Competitor market strategy generation method and device, electronic equipment and storage medium

Abstract

The application relates to the technical field of market strategy generation of competitors, and discloses a market strategy generation method and device of competitors, electronic equipment and a storage medium, wherein the method comprises the following steps: through integration of multi-platform real-time data and construction of a dynamic user portrait, the influence of market changes on user behavior can be accurately captured, the limitations of a traditional prediction model can be effectively overcome, a method combining causal reasoning and dynamic modeling is adopted, market competition situations can be intelligently analyzed, an optimal response strategy can be automatically generated, the response time is greatly shortened, and finally, through an intelligent strategy generation mechanism, a target market strategy for the target competitor is generated, and the efficiency and precision of the decision-making process are realized.The application has the beneficial effects that the strategy generation efficiency is significantly improved, the scientificity and applicability of the strategy scheme are ensured, and reliable technical support is provided for enterprises to maintain a competitive advantage in a dynamic market environment.

Description

Technical Field

[0001] This invention relates to the field of competitor market strategy generation technology, and in particular to a competitor market strategy generation method, apparatus, electronic device and storage medium. Background Technology

[0002] In the current technological environment, enterprises face two prominent technical bottlenecks in generating market strategies. First, traditional user behavior prediction models generally suffer from the limitations of static prediction, relying primarily on historical data for training. They struggle to respond in real-time to the dynamic impact of sudden market events on user decisions. The inability to capture instantaneous changes in the market environment leads to significant discrepancies between predictions and actual user behavior, severely impacting decision accuracy. Second, existing solutions also exhibit significant technical problems in competitor response mechanisms, specifically in response lag and rigid rule engines. Most current systems use engines based on preset rules to handle competitor dynamics, requiring enterprises to manually set fixed response thresholds before market changes. This rigid approach cannot adapt to the complex and ever-changing market competition environment and often requires significant time to respond to rapidly changing market conditions, causing enterprises to miss optimal market opportunities. Therefore, there is an urgent need for an innovative market strategy generation method that can flexibly adapt to dynamic market environments, thereby improving the timeliness and effectiveness of decision-making. Summary of the Invention

[0003] Based on this, it is necessary to address the existing problem of generating market strategies for competitors by proposing a method, apparatus, electronic device, and storage medium for generating market strategies for competitors.

[0004] A method for generating a competitor's market strategy, the method comprising:

[0005] Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products;

[0006] By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user.

[0007] Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0008] The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model.

[0009] Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results;

[0010] The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor.

[0011] Furthermore, the step of performing cross-domain semantic alignment and fusion of multiple real-time behavioral data using a contrastive learning algorithm to obtain a unified user profile for each user includes:

[0012] Obtain user information for each user, and label each real-time behavior data based on the user information;

[0013] A contrastive learning model based on an encoder is constructed, in which real-time behavior data from the same user but different platforms are used as positive sample pairs for semantic convergence, and real-time behavior data from different users are used as negative sample pairs for semantic distance, in order to train the encoder.

[0014] Multiple real-time behavioral data of each user are input into the trained encoder and mapped into a unified vector space to generate a multi-dimensional user feature vector for that user, which serves as the unified user profile for that user.

[0015] Furthermore, in the step of extracting feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features respectively, the step of extracting market features from the market data includes:

[0016] Using a pre-trained multilingual BERT model, multilingual competitor announcement texts from the target market are parsed in real time to identify key action types and quantification parameters in the announcements, thereby generating competitor features.

[0017] By using computer vision models to perform time-series analysis on satellite images of relevant areas of the target market, data on changes in customer flow at offline stores can be extracted.

[0018] Based on the competitor characteristics and the customer flow change data, a weighted fusion is performed to generate a comprehensive market popularity index, which serves as the market characteristic.

[0019] Furthermore, the step of converting the user characteristics, market characteristics, and competitor characteristics into structured causal variables, and constructing a dynamic causal graph based on the structured causal variables using a structured causal model, includes:

[0020] Extract the action type and intensity parameters from the competitor features and map them as treatment variables; extract the purchase intention index from the user features and map it as a core outcome variable; extract the market popularity index from the market features and map it as a moderating variable.

[0021] Construct a directed acyclic graph that takes the treatment variable as the cause, is influenced by the moderating variable, and acts on the core outcome variable, to obtain the basic graph structure of the structural causal model;

[0022] In the basic graph structure, a preset time delay parameter is introduced for the causal path between the treatment variable and the core outcome variable, thereby obtaining the dynamic causal graph.

