logarithmic decreasing strategy is used to adjust the dynamic balance search capability of inertia weights and adapt to individual particle differences. This significantly improves the efficiency and accuracy of model hyperparameter search, enhances the performance of the conversion model, and improves the accuracy and adaptability of model output results. This effectively promotes the precise optimization of film and television membership conversion strategies, ultimately achieving intelligent and personalized management of the film and television membership lifecycle.

[0005] The technical solution adopted by the present invention is as follows: The film and television membership management system based on big data provided by the present invention includes a data acquisition module, a membership lifecycle segmentation module, a membership lifecycle conversion module, and a film and television membership intelligent management module;

[0006] The data acquisition module specifically obtains optimized data for film and television membership management through data collection and optimization.

[0007] The member lifecycle segmentation module specifically improves the clustering algorithm through granular sphere secondary optimization, two-factor density calculation method and dynamic allocation strategy, establishes a member lifecycle segmentation model, and inputs the behavioral data of film and television members into the model to obtain the lifecycle stages of film and television members;

[0008] The member lifecycle conversion module specifically involves constructing a member lifecycle conversion model, training the model using historical data, improving the particle optimization algorithm using a group feedback learning strategy and a logarithmic decreasing strategy, optimizing the hyperparameters of the conversion model, obtaining the optimal combination of hyperparameters and updating the model hyperparameters to form the optimal member lifecycle conversion model, and inputting real-time data into the model to obtain a real-time film and television member conversion strategy.

[0009] The intelligent management module for film and television memberships specifically enables intelligent management of film and television memberships based on the lifecycle stage of the membership and real-time membership conversion strategies.

[0010] Furthermore, the data acquisition module specifically involves collecting data from multiple sources from the film and television management platform and optimizing the collected raw data of film and television membership management to obtain optimized data for film and television membership management.

[0011] The raw data for film and television membership management includes historical film and television membership stage data and real-time film and television membership stage data; both historical and real-time film and television membership stage conversion data include user personal data, user behavior data, user payment data, and user activity data; the historical film and television membership stage conversion data also includes film and television membership lifecycle tags, film and television membership conversion strategies, and corresponding film and television membership conversion results; the data optimization specifically involves data cleaning, data normalization, and data encoding processing of the raw data to obtain optimized film and television membership management data.

[0012] Furthermore, the member lifecycle segmentation module specifically includes the following steps:

[0013] The initialization of stable granularity balls involves first constructing a set of film and television member feature vectors, then initializing the granularity ball set to an empty set, and adding the film and television member feature vector set as the initial granularity ball to the granularity ball queue to be processed. Subsequently, the granularity ball iterative segmentation operation is performed repeatedly until no new granularity balls are added to the granularity ball queue to be processed, thus obtaining the initial stable granularity ball set.

[0014] The secondary optimization of the granular spheres involves calculating the radius *r* of each sphere in the initial stable sphere set, then sorting the spheres in the initial stable sphere set in ascending order by radius *r*, and calculating the radius increment of adjacent spheres. The radius of the sphere with the largest radius increment is set as the radius threshold. Finally, traverse the set of initial stable spheres, and if the radius of the sphere is... If the result is satisfactory, it is directly added to the final granularity sphere candidate set; otherwise, an iterative granularity sphere segmentation operation is performed, and the segmented sub-granularity spheres are added to the final granularity sphere candidate set. Indicates the first The radius of a sphere of a certain size. Indicates the first The radius of a single sphere;

[0015] The comprehensive local density is calculated specifically by using a two-factor density calculation method to calculate the internal density and neighborhood density of each sphere. These two density values ​​are then multiplied to obtain the comprehensive local density of each sphere. The formula used is as follows:

[0016] ;

[0017] ;

[0018] ;

[0019] In the formula, Indicates particle size spheres Internal density, Indicates particle size spheres neighborhood density, Indicates particle size spheres The member feature vector with median Euclidean distance from the center. Indicates the scaling factor. Represents particle size spheres The center Represents the Euclidean distance function. Indicates proximity The j-th granular sphere, Indicates particle size spheres The center Indicates particle size spheres Given a set of neighboring spheres, select m spheres at a distance of [m]. The closest granular sphere in the Euclidean distance is used as the granular sphere. The set of neighboring spheres, Represents particle size spheres The overall local density, Represents particle size spheres Number of film and television members covered Represents the i-th granular sphere;

[0020] Cluster center selection involves first calculating the relative distance of each sphere, multiplying the comprehensive local density of each sphere by the relative distance to obtain the decision value of each sphere, sorting the decision values ​​of all spheres in descending order, and selecting the Q spheres with the highest decision values ​​as cluster centers.

[0021] The dynamic allocation of non-central spheres includes the following steps:

[0022] The initial allocation is as follows: for each granular sphere that serves as a cluster center, its m neighboring granular spheres are directly assigned to the cluster to which the cluster center belongs;

[0023] The queue to be allocated is initialized by sorting the unallocated granular balls in descending order according to their neighborhood density values ​​and adding them to the queue to be allocated.

