Method for predicting and evaluating popularity of document travel consumption product based on artificial intelligence
By simulating the market dynamics of cultural and tourism consumption products through a multi-agent reinforcement learning framework, this approach solves the problem of the inability to quantify the competition and synergy effects between products in existing technologies. It improves the accuracy and reliability of popularity prediction and provides precise support for adjusting marketing strategies.
Patent Information
- Application Number
- CN202511495542.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2026-01-13
AI Technical Summary
Existing methods for predicting the popularity of cultural and tourism consumption products fail to quantify the competitive or synergistic effects between products. The prediction results deviate significantly from the actual market dynamics and lack the ability to trace the causes of popularity fluctuations, making it difficult to support precise strategy optimization.
By employing a multi-agent reinforcement learning framework, a virtual market simulation environment is constructed by collecting and integrating data on the attributes of cultural and tourism consumption products, user interactions, and environmental events. This simulates competition and synergistic effects between products, generates a popularity probability distribution, and associates it with marketing actions to achieve a quantitative assessment of popularity changes.
It improves the accuracy and reliability of popularity prediction, provides decision-making basis that is closer to the real market, can trace the root cause of popularity changes and associate them with specific marketing actions, and enhances the interpretability and operability of the evaluation results.
Smart Images
Figure CN121329487A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent decision-making technology, and in particular to an artificial intelligence-based method for predicting and evaluating the popularity of cultural and tourism consumption products. Background Technology
[0002] In recent years, with the deepening application of artificial intelligence in business analytics, methods for predicting and evaluating the popularity of cultural and tourism consumption products have gradually shifted from traditional statistical models to data-driven intelligent decision-making. These methods primarily utilize machine learning algorithms to mine historical consumption data, user behavior logs, and market environment information, generating point estimates or interval predictions of product popularity through regression analysis, time series forecasting, or deep learning models. These methods have improved the accuracy of popularity trend judgments to a certain extent and provided preliminary data support for resource allocation, becoming an important tool for current cultural and tourism consumption market analysis.
[0003] Existing methods for predicting and evaluating the popularity of cultural and tourism consumption products have limitations. Most models treat cultural and tourism consumption products as independent entities for prediction, failing to quantify the competition or synergistic effects between products, resulting in a large deviation between the prediction results and the actual market dynamics. In addition, the prediction output is mostly numerical results, lacking the ability to trace the causes of popularity fluctuations, making it difficult for the evaluation results to support accurate strategy optimization and limiting the decision support value of the prediction model in a dynamic market environment. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides an artificial intelligence-based method for predicting and evaluating the popularity of cultural and tourism consumption products, which solves the problems of being unable to simulate multi-product interaction and lacking causal traceability.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an artificial intelligence-based method for predicting and evaluating the popularity of cultural and tourism consumption products. The method includes: collecting attribute data, user interaction data, and environmental event data of cultural and tourism consumption products; cleaning, standardizing, and integrating the data to generate a market dataset; defining a state space, action space, and interaction rules based on the market dataset, and creating a corresponding dynamic agent for each cultural and tourism consumption product; constructing a multi-agent reinforcement learning framework and inputting the dynamic agents into the framework to generate a virtual market simulation environment; in the virtual market simulation environment, driving each dynamic agent to select marketing actions based on the state attribute values in the current state space, calculating the state attribute values for the next moment based on the marketing actions and interaction rules executed by all dynamic agents, and generating a market state evolution sequence; calculating the popularity probability distribution of each cultural and tourism consumption product in the market state evolution sequence during a preset target prediction period, extracting popularity change events that trigger preset popularity thresholds and their corresponding marketing actions, fusing the popularity probability distribution, popularity change events, and corresponding marketing actions to generate a preliminary evaluation result; and mapping the preliminary evaluation result with the attribute data of cultural and tourism consumption products to generate a structured evaluation report.
[0007] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the steps of collecting attribute data of cultural and tourism consumption products, user interaction data, and environmental event data, and performing data cleaning, standardization, and correlation integration to generate a market dataset are as follows. Collect data on the attributes of cultural and tourism consumption products, user interaction data, and environmental event data, and perform missing value filling and product identifier verification on the attribute data of cultural and tourism consumption products. Filter invalid sessions and unify timestamps in user interaction data; Environmental event data is mapped to a preset classification system through event type mapping; Max-min normalization and one-hot encoding are performed on the attribute data of cultural and tourism consumption products, user interaction data, and environmental event data, and all time data is converted into Unix timestamp format. After linking user interaction data with cultural and tourism consumption product attribute data through product identifiers, environmental event data is linked with the already associated user interaction data by matching timestamps to generate a market dataset.
