Cinema intelligent pricing system based on multi-source data fusion and machine learning

Abstract

The application discloses a cinema intelligent pricing system based on multi-source data fusion and machine learning, relates to the technical field of artificial intelligence, and comprises a multi-source data fusion end, a user behavior modeling end, a dynamic pricing decision end and a strategy feedback module. The system fuses film content, regional consumption, competitive scheduling and historical box office data, constructs a user fine-grained preference portrait, quantifies the viewing demand intensity, generates a differentiated ticket price strategy based on reinforcement learning, and optimizes the model parameters online in combination with real-time feedback data. The cinema intelligent pricing system based on multi-source data fusion and machine learning can realize accurate demand prediction and dynamic income balance, and improves the operation efficiency and market competitiveness of the cinema.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a smart pricing system for cinemas based on multi-source data fusion and machine learning. Background Technology

[0002] The cinema intelligent pricing system based on multi-source data fusion and machine learning is a complex system integrating multiple advanced technologies. It aims to improve the scientific nature and dynamic adaptability of cinema ticket pricing, which is one of the core functions of the system. It is responsible for integrating data from multiple dimensions such as the market, user behavior, film content, and competitive environment. The machine learning model can mine the potential patterns between ticket prices and movie-watching demand from historical and real-time data to ensure the generation of competitive and revenue-optimized ticket pricing strategies in different scenarios. In order to ensure the comprehensiveness and responsiveness of pricing decisions, the system integrates multi-source heterogeneous data to support the efficient operation of cinema intelligent pricing.

[0003] Currently, due to the numerous dynamic influencing factors involved in cinema operations, traditional pricing methods have limited ability to integrate multi-source data when dynamically adjusting ticket prices, which may affect the accuracy of ticket price strategies in responding to market changes. At the same time, when building pricing models, some methods do not sufficiently model user preferences and movie-watching behavior at a fine-grained level, which may lead to discrepancies between recommended ticket prices and actual demand. When dealing with different types of films and different time slot combinations, existing systems have room for optimization in integrating multi-dimensional features such as film popularity, screening density, and regional consumption levels, which may affect the synergistic effect and revenue performance of the overall pricing strategy.

[0004] Therefore, a smart cinema pricing system based on multi-source data fusion and machine learning is proposed to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide a smart pricing system for cinemas based on multi-source data fusion and machine learning, in order to solve the problems mentioned in the background above.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a cinema intelligent pricing system based on multi-source data fusion and machine learning, including a multi-source data fusion terminal, a user behavior modeling terminal, and a dynamic pricing decision terminal, wherein the multi-source data fusion terminal, the user behavior modeling terminal, and the dynamic pricing decision terminal are all equipped with a strategy feedback module; The multi-source data fusion terminal is used to perform time alignment, feature normalization and dimensionality compression on multi-source data by accessing heterogeneous data streams from film content database, regional consumption level database, competitor screening database and historical box office database, generate structured fusion feature vectors, and transmit the fusion feature vectors to the user behavior modeling terminal in real time. The user behavior modeling terminal is used to construct a fine-grained user preference profile based on the fused feature vector, combined with the movie viewing records of the cinema membership system and the click stream logs of the online ticketing platform, and to quantitatively predict the ticket purchase intention of different user groups under specific movie and time period combinations through a deep neural network model, and output a movie viewing demand intensity index. The dynamic pricing decision-making terminal is used to receive the movie-watching demand intensity index and the fused feature vector, call the pre-trained reinforcement learning pricing model, generate a differentiated ticket price strategy for each screening based on the current number of remaining seats in the theater, the stage of the film's release cycle, and holiday attributes, and push the ticket price strategy to the cinema ticketing system for execution. The strategy feedback module is used to collect actual occupancy rate, refund rate and user rating data after the ticketing strategy is implemented, calculate the strategy implementation deviation value, and send the deviation value back to the dynamic pricing decision terminal to trigger the online fine-tuning mechanism of the model parameters.