[0024] Dynamic threshold allocation, specifically, initializing candidate allocation thresholds. Then, iterate through each granular ball in the queue to be assigned, and calculate the number of granular balls in the set of neighboring granular balls that have been assigned to the cluster to which the cluster center belongs. ,like Then calculate the membership degree value of the granular sphere to each cluster, assign it to the cluster with the largest membership degree value according to the membership degree value, and remove the granular sphere from the assignment queue. If a granularity ball is not found in the list of granularity balls to be assigned, the process is to temporarily skip that granularity ball and continue checking other granularity balls to be assigned. Finally, the candidate assignment threshold is adjusted, and the process continues to traverse the assignment queue until any dynamic assignment stopping condition is met, where m represents the number of neighboring granularity balls for each granularity ball. The dynamic assignment stopping condition includes the assignment queue being empty and... The formula used is as follows:

[0025] ;

[0026] In the formula, Represents particle size spheres Belongs to cluster The membership value, Let q be the q-th cluster. Represents particle size spheres In the set of neighboring spheres, belonging to the cluster The number of spheres with varying particle sizes;

[0027] The grain size ball bottom allocation is specifically as follows: for grain size balls that are still unassigned, calculate the Euclidean distance to the assigned grain size balls from the set of neighboring grain size balls, select the nearest assigned grain size ball, and assign the grain size ball to the cluster to which the assigned grain size ball belongs;

[0028] The clustering results are output as follows: after all the granular balls have been assigned, they are merged into the corresponding cluster set to form multiple independent clusters. Each cluster represents a lifecycle stage of a movie and TV membership.

[0029] A member lifecycle segmentation model is constructed by initializing the stable granular sphere, performing secondary optimization of the granular sphere, calculating the comprehensive local density, selecting cluster centers, dynamically allocating non-central granular spheres, and outputting clustering results. This improved clustering algorithm is then used as the member lifecycle segmentation model.

[0030] The real-time segmentation of film and television membership lifecycle involves inputting lifecycle segmentation data into a membership lifecycle segmentation model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the clustering output results. The cluster label with the highest frequency of occurrence is selected as the film and television membership lifecycle label for that cluster, thus obtaining the lifecycle stage of the film and television membership.

[0031] Furthermore, the member lifecycle conversion module specifically includes the following steps:

[0032] A member lifecycle conversion model is established, specifically by constructing a member lifecycle conversion model based on the Q-learning algorithm and training the model using historical film and television membership stage data as training data to obtain the trained member lifecycle conversion model.

[0033] The conversion model hyperparameter optimization specifically involves obtaining the optimal hyperparameter combination of the model through an improved particle optimization algorithm, and updating the model's hyperparameters based on this optimal combination to obtain the optimal member lifecycle conversion model. This includes the following steps:

[0034] Initializing the particle swarm involves encoding the hyperparameters of the trained member lifecycle transformation model into individual particle position vectors, and generating them using a random initialization method. The position vectors of individual particles constitute the initial particle swarm;

[0035] The particle fitness value is calculated by calculating the fitness value of individual particles in the population; the performance of the member lifecycle conversion model established based on the position of individual particles is used as the fitness value of individual particles.

[0036] The population average optimal position is calculated by taking the average of the optimal positions of all individual particles based on a population feedback learning strategy; the formula used is as follows:

[0037] ;

[0038] In the formula, This represents the average optimal position of the population. This represents the optimal position of each individual particle, and k represents the index of that individual particle.

[0039] The adaptive inertia weights are calculated by adjusting them using a logarithmic decreasing strategy; the formula used is as follows:

[0040] ;

[0041] In the formula, This indicates that the k-th particle is in the... Inertia weights in the next iteration Indicates the initial inertia weight. Indicates the final inertia weight. This represents the adjustment coefficient, which controls the rate at which the inertia weight decreases. This represents the maximum number of iterations, and t represents the current number of iterations. This indicates that the k-th particle is in The optimal position in the next iteration;

[0042] Individual particle updates specifically involve updating particle velocity and position.

[0043] The particle search terminates when the optimal particle position is updated, and the search ends and the global optimal particle position is obtained when any one of the search termination conditions is met; the global optimal particle position specifically refers to the optimal combination of hyperparameters of the model.

[0044] The optimal hyperparameter update specifically involves updating the hyperparameter configuration of the model based on the optimal combination of hyperparameters to obtain the optimal member lifecycle conversion model.

[0045] The acquisition of member lifecycle conversion strategies involves inputting the current lifecycle stage of film and television members and real-time film and television member stage conversion data into the optimal member lifecycle conversion model to obtain real-time film and television member conversion strategies.

[0046] Furthermore, the intelligent management module for film and television memberships specifically implements personalized conversion strategies based on the lifecycle stages of film and television members and in conjunction with real-time film and television membership conversion strategies, thereby achieving intelligent and comprehensive management of the lifecycle of film and television members.

[0047] The beneficial effects achieved by the present invention using the above solution are as follows:

[0048] (1) In view of the technical problems in traditional film and television membership management systems, which are unable to dynamically manage film and television members in real time according to their needs, resulting in low management efficiency, high member churn rate and insufficient member activity, this solution innovatively combines member life cycle division with member life cycle conversion. First, an improved clustering algorithm is used to accurately divide the member life cycle stages, and the conversion strategy is dynamically adjusted according to these stages. The system can dynamically adjust management and conversion strategies in real time according to the data of film and television members, which improves the real-time adaptability of the system, enhances the real-time and dynamic nature of film and television member conversion, effectively improves member activity and loyalty, and realizes efficient and intelligent management of film and television members.