[0008] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the specific steps of defining the state space, action space, and interaction rules based on the market dataset, and creating a corresponding dynamic intelligent agent for each cultural and tourism consumption product, are as follows. The state space is defined based on the attribute dimensions of Chinese tourism consumption products in the market dataset, the action space is defined by enumerating executable marketing actions, and the interaction rules are defined by state transition functions and reward functions. Create a corresponding intelligent agent for each cultural and tourism consumption product, and bind it with a strategy function and a value function to generate a dynamic intelligent agent.
[0009] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the specific steps for constructing a multi-agent reinforcement learning framework and inputting dynamic agents into the framework to generate a virtual market simulation environment are as follows. A multi-agent reinforcement learning framework is constructed by using a centralized training framework and a distributed execution framework. By loading the state space, action space, interaction rules, and dynamic agents into a multi-agent reinforcement learning framework, a virtual market simulation environment is generated.
[0010] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the step of driving each dynamic intelligent agent to select marketing actions based on the state attribute values in the current state space in a virtual market simulation environment includes the following specific steps. Create an empty-market state evolution sequence using list initialization operations; Read the state attribute values of all dynamic agents in the virtual market simulation environment at the current moment, input the state attribute values at the current moment into the corresponding policy function, and generate the corresponding marketing actions.
[0011] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the step of calculating the state attribute value at the next moment and generating a market state evolution sequence based on the marketing actions and interaction rules executed by all dynamic intelligent agents is as follows: Collect the marketing actions of each dynamic intelligent agent and form a marketing action record table; Add the current state attribute value and marketing action record table to the empty-market state evolution sequence; Apply the state transition function to the current state attribute value and marketing action record table, calculate the state attribute value for the next moment, and load it into the virtual market simulation environment; Iteratively execute read, record, and append operations until the preset number of time steps is reached, and output the market state evolution sequence.
[0012] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the specific steps for calculating the probability distribution of the popularity of each cultural and tourism consumption product in the market state evolution sequence during the preset target prediction period are as follows: By traversing the market state evolution sequence, for each cultural and tourism consumption product, all popularity values within the preset target prediction period are extracted to form a popularity data sequence. The kernel density estimation method is applied to the heat data sequence to generate a heat probability distribution.
[0013] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the steps of extracting popularity change events that trigger preset popularity thresholds and corresponding marketing actions, and fusing the popularity probability distribution, popularity change events, and corresponding marketing actions to generate preliminary evaluation results are as follows. Calculate the change in popularity for each cultural and tourism consumption product, mark events where the change in popularity exceeds a preset popularity threshold as popularity change events, and record popularity change events and corresponding marketing actions. By integrating the probability distribution of popularity, events that change popularity, and corresponding marketing actions, preliminary evaluation results are generated.
[0014] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products according to the present invention, the specific steps for associating and mapping the preliminary evaluation results with the attribute data of cultural and tourism consumption products to generate a structured evaluation message are as follows. The evaluation data of each cultural and tourism consumption product is extracted from the preliminary evaluation results. The evaluation data is then matched with the attribute data of the cultural and tourism consumption product through product identifiers to generate associated data records. The structured data sequence method is applied to the associated data records to generate a structured evaluation message.
[0015] As a preferred embodiment of the AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products, the structured data sequence method converts all related data records into strings in JSON standard format and integrates the structure using predefined key-value pairs.
[0016] The beneficial effects of this invention are as follows: by simulating market dynamics through a multi-agent reinforcement learning framework, it can capture the competition and synergy effects between different cultural and tourism consumption products, improve the accuracy and reliability of popularity prediction, and provide decision-making basis that is closer to the real market; by integrating popularity probability distribution and causal event analysis, it can achieve quantitative evaluation of popularity changes, trace the root cause of fluctuations and associate them with specific marketing actions, enhance the interpretability and operability of the evaluation results, and achieve the goal of accurately adjusting marketing strategies. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of an AI-based method for predicting and evaluating the popularity of cultural and tourism consumption products.
[0019] Figure 2 A flowchart for generating a virtual simulation environment.
[0020] Figure 3 A flowchart for generating the probability density distribution.