[0023] Furthermore, the step of performing counterfactual reasoning based on the dynamic causal graph to generate causal reasoning results includes:

[0024] Based on the dynamic causal graph, a structural causal model computation framework including the time delay parameters is constructed;

[0025] Obtain the current market status and apply preset intervention measures to the treatment variables in the calculation framework based on the current market status;

[0026] Based on the preset intervention measures, the probability distribution changes of the core outcome variables within a set time window are deduced through the computational framework.

[0027] The expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures are output as the causal inference result.

[0028] Furthermore, the step of inputting the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor includes:

[0029] The expected change and confidence interval under the preset intervention measures in the causal reasoning results are combined with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt.

[0030] The strategy generation prompts are input into a preset large language model to generate multiple strategy options; each strategy option includes an action plan, expected quantitative performance indicators, and corresponding implementation risk assessment.

[0031] A weighted score is assigned based on the expected quantitative performance indicators in each of the strategy options and the implementation risk assessment.

[0032] Select the strategy option with the highest score as the target market strategy.

[0033] Furthermore, before the step of inputting the causal reasoning result into a preset large-scale language model to generate a target market strategy for the target competitor, the method further includes:

[0034] Real-time acquisition of regional cultural activities related to the target market;

[0035] Extract the regional cultural characteristics of the aforementioned regional cultural activities;

[0036] The regional cultural characteristics are analyzed to obtain regional cultural rules;

[0037] The regional cultural rules are matched with a preset general strategy parameter baseline, and the parameter adjustment amount is calculated based on the rule content.

[0038] The initial large-scale language model is adjusted according to the parameter adjustment amount to obtain the preset large-scale language model.

[0039] A competitor's market strategy generation device, the device comprising:

[0040] The acquisition module is used to acquire real-time behavioral data of multiple users on various platforms, as well as market data of the target market and competitor data of the target competitors;

[0041] The comparison module is used to perform cross-domain semantic alignment and fusion of multiple real-time behavioral data through a comparison learning algorithm to obtain a unified user profile for each user.

[0042] The extraction module is used to extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0043] The conversion module is used to convert the user characteristics, market characteristics, and competitor characteristics into structured causal variables, and to construct a dynamic causal graph based on the structured causal variables using a structured causal model.

[0044] The reasoning module is used to perform counterfactual reasoning based on the dynamic causal graph and generate causal reasoning results.

[0045] The generation module is used to input the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor.

[0046] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps:

[0047] Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products;

[0048] By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user.

[0049] Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0050] The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model.

[0051] Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results;

[0052] The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor.

[0053] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0054] Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products;

[0055] By using a contrastive learning algorithm, multiple real-time behavioral data are semantically aligned and fused across domains to obtain a unified user profile for each user.

[0056] Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0057] The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model.

[0058] Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results;

[0059] The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor.

[0060] The beneficial effects of this invention are as follows: By combining contrastive learning, causal inference, and large-scale language models, significant technical effects are achieved, including improved data processing efficiency and reduced computational latency. First, by integrating real-time data from multiple platforms through contrastive learning algorithms, a dynamic user profile is constructed, enabling the system to quickly and accurately capture the impact of market changes on user behavior. This improved data processing efficiency allows for the extraction of multi-dimensional user characteristics in a short time, providing rich input data for subsequent analysis. Second, this invention employs a method combining causal inference and dynamic modeling, breaking through the rigid processing mode of traditional rule engines. It can intelligently analyze market competition and automatically generate optimal response strategies. With the help of structural causal models, the system improves its responsiveness to changes in complex market environments while significantly reducing computational latency. Traditional methods often require cumbersome manual intervention and threshold setting, while this invention enables instant decision-making based on real-time data, quickly adapting to market changes, shortening response time, and ensuring that enterprises can act quickly in competition. Finally, through an intelligent strategy generation mechanism, large-scale language models can efficiently generate strategy solutions that meet current market demands. This process not only improves the efficiency and accuracy of decision-making but also ensures the scientific validity and applicability of the strategy solutions. In summary, this invention provides strong technical support for enterprises to maintain a competitive advantage in a dynamic market environment, improves overall response speed, and enhances the flexibility and accuracy of market responses. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] in:

[0063] Figure 1 This is an application environment diagram of a competitor's market strategy generation method in one embodiment;

[0064] Figure 2 A flowchart of a method for generating a competitor's market strategy in one embodiment;

[0065] Figure 3 A structural block diagram of a competitor's market strategy generation device in one embodiment;

[0066] Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Figure 1 Generate an application environment diagram for the market strategies of competitors in one embodiment. (Refer to...) Figure 1 The method for generating competitor market strategies is applied to a competitor market strategy generation system. This system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to acquire real-time behavioral data, and the server 120 is used to generate the target market strategies for the target competitor.