[0049] (2) In view of the technical problems in the existing clustering algorithms applicable to the classification of film and television members, such as the inability to accurately classify members with differentiated behaviors and the lack of consideration for the global correlation of member groups, resulting in low accuracy of the classification of film and television member life cycle stages, this solution innovatively proposes to improve the clustering algorithm by adopting granular sphere secondary optimization, two-factor density calculation method and dynamic allocation strategy, which improves the accuracy of the boundary of member life cycle stages, ensures the accurate classification of members with differentiated behaviors, solves the problem that traditional methods are limited to local information, enhances the consideration of global and local correlation, effectively improves the accuracy and adaptability of life cycle classification, and thus realizes intelligent management of film and television members.

[0050] (3) To address the technical problems of existing hyperparameter settings for member lifecycle conversion models being unreasonable, relying on linear inertia weight adjustment and insufficient inter-particle information interaction during hyperparameter optimization, which makes individuals prone to getting trapped in local optima and thus affecting the accuracy of model output results, this solution innovatively adopts a group feedback learning strategy in the particle swarm optimization algorithm, which increases the dimension of inter-particle information interaction and avoids excessive particle aggregation near local optima; it adopts a logarithmic decreasing strategy to adjust the dynamic balance search capability of inertia weight and adapt to individual particle differences; it significantly improves the efficiency and accuracy of model hyperparameter search, enhances the performance of the conversion model, improves the accuracy and adaptability of model output results, effectively promotes the precise optimization of film and television member conversion strategies, and ultimately realizes intelligent and personalized management of film and television member lifecycle. Attached Figure Description

[0051] Figure 1 A schematic diagram of the modules of the big data-based film and television membership management system provided by the present invention;

[0052] Figure 2 A flowchart illustrating the process of dividing the member lifecycle into modules;

[0053] Figure 3 A flowchart illustrating the member lifecycle conversion module;

[0054] Figure 4 A flowchart illustrating the process of optimizing hyperparameters of the conversion model in the member lifecycle conversion module;

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0057] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0058] Example 1, see Figure 1 The film and television membership management system based on big data provided by the present invention includes a data acquisition module, a membership lifecycle segmentation module, a membership lifecycle conversion module, and a film and television membership intelligent management module.

[0059] The data acquisition module specifically obtains optimized data for film and television membership management through data collection and optimization, and sends the data to the membership lifecycle segmentation module and the membership lifecycle conversion module.

[0060] The member lifecycle segmentation module receives data sent by the data acquisition module and uses it to segment the lifecycle stages of film and television members. Specifically, it establishes a member lifecycle segmentation model by improving the clustering algorithm, inputs the behavioral data of film and television members into the model, obtains the lifecycle stages of film and television members, and sends the data to the member lifecycle conversion module and the film and television member intelligent management module.

[0061] The member lifecycle conversion module receives data from the data acquisition module and the member lifecycle segmentation module, and provides personalized conversion strategies for each film and television member's lifecycle stage. Specifically, it constructs a member lifecycle conversion model, trains the model using historical data, improves the particle optimization algorithm, optimizes the hyperparameters of the conversion model, obtains the optimal hyperparameter combination of the model and updates the model hyperparameters to form the optimal member lifecycle conversion model, and inputs real-time data into the model to obtain a real-time film and television member conversion strategy, and sends the data to the film and television member intelligent management module.

[0062] The intelligent management module for film and television memberships receives data from the membership lifecycle segmentation module and the membership lifecycle conversion module, and intelligently manages film and television memberships based on the lifecycle stages of film and television members and real-time conversion strategies.

[0063] By performing the above operations, this solution addresses the technical problems inherent in traditional film and television membership management systems, which are unable to dynamically manage film and television memberships in real time based on member needs, resulting in low management efficiency, high member churn rates, and insufficient member activity. It innovatively combines member lifecycle segmentation with member lifecycle conversion. First, an improved clustering algorithm is used to accurately segment the member lifecycle stages, and then the conversion strategy is dynamically adjusted based on these stages. This allows for real-time adjustments to management and conversion strategies based on film and television member data, improving the system's real-time adaptability, enhancing the real-time and dynamic nature of member conversion, effectively increasing member activity and loyalty, and achieving efficient and intelligent management of film and television members.

[0064] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, the data acquisition module collects data from multiple sources from the film and television management platform and optimizes the collected raw film and television membership management data to obtain optimized film and television membership management data. The raw film and television membership management data includes historical film and television membership stage data and real-time film and television membership stage data. Both the historical film and television membership stage conversion data and the real-time film and television membership stage conversion data include user personal data, user behavior data, user payment data, and user activity data. The historical film and television membership stage conversion data also includes film and television membership lifecycle tags, film and television membership conversion strategies, and corresponding film and television membership conversion results.

[0065] The user's personal data includes gender, region, age, film and television preferences, and whether they are a member;

[0066] The user behavior data includes the user's viewing history, interaction data, comments, movie and TV ratings, and social sharing;

[0067] The user payment data includes subscription membership, payment frequency, and payment amount;

[0068] The user activity data includes login frequency, viewing duration, types of content viewed, and viewing frequency.