[0021] Figure 4 A flowchart for generating structured evaluation messages. Detailed Implementation
[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides an artificial intelligence-based method for predicting and evaluating the popularity of cultural and tourism consumption products, including the following steps: S1. Collect data on the attributes of cultural and tourism consumption products, user interaction data, and environmental event data, and perform data cleaning, standardization, and correlation integration to generate a market dataset; S1.1, Fill in missing values and validate product identifiers for cultural and tourism consumption product attribute data; filter invalid sessions and unify timestamps for user interaction data; map environmental event data to a preset classification system through event type mapping; It should be noted that the attribute data of cultural and tourism consumption products are grouped according to product category. Within each group, the mode of the attribute field (a single data column used to describe specific characteristics of the product) with missing values is counted, and the mode is used as the fill value to fill all missing values in the current group. Extract the product identifier for each cultural and tourism consumption product and check for duplicate product identifiers. If two or more different cultural and tourism consumption products are found to have the same product identifier, they are identified as duplicate identifiers and need to be checked and corrected to ensure that each product identifier corresponds to only one cultural and tourism consumption product. Check the session duration of each user interaction data entry and remove user interaction data with session durations lower than the minimum time unit; define the minimum time unit based on the timestamp recording precision of the data collection, usually set to 0.01 seconds. The value is based on the fact that the minimum precision of the timestamp of common data collection components is one-hundredth of a second. Session durations can only be effectively recorded and distinguished if they reach or exceed the minimum precision. Session durations with less than the minimum precision cannot be reliably measured; convert the timestamp information in all user interaction data into a unified "year-month-day-hour:minute:second" format. By analyzing common event types in the cultural and tourism consumption sector, such as weather changes, holidays, and promotional activities, a standard category list is developed. Each category is clearly defined in terms of name and scope based on domain knowledge and historical data. For example, "heavy rain" is classified as "abnormal weather," and "National Day" is classified as "holiday." Keyword matching rules are also set. For example, if the event type description contains "discount," it is mapped to the "promotional activity" category. By comparing the event descriptions in the environmental event data with the keywords in the classification system, each environmental event data is classified into the most matching event category.
[0026] S1.2 Perform max-min normalization and one-hot encoding on the attribute data of cultural and tourism consumption products, user interaction data, and environmental event data, and convert all time data into Unix timestamp format; It should be noted that the maximum-minimum normalization operation is performed on the numerical attribute fields (such as product price) in the cultural and tourism consumption product attribute data, user interaction data, and environmental event data: find the maximum and minimum values in each numerical attribute field, calculate the numerical difference between each value and the minimum value, calculate the normalized difference between the maximum and the minimum values, and calculate the ratio of the numerical difference to the normalized difference, thus completing the maximum-minimum normalization operation; Perform one-hot encoding on categorical attribute fields (such as product type) in cultural and tourism consumption product attribute data, user interaction data, and environmental event data. That is, create a new binary field (0 or 1) for each product category, so that each category has an independent identifier. All time data is parsed into structured time information containing components such as year, month, day, hour, minute, and second; the total number of seconds between the structured time information and the epoch time (based on the Coordinated Universal Time standard definition, i.e., January 1, 1970, 00:00:00 UTC) is calculated, which is the corresponding Unix timestamp value.
[0027] S1.3 After associating user interaction data with cultural and tourism consumption product attribute data through product identifiers, environmental event data is associated with the associated user interaction data by matching timestamps to generate a market dataset.
[0028] It should be noted that the process involves iterating through each user interaction record in the user interaction data, using the product identifier contained in each record, searching for attribute records with the same product identifier in the cultural tourism consumption product attribute data, and merging the found attribute records with the current user interaction record to form associated user interaction data. Timestamp matching is used to compare the timestamps of each record in the environmental event data with the timestamps of the associated user interaction data. When the timestamps match or the timestamp difference is less than the effective association time interval, the environmental event data is associated with the corresponding user interaction data. All associated data records are then integrated into a market dataset. The effective association time interval is defined based on the timeliness of user behavior in response to environmental events in cultural and tourism consumption scenarios; the value is 1 hour. If it exceeds 1 hour, it will lead to an increase in other interfering factors and generate false associations. If it does not exceed 1 hour, it will lead to the omission of actual associations due to the short time interval.