[0069] like Figure 2 As shown, in one embodiment, a method for generating a competitor's market strategy is provided. This method can be applied to both terminals and servers; this embodiment uses a terminal application as an example. The method for generating a competitor's market strategy specifically includes the following steps:

[0070] S1: Obtain real-time behavioral data of multiple users on various platforms, as well as market data of the target market and competitor data of the target competitors;

[0071] S2: By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user;

[0072] S3: Extract the feature data of the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively;

[0073] S4: Convert the user characteristics, market characteristics, and competitor characteristics into structured causal variables, and construct a dynamic causal graph based on the structured causal variables using a structured causal model;

[0074] S5: Based on the dynamic causal graph, perform counterfactual reasoning to generate causal reasoning results;

[0075] S6: Input the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor.

[0076] As described in step S1 above, real-time behavioral data of multiple users on various platforms, as well as market data of the target market and competitor data of the target products, are acquired. Real-time behavioral data from different users is collected through various channels, including but not limited to e-commerce websites, social media platforms, and applications. E-commerce websites can provide information such as user purchase records, browsing behavior, and product reviews; social media can provide data on user interaction behavior, likes, comments, and shares; and applications can provide user usage frequency and behavioral patterns within the application. Simultaneously, market data of the target market also needs to be collected, typically including competitors' promotional activities, product pricing, and changes in market demand. Competitor data includes competitors' sales records, user feedback, and market share. This data provides the foundation for subsequent user profiling, market feature extraction, and competitor analysis, enabling subsequent analysis and predictions to be based on a comprehensive and accurate reality. Specifically, real-time behavioral data can be limited to behavioral data within a preset time period from the current time, such as behavioral data within one hour of the current time. It should be noted that this user is a user of the target market, and each user can be segmented using a location device.

[0077] As described in step S2 above, multiple real-time behavioral data are semantically aligned and fused across domains using a contrastive learning algorithm to obtain a unified user profile for each user. After data acquisition, the collected real-time behavioral data needs to be processed using a contrastive learning algorithm to achieve cross-domain semantic alignment and fusion. The core of this process is to effectively integrate real-time behavioral data from different platforms, ensuring that data from different sources exhibit consistency within the same semantic space. First, through contrastive learning, data from the same user on different platforms can be considered positive samples, while data from different users can be considered negative samples. When training with a neural network model (such as a neural network or multilayer perceptron), the model learns how to bring the behavioral data of the same user from different platforms closer together, while pushing data from different users further apart, within a fixed low-dimensional space. Through continuous iterative training, a unified user profile is generated. This unified user profile not only covers the user's basic information (age, gender, etc.) but also includes deeper features such as user preferences, spending power, and brand loyalty.

[0078] As described in step S3 above, feature data from the unified user profile, market data, and competitor data are extracted to obtain user features, market features, and competitor features, respectively. User features are extracted from the unified user profile, including but not limited to user consumption habits, purchase history, brand preferences, and sensitivity to price changes. These features help the system understand the behavioral patterns of target users, providing personalized information for subsequent market strategy design. Market features are extracted from the target market data, including information on market demand fluctuations, price trends, and dynamic changes in the competitive environment. Market feature data helps companies grasp market trends when making decisions and rationally formulate product launch dates, pricing strategies, and promotional activities. Competitor features are extracted from the competitor data, including competitors' price adjustment strategies, promotional methods, user reviews, and feedback. Through competitor analysis, the position of one's own product in the market can be identified, as well as how to stand out from the competition. In this process, feature extraction techniques (such as principal component analysis (PCA), feature selection techniques, etc.) can be used to enhance the effectiveness and interpretability of the model, forming a multi-dimensional feature set.

[0079] As described in step S4 above, the user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables. Based on these structured causal variables, a dynamic causal graph is constructed using a structural causal model. Converting the extracted user characteristics, market characteristics, and competitor characteristics into structured variables suitable for causal inference allows qualitative and quantitative features to be transformed into a format that can be processed within the structural causal model. Firstly, user characteristics include "purchase intention" and "brand loyalty," market characteristics can be "market popularity index" and "demand change rate," while competitor characteristics can be their price, promotion frequency, etc. Constructing a dynamic causal graph using a structural causal model (SCM) first requires defining the causal relationships between different variables. For example, how do user characteristics and competitor characteristics affect market characteristics, and how changes in the market environment, in turn, affect user decisions? Constructing a directed acyclic graph (DAG) can visually demonstrate these causal relationships. Based on this, counterfactual reasoning can be introduced to support the simulation of strategy effects under different scenarios.