[0007] The multi-source data fusion terminal includes a data access module, a feature alignment module, and a vector generation module; The data access module includes a film content unit, a regional consumption unit, a competitive screening unit, and a historical box office unit. The film content unit is used to obtain film type, director information, cast, trailer popularity index and professional rating data from the metadata interface provided by the film distributor; The regional consumption unit is used to extract the per capita disposable income of residents in the city where the cinema is located, the traffic flow index of the business district, and the historical average ticket price level from a third-party macroeconomic database. The competitive scheduling unit is used to periodically crawl the daily schedules, ticket prices, and prime-time occupancy rates of other cinemas within a five-kilometer radius using a web crawler program. The historical box office unit is used to read the historical occupancy rate curve and ticket price elasticity coefficient of similar films in the same season and release period over the past three years from the cinema's local database. The feature alignment module includes a timestamp synchronization unit and a missing value interpolation unit; The timestamp synchronization unit is used to align the original records from different data sources according to a unified UTC time standard, ensuring that all features have semantic consistency within the same time window; The missing value imputation unit is used to fill in missing feature fields using a mean-filling strategy based on neighboring videos or similar regions. The vector generation module includes a normalization processing unit and a principal component compression unit; The normalization processing unit is used to map the aligned features of each dimension to the zero-to-one interval to eliminate dimensional differences. The principal component compression unit is used to perform principal component analysis on the normalized high-dimensional feature matrix, retain the principal components with a cumulative contribution rate greater than a preset threshold, and form a low-dimensional fusion feature vector.

[0008] The user behavior modeling module includes a profile building module, a willingness prediction module, and a demand quantification module. The profile construction module includes a movie viewing trajectory unit and an interactive behavior unit; The movie viewing trajectory unit is used to analyze the historical ticket purchase records in the cinema membership system and extract users' long-term preference patterns for movie types, screening times and seating areas. The interactive behavior unit is used to analyze the user's page dwell time, session switching frequency, and coupon usage tendency on the ticketing platform, and to identify their short-term decision sensitivity points. The intention prediction module includes an embedding encoding unit and an attention weighting unit; The embedding coding unit is used to convert discrete user attributes and video features into dense vector representations; The attention weighting unit is used to calculate the relevance weight of each movie-watching event in the user's historical behavior sequence to the current show, and the weighted aggregation is then input into the multilayer perceptron network. The demand quantification module includes a probability output unit and an intensity mapping unit; The probability output unit is used to output the probability value of a user completing a ticket purchase for a specified session; The intensity mapping unit is used to map the probability value into a movie-watching demand intensity index according to a nonlinear function. This index increases monotonically as the probability value increases.

[0009] The dynamic pricing decision-making terminal includes a model invocation module, a strategy generation module, and an execution push module; The model invocation module includes a state encoding unit and an action space unit; The state encoding unit is used to concatenate the fusion feature vector, the movie-watching demand intensity index, the remaining seat ratio, and the number of screening days into the state input of the reinforcement learning model. The action space unit is used to define a discrete set of actions for fare adjustment, including multiple preset levels for fluctuations above and below the base price; The strategy generation module includes a Q-value evaluation unit and an optimal action unit; The Q-value evaluation unit is used to calculate the expected long-term benefit of each action in the current state through a deep Q-network; The optimal action unit is used to select the action with the largest Q value as the target fare level, and calculate the final fare value in combination with the benchmark fare. The execution push module includes a format conversion unit and an interface call unit; The format conversion unit is used to encapsulate the final ticket price value into a data packet that conforms to the communication protocol of the cinema ticketing system; The interface calling unit is used to push data packets to the ticketing system's ticket price update interface via the HTTPS protocol to complete real-time price synchronization.