[0069] The film and television member lifecycle tag specifically refers to the current lifecycle stage of the film and television member, and this is used as a cluster tag, including potential member stage, new member stage, active member stage, high-value member stage, churn risk member stage, churned member stage, and reactivated member stage.

[0070] The film and television membership conversion strategy is a personalized conversion strategy designed for each stage of the membership lifecycle to convert members into high-value members. It aims to improve member activity, retention rate and overall lifetime value, including film and television membership subscription coupons and personalized incentives.

[0071] The data optimization is used to improve the integrity and consistency of the original data. Specifically, it involves data cleaning, data normalization, and data encoding of the original data to obtain optimized data for film and television membership management. This includes the following steps:

[0072] Data cleaning is used to eliminate errors, missing values, and inconsistencies in data, specifically by imputing missing values ​​and removing outliers from the original data.

[0073] The missing value imputation specifically involves filling in missing values ​​using the mean imputation method; the outlier removal specifically involves detecting and removing extreme values ​​and logical outliers in the original data using the IQR method.

[0074] Data normalization is used to adjust the scale and range of data. Specifically, the Z-Score standardization method standardizes all continuous variables to ensure that the data is within a uniform range.

[0075] Data encoding processing is used to convert non-numerical data into a numerical format. Specifically, it uses a label encoding method to encode the category fields in the original data and map them to integer values.

[0076] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. Specifically, the member lifecycle segmentation module establishes a member lifecycle segmentation model using an improved clustering algorithm, inputs data into the model, and obtains the lifecycle stage of each film and television member; it includes the following steps:

[0077] The initialization of stable granular spheres is used to generate initial stable granular spheres that cover similar film and television membership lifecycles, providing basic data units for subsequent membership lifecycle stage segmentation. Specifically, firstly, based on user personal data, user behavior data, user payment data, and user activity data in the film and television membership management optimization data, a film and television membership feature vector set is constructed. Secondly, the granular sphere set is initialized to an empty set, and the film and television membership feature vector set is added to the queue of granular spheres to be processed as the initial granular spheres. Subsequently, the granular sphere iterative segmentation operation is performed repeatedly until no new granular spheres are added to the queue of granular spheres to be processed, that is, all spheres to be processed have been split to a stable state or do not need to be split, resulting in the initial stable granular sphere set.

[0078] The granular sphere iterative segmentation operation includes confirming the center of the granular sphere, allocating film and television member samples, and updating the set state;

[0079] The process of identifying the center of the granular sphere specifically involves taking the first granular sphere from the granular sphere queue according to the first-in-first-out principle, calculating the mean of all member feature vectors within that granular sphere to obtain the distance o from the center of the granular sphere, and then identifying the film and television member feature vector within that granular sphere that is furthest from the center o. The particle size inside the sphere and The feature vector of the movie / TV membership with the furthest feature distance Finally, we take o and... , The midpoint is taken as the center of the two sub-granular spheres; the characteristic distance is calculated using Euclidean distance; the formula used is as follows:

[0080] ;

[0081] ;

[0082] ;

[0083] In the formula, N represents the number of film and television member samples covered by the granular sphere. This represents the feature vector of the nth film and television member within this granularity sphere. and These represent the centers of the two sub-spheres;

[0084] The specific process of allocating film and television membership samples is to calculate the first granularity sphere. Each film and television member's feature vector is then mapped to... and Using the Euclidean distance, each film and television member is assigned to a sub-sphere that is closer in distance, forming a sub-granular sphere. and ;

[0085] The updated set state is specifically calculated separately. , and The distribution metric is calculated, and the weighted distribution metric of the sub-granular spheres is calculated. ,like This indicates that the compactness of the daughter balls did not improve after splitting. Add to the initial stable set of grain size spheres; otherwise, it indicates that the grain size sphere splitting is effective, and the sub-grain size spheres are... and Add the teams to the list of particle size to be processed; the formula used is as follows:

[0086] ;

[0087] ;

[0088] In the formula, Indicates sub-granular spheres Distribution measure, Indicates the first particle size sphere Distribution measure, Sub-sized balls Distribution measure, Indicates sub-granular spheres Number of film and television members covered Sub-sized balls Number of film and television members covered;

[0089] The granularity sphere secondary optimization is used to improve the distribution accuracy of the initial stable granularity spheres, address the issue of mixing across member lifecycle stages in the initial granularity spheres, and enhance the ability of the granularity spheres to characterize the boundaries of member lifecycle stages. Specifically, it involves calculating the radius r of each granularity sphere in the initial stable granularity sphere set, then sorting the granularity spheres in the initial stable granularity sphere set in ascending order by radius r, and calculating the radius increment of adjacent granularity spheres. The radius of the sphere with the largest radius increment is set as the radius threshold. Finally, traverse the set of initial stable spheres, and if the radius of the sphere is... If the result is positive, it is directly added to the final granularity sphere candidate set; otherwise, the granularity sphere iterative segmentation operation in the stable granularity sphere initialization is performed, and the segmented sub-granularity spheres are added to the final granularity sphere candidate set. Indicates the first The radius of a sphere of a certain size. Indicates the first The radius of a single sphere;

[0090] The comprehensive local density is calculated to quantify the strength of the film and television membership stage features of each final granular sphere in the candidate set of final granular spheres, accurately identify the core stage of the membership life cycle, and characterize the concentration of membership behavior within the granular sphere and the neighborhood correlation between spheres. Specifically, the internal density and neighborhood density of each granular sphere are calculated separately using the two-factor density calculation method, and finally the two density values ​​are multiplied to obtain the comprehensive local density of each granular sphere.

