Aactor recommendation system based on artificial intelligence

Abstract

The invention discloses an actor recommendation system based on artificial intelligence, and relates to the field of film and television intelligent analysis, and the system comprises a network model building unit which is used for collecting basic information based on an optimal data set range, determining episode evaluation indexes, and building a play relation network model; the dynamic optimization algorithm unit is used for optimizing the drama relation network model by utilizing a dynamic multi-objective genetic algorithm to obtain optimal parameters; and the recommendation score output unit is used for inputting the to-be-evaluated information and outputting a recommendation score based on the drama relation network model. According to the method, the problems of high subjectivity, non-uniform evaluation standards and neglect of role matching degree in a traditional actor selection process are effectively solved, automatic analysis is performed by utilizing an advanced artificial intelligence algorithm, the actors are scored, a more objective, comprehensive and accurate actor recommendation basis is provided for a film and television production team, and the angle selection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent film and television analysis, and in particular to an actor recommendation system based on artificial intelligence. Background Art

[0002] In today's fiercely competitive film and television market, accurate actor recommendations have become a key link in the production process of film and television works. An actor who is highly matched with the role can make the character image more full and three-dimensional, enhance the persuasiveness and appeal of the plot, and thus attract more audiences; at the same time, a suitable cast can also improve the overall quality of the TV series, increase the market appeal and commercial value of the work; in addition, with the continuous improvement of the audience's aesthetic level and the increasing diversification of market demand, the selection of actors also needs to be more accurate and personalized to meet the needs of different types of repertoires and target audiences.

[0003] At present, the film and television industry mainly relies on traditional evaluation methods and industry experience when recommending actors. These methods include actor auditions, personal judgments of directors and producers, performances of actors in previous works, and market feedback. During the audition process, actors demonstrate their acting skills and understanding of the role by performing specific scenes, while directors and producers make choices based on audition performance, the actor's appearance, acting skills, and fit with the role. At the same time, the actor's popularity and market appeal are also important considerations when recommending casting. Recommending suitable actors is crucial to the success of film and television works. It not only directly affects the presentation of the plot, but also affects the ratings, reputation, and market competitiveness of film and television works.

[0004] However, existing actor recommendation technologies have obvious shortcomings. First, traditional actor recommendation technologies often rely on subjective judgments and only consider single-dimensional data (such as ratings or social media popularity), which makes it difficult to effectively capture and analyze the complex relationship network between actors, roles, and repertoires, resulting in an incomplete evaluation of the matching degree between actors and specific roles and a lack of objective and systematic recommendation criteria. Second, existing technologies lack the ability to effectively screen and process historical data, which may introduce outdated or irrelevant information, affecting the accuracy and timeliness of actor recommendations. In addition, existing recommendation technologies lack in-depth analysis of the matching degree between actors and specific roles, are inefficient in processing large-scale data, and are difficult to meet the needs of rapid decision-making. These problems seriously restrict the scientificity and effectiveness of the actor recommendation process.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In response to the problems in the related technology, the present invention proposes an actor recommendation system based on artificial intelligence, which is capable of using advanced artificial intelligence algorithms to analyze and score actors, providing film and television production teams with a more objective, comprehensive and accurate basis for actor recommendation, thereby improving the efficiency of casting, and thus solving the problem that it is difficult in the prior art to effectively capture and analyze the complex relationship network between actors, roles and repertoires, resulting in an incomplete assessment of the match between actors and specific roles, and a lack of effective screening and processing capabilities for historical data, which may introduce outdated or irrelevant information, affecting the accuracy and timeliness of actor recommendations.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] An actor recommendation system based on artificial intelligence, the actor recommendation system based on artificial intelligence comprising:

[0009] A network model building unit is used to collect basic information based on the optimal data set range, determine the drama evaluation index, and establish a drama relationship network model;

[0010] A dynamic optimization algorithm unit is used to optimize the play relationship network model using a dynamic multi-objective genetic algorithm to obtain optimal parameters;

[0011] The recommendation score output unit is used to input the information to be evaluated and output the recommendation score based on the play relationship network model.

[0012] Furthermore, the network model building unit includes a data collection module, an indicator analysis module and a model building module;

[0013] Among them, the data collection module is used to collect basic information and establish a high-quality data set according to the optimal data set range;

[0014] The indicator analysis module is used to analyze the relationship between actor influence and ratings based on high-quality data sets and determine comprehensive evaluation indicators;

[0015] The model building module is used to establish a play relationship network model based on the determined comprehensive evaluation indicators using graph neural network technology.