[0010] The strategy feedback module includes an effect monitoring module, a deviation calculation module, and a parameter update module; The performance monitoring module includes an occupancy rate collection unit, a ticket refund rate collection unit, and a rating collection unit. The occupancy rate collection unit is used to calculate the ratio of the actual number of seats sold to the total number of seats after the event ends. The ticket refund rate collection unit is used to count the proportion of ticket refund orders that occurred within two hours before the start of the performance to the total number of orders. The rating collection unit is used to aggregate the average star ratings submitted by users within 24 hours after watching the movie; The deviation calculation module includes a target comparison unit and an error integration unit; The target comparison unit is used to perform a difference calculation between the actual movie-watching demand intensity index and the predicted value. The error integration unit is used to accumulate the deviation values ​​of multiple consecutive games through a sliding window to generate a comprehensive strategy deviation signal. The parameter update module includes a gradient calculation unit and a weight adjustment unit; The gradient calculation unit is used to calculate the gradient direction of the model parameters based on the backpropagation of the comprehensive strategy deviation signal; The weight adjustment unit is used to update the network weights of the reinforcement learning model in small steps along the gradient direction to complete the online learning process.

[0011] The present invention has the following beneficial effects: By constructing a multi-source data fusion platform, the structured integration of multi-dimensional heterogeneous data, including film content, regional economy, competitive landscape, and historical performance, was achieved, overcoming the market perception lag problem caused by data silos in traditional methods. Through the joint modeling of long-term preferences and short-term interaction behaviors in the user behavior modeling platform, the granularity and accuracy of movie-going demand prediction were significantly improved, avoiding ticket price recommendation bias caused by coarse user profiles. By introducing a reinforcement learning mechanism in the dynamic pricing decision-making platform, the ticketing strategy can achieve a dynamic balance between maximizing revenue and ensuring occupancy rates. Furthermore, it can continuously optimize model parameters based on real-time effect data provided by the strategy feedback module, thereby generating highly collaborative and adaptable pricing schemes for different types of films, different time periods, and different regional scenarios, effectively improving the overall operating revenue and market competitiveness of cinemas. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall system architecture of a cinema intelligent pricing system based on multi-source data fusion and machine learning according to the present invention; Figure 2 This is a schematic diagram of the architecture of the multi-source data fusion terminal of a cinema intelligent pricing system based on multi-source data fusion and machine learning according to the present invention. Figure 3 This is a schematic diagram of the architecture of the user behavior modeling terminal of a cinema intelligent pricing system based on multi-source data fusion and machine learning according to the present invention. Figure 4 This is a schematic diagram of the architecture of the dynamic pricing decision-making terminal of a cinema intelligent pricing system based on multi-source data fusion and machine learning according to the present invention. Figure 5 This is a schematic diagram of the architecture of the strategy feedback module of a cinema intelligent pricing system based on multi-source data fusion and machine learning according to the present invention. Detailed Implementation

[0013] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0014] Example 1, please refer to Figures 1 to 2 As shown: A cinema intelligent pricing system based on multi-source data fusion and machine learning. The system includes a multi-source data fusion terminal, a user behavior modeling terminal, and a dynamic pricing decision terminal. The multi-source data fusion terminal, the user behavior modeling terminal, and the dynamic pricing decision terminal are all equipped with a strategy feedback module. The multi-source data fusion terminal is used to access heterogeneous data streams from film content databases, regional consumption level databases, competitor screening databases, and historical box office databases. It performs time alignment, feature normalization, and dimensionality compression on the multi-source data to generate structured fusion feature vectors, and transmits the fusion feature vectors to the user behavior modeling terminal in real time. The user behavior modeling end is used to construct a fine-grained user preference profile based on the fused feature vector, combined with the movie viewing records of the cinema membership system and the click stream logs of the online ticketing platform. It also uses a deep neural network model to quantitatively predict the ticket purchase intention of different user groups under specific movie and time period combinations, and outputs a movie viewing demand intensity index. The dynamic pricing decision-making end receives the movie-watching demand intensity index and fused feature vector, calls the pre-trained reinforcement learning pricing model, generates a differentiated ticket price strategy for each screening based on the current number of remaining seats in the theater, the stage of the film's release cycle, and holiday attributes, and pushes the ticket price strategy to the cinema ticketing system for execution. The strategy feedback module is used to collect actual occupancy rate, refund rate and user rating data after the ticketing strategy is implemented, calculate the strategy implementation deviation value, and send the deviation value back to the dynamic pricing decision terminal to trigger the online fine-tuning mechanism of model parameters.