[0029] S2. Define the state space, action space, and interaction rules based on the market dataset, and create a corresponding dynamic intelligent agent for each cultural and tourism consumption product; construct a multi-agent reinforcement learning framework, and input the dynamic intelligent agent into the multi-agent reinforcement learning framework to generate a virtual market simulation environment; S2.1. Define the state space based on the attribute dimensions of tourism consumption products in the market dataset, enumerate the executable marketing actions to define the action space, and define the interaction rules through state transition functions and reward functions; It should be noted that the attribute dimensions of cultural and tourism consumption products are extracted from the market dataset as constituent elements of the state space, such as continuous or discrete variables like product price, real-time popularity, and geographical location. These attribute dimensions are combined into a state vector representing the market state. All possible marketing actions for cultural and tourism consumption products are enumerated, such as price reductions, advertising intensity, and package combination strategies, forming the action space. Interaction rules are established using state transition functions and reward functions. The state transition function describes how the market state changes after executing a marketing action, while the reward function calculates immediate returns based on the state change and the cost of the marketing action. The expression for the state transition function is as follows: ; in, This represents the state vector at the next moment; Represents the state vector at the current moment; Represents the state change function; This indicates that all cultural and tourism consumption products are in A collection of marketing actions performed at all times; express Momentary environmental events; Indicates a specific moment in time; The expression for the reward function is, ; in, Indicates the first The reward value obtained from each cultural and tourism consumption product; This represents the popularity weight parameter, defined based on the commercial goal priority of cultural and tourism consumption products. It is usually set to 0.7. If it is higher than 0.7, it will lead to an excessive pursuit of popularity, which will increase costs. If it is lower than 0.7, it will lead to insufficient exposure of cultural and tourism consumption products and a decrease in user attention. Indicates the first Cultural and tourism consumption products Real-time popularity at any given moment; Indicates the first Cultural and tourism consumption products Real-time popularity at any given moment; Indicates the first The numerical values of marketing activities executed for each cultural tourism consumption product; This represents the action cost function.
[0030] S2.2 Create a corresponding intelligent agent for each cultural and tourism consumption product, and bind the strategy function and value function to generate a dynamic intelligent agent; It should be noted that the strategy function and value function are defined based on the state transition function and reward function; a corresponding intelligent agent is created for each cultural tourism consumption product, and a strategy function is assigned to each intelligent agent. The strategy function selects marketing actions from the action space based on the current state vector of the cultural tourism consumption product represented by the intelligent agent, and assigns a value function. The value function calculates the long-term expected return of the intelligent agent under the current state vector based on the reward function, generating a dynamic intelligent agent capable of autonomous decision-making and market state evaluation; the expression of the strategy function is as follows: ; in, In the state vector The probability of choosing a marketing action; Represents the strategy function; Numerical representation of marketing actions; The state vector is represented by the policy function, which is essentially a probability distribution mapping from the state space to the action space. It defines the probability distribution given the current state vector. Under these conditions, the agent selects the probability value of each possible marketing action; The expression for the value function is: ; in, In the state vector The long-term expected cumulative reward; The mathematical expectation operator refers to averaging over all possible sequences of state vectors and marketing action sequences. Representing infinity, it refers to the accumulation of all future rewards; This represents the discount factor, defined based on the agent's emphasis on future rewards, and its value range is... If the value is greater than 1, it will cause the value function to diverge to infinity, causing the value function to lose convergence. If it is less than 0, it will cause the long-term expected cumulative reward to alternate between positive and negative, making the value function meaningless. express The reward value obtained at time step (k) is calculated by the reward function. Represents the initial state vector; Represents the state vector Start calculating cumulative rewards; In This indicates the initial state, meaning the current parameter is in its initial state. The value function is essentially a function that maps a state vector to a real number. It calculates the expected value of the cumulative sum of all future rewards of an agent under a given state vector, after being weighted by a discount factor, and is used to evaluate the long-term return potential of a given state vector.
[0031] S2.3 Construct a multi-agent reinforcement learning framework through a centralized training framework and a distributed execution framework. Load the state space, action space, interaction rules, and dynamic agents into the multi-agent reinforcement learning framework to generate a virtual market simulation environment.
[0032] It should be noted that, by collecting agent experience data (including state vectors, executed marketing actions, and obtained reward values) from all dynamic agents through existing centralized processing capabilities, the optimization adjustment amounts for the policy function and value function are calculated using this agent experience data. The expression for calculating the optimization adjustment amounts is as follows: ; ; ; in, This represents the optimization adjustment amount of the policy function; This represents the optimization adjustment amount of the value function; TD error is used to measure the difference between the long-term expected cumulative reward of the current state vector and the long-term expected cumulative reward of the next state vector. The function parameters of the strategy function; This represents the function parameters of the policy function. The gradient; The function parameters represent the value function; Represents the value function with respect to function parameters The gradient; This represents the reward value obtained from cultural and tourism consumption products; Represents the next state vector Long-term expected cumulative rewards; The optimization adjustment is applied to the function parameters of the current policy function and value function of the dynamic agent to generate updated function parameters. The updated function parameters are then distributed synchronously to all dynamic agents to complete the construction of the centralized training framework. By binding a policy function and a value function to each dynamic agent, independent decision-making and execution capabilities are configured for each dynamic agent. This enables each dynamic agent to read its own state vector in real time from the virtual market simulation environment and input the state vector into the bound policy function. The policy function outputs the probability distribution of marketing actions based on the state vector. The dynamic agent selects specific marketing actions based on the probability distribution and executes the marketing actions autonomously. This establishes an independent execution function for each dynamic agent and completes the construction of a distributed execution framework. By connecting the parameter update function of the centralized training framework with the action execution function of the distributed execution framework through data flow, the parameter update output by the centralized training framework is directly applied to the dynamic agent in the distributed execution framework, thus completing the construction of the multi-agent reinforcement learning framework. The state space, action space, interaction rules, and dynamic agents are loaded into a multi-agent reinforcement learning framework. The current value of the market dataset (which refers to the latest attribute data of each cultural and tourism consumer product recorded in the market dataset at the data collection time point, such as the latest popularity value, the latest price, and the latest inventory) is set as the initial state of the environment to generate a virtual market simulation environment.