[0016] Furthermore, the data acquisition module includes a range analysis module, an information capture module and a data cleaning module;

[0017] Among them, the range analysis module is used to determine the optimal data collection range using a density clustering algorithm;

[0018] The information capture module is used to collect basic information using web crawler technology according to the optimal data collection scope;

[0019] The data cleaning module is used to use the local sensitive hashing algorithm to clean basic information and generate high-quality data sets.

[0020] Furthermore, using density clustering algorithm, the optimal data collection range is determined to include:

[0021] Based on multi-source data collection and natural language processing technology, obtain text feature vectors of the film and television industry;

[0022] Using density clustering algorithm, the time decay factor is introduced to generate time-sensitive text clustering results;

[0023] Determine the optimal data collection scope based on clustering result analysis and industry background knowledge.

[0024] Furthermore, the expression of the time attenuation factor is introduced as follows:

[0025] λ(t)=e-α(t now -t)

[0026] d'(x,y)=d(x,y) / (λ(tx)×λ(ty));

[0027] Wherein, λ(t) represents the time decay factor; α represents an adjustable parameter (in this embodiment, α is 0.001) used to control the speed of time decay; t now represents the current time, t represents the release time of actor-related data; d'(x,y) represents the corrected distance after the introduction of the time decay factor; d(x,y) represents the original Euclidean distance, which is used to measure the similarity between two actor data points in the feature space; t x With t y Respectively represent the time corresponding to actor data points x and y.

[0028] Furthermore, the indicator analysis module includes an indicator extraction module, a standardization processing module and an indicator generation module;

[0029] Among them, the indicator extraction module is used to extract evaluation indicator data from high-quality data sets using variance contribution rate and correlation coefficient matrix. The evaluation indicator data includes audience rating, word-of-mouth rating and drama type.

[0030] A standardization processing module is used to perform standardization processing based on the extracted evaluation index data using statistical methods;

[0031] The indicator generation module is used to generate comprehensive evaluation indicators using the weighted average method.

[0032] Furthermore, the model building module includes a network structure building module, a node embedding generation module and a play relationship network module;

[0033] Among them, the network structure building module is used to determine the play nodes and actor nodes based on comprehensive evaluation indicators and establish a heterogeneous graph network structure;

[0034] The node embedding generation module is used to generate node embedding representations using the graph attention network;

[0035] The play relationship network module is used to select node embedding representation as input feature based on graph convolutional network and build a play relationship network model.

[0036] Furthermore, based on the graph convolutional network, node embedding representation is selected as the input feature to construct the play relationship network model including:

[0037] Utilize graph convolutional networks to extract local structural features represented by node embeddings;

[0038] Based on the multi-layer perceptron, the predicted score is calculated to generate the predicted value of the play score;

[0039] The expression for calculating the prediction score is:

[0040]

[0041] In the formula, represents the predicted score of the i-th play sample; n represents the number of plays; m represents the number of actors participating in the play; y xj represents the score obtained by actor x participating in the i-th play sample in the previous play sample j; r xj represents the role coefficient of actor x in the previous play sample j; f(a,p) represents the influence of the actor's age a and fan coefficient p;

[0042] Based on the mean square error loss, the objective function is constructed, and the model parameters are optimized to establish the play relationship network model;

[0043] The expression for constructing the objective function is:

[0044]

[0045] Where, Obj represents the objective function; n represents the number of plays; Represents the prediction error of the i-th play sample.

[0046] Furthermore, the dynamic optimization algorithm unit includes a recommended solution design module, a population optimization evaluation module, and an optimal solution selection module;

[0047] The recommendation scheme design module is used to design the encoding scheme of the recommendation scheme and initialize the candidate recommendation set based on the play type and actor characteristics;

[0048] Population optimization evaluation module, which is used to optimize the recommended scheme and evaluate the population fitness by using the dynamic crossover rate and mutation rate adjustment mechanism;

[0049] The optimal solution selection module is used to select the optimal recommendation solution based on the Pareto optimal set and the play type, and update the play relationship network model.

[0050] Furthermore, the population optimization evaluation module includes an elite matching acquisition module, a genetic operation execution module, and a fitness evaluation module;

[0051] Among them, the elite matching acquisition module is used to obtain candidate actor combinations based on actor fitness scores and elite retention strategies;

[0052] A genetic operation execution module is used to perform crossover and mutation operations on actor combinations using dynamic crossover rate and mutation rate to generate optimized recommendation solutions;

[0053] The fitness evaluation module is used to evaluate the fitness of the optimized recommendation scheme based on multi-objective evaluation and matching calculation.

