Market investigation data analysis method and system

Abstract

The invention relates to the technical field of market research data analysis, in particular to a market research data analysis method and system.The method comprises the steps that multi-source data is collected and cleaned through a multi-channel self-adaptive data integration technology, and a comprehensive data set is generated; constructing a graph network structure based on a graph neural network, updating node features through a message passing mechanism, generating market prediction data, and optimizing an investigation path based on a reinforcement learning algorithm; comparing the market feedback data with the prediction data through a feedback self-learning mechanism, and adjusting an investigation strategy to obtain an optimized investigation path; the performance of the market under extreme conditions is simulated through Monte Carlo simulation, the abnormal behaviors of the market are identified through dynamic outlier detection, a coping strategy is generated, and an investigation path is adjusted. The method effectively improves the accuracy of investigation data and the allocation efficiency of investigation resources, and adapts to a complex market environment.

Description

Technical Field

[0001] The present invention relates to the technical field of market research data analysis, and particularly to a market research data analysis method and system. Background Art

[0002] In market research, accurate research data analysis is crucial for enterprises to formulate market strategies. Traditional market research methods mainly rely on manual processing or limited automation tools and cannot cope with the complex and changeable data in the modern market. Current technologies usually based on simple data collection and statistical methods, which are difficult to process complex data from multiple channels and dimensions and are prone to distortion of research results. The prior art (Chinese invention patent, publication number: CN117541293B, title: A Market Research Data Analysis Method and System Based on Big Data) uses data templates and channel screening for data collection and analysis, but performs poorly in processing the integration and optimization of different data sources, resulting in the lack of accuracy and adaptability of research results.

[0003] The means adopted by the prior art include collecting and standardizing market research data through preset data templates, and its main problems are as follows:

[0004] In the existing methods, the credible value and energy efficiency value of the data source are fixed and cannot be adaptively adjusted according to real-time market changes, resulting in poor data accuracy; the prior art relies on static channel screening and statistical analysis and cannot dynamically adjust research resources; the processing speed of research feedback data in the existing methods is slow and cannot adjust research strategies in real time. Summary of the Invention

[0005] In view of the above-mentioned many problems existing in the prior art, the present invention provides a market research data analysis method and system. The present invention establishes an efficient and intelligent market research data analysis system through multi-channel adaptive data integration, graph neural network and reinforcement learning algorithm. The present invention can dynamically integrate multi-source data, generate accurate market predictions, and optimize the research path through a real-time feedback self-learning mechanism, making the allocation of research resources more reasonable and the research results more accurate and reliable.

[0006] A market research data analysis method includes the following steps:

[0007] Collect market research data, environmental data and social behavior data from multiple data sources through multi-channel adaptive data integration technology, and clean and standardize the collected data to generate a comprehensive data set;

[0008] Based on the graph neural network, a graph network structure is constructed by inputting a comprehensive dataset. The message passing mechanism of the graph neural network is used to update the node features in the graph network, generate market prediction data, and dynamically generate and optimize the research path based on the market prediction data and the reinforcement learning algorithm to obtain the first optimized path data;

[0009] Based on the feedback self-learning mechanism, the market feedback data is compared with the market prediction data, the deviation of the research execution is calculated to generate a feedback matrix, and the feedback matrix is learned and optimized through a recurrent neural network to generate an optimized research strategy, and the research path is further adjusted based on the optimized research strategy to obtain the second optimized path data;

[0010] Based on the Monte Carlo simulation method and the graph neural network, the performance of the market under extreme conditions is simulated to generate market simulation data, and the dynamic outlier detection algorithm is used to identify abnormal behaviors in the market to generate anomaly detection data. A coping strategy is generated based on the anomaly detection data, and the research path is dynamically adjusted to obtain the third optimized path data.

[0011] Preferably, the multi-channel adaptive data integration technology adaptively adjusts the weights of each data source of market research data, environmental data, and social behavior data through a dynamic Bayesian network, and the weight allocation is automatically optimized based on the historical performance and data quality of each data source.

[0012] Preferably, the message passing mechanism of the graph neural network updates the node features in the graph network, and the node feature update formula is:

[0013]

[0014] Among them, represents the eigenvalue of node v in the (k + 1)-th iteration; N(v) represents the set of neighbor nodes of node v; W k represents the weight matrix in the k-th iteration; σ represents the activation function used for non-linear mapping of the updated node eigenvalue; represents the eigenvalue of neighbor node u in the k-th iteration; N(u) represents the set of neighbor nodes of node u.

[0015] Preferably, the optimization of the research path of the graph neural network is realized through the reinforcement learning algorithm. The generation of the research path is based on the value function and the reward mechanism. The value function is dynamically adjusted according to the accuracy of the research data and the efficiency of the use of research resources, and the generated first optimized path data is used to guide the allocation of research resources and path selection.

[0016] Preferably, the recurrent neural network is used to learn and optimize the market feedback data and the market prediction data. The hidden state update formula of the recurrent neural network is:

[0017] h t = σ(W x x t + W h h t-1 + b h )

[0018] where h t represents the hidden state at time step t; x t represents the current input data; W x represents the weight matrix of the input data; W h represents the weight matrix of the hidden state; h t-1 represents the hidden state at time step t - 1; b h represents the bias term; σ represents the activation function used for non - linear mapping of the hidden state.

[0019] Preferably, the feedback matrix includes multiple - dimensional information of market feedback data. The information of each dimension includes market performance, market region, and consumer behavior. The feedback matrix is used for weighted processing during the optimization of the research strategy to adjust the priority of the research strategy. The generated second - optimized path data is used for dynamically adjusting the research strategy and further optimizing the research path level.

[0020] Preferably, the Monte Carlo simulation method generates market prediction results under multiple extreme conditions through multiple random samplings. The simulation conditions include market demand fluctuations, price fluctuations, and supply chain stability. The simulation results are used to adjust the research path and the allocation of research resources.

[0021] Preferably, the dynamic outlier detection algorithm identifies abnormal behaviors by analyzing the density distribution of market simulation data. The abnormal behaviors include abnormal demand fluctuations, supply chain interruptions, and sharp price fluctuations.

[0022] Preferably, the abnormal behavior data identified by the dynamic outlier detection algorithm is used to dynamically adjust the research path. The system automatically re - allocates research resources according to the type of the abnormal behavior, generating third - optimized path data. The third - optimized path data is used for optimizing the research path under extreme market conditions.

[0023] A system for implementing the market research data analysis method, the system includes:

[0024] A data acquisition module, which is used to collect market research data, environmental data, and social behavior data from multiple data sources through multi - channel adaptive data integration technology, and clean and standardize the collected data to generate a comprehensive data set;

[0025] A graph neural network module is used to receive the comprehensive dataset, construct a graph network structure, update the node features in the graph network using the message passing mechanism of the graph neural network, generate market prediction data, and dynamically generate and optimize the research path based on the market prediction data and the reinforcement learning algorithm to obtain the first optimized path data;

[0026] A feedback self-learning module is used to compare the market feedback data with the market prediction data, calculate the deviation of the research execution, generate a feedback matrix, learn and optimize the feedback matrix through a recurrent neural network, generate an optimized research strategy, and further adjust the research path based on the optimized research strategy to obtain the second optimized path data;

[0027] A simulation and anomaly detection module is used to simulate the performance of the market under extreme conditions based on the Monte Carlo simulation method and the graph neural network, generate market simulation data, identify abnormal behaviors in the market using a dynamic outlier detection algorithm, generate anomaly detection data, generate a coping strategy based on the anomaly detection data, and dynamically adjust the research path to obtain the third optimized path data.

[0028] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0029] The present invention realizes the dynamic adjustment of the weights of different data sources through an adaptive data integration technology, ensuring the accuracy and credibility of the research data;

[0030] The present invention optimizes the research path through the message passing mechanism of the graph neural network, generates a more adaptable research strategy, and realizes the efficient allocation of research resources;

[0031] The present invention can quickly respond to market feedback data and dynamically adjust the research path through a feedback self-learning mechanism, improving the efficiency of research execution;

[0032] The present invention simulates the performance of the market under extreme conditions through the Monte Carlo simulation method, generates an optimized research path strategy under extreme market scenarios, and improves the system's ability to cope with market fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow schematic diagram of the method of the present invention;

[0034] Figure 2 It is a flow schematic diagram of data collection and integration in the present invention;

[0035] Figure 3 It is a flow schematic diagram of path optimization of the graph neural network in the present invention;

[0036] Figure 4 It is a flow schematic diagram of feedback self-learning in the present invention;

[0037] Figure 5 This is a schematic diagram of the abnormal detection and path adjustment process in the present invention;

[0038] Figure 6 This is a structural block diagram of the system of the present invention. Detailed implementation manners

[0039] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0040] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising" and the like used herein indicate the presence of the described features, steps, operations and / or components, but do not exclude the presence or addition of one or more other features, steps, operations or components.

[0041] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0042] As Figure 1 shown, a method for analyzing market research data includes the following steps:

[0043] Collect market research data, environmental data, and social behavior data from multiple data sources through multi-channel adaptive data integration technology, and clean and standardize the collected data to generate a comprehensive data set;

[0044] Preferably, as Figure 2 shown, the multi-channel adaptive data integration technology adaptively adjusts the weights of each data source of market research data, environmental data, and social behavior data through a dynamic Bayesian network, and the weight allocation is automatically optimized based on the historical performance and data quality of each data source.