[0033] It should also be noted that existing technologies typically use statistical models or single-agent models to statically predict the popularity of cultural and tourism consumption products. These methods fail to depict the dynamic interactive behavior of multiple products in market competition, as well as the collaborative impact of environmental events on the overall market, resulting in predictions that do not accurately reflect the complex game process. This solution generates a virtual market simulation environment by constructing a multi-agent reinforcement learning framework. Each cultural and tourism consumption product is mapped to a dynamic agent with autonomous decision-making capabilities. Within this unified environment, a continuous game process based on state vectors, marketing actions, and interaction rules is simulated, achieving a microscopic reconstruction of market competition dynamics and solving the problem of multi-agent interaction modeling failure.
[0034] S3. In the virtual market simulation environment, drive each dynamic agent to select marketing actions based on the state attribute values in the current state space, calculate the state attribute values at the next moment based on the marketing actions and interaction rules executed by all dynamic agents, and generate a market state evolution sequence. S3.1. Create an empty-market state evolution sequence through list initialization operation; read the state attribute values of all dynamic agents at the current moment from the virtual market simulation environment, input the state attribute values at the current moment into the corresponding policy function, and generate the corresponding marketing action; It should be noted that the defined data sequence that can store multiple records is essentially a dynamically expandable ordered set used to record the state attribute values and marketing actions of all dynamic agents at each moment. The data sequence does not contain any data records in its initial state. The data sequence is configured to store the state attribute values and marketing actions of all dynamic agents at each moment in chronological order, thus completing the creation of the empty-market state evolution sequence. The system reads the state attribute values of all dynamic agents at the current moment from the virtual market simulation environment. These state attribute values include the popularity and price values of cultural and tourism consumer products. The system then inputs the current state attribute value of each dynamic agent into the policy function bound to the dynamic agent. The policy function outputs the marketing action with the highest probability based on the input state attribute value.
[0035] S3.2 Collect the marketing actions of each dynamic agent to form a marketing action record table; append the current state attribute value and the marketing action record table to the empty-market state evolution sequence; It should be noted that the marketing actions generated by each dynamic agent through the policy function are collected and organized into a marketing action record table according to the structure of product identifiers, marketing action types and marketing action parameters in the cultural and tourism consumption product attribute data. All dynamic agent state attribute values read from the virtual market simulation environment at the current moment are combined with the marketing action record table to form a complete record and appended to the empty-market state evolution sequence.
[0036] S3.3 Apply the state transition function to the current state attribute value and marketing action record table, calculate the state attribute value for the next moment, and load it into the virtual market simulation environment; iteratively execute the read, record, and append operations until the preset time step is reached, and output the market state evolution sequence.
[0037] It should be noted that a state transition function is applied to the state attribute values read from the virtual market simulation environment at the current moment and the marketing action record table to calculate the state attribute values at the next moment; the state attribute values at the next moment are then written into the virtual market simulation environment to update the state attribute values of the virtual market simulation environment. Repeat the following steps: Read the state attribute values of all dynamic agents at the current moment from the virtual market simulation environment; input the state attribute values into the corresponding policy function to generate marketing actions; collect the marketing actions to form a marketing action record table; append the state attribute values and the marketing action record table to the empty-market state evolution sequence; apply the state transition function to the current state attribute values and the marketing action record table to calculate the state attribute values at the next moment and write them into the virtual market simulation environment, until the time step is reached; the empty-market state evolution sequence after all append operations are completed is the final market state evolution sequence; The business objective cycle for predicting the popularity of cultural and tourism consumption products defines the time step, which is set to 7 days. The basis for this value is that 7 days is a common short-term business cycle for predicting the popularity of cultural and tourism consumption products. This can cover the complete impact cycle of short-term hot events such as holidays, while avoiding the increase in extrapolation and calculation due to an excessively long cycle.