[0054] The beneficial effects of the present invention are:

[0055] (1) The present invention effectively solves the problems of strong subjectivity, inconsistent evaluation standards, and neglect of role matching in the traditional actor selection process through innovative data processing and model building methods. At the same time, the present invention comprehensively considers multi-dimensional factors such as the actor's historical performance, audience feedback, and role requirements, and uses advanced artificial intelligence algorithms to automatically analyze and score the actors, providing film and television production teams with a more objective, comprehensive, and accurate basis for actor recommendation, thereby improving casting efficiency.

[0056] (2) The present invention can effectively screen out the most representative and timely actor-related data by using a density clustering algorithm to determine the optimal data collection range, thus avoiding the problem of over-reliance on outdated information in traditional methods. At the same time, the system uses web crawler technology and local sensitive hashing algorithm for data collection and cleaning, which greatly improves data quality and reliability. In addition, the system innovatively introduces comprehensive evaluation indicators, including multi-dimensional factors such as ratings, word-of-mouth ratings, and drama types, and uses graph neural network technology to construct a drama relationship network model, which effectively captures the complex relationship between actors, roles, and dramas. It not only overcomes the defect of traditional evaluation methods being too one-sided, but also can more comprehensively reflect the actual performance and potential of actors.

[0057] (3) By introducing a dynamic crossover rate and mutation rate adjustment mechanism, the present invention can more effectively explore the solution space and avoid falling into a local optimal solution. This optimization method not only improves the prediction accuracy of the model, but also enhances the system's adaptability to different types of plays and role requirements. At the same time, the optimal recommendation scheme is selected based on the Pareto optimal set and the type of play, ensuring that the system can achieve a balance between multiple goals, such as the actor's performance ability, market appeal, and matching degree with the role. This multi-objective optimization method effectively solves the problem of over-reliance on a single indicator or subjective judgment in the traditional casting process, providing the production team with a more comprehensive and objective decision-making basis, which helps to recommend the actors who best suit specific roles and plot requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 is a principle block diagram of an actor recommendation system based on artificial intelligence according to an embodiment of the present invention;

[0060] Figure 2 is an interaction relationship diagram between actors and participating plays according to an embodiment of the present invention;

[0061] Figure 3 is a complex network relationship diagram composed of actors and plays according to an embodiment of the present invention;

[0062] Figure 4 is a specific implementation diagram of a genetic operation execution module in an actor recommendation system based on artificial intelligence according to an embodiment of the present invention;

[0063] Figure 5 It is a principle flow chart of an actor recommendation system based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0065] According to an embodiment of the present invention, an actor recommendation system based on artificial intelligence is provided.

[0066] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1-Figure 5 As shown, according to one embodiment of the present invention, an actor recommendation system based on artificial intelligence is provided, and the actor recommendation system based on artificial intelligence includes:

[0067] The network model building unit 1 is used to collect basic information based on the optimal data set range, determine the drama evaluation index, and establish a drama relationship network model;

[0068] Dynamic optimization algorithm unit 2, used to optimize the play relationship network model using a dynamic multi-objective genetic algorithm to obtain optimal parameters;

[0069] The recommendation score output unit 3 is used to input the information to be evaluated and output the recommendation score based on the play relationship network model.

[0070] Specifically, based on the drama relationship network model constructed by the dynamic optimization algorithm unit 2, the recommendation score output unit 3 first determines the type of the drama to be evaluated. In this embodiment, the drama type in the information to be evaluated is an urban life ethics drama, and then selects the established drama relationship network model corresponding to the type (constructed by the dynamic optimization algorithm unit 2); next, for each actor to be evaluated, input includes name, age, gender, the number of dramas of the same type that have participated in the past five years, the ratings of the most recent drama of the same type, and the number of fans; the drama relationship network model will automatically calculate the actor's score based on this information and the best parameter combination obtained by the previous optimization, and each actor will get a score of 0-10. In this embodiment, actor x A 8.5 points, actor x B Scored 7.2, actor x C Finally, the system generates a scoring report that explains the scoring results in detail. In the above example, it indicates that actor x C The recommended score in the urban life ethics drama is 9.1 points, which means he performed well. His recent performance in this type of drama has been stable, and he has a high audience recognition and fan base. He is very suitable for the protagonist positioning of this drama, so the system determines the recommended actor candidates; when it is necessary to evaluate the performance of actors in different types of dramas, different types of drama relationship network models can be repeatedly used for evaluation to fully understand the adaptability and diversity of the actors.