[0091] The internal density reflects the distribution and concentration of the life cycle stage characteristics of film and television members within the granular sphere. A higher internal density indicates that the member behavior within the granular sphere is highly consistent, suggesting that this granular sphere represents a member group in a stable life cycle stage.

[0092] The neighborhood density reflects the correlation between the granular sphere and its neighboring granular spheres, that is, the similarity of the member behavior characteristics within the granular sphere to other member groups. A higher neighborhood density indicates that the granular sphere is similar to the characteristics of its neighboring groups, which helps to identify the transition zone with other life cycle stages.

[0093] The formula used is as follows:

[0094] ;

[0095] ;

[0096] ;

[0097] In the formula, Represents particle size spheres Internal density, Represents particle size spheres neighborhood density, Represents particle size spheres The member feature vector with median Euclidean distance from the center. This represents the scaling factor, and its value range is... between, Represents particle size spheres The center Represents the Euclidean distance function. Indicates proximity The j-th granular sphere, Represents particle size spheres The center Represents particle size spheres Given a set of neighboring spheres, select m spheres at a distance of [m]. The closest granular sphere in the Euclidean distance is used as the granular sphere. The set of neighboring spheres, Represents particle size spheres The overall local density, Represents particle size spheres Number of film and television members covered Represents the i-th granular sphere;

[0098] Cluster center selection is used to identify cluster centers for the core stages of the film and television membership lifecycle. Specifically, the relative distance of each granular sphere is first calculated, and the comprehensive local density of each granular sphere is multiplied by the relative distance to obtain the decision value of each granular sphere. The decision values ​​of all granular spheres are sorted in descending order, and the top Q granular spheres with the highest decision values ​​are selected as cluster centers.

[0099] Non-centralized granularity sphere dynamic allocation is used to dynamically assign film and television members who are not clustered to the corresponding lifecycle stage, ensuring that each member is accurately assigned to the most suitable lifecycle stage; specifically, it includes the following steps:

[0100] The initial allocation is as follows: for each granular sphere that serves as a cluster center, its m neighboring granular spheres are directly assigned to the cluster to which the cluster center belongs;

[0101] The queue to be allocated is initialized by sorting the unallocated granular balls in descending order according to their neighborhood density values ​​and adding them to the queue to be allocated.

[0102] Dynamic threshold allocation, specifically, initializing candidate allocation thresholds. Then, iterate through each granular ball in the queue to be assigned, and calculate the number of granular balls in the set of neighboring granular balls that have been assigned to the cluster to which the cluster center belongs. ,like Then calculate the membership degree value of the granular sphere to each cluster, assign it to the cluster with the largest membership degree value according to the membership degree value, and remove the granular sphere from the assignment queue. If a certain value is not found, the current granularity ball is temporarily skipped, and other granularity balls to be assigned are checked. Finally, the candidate assignment threshold is adjusted, and the assignment queue is traversed again until any dynamic assignment stopping condition is met, where m represents the number of neighboring granularity balls for each granularity ball. This indicates rounding up; the cluster to which the cluster center belongs represents a set of granular spheres similar to that cluster center, with granular spheres within the cluster exhibiting similarity in lifecycle characteristics, and the cluster center being the representative point of these granular spheres; the dynamic allocation stopping condition includes the queue to be allocated being empty and... ;

[0103] The candidate allocation threshold adjustment specifically refers to the adjustment of the allocation queue if there is no change and If so, then subtract 1 from the candidate allocation threshold;

[0104] The formula used is as follows:

[0105] ;

[0106] In the formula, Represents particle size spheres Belongs to cluster The membership value, Let q be the q-th cluster. Represents particle size spheres In the set of neighboring spheres, belonging to the cluster The number of spheres with varying particle sizes;

[0107] The grain size ball bottom allocation is specifically as follows: for grain size balls that are still unassigned, calculate the Euclidean distance to the assigned grain size balls from the set of neighboring grain size balls, select the nearest assigned grain size ball, and assign the grain size ball to the cluster to which the assigned grain size ball belongs;

[0108] The clustering results are output as follows: after all the granular balls have been assigned, they are merged into the corresponding cluster set to form multiple independent clusters. Each cluster represents a lifecycle stage of a movie and TV membership.

[0109] A member lifecycle segmentation model is constructed by initializing the stable granular sphere, performing secondary optimization of the granular sphere, calculating the comprehensive local density, selecting cluster centers, dynamically allocating non-central granular spheres, and outputting clustering results. This improved clustering algorithm is then used as the member lifecycle segmentation model.

[0110] The real-time segmentation of the film and television membership lifecycle involves inputting lifecycle segmentation data into the membership lifecycle segmentation model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the clustering output results. The cluster label with the highest frequency of occurrence is selected as the film and television membership lifecycle label for that cluster, thus obtaining the lifecycle stage of the film and television membership.

[0111] The lifecycle segmentation data includes user personal data, user behavior data, user payment data, user activity data, and film and television membership lifecycle tags from historical film and television membership stage conversion data and real-time film and television membership stage conversion data.