[0038] S4. Calculate the popularity probability distribution of each cultural and tourism consumption product in the market state evolution sequence during the preset target prediction period, extract the popularity change events that trigger the preset popularity threshold and the corresponding marketing actions, and integrate the popularity probability distribution, popularity change events and corresponding marketing actions to generate preliminary evaluation results. S4.1. Traverse the market state evolution sequence and extract all popularity values within the preset target prediction period for each cultural and tourism consumption product to form a popularity data sequence. It should be noted that, by traversing all data entries in the market state evolution sequence, for each cultural and tourism consumption product, data entries within the target prediction period are selected based on the product identifier. The real-time popularity of the current cultural and tourism consumption product is extracted from the data entries, and all the extracted real-time popularity is arranged in chronological order to form the popularity data sequence of the current cultural and tourism consumption product within the target prediction period. Based on the monthly marketing review cycle commonly used in the cultural and tourism consumption industry, the target forecast period is defined as 30 days. The basis for this value is that the evaluation of the marketing activities of cultural and tourism consumption products is usually carried out on a monthly basis. 30 days can cover the complete marketing cycle and include consumer behavior data for at least 4 weekend periods. At the same time, it avoids insufficient data due to a cycle that is too short or too much market noise due to a cycle that is too long.
[0039] S4.2 Apply the kernel density estimation method to the heat data sequence to generate a heat probability distribution; It should be noted that the bandwidth parameter is determined based on the variance and sample size of the heat data sequence. This bandwidth parameter is then substituted into the kernel density estimation formula to calculate the probability density of each heat data point. The expression for the kernel density estimation formula is as follows: ; ; in, Indicates the target popularity data The probability density at a location refers to the target heat data. The relative probability of occurrence; This indicates the number of samples in the heat data sequence; This represents the bandwidth parameter, which controls the smoothness of the kernel density estimation. This represents the sample index in the heat data sequence, with a value ranging from 1 to... ; The kernel function is represented by the Gaussian kernel function in this scheme. This indicates the target heat data for which the probability density needs to be calculated. It is not included in the heat data sequence and is an arbitrarily specified target heat data. Indicates the number of times in the heat data sequence The real-time popularity of each sample is used to calculate the target popularity data. The probability of its occurrence; The standard deviation represents the sequence of heat data and is used to measure the dispersion of heat values. Represents an empirical constant; Indicates the convergence index; The probability density of the target heat data is calculated based on all the heat data in the heat data sequence. All the obtained probability densities are integrated to form a continuous function curve, which is the probability density distribution. Empirical constants are defined based on the optimal bandwidth selection rule of the Gaussian kernel function. The value is usually set to 1.06. The basis for this value is to minimize the mean square error of the kernel density estimation and provide the best smoothing effect when the data is approximately normally distributed. If it is greater than 1.06, it will lead to over-smoothing, thereby masking the true details and peaks of heat fluctuations. If it is less than 1.06, it will lead to insufficient smoothing, making the generated probability distribution unstable and unreliable. The convergence exponent is defined in the convergence theory based on kernel density estimation. The value is usually taken as -1 / 5. The value is based on the theoretical optimal rate at which the bandwidth parameter decays as the number of samples increases. If the absolute value is greater than 1 / 5, it will lead to a decrease in the bias of the kernel density estimation, but an increase in the variance, which will increase the difference in the probability density distribution. If the absolute value is less than 1 / 5, it will lead to a decrease in the variance of the kernel density estimation, but an increase in the bias, which will make the probability density distribution result less accurate.
[0040] S4.3 Calculate the change in popularity of each cultural and tourism consumption product, mark events where the change in popularity exceeds the preset popularity threshold as popularity change events, and record the popularity change events and corresponding marketing actions. It should be noted that, by traversing each consecutive time point in the market state evolution sequence, the difference between the popularity data of each cultural and tourism consumption product at the current time point and the popularity data at the previous adjacent time point is taken as the popularity change. The popularity change is compared with the popularity threshold. When the absolute value of the popularity change exceeds the popularity threshold, the current time point is marked as a popularity change event. The marketing action corresponding to the current time point is extracted from the market state evolution sequence. The value of each marked popularity change event, the popularity change, and the corresponding marketing action are collected and organized. The popularity threshold is defined based on the significance level of fluctuations in the popularity of cultural and tourism consumption products: the fluctuations in the popularity of cultural and tourism consumption products need to balance the sensitivity and specificity of event detection. The significance level represents the allowable false alarm probability, that is, the probability of incorrectly labeling noisy events. By analyzing the characteristics of cultural and tourism consumption scenarios and historical data, the range of significance level values is obtained, and the range of significance level values is directly defined as the range of popularity threshold values, which is 5% to 20%. If the popularity threshold is lower than 5%, it will be too sensitive and label too many small fluctuations and noisy events, resulting in redundant collection of popularity change events and difficulty in focusing on key changes. If the popularity threshold is higher than 20%, it will be too lenient and miss many medium-amplitude popularity change events with actual impact, reducing the comprehensiveness of the assessment.