[0071] In one embodiment, the network model building unit 1 includes a data collection module, an indicator analysis module and a model building module;

[0072] Among them, the data collection module is used to collect basic information and establish a high-quality data set according to the optimal data set range;

[0073] The indicator analysis module is used to analyze the relationship between actor influence and ratings based on high-quality data sets and determine comprehensive evaluation indicators;

[0074] The model building module is used to establish a play relationship network model based on the determined comprehensive evaluation indicators using graph neural network technology.

[0075] In one embodiment, the data acquisition module includes a range analysis module, an information capture module, and a data cleaning module;

[0076] Among them, the range analysis module is used to determine the optimal data collection range using a density clustering algorithm;

[0077] The information capture module is used to collect basic information using web crawler technology according to the optimal data collection scope;

[0078] The data cleaning module is used to use the local sensitive hashing algorithm to clean basic information and generate high-quality data sets.

[0079] Specifically, the collected basic information is preprocessed to extract key features, including the title, leading actors, and directors; then the local sensitive hashing algorithm (LSH) is used to perform similarity detection on the preprocessed data to identify and remove duplicate data; then, the detected duplicate data is deduplicated according to preset rules (in this embodiment, the earliest collected data is selected); finally, the cleaned data is further checked for outliers or erroneous data, such as unreasonable release time, wrong cast list, etc., and corrected, thereby generating a high-quality data set without duplication and error, providing accurate basic data for subsequent analysis and processing.

[0080] In one embodiment, using a density clustering algorithm, determining the optimal data collection range includes:

[0081] Based on multi-source data collection and natural language processing technology, obtain text feature vectors of the film and television industry;

[0082] Specifically, we used web crawlers and database query technology to collect academic papers, industry reports and news articles related to the film and television industry in the past 10 years from academic databases, industry websites and mainstream media websites; then we used a word segmentation tool to segment the text, and used the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm to extract keywords and weights. Finally, based on the extracted keywords and document release time, we generated a feature vector containing the top N keywords with the highest TF-IDF weights and time information.

[0083] Using density clustering algorithm, the time decay factor is introduced to generate time-sensitive text clustering results;

[0084] Specifically, based on the traditional density-based clustering algorithm (DBSCAN algorithm is used in this embodiment), the time decay factor λ(t) is introduced, and the distance calculation formula is modified to d'(x, y). Then, the improved density-based clustering algorithm is used to perform clustering analysis on the acquired feature vectors of the film and television industry texts to obtain the text clustering results that consider the time factor.

[0085] Determine the optimal data collection scope based on clustering result analysis and industry background knowledge.

[0086] Specifically, the word frequency statistics method is used to analyze the time distribution of clusters and keyword changes, combined with industry background knowledge, to identify industry turning points; in this embodiment, based on the identified industry turning points and cluster analysis results, 2014-2023 is determined as the optimal data collection range.

[0087] Specifically, the industry background knowledge includes: with the development and popularization of the Internet, the film and television industry has undergone a major transformation, and high-quality content and users have gradually flowed from traditional TV media to emerging network platforms. Today, Internet platforms have gradually replaced TV platforms and become the most important TV drama broadcasting media; and compared with TV stations, network platforms have significant differences in audience aesthetic tastes, viewing habits, etc. For example, dramas broadcast by traditional TV stations target the entire family as target users, are broadcast at fixed time periods, have a slower pace, and use simpler camera language, while dramas broadcast on network platforms are usually watched by one person at a flexible viewing time, so they emphasize a fast pace and immersion; at the same time, the development of self-media has led to more frequent interactions between actors and audiences, and the audience's aesthetic appreciation of actors has also entered the Internet age.

[0088] In one embodiment, the expression for introducing the time decay factor is:

[0089] λ(t)=e-α(t now -t)

[0090] d'(x,y)=d(x,y) / (λ(tx)×λ(ty));

[0091] Wherein, λ(t) represents the time decay factor; α represents an adjustable parameter (in this embodiment, α is 0.001) used to control the speed of time decay; t now represents the current time, t represents the release time of actor-related data; d'(x,y) represents the corrected distance after the introduction of the time decay factor; d(x,y) represents the original Euclidean distance, which is used to measure the similarity between two actor data points in the feature space; t x With t y Respectively represent the time corresponding to actor data points x and y.