[0112] By performing the above operations, this solution addresses the technical problems of existing clustering algorithms applicable to film and television membership segmentation, such as the inability to accurately segment members with differentiated behaviors and the lack of consideration for the global correlation of member groups, resulting in low accuracy in segmenting the life cycle stages of film and television members. This solution innovatively proposes an improved clustering algorithm using granular sphere quadratic optimization, a two-factor density calculation method, and a dynamic allocation strategy. This improves the accuracy of the member life cycle stage boundaries, ensures the accurate segmentation of members with differentiated behaviors, solves the problem of traditional methods being limited to local information, enhances the consideration of global and local correlations, and effectively improves the accuracy and adaptability of life cycle segmentation, thereby realizing intelligent management of film and television members.

[0113] Example 4, see Figure 1 , Figure 3 and Figure 4 This embodiment is based on the above embodiment, and the member lifecycle conversion module specifically includes the following steps:

[0114] A member lifecycle conversion model is established, specifically by constructing a member lifecycle conversion model based on the Q-learning algorithm, and training the model using historical film and television membership stage data as training data to obtain the trained member lifecycle conversion model; the construction of the member lifecycle conversion model based on the Q-learning algorithm specifically includes the following steps:

[0115] Define a state space, specifically by forming a state vector from the lifecycle stage and the converted data of each film and television member's stage.

[0116] Define the action space, specifically by treating each movie / TV membership conversion strategy as a specific action. The action space is the set of all possible strategy actions.

[0117] Design a reward function to measure members' response to the conversion strategy, specifically by rewarding members based on the conversion results of corresponding movie and TV show memberships;

[0118] Q-value updates are specifically performed at each time step by updating the Q-value based on the current state, the selected action, the observed reward, and the next state.

[0119] The model training specifically involves interacting with the environment based on training data and continuously updating the Q-value. First, the Q-value table is initialized, an action is selected based on the current state, and the reward is observed. The Q-value is updated based on the feedback. As training progresses, the model gradually adjusts its strategy, selecting the optimal conversion strategy at different lifecycle stages. Training terminates when the Q-value converges or reaches the maximum number of training rounds.

[0120] The conversion model hyperparameter optimization specifically involves obtaining the optimal hyperparameter combination of the model through an improved particle optimization algorithm, and updating the model's hyperparameters based on this optimal combination to obtain the optimal member lifecycle conversion model. This includes the following steps:

[0121] Initializing the particle swarm involves encoding the hyperparameters of the trained member lifecycle transformation model into individual particle position vectors, and generating them using a random initialization method. The position vectors of individual particles, each encoding a candidate combination of model hyperparameters, constitute the initial particle swarm; the hyperparameters of the model include learning rate, discount factor, exploration rate, and batch size;

[0122] The particle fitness value is calculated by calculating the fitness value of individual particles in the population; the performance of the member lifecycle conversion model established based on the position of individual particles is used as the fitness value of individual particles.

[0123] The population average optimal position is calculated by taking the average of the optimal positions of all individual particles based on a population feedback learning strategy; the formula used is as follows:

[0124] ;

[0125] In the formula, This represents the average optimal position of the population. This represents the optimal position of each individual particle, and k represents the index of that individual particle.

[0126] The adaptive inertia weights are calculated by adjusting them using a logarithmic decreasing strategy; the formula used is as follows:

[0127] ;

[0128] In the formula, This indicates that the k-th particle is in the... Inertia weights in the next iteration Indicates the initial inertia weight. Indicates the final inertia weight. This represents the adjustment coefficient, which controls the rate at which the inertia weight decreases; its value ranges from [value range missing]. between, This represents the maximum number of iterations, and t represents the current number of iterations. This indicates that the k-th particle is in The optimal position in the next iteration;

[0129] Individual particle updates specifically involve updating particle velocity and position; the formulas used are as follows:

[0130] ;

[0131] In the formula, This indicates that the k-th particle is in the... Speed ​​in the next iteration This indicates that the k-th particle is in the... Speed ​​in the next iteration This represents the position of the k-th particle in the t-th iteration. This represents the local optimal position of an individual particle. This represents the global optimal position of the particle. and Represents a random number in the range [0,1]. This indicates that the k-th particle is in the... Position in the next iteration and These represent individual learning factors and group learning factors, respectively.

[0132] The particle search terminates when any one of the following conditions is met: the optimal particle position is updated, and the global optimal particle position is obtained. The global optimal particle position specifically refers to the optimal combination of hyperparameters of the model. The search termination conditions include the global optimal particle position being higher than the fitness threshold and the number of iterations reaching the maximum number of iterations.

[0133] The optimal position update of the particles specifically involves re-evaluating the fitness values ​​of all updated particles and comparing the current particle's fitness value with its optimal position. If the current particle's fitness value is better, the optimal position of the current individual particle is updated. At the same time, the global optimal position of the particles is updated based on the particle with the best fitness among all individual particles.

[0134] The optimal hyperparameter update specifically involves updating the hyperparameter configuration of the model based on the optimal combination of hyperparameters to obtain the optimal member lifecycle conversion model.

[0135] The acquisition of member lifecycle conversion strategies involves inputting the current lifecycle stage of film and television members and real-time film and television member stage conversion data into the optimal member lifecycle conversion model to obtain real-time film and television member conversion strategies.