[0080] As described in step S5 above, counterfactual reasoning is performed based on the dynamic causal graph to generate causal reasoning results. Counterfactual reasoning is an important method in causal inference, allowing the system to speculate how other variables would change if a certain event had not occurred. For example, the system can simulate how user purchasing decisions and market conditions would change if a competitor adopted different promotional strategies or adjusted prices. Specifically, different intervention conditions need to be set first, involving "intervention" or "manipulation" of specific variables. Then, based on the dynamic causal graph, calculations and deductions are performed to evaluate the impact of these interventions on outcome variables (such as market share, sales, etc.). In this process, using the formulas and paths provided by the structural causal model, the system can calculate the relationships between different variables in real time, predict and quantify the degree of influence on the outcome.

[0081] As described in step S6 above, the causal inference results are input into a pre-set large-scale language model (LLM) to generate a target market strategy for the target competitor. In this step, the system uses the causal inference results generated in the previous steps, through a pre-trained large-scale language model (LLM), to generate a specific market strategy for the target competitor. First, the causal inference results provide the model with detailed data information, including user preferences, market changes, and the influence of each competitor. This information provides rich context for the LLM's strategy generation input. Next, after appropriate formatting, the causal inference results are integrated into structured prompts to guide the LLM in more intelligent decision generation. These prompts should include clear guidance on strategy objectives, potential market opportunities, risk assessment, and specific implementation plans, thereby ensuring that the generated market strategy is targeted and actionable. After the generation process is complete, the LLM will output a strategy combination, which should include multiple independent and selectable market behavior options, each taking into account real-time market changes and user preferences.

[0082] By combining contrastive learning, causal inference, and large language models (LLM), significant technical effects are achieved, including improved data processing efficiency and reduced computational latency. First, by integrating real-time data from multiple platforms through contrastive learning algorithms, a dynamic user profile is constructed, enabling the system to quickly and accurately capture the impact of market changes on user behavior. This improved data processing efficiency allows for the extraction of multi-dimensional user features in a short time, providing rich input data for subsequent analysis. Second, this invention employs a method combining causal inference and dynamic modeling, breaking through the rigid processing mode of traditional rule engines. It can intelligently analyze market competition and automatically generate optimal response strategies. With the help of structural causal models, the system improves its responsiveness to changes in complex market environments while significantly reducing computational latency. Traditional methods often require cumbersome manual intervention and threshold setting, while this invention enables instant decision-making based on real-time data, quickly adapting to market changes, shortening response time, and ensuring that enterprises can act swiftly in competition. Finally, through an intelligent strategy generation mechanism, LLM can efficiently generate strategy solutions that meet current market demands. This process not only improves the efficiency and accuracy of decision-making but also ensures the scientific validity and applicability of strategy solutions. In summary, this invention provides strong technical support for enterprises to maintain a competitive advantage in a dynamic market environment, improves overall response speed, and enhances the flexibility and accuracy of market responses.

[0083] In one embodiment, step S2, which involves performing cross-domain semantic alignment and fusion of multiple real-time behavioral data using a contrastive learning algorithm to obtain a unified user profile for each user, includes:

[0084] S201: Obtain user information for each user, and mark each of the real-time behavioral data based on the user information;

[0085] S202: Construct a contrastive learning model based on the encoder, using real-time behavior data from the same user but different platforms as positive sample pairs for semantic convergence, and using real-time behavior data from different users as negative sample pairs for semantic distance, in order to train the encoder.

[0086] S203: Input multiple real-time behavioral data of each user into the trained encoder, map them into a unified vector space, and generate a multi-dimensional user feature vector for the user, which serves as the unified user profile for the user.

[0087] As described in step S201 above, user information for each user is obtained, and each piece of real-time behavioral data is labeled based on the user information. Basic attributes of each user are identified, such as age, gender, geographical location, and registration time. User information may also include deeper information about their consumption behavior, such as historical behavior records, consumption preferences, and brand loyalty. User information can be obtained through various means, including extracting stored data from a database, obtaining third-party data through API interfaces, or collecting background information during user registration or login. Labeling behavioral data based on user information helps the subsequent comparative learning model more accurately identify user characteristics. During the labeling process, each piece of real-time behavioral data is accompanied by a label indicating its corresponding user ID and characteristics, enabling the model to clearly identify which behavioral data comes from the same user during subsequent training.