[0092] In one embodiment, the indicator analysis module includes an indicator extraction module, a standardization processing module and an indicator generation module;

[0093] Among them, the indicator extraction module is used to extract evaluation indicator data from high-quality data sets using variance contribution rate and correlation coefficient matrix. The evaluation indicator data includes audience rating, word-of-mouth rating and drama type.

[0094] Specifically, the audience rating, word-of-mouth rating and drama type are extracted from the high-quality data set as the main evaluation indicators. In this embodiment, the audience rating takes into account the TV station ratings and network clicks, the word-of-mouth rating adopts Douban rating, and the drama types include urban life ethics, revolutionary spy war, ancient costume legend, military life, era emotions, rural areas, youth idols, children's science fiction, case-related suspense, and martial arts detective stories. The variance contribution rate of each indicator is calculated using the principal component analysis method. The results show that the variance contribution rates of the audience rating, word-of-mouth rating and drama type are 45%, 35% and 20% respectively. Constructing the correlation coefficient matrix, it is found that the correlation coefficient between the audience rating and the word-of-mouth rating is 0.6, and the correlation coefficient with the drama type is 0.3.

[0095] A standardization processing module is used to perform standardization processing based on the extracted evaluation index data using statistical methods;

[0096] Specifically, the ratings are processed into percentiles, and the dramas of each year are ranked according to the ratings, and scored according to the following preset rules:

[0097] 5 points — ranked in the top 5%;

[0098] 4 points - ranked in the top 5%-10%;

[0099] 3 points - ranked in the top 10%-30%;

[0100] 2 points - ranked in the top 30%-50%;

[0101] 1 point - ranked in the bottom 50%;

[0102] For the word-of-mouth rating, this embodiment uses the rating of a certain video portal website (1-10 point system). If the drama has no rating on the video portal website, the average score of all dramas released in that year is used instead. In this embodiment, the word-of-mouth rating in 2022 is 7.2 points; finally, for the drama type, One-Hot Encoding is used to convert the 10 types into a 10-dimensional vector. In the above embodiment, the urban life ethics type is represented as [1,0,0,0,0,0,0,0,0,0].

[0103] The indicator generation module is used to generate comprehensive evaluation indicators using the weighted average method.

[0104] Specifically, according to the calculated variance contribution rate, a weight is assigned to each indicator, including a ratings weight of 0.45, a word-of-mouth score weight of 0.35, and a drama type weight of 0.20; then for each actor, the weighted average score of all the dramas in which he has participated is calculated as a comprehensive evaluation indicator, and the calculation expression is comprehensive evaluation index = 0.45*ratings+0.35*(word-of-mouth score / 2)+0.20*drama type score; where the drama type score is obtained by calculating the proportion of different types of dramas in which the actor has participated, reflecting the diversity of the actor; in this embodiment, an actor has participated in 5 dramas, including 3 urban dramas and 2 costume dramas, and the drama type score is: (3 / 5)^2+(2 / 5)^2=0.52; finally, a comprehensive evaluation index ranging from 1 to 5 is obtained, which is used for subsequent graph neural network model construction.

[0105] In one embodiment, the model building module includes a network structure building module, a node embedding generation module, and a play relationship network module;

[0106] Among them, the network structure building module is used to determine the play nodes and actor nodes based on comprehensive evaluation indicators and establish a heterogeneous graph network structure;

[0107] Specifically, a graphics processing library (NetworkX library in Python language is used in this embodiment) is used to create a bipartite graph containing play nodes and actor nodes. Each play node is given attributes, including the determined comprehensive evaluation index, comprehensive score, audience rating, word-of-mouth score and drama type; each actor node is given characteristic attributes, that is, the actor's personal characteristics, including age, number of fans and cumulative influence index; edges are used to represent the relationship between actors participating in the play, and the attributes of the edges are role weights (such as male lead 1.0, female lead 1.0, male second 0.8, female second 0.8, and other 0.5); in this way, a complete heterogeneous graph network structure is obtained, in which the play node contains comprehensive evaluation indicators, the actor node contains personal characteristics, and the edge contains role weights, which fully reflects the complex relationship between the play and the actor.

[0108] The node embedding generation module is used to generate node embedding representations using the graph attention network;

[0109] Specifically, based on the geometric deep learning library (in this embodiment, the PyTorch Geometric library in Python language is used), a graph attention network model is implemented. The constructed heterogeneous graph network structure is used as input, and different attention heads are used to capture the complex interactions between play nodes and actor nodes; for play nodes, information from related actor nodes is aggregated; for actor nodes, information from the plays they participate in is aggregated; then, through a multi-layer graph attention mechanism, the embedding table of the node is gradually updated, and finally a low-dimensional vector representation of each play and actor is obtained, which encodes the characteristics of the node and the graph structure information.