[0015] The multi-source data fusion module includes a data access module, a feature alignment module, and a vector generation module; The data access module includes film content units, regional consumption units, competitive screening units, and historical box office units; The film content unit is used to obtain film type, director information, cast, trailer popularity index and professional rating data from the metadata interface provided by the film distributor; Regional consumption units are used to extract data on per capita disposable income of residents in the city where the cinema is located, the foot traffic index of the business district, and the historical average ticket price level from a third-party macroeconomic database. The competitive scheduling unit is used to periodically crawl the daily schedules, ticket prices, and prime-time occupancy rates of other cinemas within a five-kilometer radius using a web crawler program; The historical box office unit is used to retrieve the historical occupancy rate curves and ticket price elasticity coefficients of similar films in the same season and release period over the past three years from the cinema's local database. The feature alignment module includes a timestamp synchronization unit and a missing value interpolation unit; The timestamp synchronization unit is used to align raw records from different data sources according to a unified UTC time standard, ensuring that all features have semantic consistency within the same time window; The missing value imputation unit is used to fill in missing feature fields using a mean-based strategy based on neighboring videos or similar regions. The vector generation module includes a normalization processing unit and a principal component compression unit; The normalization processing unit is used to map the aligned features of each dimension to the zero-to-one interval to eliminate dimensional differences. The principal component compression unit is used to perform principal component analysis on the normalized high-dimensional feature matrix, retaining the principal components whose cumulative contribution rate is greater than a preset threshold, and forming a low-dimensional fused feature vector.

[0016] Example 2, please refer to Figure 3 As shown: Based on Example 1, the user behavior modeling module includes a profile building module, a willingness prediction module, and a demand quantification module; The profile building module includes a viewing trajectory unit and an interactive behavior unit; The movie viewing trajectory unit is used to analyze historical ticket purchase records in the cinema membership system and extract long-term preference patterns of users for movie types, screening times and seating areas. The interactive behavior unit is used to analyze users' page dwell time, session switching frequency, and coupon usage tendencies on the ticketing platform to identify their short-term decision-making sensitivities. The intention prediction module includes an embedding encoding unit and an attention weighting unit; Embedded coding units are used to convert discrete user attributes and video features into dense vector representations; The attention weighting unit is used to calculate the relevance weight of each movie-watching event in the user's historical behavior sequence to the current show, and the weighted aggregation is then input into the multilayer perceptron network. The demand quantification module includes a probability output unit and an intensity mapping unit; The probability output unit is used to output the probability value of a user completing a ticket purchase for a specified session; The intensity mapping unit is used to map probability values ​​to a movie-watching demand intensity index using a non-linear function. This index monotonically increases with the probability value, and the non-linear function is... ,in For the probability of purchasing tickets, This is the steepness coefficient. This is the midpoint threshold.