[0041] S4.4. Integrate the data on the probability distribution of popularity, the events of popularity change, and the corresponding marketing actions to generate preliminary evaluation results.
[0042] It should be noted that the popularity probability distribution of each cultural and tourism consumption product is used as the basis for quantitative evaluation. All collected popularity change events, popularity change values, and corresponding marketing actions of the current cultural and tourism consumption products are used as event attribution elements and combined into a structured product evaluation record. The product evaluation records of all cultural and tourism consumption products are summarized to generate a preliminary evaluation result that includes the popularity probability distribution and product evaluation records. It should also be noted that existing technologies typically predict the popularity of cultural and tourism consumption products through statistical models or time series analysis, but can only output a single popularity point estimate or a simple probability distribution, failing to identify the specific triggering reasons behind popularity changes (such as which marketing actions caused popularity fluctuations), making it difficult for the evaluation results to support accurate marketing decisions. This solution quantifies prediction uncertainty by traversing the market state evolution sequence and calculating the popularity probability distribution of each product during the target prediction period. Simultaneously, it extracts popularity change events exceeding popularity thresholds and associates them with corresponding marketing actions, integrating the probability distribution with product evaluation records to generate preliminary evaluation results. This achieves a combination of probabilistic description of popularity prediction and causal tracing of popularity change events, improving the interpretability and decision support of evaluation results, and solving the problem of being unable to associate popularity changes with specific marketing actions.
[0043] S5. Associate and map the preliminary assessment results with the attribute data of cultural and tourism consumption products to generate a structured assessment message.
[0044] S5.1 Extract the evaluation data of each cultural and tourism consumption product from the preliminary evaluation results, and use product identifiers to associate and match the evaluation data with the attribute data of cultural and tourism consumption products to generate associated data records; It should be noted that the evaluation data for each cultural and tourism consumption product is extracted from the preliminary evaluation results. This evaluation data is the product evaluation record, which includes the popularity probability distribution and event attribution elements. The product identifier in the cultural and tourism consumption product attribute data is used as the association key to match the evaluation data with records in the cultural and tourism consumption product attribute data that have the same product identifier. The successfully matched evaluation data is merged with the corresponding cultural and tourism consumption product attribute data to generate a related data record that includes cultural and tourism consumption product attribute data, popularity probability distribution, and event attribution elements.
[0045] S5.2 Apply the structured data sequence method to the associated data records to generate a structured evaluation message.
[0046] It should be noted that, in order to combine the cultural and tourism consumption product attribute data, popularity probability distribution and event attribution elements contained in each associated data record according to the predefined key-value pair structure, and convert all associated data records into strings in JSON standard format, a structured evaluation message is generated. The key-value pair structure is based on the field names and data types of cultural and tourism consumption product attribute data, heat probability distribution, and event attribution elements, and predefined mapping rules for key names and value formats. Specifically, the key name "product_id" corresponds to the product identifier in the cultural and tourism consumption product attribute data; the key name "heat_distribution" corresponds to the probability density data of the heat probability distribution; and the key name "events" corresponds to the event attribution elements.
[0047] In summary, this invention uses a multi-agent reinforcement learning framework to simulate market dynamics, capturing the competition and synergy between different cultural and tourism consumption products, improving the accuracy and reliability of popularity prediction, and providing decision-making basis that is closer to the real market. By integrating popularity probability distribution and causal event analysis, it achieves quantitative assessment of popularity changes, traces the root causes of fluctuations and associates them with specific marketing actions, enhancing the interpretability and operability of the assessment results, and achieving the goal of accurately adjusting marketing strategies.
[0048] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence, characterized in that: include, Collect data on the attributes of cultural and tourism consumption products, user interaction data, and environmental event data, and perform data cleaning, standardization, and correlation integration to generate a market dataset; Define the state space, action space, and interaction rules based on the market dataset, and create a corresponding dynamic intelligent agent for each cultural and tourism consumption product; A multi-agent reinforcement learning framework is constructed, and dynamic agents are input into the multi-agent reinforcement learning framework to generate a virtual market simulation environment. In a virtual market simulation environment, each dynamic agent is driven to select a marketing action based on the state attribute value in the current state space. Based on the marketing actions and interaction rules executed by all dynamic agents, the state attribute value at the next moment is calculated, and a market state evolution sequence is generated. Calculate the probability distribution of popularity of each cultural and tourism consumption product in the market state evolution sequence during the preset target prediction period, extract the popularity change events that trigger the preset popularity threshold and the corresponding marketing actions, and integrate the popularity probability distribution, popularity change events and corresponding marketing actions to generate preliminary evaluation results. The preliminary assessment results are correlated and mapped with the attribute data of cultural and tourism consumption products to generate a structured assessment report.
2. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 1, characterized in that: The process involves collecting data on the attributes of cultural and tourism consumption products, user interaction data, and environmental event data, followed by data cleaning, standardization, and correlation integration to generate a market dataset. The specific steps are as follows: Collect data on the attributes of cultural and tourism consumption products, user interaction data, and environmental event data, and perform missing value filling and product identifier verification on the attribute data of cultural and tourism consumption products. Filter invalid sessions and unify timestamps in user interaction data; Environmental event data is mapped to a preset classification system through event type mapping; Max-min normalization and one-hot encoding are performed on the attribute data of cultural and tourism consumption products, user interaction data, and environmental event data, and all time data is converted into Unix timestamp format. After linking user interaction data with cultural and tourism consumption product attribute data through product identifiers, environmental event data is linked with the already associated user interaction data by matching timestamps to generate a market dataset.
3. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 2, characterized in that: The steps for defining the state space, action space, and interaction rules based on the market dataset, and creating a corresponding dynamic intelligent agent for each cultural tourism consumption product, are as follows: The state space is defined based on the attribute dimensions of Chinese tourism consumption products in the market dataset, the action space is defined by enumerating executable marketing actions, and the interaction rules are defined by state transition functions and reward functions. Create a corresponding intelligent agent for each cultural and tourism consumption product, and bind it with a strategy function and a value function to generate a dynamic intelligent agent.
4. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 3, characterized in that: The specific steps for constructing a multi-agent reinforcement learning framework and inputting dynamic agents into the framework to generate a virtual market simulation environment are as follows. A multi-agent reinforcement learning framework is constructed by using a centralized training framework and a distributed execution framework. By loading the state space, action space, interaction rules, and dynamic agents into a multi-agent reinforcement learning framework, a virtual market simulation environment is generated.
5. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 4, characterized in that: In the virtual market simulation environment, each dynamic agent is driven to select marketing actions based on the state attribute values in the current state space. The specific steps are as follows: Create an empty-market state evolution sequence using list initialization operations; Read the state attribute values of all dynamic agents in the virtual market simulation environment at the current moment, input the state attribute values at the current moment into the corresponding policy function, and generate the corresponding marketing actions.
6. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 5, characterized in that: The process of calculating the state attribute values for the next moment and generating a market state evolution sequence based on the marketing actions and interaction rules executed by all dynamic agents is as follows: Collect the marketing actions of each dynamic intelligent agent and form a marketing action record table; Add the current state attribute value and marketing action record table to the empty-market state evolution sequence; Apply the state transition function to the current state attribute value and marketing action record table, calculate the state attribute value for the next moment, and load it into the virtual market simulation environment; Iteratively execute read, record, and append operations until the preset number of time steps is reached, and output the market state evolution sequence.
7. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 6, characterized in that: The specific steps for calculating the probability distribution of popularity of each cultural and tourism consumption product in the market state evolution sequence during the preset target prediction period are as follows. By traversing the market state evolution sequence, for each cultural and tourism consumption product, all popularity values within the preset target prediction period are extracted to form a popularity data sequence. The kernel density estimation method is applied to the heat data sequence to generate a heat probability distribution.
8. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 7, characterized in that: The process involves extracting popularity change events that trigger preset popularity thresholds and their corresponding marketing actions, then fusing the popularity probability distribution, popularity change events, and corresponding marketing actions to generate preliminary evaluation results. The specific steps are as follows: Calculate the change in popularity for each cultural and tourism consumption product, mark events where the change in popularity exceeds a preset popularity threshold as popularity change events, and record popularity change events and corresponding marketing actions. By integrating the probability distribution of popularity, events that change popularity, and corresponding marketing actions, preliminary evaluation results are generated.
9. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 8, characterized in that: The steps for associating and mapping the preliminary assessment results with the attribute data of cultural and tourism consumption products to generate a structured assessment report are as follows: The evaluation data of each cultural and tourism consumption product is extracted from the preliminary evaluation results. The evaluation data is then matched with the attribute data of the cultural and tourism consumption product through product identifiers to generate associated data records. The structured data sequence method is applied to the associated data records to generate a structured evaluation message.
10. The method for predicting and evaluating the popularity of cultural and tourism consumption products based on artificial intelligence as described in claim 9, characterized in that: The structured data sequence method converts all related data records into strings in the JSON standard format and integrates the structure using predefined key-value pairs.