[0110] The play relationship network module is used to select node embedding representation as input feature based on graph convolutional network and build a play relationship network model.

[0111] Specifically, the generated node embeddings are used as input features and a graph convolutional network is used to build a drama rating prediction model. The objective function of the model is defined as the mean square error loss where y i is the true score of the i-th play sample, is the predicted score of the i-th play sample; then the optimizer is used to train the model, and the parameters of the graph convolution layer are updated through back propagation, and finally a graph neural network model that can accurately predict the play score is obtained.

[0112] In one embodiment, based on a graph convolutional network, node embedding representation is selected as an input feature, and constructing a play relationship network model includes:

[0113] Utilize graph convolutional networks to extract local structural features represented by node embeddings;

[0114] Based on the multi-layer perceptron, the predicted score is calculated to generate the predicted value of the play score;

[0115] The expression for calculating the prediction score is:

[0116]

[0117] In the formula, represents the predicted score of the i-th play sample; n represents the number of plays; m represents the number of actors participating in the play; y xj represents the score of actor x who participated in the i-th play sample in the previous play sample j; r xj represents the role coefficient of actor x in the previous play sample j; f(a,p) represents the influence of the actor's age a and fan coefficient p, which is a nonlinear function learned by MLP;

[0118] Based on the mean square error loss, the objective function is constructed, and the model parameters are optimized to establish the play relationship network model;

[0119] The expression for constructing the objective function is:

[0120]

[0121] Where, Obj represents the objective function; n represents the number of plays; Represents the prediction error of the i-th play sample.

[0122] In one embodiment, the dynamic optimization algorithm unit 2 includes a recommended solution design module, a population optimization evaluation module, and an optimal solution selection module;

[0123] The recommendation scheme design module is used to design the encoding scheme of the recommendation scheme and initialize the candidate recommendation set based on the play type and actor characteristics;

[0124] Specifically, the real number coding method is used to encode the actor characteristics and the play requirements into the chromosomes of the recommendation scheme. For different types of plays, the corresponding initial candidate recommendation sets are generated respectively; based on the Latin hypercube sampling technology, the initial recommendation schemes are evenly generated in the recommendation space to ensure the diversity of the recommendations; in this way, an initial set containing multiple candidate recommendation schemes is obtained, laying the foundation for subsequent optimization.

[0125] Population optimization evaluation module, which is used to optimize the recommended scheme and evaluate the population fitness by using the dynamic crossover rate and mutation rate adjustment mechanism;

[0126] Specifically, based on the diversity of recommendation schemes and the number of iterations, the crossover rate and mutation rate are dynamically adjusted; simulated binary crossover (SBX) and polynomial mutation operators are used to generate new recommendation schemes, and the matching degree of each recommendation scheme is evaluated using non-dominated sorting and crowding distance calculation; through this process, a new generation of recommendation scheme sets with high matching degree are obtained to promote the exploration of recommendation space.

[0127] The optimal solution selection module is used to select the optimal recommendation solution based on the Pareto optimal set and the play type, and update the play relationship network model.

[0128] Specifically, a fast non-dominated sorting algorithm is used to screen out the Pareto optimal solution set from the final set of recommended solutions; for different types of plays, the optimal recommendation solution is selected from the Pareto frontier, and the selected optimal solution is used to update the actor-play matching model parameters of the corresponding type of play; in this way, recommendation parameters that are balanced and optimized on multiple objectives and suitable for specific types of plays are obtained, thereby improving the recommendation accuracy of the model on different types of plays.

[0129] In one embodiment, the population optimization evaluation module includes an elite matching acquisition module, a genetic operation execution module, and a fitness evaluation module;

[0130] Among them, the elite matching acquisition module is used to obtain candidate actor combinations based on actor fitness scores and elite retention strategies;

[0131] Specifically, first, calculate the suitability score of each actor for a specific play. The suitability score takes into account the actor's comprehensive evaluation index (weight 0.3), the actor's performance history in the type of play (weight 0.3), the actor's role matching with the play's requirements (weight 0.2), and the actor's current popularity index (weight 0.2); the suitability score calculation formula is obtained, suitability score = 0.3*comprehensive evaluation index + 0.3*type performance history + 0.2*role matching + 0.2*popularity index.