[0017] Example 3, please refer to Figure 4 As shown: Based on Embodiment 1, the dynamic pricing decision-making terminal includes a model invocation module, a strategy generation module, and an execution push module; The model invocation module includes a state encoding unit and an action space unit; The state encoding unit is used to concatenate the fused feature vector, the intensity index of movie-watching demand, the proportion of remaining seats, and the number of days of screening into the state input of the reinforcement learning model; The action space unit is used to define a discrete set of actions for fare adjustment, including multiple preset levels for fluctuations above and below the base price. The specific level configurations are shown in the table below: Table 1 - Specific Gear Configuration in This Embodiment

[0018] The strategy generation module includes a Q-value evaluation unit and an optimal action unit; The Q-value evaluation unit is used to calculate the expected long-term benefit of each action in the current state using a deep Q-network; The optimal action unit is used to select the action with the largest Q value as the target fare level, and the final fare value is calculated in combination with the benchmark fare. The push module includes a format conversion unit and an interface call unit; The format conversion unit is used to encapsulate the final ticket price value into a data packet that conforms to the communication protocol of the cinema ticketing system; The interface call unit is used to push data packets to the ticketing system's ticket price update interface via the HTTPS protocol to complete real-time price synchronization.

[0019] Example 4, please refer to Figure 5 As shown: Based on Embodiment 1, the strategy feedback module includes an effect monitoring module, a deviation calculation module, and a parameter update module; The performance monitoring module includes a seating capacity data collection unit, a ticket refund rate data collection unit, and a rating data collection unit. The occupancy rate data collection unit is used to calculate the ratio of the actual number of seats sold to the total number of seats after the event ends. The ticket refund rate collection unit is used to count the proportion of ticket refund orders that occurred within two hours before the start of the performance to the total number of orders. The rating collection unit is used to aggregate the average star ratings submitted by users within 24 hours after watching the movie; The deviation calculation module includes a target comparison unit and an error integration unit; The target comparison unit is used to calculate the difference between the actual movie-watching demand intensity index and the predicted value. The error integration unit is used to accumulate the deviation values ​​of multiple consecutive games through a sliding window to generate a comprehensive strategy deviation signal; The parameter update module includes a gradient calculation unit and a weight adjustment unit; The gradient calculation unit is used to calculate the gradient direction of the model parameters based on the backpropagation of the comprehensive strategy deviation signal; The weight adjustment unit is used to update the network weights of the reinforcement learning model in small steps along the gradient direction, thus completing the online learning process.

[0020] This invention discloses a cinema intelligent pricing system based on multi-source data fusion and machine learning. During operation, the system first connects to four heterogeneous data sources: film content, regional consumption, competitive scheduling, and historical box office data. After timestamp synchronization and missing value imputation, a low-dimensional fusion feature vector is generated through normalization and principal component analysis. Subsequently, the user behavior modeling end uses this feature vector to integrate member viewing trajectories and platform interaction behaviors, constructing user profiles using embedding encoding and attention mechanisms, and outputting a viewing demand intensity index. Finally, the dynamic pricing decision-making end integrates the fusion feature vector, demand intensity index, and remaining data... The remaining seat ratio and the number of screening days are jointly encoded as reinforcement learning states. A deep Q-network is called to evaluate the long-term benefits of each ticketing action, select the optimal action to generate the final ticket price, and push it to the ticketing system. Finally, the strategy feedback module collects the occupancy rate, refund rate, and user ratings, calculates the strategy execution deviation, and drives the online fine-tuning of model parameters to form a closed-loop optimization mechanism. Through the above process, the system realizes the structured integration of multi-source market signals, fine-grained characterization of user preferences, and dynamic adaptive adjustment of ticketing strategies, effectively solving the pricing inaccuracies caused by data fragmentation, coarse profiling, and rigid strategies in traditional methods.

[0021] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

[0022] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principles of this invention are further supplemented below with a specific application scenario.