[0136] By performing the above operations, this solution addresses the technical problems of existing member lifecycle conversion models, such as unreasonable hyperparameter settings, reliance on linear inertia weight adjustment and insufficient inter-particle information interaction during hyperparameter optimization, which easily leads to individuals getting trapped in local optima and thus affecting the accuracy of model output results. This solution innovatively adopts a swarm feedback learning strategy in the particle swarm optimization algorithm, increasing the dimension of inter-particle information interaction and preventing particles from excessively clustering near local optima. A logarithmic decreasing strategy is used to adjust the dynamic balance search capability of the inertia weight and adapt to individual particle differences. This significantly improves the efficiency and accuracy of the model's hyperparameter search, enhances the performance of the conversion model, and improves the accuracy and adaptability of the model's output results. It effectively promotes the precise optimization of film and television membership conversion strategies, ultimately achieving intelligent and personalized management of the film and television membership lifecycle.

[0137] Example 5, see Figure 1 This embodiment is based on the above embodiment. Specifically, the intelligent management module for film and television members automatically implements personalized conversion strategies based on the life cycle stage of film and television members and in combination with real-time film and television member conversion strategies, so as to realize intelligent and comprehensive management of the life cycle of film and television members and provide continuous member growth for film and television platforms.

[0138] 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.

[0139] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0140] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A film and television membership management system based on big data, characterized in that: It includes a data acquisition module, a member lifecycle segmentation module, a member lifecycle conversion module, and a film and television membership intelligent management module; The data acquisition module specifically obtains optimized data for film and television membership management through data collection and optimization. The member lifecycle segmentation module specifically establishes a member lifecycle segmentation model by using granular sphere secondary optimization, a two-factor density calculation method, and a non-central granular sphere dynamic allocation improved clustering algorithm. The behavioral data of film and television members is then input into the model to obtain the lifecycle stages of film and television members. The granular sphere secondary optimization specifically involves calculating the radius r of each granular sphere in the initial stable granular sphere set, then sorting the granular spheres in the initial stable granular sphere set in ascending order by radius r, and calculating the radius increment of adjacent granular spheres. And set the radius of the sphere with the largest radius increment as the radius threshold. Finally, traverse the set of initial stable spheres, and if the radius of the sphere is... If the result is positive, add it to the final granularity sphere candidate set; otherwise, perform granularity sphere iterative segmentation, adding the segmented sub-granularities to the final granularity sphere candidate set. Indicates the first The radius of a sphere of a certain size. Indicates the first The radius of a single sphere; The two-factor density calculation method specifically involves calculating the internal density and neighborhood density of each particle size sphere separately. The dynamic allocation of non-centralized spheres specifically includes the following steps: The initial allocation is as follows: for each granular sphere that serves as a cluster center, its m neighboring granular spheres are directly assigned to the cluster to which the cluster center belongs; The queue to be allocated is initialized by sorting the unallocated granular balls in descending order according to their neighborhood density values ​​and adding them to the queue to be allocated. Dynamic threshold allocation, specifically, initializing candidate allocation thresholds. Then, iterate through each granular ball in the queue to be assigned, and calculate the number of granular balls in the set of neighboring granular balls that have been assigned to the cluster to which the cluster center belongs. ,like Then calculate the membership degree value of the granular sphere to each cluster, assign it to the cluster with the largest membership degree value according to the membership degree value, and remove the granular sphere from the assignment queue. If a granularity ball is not found in the list of granularity balls to be assigned, the process is to temporarily skip that granularity ball and continue checking other granularity balls to be assigned. Finally, the candidate assignment threshold is adjusted, and the process continues to traverse the assignment queue until any dynamic assignment stopping condition is met, where m represents the number of neighboring granularity balls for each granularity ball. The dynamic assignment stopping condition includes the assignment queue being empty and... The formula used is as follows: ； In the formula, Represents particle size spheres Belongs to cluster The membership value, Let q be the q-th cluster. Represents particle size spheres In the set of neighboring spheres, belonging to the cluster The number of spheres with varying particle sizes; The grain size ball bottom allocation is specifically as follows: for grain size balls that are still unassigned, calculate the Euclidean distance to the assigned grain size balls from the set of neighboring grain size balls, select the nearest assigned grain size ball, and assign the grain size ball to the cluster to which the assigned grain size ball belongs; The membership lifecycle conversion module specifically involves constructing a membership lifecycle conversion model, training the model using historical data, and improving the particle optimization algorithm using a herd feedback learning strategy and a logarithmic decreasing strategy to optimize the hyperparameters of the conversion model. The optimal hyperparameter combination is obtained and the model hyperparameters are updated to form the optimal membership lifecycle conversion model. Real-time data is then input into this model to obtain a real-time film and television membership conversion strategy. The module includes the following steps: A member lifecycle conversion model is established, specifically by constructing a member lifecycle conversion model based on the Q-learning algorithm and training the model using historical film and television membership stage conversion data as training data to obtain the trained member lifecycle conversion model. The conversion model hyperparameter optimization specifically involves obtaining the optimal hyperparameter combination of the model through an improved particle optimization algorithm, and updating the model's hyperparameters based on this optimal combination to obtain the optimal member lifecycle conversion model. This includes the following steps: Initializing the particle swarm involves encoding the hyperparameters of the trained member lifecycle transformation model into individual particle position vectors, and generating them using a random initialization method. The position vectors of individual particles constitute the initial particle swarm; The particle fitness value is calculated by calculating the fitness value of individual particles in the population; the performance of the member lifecycle conversion model established based on the position of individual particles is used as the fitness value of individual particles. The population average optimal position is calculated by taking the average of the optimal positions of all individual particles based on a population feedback learning strategy; the formula used is as follows: ； In the formula, This represents the average optimal position of the population. This represents the optimal position of each individual particle, and k represents the index of that individual particle. The adaptive inertia weights are calculated by adjusting them using a logarithmic decreasing strategy; the formula used is as follows: ； In the formula, This indicates that the k-th particle is in the... Inertia weights in the next iteration Indicates the initial inertia weight. Indicates the final inertia weight. This represents the adjustment coefficient, which controls the rate at which the inertia weight decreases. This represents the maximum number of iterations, and t represents the current number of iterations. Indicates that the k-th particle is in The optimal position in the next iteration; Individual particle updates specifically involve updating particle velocity and position. The particle search terminates when the optimal particle position is updated, and the search ends and the global optimal particle position is obtained when any one of the search termination conditions is met; the global optimal particle position specifically refers to the optimal combination of hyperparameters of the model. The optimal hyperparameter update specifically involves updating the hyperparameter configuration of the model based on the optimal combination of hyperparameters to obtain the optimal member lifecycle conversion model. The acquisition of member lifecycle conversion strategy involves inputting the current lifecycle stage of film and television members and real-time film and television member stage conversion data into the optimal member lifecycle conversion model to obtain the real-time film and television member conversion strategy. The intelligent management module for film and television memberships specifically implements personalized conversion strategies based on the lifecycle stage of film and television members and in conjunction with real-time film and television membership conversion strategies, thereby achieving intelligent and comprehensive management of the film and television membership lifecycle; the film and television membership conversion strategies are specifically personalized conversion strategies formulated for the conversion of members at each lifecycle stage into high-value members.