[0132] Specifically, according to the basic principle of genetic algorithm, a large number of parameter groups are randomly generated first, and these parameter groups represent different combinations of actors. In view of the need to select actors for different types of plays, when building the model, only the corresponding types of plays can be selected to build the model, so as to obtain the parameter combination that minimizes the objective function under different types.

[0133] Specifically, according to the fitness score, the top 50% of actors are selected as the candidate pool; then, using the elite retention strategy, the 10% actors with the highest fitness scores are directly retained to the next generation; the remaining candidate actors will enter the crossover and mutation operations.

[0134] A genetic operation execution module is used to perform crossover and mutation operations on actor combinations using dynamic crossover rate and mutation rate to generate optimized recommendation solutions;

[0135] Specifically, the crossover rate and mutation rate are dynamically adjusted to balance exploration and utilization. The calculation formulas for the crossover rate and mutation rate are as follows:

[0136] Crossover rate = 0.7-0.5*(current generation / maximum generation);

[0137] Mutation rate = 0.1 + 0.2 * (current generation / maximum generation);

[0138] In the above embodiment, the maximum number of generations is 100, and the current generation is the 50th generation, so the crossover rate = 0.7-0.5*(50 / 100) = 0.45; the mutation rate = 0.1+0.2*(50 / 100) = 0.2;

[0139] Specifically, a crossover operation is performed to randomly select two candidate actors and exchange some of their features (such as type performance history, role matching, etc.) with a probability of the crossover rate; this is equivalent to the crossover operation process in the genetic algorithm.

[0140] Specifically, a mutation operation is performed to randomly change a feature of the candidate actor (such as increasing or decreasing the popularity index) with a probability of the mutation rate, which is equivalent to the mutation operation process in the genetic algorithm.

[0141] Specifically, through these operations, a new generation of candidate actor combinations is generated as the optimized recommendation scheme. This process is equivalent to generating new parameter combinations in genetic algorithms.

[0142] The fitness evaluation module is used to evaluate the fitness of the optimized recommendation scheme based on multi-objective evaluation and matching calculation.

[0143] Specifically, a multi-objective evaluation is performed on the optimized recommendation scheme, considering objectives including the overall fitness of the actor combination (weight 0.4), the diversity of the actor combination (weight 0.3), and the satisfaction of the budget constraint (weight 0.3); then the score of each objective is calculated, including the overall fitness (the average of the fitness scores of all actors), diversity (using Shannon entropy to calculate the distribution diversity of the actors' age groups), and the satisfaction of the budget constraint (1-total cost / budget upper limit, if the budget is exceeded, it is 0), and the final matching calculation formula is obtained, matching = 0.4*overall fitness + 0.3*diversity + 0.3*budget constraint satisfaction.

[0144] Specifically, in the above embodiment, the score of an actor's recommendation scheme is: overall fitness 0.85, diversity 0.7, budget constraint satisfaction 0.9, then the matching degree of the scheme is 0.4*0.85+0.3*0.7+0.3*0.9=0.82.

[0145] Specifically, the optimized recommended solutions are sorted according to the calculated matching degree, and the solution with the highest matching degree is selected as the optimal solution of the current generation. This process is equivalent to substituting the model into the genetic algorithm to calculate the error value, so as to find the optimal parameter combination. This process will be repeated in the iteration of the genetic algorithm until the termination condition (the preset maximum number of generations) is reached.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An actor recommendation system based on artificial intelligence, characterized in that: The AI-based actor recommendation system includes: A network model building unit is used to collect basic information based on the optimal data set range, determine the drama evaluation index, and establish a drama relationship network model; A dynamic optimization algorithm unit is used to optimize the play relationship network model using a dynamic multi-objective genetic algorithm to obtain optimal parameters; The recommendation score output unit is used to input the information to be evaluated and output the recommendation score based on the play relationship network model.

2. The actor recommendation system based on artificial intelligence according to claim 1, characterized in that: The network model building unit includes a data acquisition module, an indicator analysis module and a model building module; Wherein, the data collection module is used to collect basic information according to the optimal data set range and establish a high-quality data set; The indicator analysis module is used to analyze the relationship between the actor's influence and the ratings based on the high-quality data set and determine the comprehensive evaluation indicators; The model building module is used to establish a play relationship network model based on the determined comprehensive evaluation indicators using graph neural network technology.