[0023] Suppose a multiplex cinema in a prime commercial area of ​​a first-tier city plans to release a highly anticipated sci-fi blockbuster during the summer season. The film is directed by a renowned director, stars top actors, and its trailer has already garnered over 50 million views on social media platforms, with an initial rating of 8.2 on professional film review websites. At this point, the system initiates its intelligent pricing process: First, if... Figure 1 and Figure 2As shown, the film content unit in the multi-source data fusion terminal obtains the aforementioned film metadata in real time through the film distributor's API; the regional consumption unit retrieves data from a third-party macroeconomic database, showing that the per capita disposable income of the city where the cinema is located in the previous quarter was 7,800 yuan, the average daily foot traffic index of the business district on weekends reached 92 (out of 100), and the historical average ticket price was 58 yuan; the competitive screening unit uses a web crawler to scrape the screening information of the other 6 cinemas within a 5-kilometer radius every hour, finding that 3 of them have already scheduled the film during prime time, with an average ticket price of 62 yuan and a prime time screening occupancy rate of 85%; the historical box office unit... Retrieving data from the local database, similar sci-fi films released during the summer holidays of the past two years had an average occupancy rate of 68% on their opening day, with a ticket price elasticity coefficient of -1.3. Subsequently, the timestamp synchronization unit mapped the heterogeneous data to the 14:00 time window in the UTC+8 time zone. The missing value imputation unit filled in the ticket price field for individual cinemas that were not publicly disclosed using the average price of 61 yuan in the same business district. The normalization unit linearly scaled each feature to the [0,1] interval, and the principal component compression unit performed PCA dimensionality reduction on the 28-dimensional original features, retaining the top 8 principal components with a cumulative contribution rate of 92%, generating a low-dimensional fused feature vector. .

[0024] Next, as Figure 3 As shown, the user behavior modeling terminal receives... Subsequently, the viewing trajectory unit in the profile construction module analyzed the historical records of the cinema's 120,000 members, identifying that approximately 35% of users had watched at least two science fiction films in the past year and preferred the 20:00–22:00 time slot. The interaction behavior unit analyzed the clickstream logs of the ticketing platform over the past 24 hours, finding that the average dwell time on the film's details page was 128 seconds, the frequency of showtime switching was lower than the average, and the coupon redemption rate was only 18%, indicating that users had low price sensitivity. The embedding encoding unit mapped discrete features such as user ID, film type, and time slot into a 128-dimensional dense vector. The attention weight unit calculated the relevance weights between the user's three most recent viewing events and the current showtime based on the Transformer mechanism, and after weighting, input the data into a three-layer MLP network. The probability output unit outputs the probability of purchasing tickets for this showtime. Substituting the intensity mapping element into the nonlinear function: ; Obtain movie viewing demand intensity index .

[0025] Subsequently, as Figure 4 As shown, the state coding unit of the dynamic pricing decision-making end will , Current percentage of remaining seats in the theater (Total seating capacity: 180, 50 tickets sold) and number of screening days (On the premiere day) concatenated into a 42-dimensional state vector The motion space unit defines six fare levels for motion. The Q-value evaluation unit loads a pre-trained deep Q-network to... Given the input, output the expected cumulative reward Q for each action, where The corresponding Q value is the highest; the optimal action unit selects a 10% premium strategy accordingly. If the base ticket price is 60 yuan, the final ticket price will be 66 yuan; the format conversion unit encapsulates the ticket price into a JSON format data packet, and the interface call unit pushes it to the RESTful API of the cinema ticketing system through an HTTPS POST request to achieve price synchronization across all channels within 5 minutes.

[0026] After the show concluded, the strategy feedback module was activated: the attendance rate collection unit recorded 132 tickets sold, representing an attendance rate of 73.3%; the refund rate collection unit detected 3 refunds within two hours of the show's start, representing a refund rate of 2.3%; the rating collection unit aggregated the average of 4.6 stars (out of 5) submitted by users over the past 24 hours; and the target comparison unit calculated the actual demand intensity. Compared with the predicted value deviation The error integration unit performs a sliding window accumulation of the deviations from five consecutive sessions to generate a comprehensive deviation signal. Gradient calculation unit based on Backpropagation calculates the gradient of the output layer weights of the DQN network. The weight adjustment unit is based on the learning rate. implement The model was fine-tuned online.