[0045] The multi-channel adaptive data integration technology first collects data from different sources (including market research data, environmental data, and social behavior data). These data may vary greatly in characteristics, formats, precision, and frequency. Therefore, the first step in data integration is to clean and standardize these data through a preprocessing module. The cleaning process mainly removes noise data, redundant data, and abnormal data to ensure the validity and consistency of the data. The standardization process converts data from different sources into a unified format for subsequent unified processing and analysis.

[0046] In the present invention, the dynamic Bayesian network is the core technology used for adaptive adjustment of each data source. The dynamic Bayesian network is a structured network based on a probability model. It can model time series data and deduce the importance and weight allocation of the current data source through historical data performance. The dynamic Bayesian network captures the mutual relationships between data sources and the performance of each data source at different times, and adjusts their weights accordingly. Specifically, market research data may be more important during certain time periods, while in other time periods, environmental data or social behavior data may have a greater impact on market prediction. Therefore, the dynamic Bayesian network dynamically adjusts the weight of each data source by continuously updating the conditional probability of the data source.

[0047] For example, assume that the system is processing the market feedback data of a certain brand's product. During a specific time period, market research data may provide key consumer purchase trends, while in another time period, environmental factors (such as weather and seasonal changes) may become the dominant factors. At this time, the dynamic Bayesian network will automatically calculate the impact of each data source on the prediction result based on historical records and the current situation, and adjust the weight accordingly. For example, during seasons when weather has an important impact on consumer behavior, the weight of environmental data may increase significantly to ensure that subsequent analysis can fully consider the impact of environmental changes.

[0048] Throughout the process, the dynamic Bayesian network not only considers the data performance at the current moment but also combines historical performance for weight optimization. For example, some data sources may often generate noise during past data processing, resulting in a large deviation in the prediction result. The system will identify this phenomenon through the Bayesian network and gradually reduce the weight of this data source, thereby improving the reliability and accuracy of the overall data.

[0049] The dynamic Bayesian network in the present invention regularly evaluates each data source. The data of each data source is represented by nodes in the network structure, and the edges between nodes represent the relevance between data sources. Through the update of the time series, the probability distribution in the dynamic Bayesian network changes in real time, thereby generating the optimal weight configuration at the current moment.

[0050] To achieve this process, the system first establishes an initial Bayesian network structure for each data source and trains it based on historical data. During the training process, the system analyzes the contribution degrees of different data sources to the results at different times, and learns the optimal network structure and parameters by maximizing the posterior probability. As data flows in real time, the network will be dynamically adjusted, and the weight allocation will be gradually optimized. The weight adjustment process is implemented through the Bayesian update formula, and the core is to utilize the conditional probability formula:

[0051]

[0052] Among them, P(X|Y) represents the probability distribution of data source X given data source Y. This formula ensures that the weights of each data source in the integrated dataset are dynamically changing and make optimal decisions based on the current situation.

[0053] Through the above mechanism, the system can effectively integrate and process data from different sources, maximizing the utilization rate of data. Through the adaptive adjustment of the dynamic Bayesian network, the system can automatically optimize the weights of data sources according to the current market conditions, environmental changes, and social behaviors, thus ensuring that the finally generated integrated dataset is more representative and accurate.

[0054] For example, in market research, the survey data for a certain product, the user discussions in the social network, and the external data related to the environment (such as economic changes, policy changes, etc.) may all affect the market performance of the product simultaneously. The multi-channel adaptive data integration technology can dynamically adjust the importance of data sources according to the changes of these factors through the dynamic Bayesian network. For example, if the user discussions in the social network suddenly increase and the content is related to the product, the weight of this part of the data will automatically increase, and the integrated dataset generated by the system subsequently will focus more on the social data, ensuring that the subsequent market prediction and research analysis are more in line with the actual situation.

[0055] As Figure 3 shown, based on the graph neural network, a graph network structure is constructed by inputting the integrated dataset, the node features in the graph network are updated using the message passing mechanism of the graph neural network, market prediction data is generated, and based on the market prediction data and the reinforcement learning algorithm, the research path is dynamically generated and optimized to obtain the first optimized path data;

[0056] Graph Neural Network (GNN) is a type of neural network used to process graph-structured data, and its advantage lies in being able to capture the complex relationships between nodes on the graph structure. In the present invention, the input of the graph neural network is a comprehensive dataset processed by multi-channel adaptive data integration technology, which includes various data sources such as market research data, environmental data, and social behavior data. Different types of data will be characterized through the nodes and edges of the graph neural network. Specifically, each data source or data point can be regarded as a node in the graph, and the relationships between nodes are represented by edges, which can represent the correlations between different data sources, the interactions between market variables, etc. The task of the graph neural network is to learn the complex relationships between these nodes and edges through the graph structure, so as to generate more accurate market prediction data.

[0057] The message passing mechanism of the graph neural network is its core operation process. In this process, nodes will update information through the connections between their neighbor nodes. The features of each node not only depend on its own data, but also are updated based on the data of the nodes connected to it. This mechanism can capture the interactions between market variables. For example, there may be a certain correlation between the market research data node and the social behavior data node. For instance, when the social behavior of consumers changes, the market performance will also fluctuate accordingly. In the graph neural network, the market research data node will update its features according to the information of its neighbor nodes (such as the social behavior data node) to generate comprehensive features reflecting the overall market trend.

[0058] Through multiple rounds of message passing, the features of all nodes will gradually integrate the information from their neighbor nodes to generate a global market feature representation.

[0059] In practical applications, the generated market prediction data is used to predict the performance of the market in the future for a period of time, such as sales trends, changes in consumer behavior, etc. This data is not only based on historical market data, but also integrates environmental and social behavior data, so as to be able to generate more accurate and comprehensive market predictions.

[0060] After generating the market prediction data, the solution continues to use the reinforcement learning algorithm to dynamically optimize the research path. The core of reinforcement learning is to continuously explore different research paths and give feedback according to the performance of each path, so as to find the optimal path. In the present invention, the reward mechanism of reinforcement learning is mainly based on the effectiveness of the research path, data accuracy, and the utilization efficiency of research resources. The optimization of the path is achieved through the Deep Q-Learning algorithm. The system will continuously evaluate the market prediction accuracy brought by different paths and their impact on the overall research strategy, and gradually adjust the path selection to generate the first optimized path data.

[0061] Suppose in a market research project, the system needs to evaluate the sales performance of a certain product in different regions. First, market research data, environmental data (such as economic conditions, weather changes, etc.) and social behavior data (such as discussions among consumers on social media) are input into the graph neural network. Nodes in the network may represent market data of different regions, and the edges between nodes may represent the market correlation or economic connection between these regions. Through multiple rounds of message passing, the system can capture the market linkage between different regions and the potential impact of the environment and social behavior on the sales performance. For example, if the weather condition in a certain region suddenly changes and there are a large number of relevant discussions on social media, the graph neural network will transmit this information to adjacent nodes (such as other regions that are closely related to the economy or market of this region) to generate an overall market prediction.

[0062] After the market prediction data is generated, the reinforcement learning algorithm will dynamically generate the research path. By evaluating the market prediction accuracy of each region, the system can find out which regions may have market anomalies or require more research resources. Therefore, the system will adjust the research path, concentrating more resources on regions with lower prediction accuracy or greater market volatility, thereby generating optimized research path data. This process is repeated until the optimal research path is found, which is the first optimized path data generated.

[0063] Through this combination of graph neural network and reinforcement learning, the system can capture the complex relationships in market data, generate more accurate market predictions, and continuously optimize the research path through reinforcement learning. Compared with traditional linear models or rule-driven research methods, the advantage of the present invention is that it can dynamically adapt to market changes and automatically adjust the research strategy. This means that in a complex and ever-changing market environment, the system can respond in real time to new market data, environmental data and social behavior data, and continuously optimize the research path according to these data. Ultimately, the accuracy of the research results and the resource utilization efficiency will be significantly improved.

[0064] Preferably, the message passing mechanism of the graph neural network updates the node features in the graph network, and the node feature update formula is:

[0065]

[0066] where represents the feature value of node v in the (k + 1)-th iteration; N(v) represents the set of neighbor nodes of node v; W k represents the weight matrix in the k-th iteration; σ represents the activation function, which is used to perform a non-linear mapping on the updated node feature value; represents the feature value of neighbor node u in the k-th iteration; N(u) represents the set of neighbor nodes of node u.

[0067] The message passing mechanism of the graph neural network is used to update the feature values of nodes in the graph network, so as to capture the complex relationships among market, environmental, and social data through deep learning of market research data, and achieve accurate prediction of market dynamics. This mechanism passes and fuses information among nodes in an iterative manner to generate node features that reflect the overall market trend, and finally provides data support for the optimization of the market research path.

[0068] The core of the message passing mechanism of the graph neural network (GNN) is that nodes not only rely on their own information but also continuously update their features through information exchange with their neighbor nodes (i.e., the nodes connected to them by edges in the graph). This feature enables GNN to capture complex relationships among data well and is particularly suitable for processing multi-dimensional data and variables in market research. In the present invention, each node can represent a certain type of market research data (such as sales data in a specific region, user behavior data, etc.), and the edges between nodes represent the relevance between these data (for example, the mutual influence between market trends in two different regions, or the common influence of a certain environmental factor on different markets).