3. The actor recommendation system based on artificial intelligence according to claim 2, characterized in that: The data acquisition module includes a range analysis module, an information capture module and a data cleaning module; Wherein, the range analysis module is used to determine the optimal data collection range using a density clustering algorithm; The information capture module is used to collect basic information using web crawler technology according to the optimal data collection range; The data cleaning module is used to use the local sensitive hashing algorithm to clean basic information and generate a high-quality data set.

4. The artificial intelligence-based actor recommendation system according to claim 3, characterized in that: Determining the optimal data collection range by using a density clustering algorithm includes: Based on multi-source data collection and natural language processing technology, obtain text feature vectors of the film and television industry; Using density clustering algorithm, the time decay factor is introduced to generate time-sensitive text clustering results; Determine the optimal data collection scope based on clustering result analysis and industry background knowledge.

5. The artificial intelligence-based actor recommendation system according to claim 4, characterized in that: The expression of the introduced time attenuation factor is: d'(x,y)=d(x,y) / (λ(tx)×λ(ty)); Wherein, λ(t) represents the time decay factor; α represents an adjustable parameter (in this embodiment, α is 0.001) used to control the speed of time decay; t now represents the current time, t represents the release time of actor-related data; d'(x,y) represents the corrected distance after the introduction of the time decay factor; d(x,y) represents the original Euclidean distance, which is used to measure the similarity between two actor data points in the feature space; t x With t y Respectively represent the time corresponding to actor data points x and y.

6. The artificial intelligence-based actor recommendation system according to claim 2, characterized in that: The index analysis module includes an index extraction module, a standardization processing module and an index generation module; The index extraction module is used to extract evaluation index data from the high-quality data set using the variance contribution rate and the correlation coefficient matrix, wherein the evaluation index data includes audience ratings, word-of-mouth ratings, and drama types; The standardization processing module is used to perform standardization processing based on the extracted evaluation index data using statistical methods; The index generation module is used to generate comprehensive evaluation indexes using a weighted average method.

7. The artificial intelligence-based actor recommendation system according to claim 2, characterized in that: The model building module includes a network structure building module, a node embedding generation module and a play relationship network module; The network structure building module is used to determine the play nodes and actor nodes based on the comprehensive evaluation index and establish a heterogeneous graph network structure; The node embedding generation module is used to generate node embedding representation using a graph attention network; The play relationship network module is used to select node embedding representation as input feature based on the graph convolutional network to construct a play relationship network model.

8. The artificial intelligence-based actor recommendation system according to claim 1, characterized in that: The method of selecting node embedding representation as input feature based on graph convolutional network and constructing a play relationship network model includes: Utilize graph convolutional networks to extract local structural features represented by node embeddings; Based on the multi-layer perceptron, the predicted score is calculated to generate the predicted value of the play score; The expression for calculating the prediction score is: In the formula, represents the predicted score of the i-th play sample; n represents the number of plays; m represents the number of actors participating in the play; y xj represents the score obtained by actor x participating in the i-th play sample in the previous play sample j; r xj represents the role coefficient of actor x in the previous play sample j; f(a,p) represents the influence of the actor's age a and fan coefficient p; Based on the mean square error loss, the objective function is constructed, and the model parameters are optimized to establish the play relationship network model; The expression of constructing the objective function is: Where, Obj represents the objective function; n represents the number of plays; Represents the prediction error of the i-th play sample.

9. The artificial intelligence-based actor recommendation system according to claim 8, characterized in that: The dynamic optimization algorithm unit includes a recommended solution design module, a population optimization evaluation module and an optimal solution selection module; The recommendation scheme design module is used to design the encoding scheme of the recommendation scheme and initialize the candidate recommendation set based on the play type and actor characteristics; The population optimization evaluation module is used to optimize the recommended scheme and evaluate the population fitness by using the dynamic crossover rate and mutation rate adjustment mechanism; The optimal solution selection module is used to select the optimal recommendation solution based on the Pareto optimal set and the play type, and update the play relationship network model.

10. The artificial intelligence-based actor recommendation system according to claim 9, characterized in that: The population optimization evaluation module includes an elite matching acquisition module, a genetic operation execution module and a fitness evaluation module; The elite matching acquisition module is used to acquire a candidate actor combination based on the actor suitability score and the elite retention strategy; The genetic operation execution module is used to perform crossover and mutation operations on the actor combination using dynamic crossover rate and mutation rate to generate an optimized recommendation plan; The fitness evaluation module is used to evaluate the fitness of the optimized recommendation scheme based on multi-objective evaluation and matching calculation.

Images