[0027] Through the aforementioned closed-loop process, the system successfully achieved a 10% premium on ticket prices during prime time on the premiere day while maintaining a 73.3% occupancy rate. This resulted in a 6.5 yuan increase in revenue per seat compared to the average ticket price of 62 yuan at surrounding cinemas, and the refund rate was lower than the industry warning line. This verifies the technical effectiveness of the invention in maximizing revenue and market acceptance in high-profile film scenarios. This achievement relies on the accurate capture of competitive scheduling and regional consumption levels by the multi-source data fusion end, the fine-grained identification of low-price-sensitive groups by the user behavior modeling end, and the optimization capability of the reinforcement learning model for long-term revenue in the state-to-action space. These three elements work together to solve the problems of response lag and strategy rigidity caused by data fragmentation in traditional pricing.

[0028] All contents not described in detail in the specification are existing technologies known to those skilled in the art, and the data interfaces, machine learning frameworks and communication protocols are not specifically limited and can be implemented using conventional technical means. Software modules or hardware components not mentioned in this technical solution are not shown in the accompanying drawings because they are general technologies, and will not be described in detail here.

[0029] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cinema intelligent pricing system based on multi-source data fusion and machine learning, characterized in that, The system includes a multi-source data fusion terminal, a user behavior modeling terminal, and a dynamic pricing decision terminal. The multi-source data fusion terminal, the user behavior modeling terminal, and the dynamic pricing decision terminal are all equipped with a strategy feedback module. The multi-source data fusion terminal is used to access heterogeneous data streams from film content database, regional consumption level database, competitor screening database and historical box office database, perform time alignment, feature normalization and dimensionality compression on the multi-source data, generate structured fusion feature vectors, and transmit the fusion feature vectors to the user behavior modeling terminal in real time. The user behavior modeling terminal is used to construct a fine-grained user preference profile based on the fused feature vector, combined with the movie viewing records of the cinema membership system and the click stream logs of the online ticketing platform, and to quantitatively predict the ticket purchase intention of different user groups under specific movie and time period combinations through a deep neural network model, and output a movie viewing demand intensity index. The dynamic pricing decision-making terminal is used to receive the movie-watching demand intensity index and the fused feature vector, call the pre-trained reinforcement learning pricing model, generate a differentiated ticket price strategy for each screening based on the current number of remaining seats in the theater, the stage of the film's release cycle, and holiday attributes, and push the ticket price strategy to the cinema ticketing system for execution. The strategy feedback module is used to collect actual occupancy rate, refund rate and user rating data after the ticketing strategy is implemented, calculate the strategy implementation deviation value, and send the deviation value back to the dynamic pricing decision terminal to trigger the online fine-tuning mechanism of the model parameters.

2. The system according to claim 1, characterized in that, The multi-source data fusion terminal includes a data access module, a feature alignment module, and a vector generation module; The data access module includes a film content unit, a regional consumption unit, a competitive screening unit, and a historical box office unit. The film content unit is used to obtain film type, director information, cast, trailer popularity index and professional rating data from the metadata interface provided by the film distributor; The regional consumption unit is used to extract the per capita disposable income of residents in the city where the cinema is located, the traffic flow index of the business district, and the historical average ticket price level from a third-party macroeconomic database. The competitive scheduling unit is used to periodically crawl the daily schedules, ticket prices, and prime-time occupancy rates of other cinemas within a five-kilometer radius using a web crawler program. The historical box office unit is used to read the historical occupancy rate curves and ticket price elasticity coefficients of similar films in the same season and release period over the past three years from the cinema's local database.

3. The system according to claim 2, characterized in that, The feature alignment module includes a timestamp synchronization unit and a missing value interpolation unit; The timestamp synchronization unit is used to align the original records from different data sources according to a unified UTC time standard, ensuring that all features have semantic consistency within the same time window; The missing value imputation unit is used to fill in missing feature fields using a mean-filling strategy based on neighboring videos or similar regions.