[0069] The message passing process is completed through multiple rounds of iteration. In each round of iteration, each node receives information from its neighbor nodes and updates it in combination with its own features.

[0070] This formula shows that the feature update of node v depends not only on its own information but also on the feature values of its directly connected neighbor nodes u. At the same time, in the process of feature information transmission, the scale of node connections is considered, and a normalization factor is used to balance the influence among nodes of different scales.

[0071] In the application of the present invention, the goal of this message passing mechanism is to enable each node, through multiple rounds of iteration, not only to reflect the information of its direct data source but also to integrate the data features of its neighbor nodes, thereby forming a more global market prediction data. For example, in the market research of a region, a node can represent the sales data of this region, and neighbor nodes can represent other regions that are interdependent with this region or relevant environmental data (such as the influence of weather or holidays). With each round of iteration, these relevant data are gradually integrated into the feature values of each node, so that the finally generated market prediction data can reflect more complex market dynamics.

[0072] The message passing mechanism of the graph neural network first needs to construct a graph structure based on market research data. The nodes in this graph structure represent data from different sources, such as market research results, environmental data, social behavior data, etc. Next, the system will create connections for each node according to the relationships between these nodes (such as the influence between markets, the driving factors of consumer behavior on market performance, etc.), that is, define the edges in the graph.

[0073] The initial features of each node are initialized according to the corresponding data. For example, a certain node may represent the sales volume of a market area or the discussion heat of related products on social media. During the message passing process, the node will exchange information with its neighbor nodes through its edges. After each round of iteration, the feature value of the node will be updated, incorporating the feature information of its neighbor nodes. For example, the market performance data of a certain region will gradually integrate the market data of the surrounding regions, as well as other relevant environmental and social data during the iteration process, to generate a more comprehensive market prediction.

[0074] Through several rounds of message passing iterations, the graph neural network can generate the final features of each node, that is, market prediction data. This data not only includes the market research data itself, but also incorporates data from other nodes, such as the social behavior of consumers or changes in the environment, which makes the prediction results more accurate. For example, the discussion heat of a certain brand on social media may be very high in some regions. Through the message passing mechanism, this social data will gradually affect the market nodes in related regions, making the finally generated prediction data more in line with the actual situation.

[0075] The effect of this message passing mechanism is that it can fully mine and integrate information from multiple data sources, especially in a complex market environment. Through the graph neural network, each market data point (node) can not only analyze using its own data, but also obtain information from other data sources through its neighbor nodes, which greatly enhances the prediction ability of the system.

[0076] In practical applications, this means that the system can predict market trends more accurately. For example, if the sales data in a certain region suddenly shows an anomaly, through the message passing with its neighbor nodes, this anomaly information will quickly spread to the nodes in other regions, enabling the entire system to quickly adjust its market prediction model and thus make a more accurate market prediction. This feature makes the present invention have strong robustness and adaptability when dealing with multi-source data.

[0077] In addition, by using the activation function σ for non - linear mapping, the system can capture more complex market patterns and trend changes. The role of the activation function in feature update is to perform non - linear processing on node features, enabling the system to identify more complex non - linear relationships, such as the potential impact of sudden changes in consumer behavior patterns on market performance.

[0078] Suppose the system is analyzing the sales performance of a certain emerging product in the market. Node v represents the sales data of the product in a certain city, while neighbor node u represents the social media discussion heat of consumers in that city. In the initial state, the feature value h of node v v only contains the sales data of that city, while the feature value h of node u u contains social data. After one round of message passing, node v will update its features by combining the social data of neighbor node u. This means that the sales prediction for that city will be based not only on the current sales data but also on the social discussion heat. Through multiple iterations, the system will gradually integrate market data, environmental factors, etc. of surrounding cities to form a more comprehensive sales prediction.

[0079] Preferably, the optimization of the research path of the graph neural network is achieved through a reinforcement learning algorithm. The generation of the research path is based on a value function and a reward mechanism. The value function is dynamically adjusted according to the accuracy of research data and the efficiency of research resource utilization. Among them, the generated first - optimized path data is used to guide the allocation of research resources and path selection.

[0080] Reinforcement learning is a learning method that obtains experience by interacting with the environment and optimizes decisions based on experience. In the present invention, the goal of reinforcement learning is to optimize the research path. That is, during the market research process, the system needs to decide which market areas need further research and which resources should be allocated to specific areas according to the current market prediction and historical feedback. The value function and the reward mechanism are two key components of reinforcement learning.

[0081] During the optimization process of the research path, the value function is used to evaluate the "value" of each research path. Specifically, the value function reflects the long - term benefits that the system can obtain when performing research tasks in the future on the current path. This function not only considers the accuracy of research data but also takes into account the resource utilization efficiency. For example, if a certain path can obtain accurate market data with less resource investment, then the value of this path will be higher.

[0082] The reward mechanism is used to provide feedback to the system based on the execution effect of each research path, helping the system continuously adjust path selection. The design of rewards is usually based on the accuracy of research data and the efficiency of using research resources. For example, if a certain research path successfully validates the market prediction and performs well in terms of resource consumption, the system will provide a higher reward for this path, prompting the system to choose a similar path in future similar situations. Conversely, if a path fails to effectively collect market information or consumes excessive resources, the system will reduce the reward for that path.

[0083] Through multiple rounds of research path selection and execution, the system will adjust the research strategy according to the reward mechanism after each round of actions. As the number of research times increases, the system gradually learns the optimal research path through the reinforcement learning algorithm, that is, it can obtain the most accurate market research data with the least resource consumption. The finally generated first optimized path data is the optimal research path obtained by the system according to the reinforcement learning process.

[0084] In specific implementation, the system first determines potential research regions based on market prediction data. Each region can be regarded as a state in reinforcement learning, and the system needs to select the next research path from these states. The selection is based on the value function, which calculates the research value of each region in real time and is dynamically adjusted in combination with the feedback of historical data and the current resource allocation situation.

[0085] Deep Q-Learning, a commonly used algorithm in reinforcement learning, is the core technology for path optimization. Q-Learning measures the expected return after taking a certain action (such as selecting a certain research region) by recording the "Q value" of each state-action pair. Initially, the system does not have enough experience, and the path selection may be somewhat random. As the number of research times increases, the system gradually accumulates experience, the Q values are continuously updated, and the research paths gradually tend to be optimal.

[0086] For example, assume that the system conducts market research in multiple regions and has generated preliminary market prediction data. The system needs to decide which regions should be further investigated to verify or optimize these predictions. For each region, the system calculates the Q value of each region based on its market performance (such as sales volume, consumer feedback, etc.) and resource availability. A region with a higher Q value means that the research in this region may bring higher returns (that is, more accurate market prediction data), and the investment of research resources can be effectively utilized.

[0087] After each research execution, the system assigns rewards to each path based on the research results. For example, if the research results in a certain region greatly improve the market prediction and consume fewer resources, the system will assign a higher reward to that path. This reward value will be fed back to the Q-learning algorithm to update the Q-value of that path. As the research tasks are gradually completed, the system continuously adjusts and optimizes the research path through this reward mechanism.

[0088] The first optimized path data is generated through this reinforcement learning process. This data represents the path that the system believes should be preferentially selected in future research tasks under the current conditions. This path can not only ensure the accuracy of the research data but also maximize the savings in resource investment.

[0089] By combining graph neural networks with reinforcement learning algorithms, the system can achieve dynamic optimization of the research path and has the ability to efficiently allocate research resources. Compared with traditional static research path planning methods, the reinforcement learning in the present invention can continuously adjust the research strategy according to real-time market feedback and historical experience, thus greatly improving the flexibility and effectiveness of the research.

[0090] In the process of market research, resources are limited and the market environment is complex and changeable. Therefore, finding the optimal research path is of great significance. For example, in a rapidly changing market, the behavior and preferences of consumers may change at any time. Traditional research path planning methods may not be able to adjust the research strategy in a timely manner, resulting in resource waste or inaccurate data. Through reinforcement learning, the system can dynamically adjust the path selection according to the changing market feedback to ensure that the research resources are used where they are most needed.

[0091] Suppose a company plans to launch a new product and hopes to obtain feedback from consumers in different regions through market research to optimize the product strategy. The system first generates a preliminary market prediction based on the existing data and uses graph neural networks to analyze the market data in different regions. Next, the reinforcement learning algorithm will generate an initial research path based on these predictions, such as selecting certain regions for detailed research.

[0092] After the research is executed, the system evaluates the performance of each path based on the research results. If the research feedback in a certain region significantly improves the product strategy and consumes fewer resources, that path will receive a higher reward. Subsequently, the system will update the Q-value and be more inclined to select a similar path in the next research. Through multiple iterations, the system will finally generate the optimal research path, that is, the first optimized path data. This path ensures that the company obtains the most representative market feedback with reasonable resource investment.

[0093] Such as Figure 4As shown, based on the feedback self-learning mechanism, the market feedback data is compared with the market prediction data, the deviation of the research execution is calculated, a feedback matrix is generated, and the feedback matrix is learned and optimized through a recurrent neural network to generate an optimized research strategy, and the research path is further adjusted based on the optimized research strategy to obtain the second optimized path data;

[0094] The basis of the feedback self-learning mechanism lies in identifying and calculating the deviation in the research execution by comparing the actual market feedback data with the market prediction data. The market prediction data is generated using a graph neural network and a reinforcement learning algorithm in the early stage, which provides a prediction of future market trends; while the market feedback data comes from the real data obtained during the actual research process. The comparison of the two can reveal the gap or deviation between the prediction and the actual market behavior.