2. The film and television membership management system based on big data according to claim 1, characterized in that: The member lifecycle segmentation module specifically includes the following steps: The initialization of stable granularity balls involves first constructing a set of film and television member feature vectors, then initializing the granularity ball set to an empty set, and adding the film and television member feature vector set as the initial granularity ball to the granularity ball queue to be processed. Subsequently, the granularity ball iterative segmentation operation is performed repeatedly until no new granularity balls are added to the granularity ball queue to be processed, thus obtaining the initial stable granularity ball set. Secondary optimization of particle size; Calculate the overall local density; Cluster center selection involves first calculating the relative distance of each sphere, multiplying the comprehensive local density of each sphere by the relative distance to obtain the decision value of each sphere, sorting the decision values ​​of all spheres in descending order, and selecting the Q spheres with the highest decision values ​​as cluster centers. Dynamic distribution of non-central spheres; The clustering results are output as follows: after all the granular balls have been assigned, they are merged into the corresponding cluster set to form multiple independent clusters. Each cluster represents a lifecycle stage of a movie and TV membership. A member lifecycle segmentation model is constructed by initializing the stable granular sphere, performing secondary optimization of the granular sphere, calculating the comprehensive local density, selecting cluster centers, dynamically allocating non-central granular spheres, and outputting clustering results. This improved clustering algorithm is then used as the member lifecycle segmentation model. The real-time segmentation of film and television membership lifecycle involves inputting lifecycle segmentation data into a membership lifecycle segmentation model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the clustering output results. The cluster label with the highest frequency of occurrence is selected as the film and television membership lifecycle label for that cluster, thus obtaining the lifecycle stage of the film and television membership.

3. The film and television membership management system based on big data according to claim 2, characterized in that: The calculation of the comprehensive local density specifically involves using a two-factor density calculation method to calculate the internal density and neighborhood density of each particle size sphere separately, and finally multiplying these two density values ​​to obtain the comprehensive local density of each particle size sphere; the formula used is as follows: ； ； ； In the formula, Represents particle size spheres Internal density, Represents particle size spheres neighborhood density, Represents particle size spheres The member feature vector with median Euclidean distance from the center. Indicates the scaling factor. Represents particle size spheres The center, Represents the Euclidean distance function. Indicates proximity The j-th granular sphere, Represents particle size spheres The center, Represents particle size spheres Given a set of neighboring spheres, select m spheres at a distance of [m]. The closest granular sphere in the Euclidean distance is used as the granular sphere. The set of neighboring spheres, Represents particle size spheres The overall local density, Represents particle size spheres Number of film and television members covered This represents the i-th granular sphere.

4. The film and television membership management system based on big data according to claim 1, characterized in that: The data acquisition module specifically collects data from multiple sources from the film and television management platform and optimizes the collected raw film and television membership management data to obtain optimized film and television membership management data. The raw film and television membership management data includes historical film and television membership stage data and real-time film and television membership stage data. Both the historical film and television membership stage conversion data and the real-time film and television membership stage conversion data include user personal data, user behavior data, user payment data, and user activity data. The historical film and television membership stage conversion data also includes film and television membership lifecycle tags, film and television membership conversion strategies, and corresponding film and television membership conversion results. The data optimization specifically involves data cleaning, data normalization, and data encoding processing of the raw data to obtain optimized film and television membership management data.

Images