4. The system according to claim 2, characterized in that, The vector generation module includes a normalization processing unit and a principal component compression unit; The normalization processing unit is used to map the aligned features of each dimension to the interval between zero and one. The principal component compression unit is used to perform principal component analysis on the normalized high-dimensional feature matrix, retain the principal components with a cumulative contribution rate greater than a preset threshold, and form a low-dimensional fusion feature vector.

5. The system according to claim 1, characterized in that, The user behavior modeling module includes a profile building module, a willingness prediction module, and a demand quantification module. The profile construction module includes a movie viewing trajectory unit and an interactive behavior unit; The movie viewing trajectory unit is used to analyze the historical ticket purchase records in the cinema membership system and extract users' long-term preference patterns for movie types, screening times and seating areas. The interactive behavior unit is used to analyze the user's page dwell time, session switching frequency, and coupon usage tendency on the ticketing platform, and to identify their short-term decision-making sensitivities.

6. The system according to claim 5, characterized in that, The intention prediction module includes an embedding encoding unit and an attention weighting unit; The embedding coding unit is used to convert discrete user attributes and video features into dense vector representations; The attention weighting unit is used to calculate the relevance weight of each movie-watching event in the user's historical behavior sequence to the current show, and the weighted aggregation is then input into the multilayer perceptron network.

7. The system according to claim 5, characterized in that, The demand quantification module includes a probability output unit and an intensity mapping unit; The probability output unit is used to output the probability value $p$ of a user completing a ticket purchase for a specified session; The intensity mapping unit is used to map probability values. According to nonlinear functions: ； Mapped to an index of movie-watching demand intensity, where This is the steepness coefficient. This is the midpoint threshold.

8. The system according to claim 1, characterized in that, The dynamic pricing decision-making terminal includes a model invocation module, a strategy generation module, and an execution push module; The model invocation module includes a state encoding unit and an action space unit; The state encoding unit is used to concatenate the fusion feature vector, the movie-watching demand intensity index, the remaining seat ratio, and the number of screening days into the state input of the reinforcement learning model. The action space unit is used to define a discrete set of actions for fare adjustment, including multiple preset levels for fluctuations above and below the base price.

9. The system according to claim 8, characterized in that, The strategy generation module includes a Q-value evaluation unit and an optimal action unit; The Q-value evaluation unit is used to calculate the expected long-term benefit of each action in the current state through a deep Q-network; The optimal action unit is used to select the action with the largest Q value as the target fare level, and calculate the final fare value in combination with the benchmark fare. The execution push module includes a format conversion unit and an interface call unit; The format conversion unit is used to encapsulate the final ticket price value into a data packet that conforms to the communication protocol of the cinema ticketing system; The interface calling unit is used to push data packets to the ticketing system's ticket price update interface via the HTTPS protocol.

10. The system according to claim 1, characterized in that, The strategy feedback module includes an effect monitoring module, a deviation calculation module, and a parameter update module; The performance monitoring module includes an occupancy rate collection unit, a ticket refund rate collection unit, and a rating collection unit. The occupancy rate collection unit is used to calculate the ratio of the actual number of seats sold to the total number of seats after the event ends. The ticket refund rate collection unit is used to count the proportion of ticket refund orders that occurred within two hours before the start of the performance to the total number of orders. The rating collection unit is used to aggregate the average star ratings submitted by users within 24 hours after watching the movie; The deviation calculation module includes a target comparison unit and an error integration unit; The target comparison unit is used to perform a difference calculation between the actual movie-watching demand intensity index and the predicted value. The error integration unit is used to accumulate the deviation values ​​of multiple consecutive games through a sliding window to generate a comprehensive strategy deviation signal. The parameter update module includes a gradient calculation unit and a weight adjustment unit; The gradient calculation unit is used to calculate the gradient direction of the model parameters based on the backpropagation of the comprehensive strategy deviation signal; The weight adjustment unit is used to update the network weights of the reinforcement learning model in small steps along the gradient direction.

Images