[0095] The deviation of the research execution represents the error between the market prediction and the actual situation. For example, the system may predict a significant increase in the demand for a product in a certain region, but the actual market feedback shows that the demand in this region has not changed significantly. In this case, the system can identify the error in the prediction through the calculation of the deviation and use it for the next optimization.

[0096] To better process these deviation data, the system generates a feedback matrix. The feedback matrix records the deviation information of different market regions, data sources or research indicators. Each row of the matrix can represent a certain market region or research indicator, and each column can represent different research cycles or time points. Through the feedback matrix, the system can effectively represent the deviation situation of each market region or research object at different time points.

[0097] Next, the system uses a recurrent neural network (RNN) to process the feedback matrix. The recurrent neural network is particularly suitable for processing sequential data and time-related data because it can pass the information of the previous moment to the current moment through the hidden layer to capture the time dependence in the data. In the present invention, the RNN gradually optimizes the research strategy of the system by learning the deviation between the historical market feedback and the market prediction.

[0098] The system further adjusts the research path based on the optimized research strategy to obtain the second optimized path data. This path data is not only based on the market prediction, but also combines the historical market feedback data and deviation information, making the path optimization process more accurate and flexible.

[0099] In a specific implementation, first, the system regularly collects market feedback data, which may include actual product sales, consumer feedback, market environment changes, etc. The system compares this data with previous market forecasts and calculates the deviation for each market region or research metric. For example, if there is a 10% deviation between the actual product sales volume in a certain region and the sales volume predicted by the system, this information will be recorded in the feedback matrix.

[0100] After the feedback matrix is generated, the system uses an RNN to learn and optimize it. Through its recursive structure, the RNN deeply processes the time series information in the feedback matrix (such as the change of market feedback over time), gradually updates its hidden state, and then generates an optimized research strategy.

[0101] After generating the optimized strategy, the system further adjusts the research path. For example, if the market feedback in some regions indicates a large deviation in the research forecast, the system will increase its attention to these regions in future research tasks. Conversely, if the market feedback in some regions is more consistent with the prediction, the system will correspondingly reduce the resource allocation to these regions, thereby improving the overall resource utilization efficiency of the research.

[0102] Through the feedback self - learning mechanism, the system can continuously adjust and optimize the research strategy during the research process, ensuring that the research path can dynamically adapt to market changes. Different from traditional static research path planning methods, the present invention can continuously optimize the research path through real - time market feedback, making the research results more accurate and flexible.

[0103] The greatest advantage of the feedback self - learning mechanism is that it can promptly capture the deviations in research execution and use these deviations to guide future research strategies. For example, if the market feedback data in a certain region indicates a large prediction error, the system can quickly adjust the path and allocate more research resources to this region, thereby ensuring more accurate future research data. This dynamic adjustment enables the system to maintain a high degree of adaptability in a complex and changing market environment.

[0104] Suppose a company plans to conduct a market research on a product. The initial market forecast shows that the demand for this product will increase significantly in regions A and B. However, the actual research feedback finds that the sales volume in region A meets the prediction, while the sales volume in region B is much lower than the prediction. In this case, the system calculates the prediction deviation for region B by comparing the market feedback data with the prediction data and records it in the feedback matrix.

[0105] Next, the system uses an RNN to learn and optimize the feedback matrix, generating an optimized research strategy. This strategy indicates that the investment in research resources for Region A should be reduced because the market performance in this region is in line with the prediction, while the resource allocation for Region B should be increased to further investigate the reasons for the decline in demand in this region. The system adjusts the research path according to this optimized strategy, generating second-optimized path data to ensure that future research tasks can more effectively address current market problems.

[0106] Preferably, the recurrent neural network is used to learn and optimize the market feedback data and market prediction data. The update formula for the hidden state of the recurrent neural network is:

[0107] h t =σ(W x x t +W h h t-1 +b h )

[0108] Where h t represents the hidden state at time step t; x t represents the current input data; W x represents the weight matrix of the input data; W h represents the weight matrix of the hidden state; h t-1 represents the hidden state at time step t - 1; b h represents the bias term; σ represents the activation function used for non-linearly mapping the hidden state.

[0109] The basic principle of the recurrent neural network is to use the hidden state to memorize and update the historical information in the time series. At each time step, the RNN not only processes the input data at the current time step but also combines the hidden state of the previous time step. Through this recursive mechanism, the network can "remember" the information at previous time points and continuously optimize the prediction and analysis of the current data through multiple rounds of iteration.

[0110] In the present invention, the input data of the RNN includes market feedback data and market prediction data. Each input of market feedback data may reflect the actual current market performance, while the market prediction data is the expected result based on previous analysis. By inputting these two into the recurrent neural network, the RNN can compare them, generate research deviation data, and gradually update its internal hidden state. The update formula for the hidden state is as shown above.

[0111] Through this recursive structure, RNNs can handle and learn complex dependencies in time series. Each input not only affects the output at the current time step but also influences the data analysis at subsequent time steps. This temporal dependence makes RNNs well-suited for optimizing market research strategies.

[0112] In practical applications, the system first collects market feedback data, which includes the actual market performance obtained during the research execution, such as product sales volume, user feedback, or changes in market demand. Meanwhile, the system also generates preliminary market prediction data, which is usually based on historical data, market trends, and external influencing factors. In each research cycle, the system inputs the market feedback data and prediction data of the current cycle into the recurrent neural network.

[0113] Based on the input feedback and prediction data, the RNN continuously updates its understanding of market changes through its hidden state. Through multiple rounds of recursive iteration, the network can gradually adjust the research strategy and optimize the research path. At each time step, the RNN combines the current feedback data with historical prediction data and adjusts the future research path by learning historical biases.

[0114] The comparison between the feedback data and the prediction data is crucial in this process. The system uses the RNN to capture the errors or differences between the prediction and the feedback and utilizes these differences to adjust future research strategies. For example, assume that at a certain time point, the market feedback shows that the product demand in a certain region is much lower than the prediction, indicating a large prediction bias. The RNN can gradually optimize the market prediction for this region by learning from historical feedback and adjust the research path accordingly, such as allocating more resources to this region for further investigation of the reasons or adjusting the next research plan to avoid similar biases.

[0115] Through the recurrent neural network, the system can continuously optimize the research strategy and dynamically adjust the research path during the research execution. Different from traditional static research methods, the RNN can optimize the research plan in real time according to the feedback data, making the research path gradually approach the optimal during the execution process.

[0116] The RNN can remember the relationship between historical market feedback and prediction data, providing long-term optimization of the research strategy. This long-term dependence enables the research path to be continuously adjusted according to historical biases, avoiding the recurrence of similar prediction biases.

[0117] Through recursive learning, the RNN can optimize the research strategy in real time according to the feedback information during the research execution, ensuring the optimal allocation of resources. For example, if the market feedback in some regions indicates low efficiency in the use of research resources, the system can gradually reduce the investment in research resources in these regions through historical learning, thus allocating more resources to regions with better feedback.

[0118] Through the activation function σ, the system can capture the complex non-linear relationship between market feedback and prediction. For example, some unforeseen market changes may not be linear, but are affected by the cross-influence of multiple factors. Through its recursive structure and non-linear mapping ability, the RNN can better adapt to this complex market dynamic change.

[0119] Suppose a company is promoting a new product in the market. Through previous market predictions, the system believes that the product demand in a certain region will increase, so it allocates more research resources. However, the actual market feedback shows that the sales volume in this region does not reach the expected level, indicating a deviation between the predicted data and the actual feedback. Through the recurrent neural network, the system will compare the feedback data with the predicted data in each research cycle to generate deviation information. The RNN adjusts the research strategy for this region step by step by recursively learning this deviation information.

[0120] After multiple rounds of research and feedback comparison, the system may find that the demand in this region does not continue to increase, but is greatly affected by seasons or external economic conditions. Therefore, the system reduces the research resources in this region through the optimized research strategy, allocates more resources to other regions with better performance, and generates a more accurate research path. Finally, through this recursive learning process, the system generates the second optimized path data to ensure that future research plans can use resources more effectively and improve the research accuracy.

[0121] Preferably, the feedback matrix includes multi-dimensional information of market feedback data. The information of each dimension includes market performance, market region, and consumer behavior. The feedback matrix is used for weighted processing during the optimization of the research strategy to adjust the priority of the research strategy. The generated second optimized path data is used to dynamically adjust the research strategy and further optimize the research path level.

[0122] Essentially, the feedback matrix is a multi-dimensional data structure that records various information dimensions involved in market feedback. Each dimension represents a different aspect of the market: for example, market performance may reflect indicators such as product sales volume and market penetration rate; the market region represents different geographical distributions. For example, the market performance in some regions may be stronger than that in other regions; the consumer behavior dimension reflects consumer preferences, purchase behaviors, social media interactions, etc. related to the product.

[0123] The structure of the feedback matrix can be regarded as a multi-dimensional matrix, where each data point (or cell) records the feedback data under a specific time period, specific market region, and specific behavior category. For example, a data point may represent the actual purchase volume of a specific consumer group for a certain product in a certain region in a certain quarter. The system constructs the feedback matrix by collecting this feedback information and processes it to adjust the priority of the subsequent research strategy.

[0124] When optimizing the research strategy, the system needs to perform weighted processing on different dimensional information in the feedback matrix. The purpose of weighted processing is to dynamically adjust the focus of the research strategy according to the importance and influence of the data. For example, the system may identify that the market feedback in certain regions has a greater impact on the overall research results. Therefore, when optimizing the strategy, the weights of these regions will be increased to ensure that future research paths are more inclined to focus on these key markets. The formula for weighted processing is generally as follows:

[0125]

[0126] where S represents the final strategy priority score; w i represents the weight of the i-th dimension in the feedback matrix; F i represents the feedback data of the i-th dimension in the feedback matrix; n represents the number of dimensions of the feedback matrix.

[0127] Through this weighted calculation, the system can dynamically adjust the influence of each dimension in the research strategy according to the importance and relevance of the market feedback. For example, if the importance of the market performance dimension is relatively high, the weight w i of this dimension will increase, making it play a greater role in strategy optimization.

[0128] In specific applications, the system first needs to structurally process the collected market feedback data and organize these data into the form of a feedback matrix. The data of each dimension will reflect different aspects of the market, such as sales volume, market growth rate, user feedback, etc. For the data of different dimensions, the system will perform weighted allocation according to the objectives of the market research and the current research requirements.

[0129] For example, in a research on the release of a new product, market performance may become the most important dimension, while data on consumer behavior and market regions are used as auxiliary dimensions. In this case, the system will assign a higher weight to the market performance dimension to ensure that the system focuses on those regions with strong market performance data when optimizing the research path. For those regions with relatively mediocre market feedback performance, the system will correspondingly reduce the allocation of research resources, thereby improving the utilization efficiency of resources.

[0130] After the weighted processing of the feedback matrix, the system generates an optimized strategy priority score. Based on these scores, the system dynamically adjusts the research strategy to ensure that resources can be preferentially allocated to market regions or data dimensions that have a greater impact on the research results. This dynamic adjustment can not only improve the research efficiency but also ensure that the research strategy always matches the actual needs of the market.

[0131] For example, assume that the feedback matrix collected by the system shows that the market performance in a certain region is extremely strong, and the interaction of consumers on social media has also increased significantly. These pieces of information indicate that the market potential in this region is relatively large. Therefore, when planning the next research path, the system will preferentially allocate research resources to this region and reduce the resource investment in those regions with average performance. The finally generated second optimized path data reflects this optimized research strategy.

[0132] By introducing the feedback matrix, the system can collect and analyze market feedback in real time during the research process, ensuring that the research strategy can be dynamically adjusted according to actual market changes. The multi-dimensional information in the feedback matrix, such as market performance, regional distribution, and consumer behavior, provides a comprehensive perspective on the market, enabling the research strategy to respond more precisely to market demands.

[0133] More precise strategy adjustment: By weighting the feedback matrix, the system can identify which dimensions have a higher impact on the research results and preferentially adjust the research path. For example, regions with significant market performance will receive more research resources to ensure that the research results are more representative.

[0134] Optimized allocation of research resources: The feedback matrix allows the system to adjust the resource allocation in real time according to changes in market feedback. Regions with weak market feedback will receive less research resource investment, while regions with outstanding performance will receive more attention. This dynamic resource allocation mechanism greatly improves the utilization efficiency of research resources.

[0135] Dynamic optimization of the research path: Through the second optimized path data generated by weighting the feedback matrix, the system can continuously adjust and optimize the research path to ensure the effectiveness of the research strategy in different market environments.

[0136] Suppose a brand launches a new product in multiple cities. The system collects market feedback from each city through the feedback matrix, including information such as sales volume, consumer purchase behavior, and social interaction. During the analysis process, the system discovers that the sales volume performance in City A is particularly outstanding, and the discussion heat of consumers on social media is significantly higher than that in other cities. The feedback matrix records this information in different dimensions, and the system assigns higher weights to the market performance and consumer behavior dimensions according to their importance.

[0137] After weighting, the system generates an optimized research strategy, indicating that more research efforts should be made in City A, while resource investment should be reduced in cities with poor market performance. Subsequently, the system dynamically adjusts the research path according to this strategy and generates the second optimized path data. This path data takes City A as the key research area in the future to ensure that the highest quality of market feedback can be obtained from the research with limited resources.

[0138] such asFigure 5 As shown, based on the Monte Carlo simulation method and graph neural network, the performance of the market under extreme conditions is simulated to generate market simulation data, and a dynamic outlier detection algorithm is used to identify abnormal behaviors in the market to generate anomaly detection data. A coping strategy is generated based on the anomaly detection data, and the research path is dynamically adjusted to obtain the third optimized path data.

[0139] The Monte Carlo simulation method is a numerical method that simulates the behavior of complex systems through random sampling. In the present invention, the Monte Carlo simulation is used to simulate the performance of the market under extreme conditions, which may include situations such as economic collapse, sudden policy changes, large-scale supply chain disruptions, etc. By performing multiple random samplings on different market variables (such as price, demand, supply, etc.), the Monte Carlo simulation can generate a large number of possible market scenarios and statistically analyze the performance of the market under these extreme scenarios.

[0140] Each simulation generates a market simulation data point according to different combinations of extreme conditions. As the number of simulations increases, the system can accumulate a large amount of market simulation data, forming a comprehensive understanding of the behavior of the market under extreme conditions. The simulation results are not just individual data points but the probability distribution of market behavior. For example, the system may simulate the probability that the demand for a certain product decreases by 50% under a specific economic recession, or the probability that the market demand in a specific region increases after a sudden policy change.

[0141] These simulation data are processed by a graph neural network. The role of the graph neural network here is to construct a graph structure through the complex relationships of market variables and propagate information in the graph to update market predictions. For example, there may be mutual influences between different market variables (such as price, demand, supply), and the complex associations of these variables are represented by the nodes and edges of the graph neural network. During multiple rounds of message passing, the features of the market simulation data are updated, and finally, a more accurate market prediction is generated.

[0142] Once the market simulation data is generated, the system will identify abnormal behaviors in the market through a dynamic outlier detection algorithm. The dynamic outlier detection algorithm is an algorithm used to identify abnormal and outlier data in a dataset. It analyzes the distribution of the data to identify market performances that significantly deviate from the normal pattern. For example, during the simulation process, if the system finds that the demand in a certain market area shows abnormal behavior (such as a sudden increase or decrease in demand) under extreme conditions, the market performance in that area may be marked as an abnormal behavior.

[0143] The core of outlier detection lies in identifying samples that significantly deviate from normal market performance. By performing outlier analysis on the simulation data, the system can identify potential market risks or opportunities. For example, in a certain simulation, if the system detects that the consumer demand in a certain region fluctuates greatly during an economic crisis, this may indicate that the market in this region is extremely sensitive to external shocks, and the system needs to adopt additional research strategies for this region.

[0144] In the specific implementation process, the system first uses the Monte Carlo simulation method to simulate the market performance under various extreme conditions based on the existing market data and historical trends. During the simulation process, the system randomly samples multiple market variables and calculates the possible market performance under extreme conditions according to the sampling results. Each simulation data point reflects the market state under a specific extreme condition, and these data are processed through a graph neural network to generate the final market prediction.

[0145] For example, when simulating a global economic recession, the system may randomly sample multiple market factors (such as consumption ability, commodity supply, price fluctuations, etc.). The graph neural network represents the associations between these variables in a graph structure and continuously updates the node features through a message passing mechanism, enabling the eigenvalue of each market variable to comprehensively consider the influence of its neighboring market variables.

[0146] After generating the market simulation data, the system uses a dynamic outlier detection algorithm to identify abnormal behaviors among them. The algorithm will find data points that deviate significantly from the normal market performance according to the distribution pattern of the simulation data. These abnormal data usually mean potential market risks or special opportunities, such as abnormal growth in demand in a certain region during extreme economic fluctuations, or supply chain collapse under specific market conditions.

[0147] Through anomaly detection, the system can generate anomaly detection data, which provides a basis for further strategy optimization. The system will dynamically adjust the research path according to the anomaly detection data. For example, if some market regions show abnormal behaviors during the simulation process, the system will mark these regions as high-risk or high-opportunity regions and increase the research resources for these regions in the next round of research to ensure a deep understanding of their potential market changes. Finally, this dynamic adjustment process generates the third optimized path data.

[0148] By combining Monte Carlo simulation and graph neural network, the present invention can effectively simulate the market performance under extreme conditions, ensuring that the research strategy can adapt to complex and uncertain market changes. This mechanism not only provides early warnings of potential market risks but also provides data support for strategy optimization.

[0149] Through the Monte Carlo simulation method, the system can handle various extreme conditions, simulate the possible performance of the market under various uncertain environments, and provide market performance predictions under multiple scenarios for decision-makers.

[0150] Through the message passing mechanism of the graph neural network, the system can capture the complex correlations between market variables in the simulation data, making the simulation results more accurate.

[0151] Through the dynamic outlier detection algorithm, the system can timely detect abnormal behaviors in the market. These abnormal behaviors provide key bases for the adjustment of subsequent research strategies, ensuring the effective allocation of research resources.

[0152] Generation of the third optimized path data: Through the above process, the system dynamically adjusts the research path to generate the third optimized path data, enabling the research strategy to flexibly respond to complex market conditions.

[0153] Suppose a company is conducting a global market research and hopes to simulate the market performance under extreme economic conditions. The system first uses the Monte Carlo simulation to simulate the market performance in each region under the scenario of a global economic recession and generates simulation data. By processing these simulation data through the graph neural network, the system can identify the complex relationships between various market variables. For example, the demand in a certain region is greatly affected by the interruption of the global supply chain, while the demand in another region has nothing to do with the economic environment.

[0154] Next, the system uses the dynamic outlier detection algorithm to identify abnormal behaviors in the simulation data. For example, the consumption demand in a certain region rises abnormally under extreme conditions, indicating that the potential demand for a certain type of product in this region is much higher than the prediction. The system will generate coping strategies based on this abnormal behavior and increase the research resources for this region in the next round of research path planning to ensure that this abnormal behavior can be further verified and analyzed. Finally, the system generates the third optimized path data to ensure that the adjustment of the research path can adapt to the changes under extreme market conditions.

[0155] Preferably, the Monte Carlo simulation method generates market prediction results under multiple extreme conditions through multiple random samplings. The simulation conditions include market demand fluctuations, price fluctuations, and supply chain stability. The simulation results are used to adjust the research path and the allocation of research resources.

[0156] The Monte Carlo simulation method is a numerical analysis method commonly used to simulate the behavior of complex systems. Its core idea is to generate a series of possible system behavior scenarios by performing multiple random samplings on the system's random variables, and to predict the overall performance of the system by statistically analyzing the results of these scenarios. In the analysis of market research data, the market is usually affected by various uncertain factors, especially under extreme conditions, such as sudden changes in market demand, sharp price fluctuations, supply chain disruptions, etc. Therefore, Monte Carlo simulation can help the system anticipate its performance under different market conditions and provide data support for optimizing the research path.

[0157] In the present invention, the system randomly samples multiple key market variables, including:

[0158] Market demand fluctuations: Market demand fluctuates with factors such as economic conditions, consumer preferences, and competitor dynamics. For example, assume that the demand in a market usually remains stable, but under extreme economic conditions, the demand may experience a significant decline or surge. Monte Carlo simulation will perform multiple random samplings on demand fluctuations based on historical data and extreme scenarios to generate multiple possible market demand scenarios.

[0159] Price fluctuations: Price changes are driven by multiple factors such as supply and demand relationships, raw material costs, market competition, and policy impacts. In the simulation, the system will simulate price fluctuations in the market under extreme conditions through random sampling. The simulation may generate scenarios such as price increases due to global supply chain tensions or significant price drops in certain regions.

[0160] Supply chain stability: The stability of the supply chain directly affects market supply. Disruptions or fluctuations in the supply chain may cause the market to be unable to obtain goods in a timely manner, thereby affecting market performance. During the simulation process, the system evaluates the market's response under these conditions by sampling different performances of the supply chain, such as normal operation of the supply chain, partial disruptions, or complete collapses.

[0161] Through multiple random samplings of these market conditions, the system generates a series of market simulation data, which represent the performance of the market under different extreme conditions. For example, a certain scenario may simulate a situation where, in the context of an economic crisis, the market demand in a certain region drops by 20%, and the supply chain disruption causes the price to increase by 15%. By accumulating a large number of similar simulation scenarios, the system can understand the overall performance trend of the market under uncertain conditions.

[0162] During the simulation process, the results after each sampling are recorded. The system generates a large amount of market prediction data according to different market conditions, forming a probability distribution of market performance. These results not only help identify the possible fluctuation range of the market under extreme conditions, but also enable the system to judge which market conditions are the most risky or opportunistic.

[0163] The implementation of Monte Carlo simulation involves the following key steps:

[0164] Random sampling and scenario generation: The system conducts random sampling for each set market simulation condition (such as demand fluctuations, price fluctuations, supply chain stability, etc.). For example, for market demand fluctuations, the system can set a fluctuation range (such as -30% to +30%), and then based on historical data and extreme condition assumptions, conduct multiple random samplings to generate multiple market scenarios under different demand fluctuations. A similar process is also used for the simulation of price fluctuations and supply chain fluctuations.

[0165] Accumulation of simulation data: Through multiple random samplings, the system generates a large number of simulation data points. For example, when simulating a certain regional market, the system will simulate the market performance under conditions of sharp decline in demand and drastic price fluctuations. The system accumulates these data points to form an overall prediction of the market under different extreme conditions.

[0166] Market prediction analysis: After the simulation data is generated, the system conducts statistical analysis on it to identify possible performances under different market conditions. For example, the system may identify that in an extreme economic environment, the probability of a 50% decline in market demand in a certain region is 15%, while the probability of a 30% increase in price is 10%. Through this statistical analysis, the system can understand which market conditions may have a significant impact on the research, thus providing a basis for optimizing the next research path.

[0167] After the simulation results are generated, the system uses these results to adjust the research path and resource allocation. The system will identify the market areas most likely to experience fluctuations or anomalies based on the extreme scenario data generated in the simulation, and preferentially allocate research resources to these areas. For example, if the market demand in a certain region shows abnormal fluctuations in the simulation, the system will dynamically adjust the research path and allocate more resources to this region to conduct in-depth analysis of its potential risks or opportunities.

[0168] Through the Monte Carlo simulation method, the system can predict the market performance under extreme conditions during the research process, and then optimize the research path and resource allocation. Its effects are reflected in the following aspects:

[0169] The system can identify potential risks and opportunities in the market under extreme conditions through simulation. For example, through the simulation of supply chain stability, the system can foresee the impact of supply chain disruptions on market supply in advance and better respond to this risk by adjusting the research path.

[0170] The simulation results help the system allocate research resources more effectively. For example, if the simulation results show that certain market regions exhibit significant demand fluctuations or price changes under extreme conditions, the system will prioritize allocating resources to these high-risk or high-opportunity regions to ensure the accuracy of research results and the utilization efficiency of research resources.

[0171] Through simulations of a large number of extreme scenarios, the system can generate optimization suggestions for research paths under different scenarios, ensuring that in the face of a complex market environment, the research path can be dynamically adjusted according to different market performances, maximizing the effectiveness of the research.

[0172] Suppose a brand conducts product research in the global market. The system simulates the market performances under multiple extreme conditions through Monte Carlo simulation. The simulation scenarios include a decline in market demand due to a global economic crisis and an increase in product prices due to a disruption in the raw material supply chain. In one simulation scenario, the system finds that the market demand in a certain region may decline by 30%, while in another region, the price fluctuates greatly and may increase by 20%.

[0173] Based on these simulation results, the system dynamically adjusts the research path, prioritizing the allocation of resources to regions with a significant decline in demand to further study changes in consumer behavior. At the same time, in regions with large price fluctuations, the system increases the research intensity to obtain more data related to supply chain disruptions. This process ensures that the system can make timely adjustments based on the simulation results, enabling research resources to be concentrated in the market regions that require the most attention.

[0174] Preferably, the dynamic outlier detection algorithm identifies abnormal behaviors by analyzing the density distribution of market simulation data, and the abnormal behaviors include abnormal demand fluctuations, supply chain disruptions, and sharp price fluctuations.

[0175] The core idea of the dynamic outlier detection algorithm based on density analysis is that under normal circumstances, most data points (i.e., market performances) will be concentrated in a certain region or density cluster, and those points that deviate from the normal density region are outliers. Outliers usually represent abnormal behaviors in the market, such as a sharp increase or decrease in demand, abnormal price fluctuations, or a sudden interruption in the supply chain.

[0176] In market research, market simulation data is generated through the Monte Carlo simulation method, and the simulation results contain predictions of market performances under multiple extreme conditions. The dynamic outlier detection algorithm then analyzes these simulation data, evaluates their density distribution, and detects and identifies abnormal data points. Specifically, the algorithm is implemented through the following steps:

[0177] After the simulation data is generated, the system analyzes the density distribution of the market simulation data by calculating the local density around each data point. Data points with higher local density usually represent normal market performance, while those with lower local density indicate abnormal market behavior. For example, in a market simulation result, most market demand forecasts show stable or slight fluctuations, but some data points show a significant increase or decrease in demand. These points have a lower local density, that is, their market behavior deviates from the normal market performance, and the system marks them as potential abnormal behaviors.

[0178] Based on the density distribution, the system determines which data points are outliers through dynamic thresholds or statistical methods (such as Z-score or density-based clustering algorithms). These outliers reflect behaviors that are significantly different from normal market conditions. For example, in an extreme situation, a certain market area shows an abnormal price increase, while the price fluctuations in other areas are relatively stable. This sharp price fluctuation will be identified as an outlier.

[0179] After outlier detection is completed, the system classifies these abnormal data points into different types of abnormal behaviors. In the present invention, the abnormal behaviors mainly include abnormal demand fluctuations, supply chain disruptions, and sharp price fluctuations.

[0180] Abnormal demand fluctuations: The system detects sudden changes in market demand through simulation, which may reflect rapid changes in consumer behavior or strong external factors affecting the market in the short term. For example, the demand for a certain product suddenly increases by 50% in a certain market, far higher than the performance in other markets. This is an abnormal demand fluctuation.

[0181] Supply chain disruption: Outlier detection can identify scenarios where the supply chain suddenly collapses or becomes unstable. For example, in extreme conditions, the supply chain in a certain market suddenly breaks down, resulting in a significant reduction in supply. This abnormal supply change will be identified as an abnormal behavior.

[0182] Sharp price fluctuations: The system simulates price fluctuations in extreme situations. If the price performance in a certain region is very different from that in other regions, it indicates that the price in this region is affected abnormally, which may be caused by an imbalance between supply and demand, policy changes, or other economic factors.

[0183] In the specific implementation process, first, the system constructs a multi-dimensional data space based on the market research simulation data, where each data point represents the predicted value of a certain market variable (such as demand, price, or supply chain performance) under extreme conditions. The system calculates the local density of each data point through a density analysis algorithm, that is, the distance and distribution of the data point from its neighboring data points in the data space.

[0184] For example, the system may detect that the demand forecast value in a certain market area is much lower than that in other areas. This low demand performance is different from the stable demand in other areas and belongs to a local low-density area. The system will mark this data point as an abnormal behavior and consider that there are abnormal fluctuations in demand in this market area.

[0185] To further improve the accuracy of detection, the system uses a dynamic threshold adjustment mechanism, that is, it dynamically adjusts the sensitivity of outlier detection according to the changes in data. For example, when there is an obvious fluctuation trend in the market, the system will appropriately lower the threshold to capture more abnormal behaviors; while when the market is relatively stable, the system will raise the threshold to ensure that only significant outliers are detected.

[0186] After detecting abnormal behaviors, the system generates abnormal detection data, which provides a basis for the optimization of the next research path. The system adjusts the research strategy according to the severity and occurrence frequency of abnormal behaviors. For example, if multiple market areas show sharp price fluctuations, the system may give priority to allocating research resources to these areas to investigate the reasons for price fluctuations in depth. Finally, the research path generated through this process is the third optimized path data.

[0187] The application of the dynamic outlier detection algorithm significantly improves the system's ability to respond to complex market changes. By analyzing the density distribution of simulation data, the system can timely discover those behaviors that are significantly different from the normal market performance, providing accurate basis for the adjustment of subsequent research strategies. The specific effects are reflected in the following aspects:

[0188] Through density distribution analysis and dynamic threshold adjustment, the system can quickly identify abnormal data points in the market. This means that abnormal situations such as sharp fluctuations in demand, sharp price increases, or sudden interruptions in the supply chain in the market can be captured in a timely manner.

[0189] Under extreme conditions (such as economic crises, policy changes, etc.), market performance is often difficult to predict. After generating a large amount of market data through Monte Carlo simulation, the dynamic outlier detection algorithm can deeply analyze this data, identify potential market risks and opportunities, enabling the system to provide more accurate strategic support for the allocation of research resources.

[0190] By detecting abnormal behaviors, the system can adjust the research strategy and give priority to investigating those market areas that show abnormal behaviors. This process ensures that research resources can be concentrated on the most risky or potential parts of the market, improving the effectiveness of research.

[0191] Suppose a company plans to launch a new product in different market regions. The system first simulates the market demand and price fluctuations in each region through Monte Carlo simulation. In a certain simulation scenario, the system detects that the market demand in Region A shows abnormal performance. Under extreme economic conditions, the demand in this region has increased by 50%, far higher than the demand performance in other regions. Through the dynamic outlier detection algorithm, the system marks the demand in this region as abnormal behavior.

[0192] At the same time, the system also finds that Region B shows significant price fluctuations in the simulation. Compared with other regions, the price fluctuation range in Region B reaches 20%, while the price fluctuation range in other regions basically remains within 5%. The system marks this price fluctuation as another abnormal behavior.

[0193] Based on these detected abnormal behaviors, the system dynamically adjusts the research path. Due to the abnormal increase in demand in Region A, more research resources will be allocated to further study the changes in consumer behavior. And due to the significant price fluctuations in Region B, it is also included in the key research area to ensure that the research results can timely reflect the market risks in this region.

[0194] Preferably, the abnormal behavior data identified by the dynamic outlier detection algorithm is used to dynamically adjust the research path. The system automatically reallocates research resources according to the type of the abnormal behavior, generates the third optimized path data, and the third optimized path data is used for optimizing the research path under extreme market conditions.

[0195] The dynamic outlier detection algorithm identifies data points that deviate significantly from normal market behavior by analyzing the density distribution of market simulation data. These abnormal behavior data can reflect the drastic fluctuations in the market under extreme conditions, such as abnormal increase or decrease in demand, significant price fluctuations, or supply chain disruptions. The abnormal behavior data is classified by the system, and each type of abnormal behavior will affect the optimization strategy of the research path and the reallocation of research resources. For example, abnormal fluctuations in market demand usually mean that there are potential opportunities or risks in this market region, while supply chain disruptions indicate that this region faces supply problems. When adjusting the research path, the system automatically adjusts the priority of research tasks and resource allocation according to the characteristics of each type of abnormal behavior.

[0196] When the system identifies abnormal behaviors through the dynamic outlier detection algorithm, it will classify these abnormal behaviors. The classification of abnormal behaviors includes:

[0197] Abnormal demand fluctuations: Such as a drastic increase or decrease in the quantity demanded, which may indicate that the market has been stimulated by external factors (such as competitor activities, economic changes, etc.).

[0198] Significant price fluctuations: Such as a substantial increase or decrease in price, which may be caused by supply shortages, policy changes, or market competition.

[0199] Supply chain disruptions: Disruptions in the supply chain can lead to products not reaching the market in a timely manner, affecting market supply.

[0200] After identifying abnormal behavior, the system dynamically adjusts the research path based on the type of abnormal behavior. The main principle of adjusting the research path is to ensure that research resources can be concentrated on the most potentially influential market areas.

[0201] When there are abnormal fluctuations in demand, the system will preferentially allocate more research resources in this market area, increasing investigations into aspects such as consumer behavior, market competition dynamics, and external economic conditions. For example, if the demand in a certain area suddenly increases significantly, the system will increase the research intensity to deeply understand changes in consumer preferences, market capacity, and market competition.

[0202] When a supply chain disruption is identified, the system will preferentially allocate resources to investigate the actual situation of the supply chain. The research tasks may include a detailed analysis of logistics, raw material supply, and other supply chain links to ensure that the research can accurately assess the long-term impact of the supply chain disruption on the market and potential solutions.

[0203] Abnormal price fluctuations are usually accompanied by market uncertainties and drastic changes in supply and demand. The system will increase research resources in such abnormal markets, analyze the reasons for price fluctuations, and explore key factors such as market supply and demand relationships and changes in raw material costs.

[0204] While adjusting the research path, the system generates a new research path plan, called the third optimization path data. These optimization path data are generated based on simulation data, abnormal behavior detection results, and the actual progress of research tasks to ensure that research tasks under extreme market conditions can accurately respond to market risks.

[0205] The third optimization path data is the result of adjusting the research path. It further optimizes resource allocation and research priorities based on the nature of abnormal behavior and market performance. The system evaluates abnormal behavior in different regions and dynamically ranks the urgency of research tasks according to these abnormal behaviors. For example, when there are drastic fluctuations in demand in a certain market, the system will preferentially allocate resources to that area, and when supply chain problems lead to insufficient market supply, the system will give priority to researching that market.

[0206] During the generation of the third optimized path data, the system also combines historical market data and extreme condition simulation results to ensure that the adjustment of the research path is not only based on current abnormal behavior data but also can anticipate possible future market changes. For example, if the simulation data predicts that supply chain problems may spread to neighboring regions, the system will adjust the research strategy in advance to ensure that the execution of the research task is not limited to the current market conditions and can also make corresponding preparations for possible future market fluctuations.

[0207] In practical applications, the generation of the third optimized path data can greatly improve the research efficiency and the rational allocation of resources. The system will dynamically adjust the research path according to the abnormal behaviors in different market regions and ensure that the research resources are allocated to the market regions that need them most.

[0208] The system automatically identifies abnormal market behaviors through an outlier detection algorithm and automatically adjusts the research path according to the type and severity of the abnormal behaviors. The adjustment of the research path is based on the real-time feedback of market data to ensure that the research strategy can quickly respond to market changes.

[0209] The system dynamically allocates research resources according to the type of abnormal behaviors. For example, markets with abnormal demand fluctuations will be given priority to obtain more resources, while markets with supply chain disruptions will be focused on investigating their supply chain problems and possible solutions. This way of resource allocation can not only improve the efficiency of the research task but also ensure the accuracy and depth of the research results.

[0210] The third optimized path data not only optimizes the research strategies for individual markets but also can adjust the priorities of research tasks globally. The system will ensure that the planning of the research path can handle abnormal behaviors in multiple market regions simultaneously based on the global analysis of the simulation data and coordinate the global allocation of research resources.

[0211] By dynamically adjusting the research path, the system can quickly respond to abnormal behaviors in the market and ensure that the research task can adapt to the real-time changes in the market. This enables the research strategy to remain efficient in the face of complex and extreme market conditions. Through the classification of abnormal behaviors, the system can dynamically adjust the allocation of research resources according to the specific conditions of different market regions to ensure that resources can be preferentially used to address the most urgent market problems. This not only improves the resource utilization efficiency but also makes the research results more accurate and comprehensive. The generation of the third optimized path data enables the system to optimize the research strategy globally, not only to handle problems in a single market region but also to coordinate research tasks in multiple market regions and ensure the rational allocation of research resources globally.

[0212] Suppose a company is promoting a new product in multiple market regions. The system detects through simulation that there is an abnormal increase in demand in Region A, while there is a major disruption in the supply chain in Region B. Based on these two abnormal behaviors, the system dynamically adjusts the research path. For Region A, the system will allocate additional resources to deeply understand the changes in consumer behavior and analyze the reasons for the demand increase. For Region B, the system will prioritize the allocation of resources to investigate the impact of the supply chain disruption and find possible alternative supply solutions.

[0213] Through these adjustments, the system generates the third optimized path data, enabling the planning of the research path to not only cope with the current abnormal market behaviors but also anticipate possible fluctuations in the future market, ensuring that the research tasks remain efficient and accurate in the future market environment.

[0214] As Figure 6 shown, a system for implementing the market research data analysis method, the system includes:

[0215] A data acquisition module, which is used to collect market research data, environmental data, and social behavior data from multiple data sources through multi-channel adaptive data integration technology, and clean and standardize the collected data to generate a comprehensive data set; Through multi-channel adaptive data integration technology, the present invention collects information from multiple data sources (such as market research data, environmental data, social behavior data). These data are cleaned and standardized to generate a comprehensive data set. This process ensures the effective integration of different data sources and enhances the credibility and consistency of the data through a weight adaptive adjustment mechanism. The core principle of this module is based on multi-source data fusion and dynamic adjustment technology, ensuring the comparability and consistency of data in different dimensions.

[0216] A graph neural network module, which is used to receive the comprehensive data set, construct a graph network structure, and update the node features in the graph network using the message passing mechanism of the graph neural network to generate market prediction data, and dynamically generate and optimize the research path based on the market prediction data and reinforcement learning algorithm to obtain the first optimized path data; This module receives the comprehensive data set and uses the message passing mechanism of the graph neural network to construct a market prediction model. Through the graph neural network, market data is represented as a graph structure, where nodes represent market elements and edges represent the associations between elements. The node features are continuously updated through the message passing mechanism, making the market prediction more accurate. Subsequently, based on the market prediction data and reinforcement learning algorithm, the system can dynamically generate and optimize the research path to ensure the efficient allocation of research resources. By introducing intelligent learning algorithms, this module realizes automatic path optimization under changing market conditions.

[0217] The feedback self - learning module is used to compare market feedback data with market prediction data, calculate the deviation of research execution, generate a feedback matrix, and learn and optimize the feedback matrix through a recurrent neural network to generate an optimized research strategy. Based on the optimized research strategy, it further adjusts the research path to obtain the second optimized path data. When comparing market feedback data with prediction data, the system can calculate the deviation of research execution and generate a feedback matrix. Through the recurrent neural network, the feedback matrix is learned and optimized to generate a new research strategy. The optimized research strategy is used to further adjust the research path to generate the second optimized path data. This module utilizes the feedback mechanism and self - learning algorithm to achieve the continuous improvement and dynamic adaptation of the research strategy.

[0218] The simulation and anomaly detection module is used to simulate the performance of the market under extreme conditions based on the Monte Carlo simulation method and graph neural network to generate market simulation data, and use the dynamic outlier detection algorithm to identify abnormal behaviors in the market to generate anomaly detection data. Based on the anomaly detection data, it generates coping strategies and dynamically adjusts the research path to obtain the third optimized path data. This module simulates the performance of the market under extreme conditions through the Monte Carlo simulation method and graph neural network to generate simulation data. After simulation, the dynamic outlier detection algorithm is used to identify abnormal behaviors in the market, such as a sudden surge in demand or a supply chain interruption. The anomaly detection results are used to generate coping strategies, and the system dynamically adjusts the research path based on these strategies to generate the third optimized path data. The simulation and anomaly detection functions of this module ensure the flexibility and coping ability of the research path under extreme market conditions.

[0219] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0220] The above are only the embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A market research data analysis method, characterized in that: The following steps are involved: Through multi-channel adaptive data integration technology, market research data, environmental data and social behavior data are collected from multiple data sources, and the collected data are cleaned and standardized to generate a comprehensive data set; Based on the graph neural network, a graph network structure is constructed by inputting a comprehensive data set, and the node features in the graph network are updated using the message passing mechanism of the graph neural network to generate market forecast data. Based on the market forecast data and the reinforcement learning algorithm, the survey path is dynamically generated and optimized to obtain the first optimized path data; Based on the feedback self-learning mechanism, the market feedback data is compared with the market forecast data, the deviation of the survey execution is calculated, and the feedback matrix is ​​generated. The feedback matrix is ​​learned and optimized through the recursive neural network to generate the optimized survey strategy. The survey path is further adjusted based on the optimized survey strategy to obtain the second optimized path data; Based on the Monte Carlo simulation method and graph neural network, the market performance under extreme conditions is simulated to generate market simulation data. The dynamic outlier detection algorithm is used to identify abnormal behavior in the market and generate anomaly detection data. Based on the anomaly detection data, a response strategy is generated, and the research path is dynamically adjusted to obtain the third optimized path data.

2. The market research data analysis method according to claim 1, characterized in that: The multi-channel adaptive data integration technology adaptively adjusts the weights of various data sources of market research data, environmental data, and social behavior data through a dynamic Bayesian network, and the weight distribution is automatically optimized based on the historical performance and data quality of each data source.

3. The market research data analysis method according to claim 1, characterized in that: The message passing mechanism of the graph neural network updates the node features in the graph network. The node feature update formula is: in, represents the eigenvalue of node v in the k+1th iteration; N(v) represents the set of neighbor nodes of node v; W k represents the weight matrix of the kth iteration; σ represents the activation function, which is used to perform nonlinear mapping on the updated node feature values; represents the eigenvalue of neighbor node u in the kth iteration; N(u) represents the set of neighbor nodes of node u.

4. The market research data analysis method according to claim 1, characterized in that: The survey path optimization of the graph neural network is achieved through a reinforcement learning algorithm. The generation of the survey path is based on a value function and a reward mechanism. The value function is dynamically adjusted according to the accuracy of the survey data and the efficiency of survey resource utilization. The first optimized path data generated is used to guide the allocation of survey resources and path selection.

5. The market research data analysis method according to claim 1, characterized in that: The recursive neural network is used to learn and optimize market feedback data and market forecast data. The hidden state update formula of the recursive neural network is: h t =σ(W x x t +W h h t-1 +b h ) Among them, h t represents the hidden state at time step t; x t Indicates the current input data; W x Represents the weight matrix of input data; W h The weight matrix representing the hidden state; h t-1 represents the hidden state at time step t-1; b h represents the bias term; σ represents the activation function, which is used to perform nonlinear mapping on the hidden state.

6. The market research data analysis method according to claim 1, characterized in that: The feedback matrix includes multiple dimensions of market feedback data, and the information of each dimension includes market performance, market area and consumer behavior. The feedback matrix is ​​used for weighted processing when optimizing the research strategy to adjust the priority of the research strategy, and the generated second optimization path data is used to dynamically adjust the research strategy and further optimize the research path level.

7. The market research data analysis method according to claim 1, characterized in that: The Monte Carlo simulation method generates market forecast results under multiple extreme conditions through multiple random samplings. The simulation conditions include market demand fluctuations, price fluctuations and supply chain stability. The simulation results are used to adjust the research path and research resource allocation.

8. The market research data analysis method according to claim 1, characterized in that: The dynamic outlier detection algorithm identifies abnormal behaviors by analyzing the density distribution of market simulation data, including abnormal demand fluctuations, supply chain disruptions, and sharp price fluctuations.

9. The market research data analysis method according to claim 8, characterized in that: The abnormal behavior data identified by the dynamic outlier detection algorithm is used to dynamically adjust the research path. The system automatically reallocates research resources according to the type of abnormal behavior and generates third optimized path data. The third optimized path data is used to optimize the research path under extreme market conditions.

10. A system for implementing the market research data analysis method according to any one of claims 1 to 9, characterized in that: The system comprises: The data collection module is used to collect market research data, environmental data and social behavior data from multiple data sources through multi-channel adaptive data integration technology, and clean and standardize the collected data to generate a comprehensive data set; A graph neural network module, which is used to receive the comprehensive data set, construct a graph network structure, and use the message passing mechanism of the graph neural network to update the node features in the graph network, generate market forecast data, and dynamically generate and optimize the survey path based on the market forecast data and the reinforcement learning algorithm to obtain the first optimized path data; A feedback self-learning module is used to compare the market feedback data with the market forecast data, calculate the deviation of the survey execution, generate a feedback matrix, and learn and optimize the feedback matrix through a recursive neural network to generate an optimized survey strategy, and further adjust the survey path based on the optimized survey strategy to obtain second optimized path data; The simulation and anomaly detection module is used to simulate the performance of the market under extreme conditions based on the Monte Carlo simulation method and graph neural network, generate market simulation data, and use a dynamic outlier detection algorithm to identify abnormal behavior in the market, generate anomaly detection data, and generate a response strategy based on the anomaly detection data, and dynamically adjust the research path to obtain the third optimized path data.

Images