[0088] As described in step S202 above, a contrastive learning model based on an encoder is constructed. Real-time behavioral data from the same user but different platforms are treated as positive sample pairs for semantic convergence, while real-time behavioral data from different users are treated as negative sample pairs for semantic distance reduction, in order to train the encoder. This encoder-based contrastive learning model aims to extract more representative user features by comparing user behavioral data on different platforms. For behavioral data collected from the same user on different platforms, these data are treated as positive sample pairs to semantically converge the behavioral features of the same user, increasing their similarity in the vector space. For behavioral data from different users, these data are treated as negative sample pairs for semantic distance reduction, thereby improving the model's ability to identify differences between different users. The core of contrastive learning lies in using a loss function (such as Triplet Loss or Contrastive Loss) to calculate similarity. During training, the model continuously adjusts its weights to minimize the distance between positive samples from the same user while maximizing the distance between negative samples from different users. This process is repeated until the encoder can effectively map real-time behavioral data into a unified vector space that reflects user features.

[0089] As described in step S203 above, multiple real-time behavioral data points for each user are input into the trained encoder and mapped to a unified vector space to generate a multi-dimensional user feature vector for that user, serving as the unified user profile. The core task of applying the trained encoder to each user's multiple real-time behavioral data points is to uniformly map real-time behavioral data collected from different platforms into a fixed vector space. This space is learned through the aforementioned contrastive learning model. Through this mapping, behavioral data of the same user from different platforms can be merged and standardized, thus forming the basis for a unified user profile. Each user's multi-dimensional feature vector consists of multiple feature values ​​that reflect the user's behavioral patterns, consumption habits, and preferences. For example, the feature vector may contain information such as interest in specific products, average spending, and access frequency. By integrating these user features into a single vector, the system can not only quickly identify similar features in the user profile but also provide strong support for data analysis, market decision-making, and strategy generation.

[0090] In one embodiment, in step S3, which involves extracting feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features respectively, the step of extracting market features from the market data includes:

[0091] S301: Using a pre-trained multilingual BERT model, the multilingual competitor announcement text from the target market is parsed in real time to identify key action types and quantification parameters in the announcement, so as to generate competitor features;

[0092] S302: Perform time-series analysis on satellite images of the relevant areas of the target market using a computer vision model to extract customer flow change data for offline stores;

[0093] S303: Based on the competitor characteristics and the customer flow change data, perform weighted fusion to generate a comprehensive market popularity index, which is used as the market characteristic.

[0094] As described in step S301 above, a pre-trained multilingual BERT (Multilingual Bidirectional Encoder Representations from Transformers) model is used to parse multilingual competitor announcement texts from the target market in real time, identifying key action types and quantification parameters in the announcements to generate competitor features. The multilingual BERT model possesses powerful natural language processing capabilities, capable of handling text data in multiple languages. First, competitor announcement texts from the target market need to be input into the multilingual BERT model. These announcements may involve key information such as competitor pricing strategies, promotional activities, and new product launches. Through in-depth text mining, the BERT model can identify key action types contained in the announcements, such as "price reduction," "launching new products," and "starting marketing campaigns." Simultaneously, the model can also extract relevant quantification parameters, such as price change magnitude, promotion duration, and target user group.

[0095] As described in step S302 above, time-series analysis is performed on satellite images of the relevant areas of the target market using a computer vision model to extract data on changes in customer flow at offline stores. Relevant satellite images (such as those from Google Earth Engine) are collected chronologically, with updates potentially daily to capture changes within a specific time period. Advanced image processing algorithms and deep learning models (e.g., Convolutional Neural Networks, CNNs) are employed to extract the specific locations, on / off status, and other background information of offline stores from static satellite images. Customer flow estimation generally relies on detecting and counting the number of people in the images. Through time-series analysis, changes in store customer flow within a specific area and time period can be quantified, and changes in customer flow density at different points in time can be compared. This data not only reflects market activity during a specific period but can also be cross-validated with competitors' marketing activities, enabling companies to better understand the real-time dynamics of the market. Specifically, high-resolution satellite images are used to acquire views of the target market area, ensuring the images are clear and have sufficient resolution to capture crowd dynamics. The satellite images should cover a specific time period (e.g., daytime) to facilitate accurate calculation of customer flow. The YOLO (You Only Look Once) algorithm is used to process and detect individuals in images in real time. The core of YOLO lies in dividing the image into a grid and predicting bounding boxes and corresponding class probabilities for each grid. This model can accurately identify pedestrians in images, is fast, and suitable for real-time analysis scenarios, such as rapid calculation of pedestrian traffic. By inputting satellite imagery into the YOLO model, the location of pedestrians in the image is automatically detected, and the bounding box of each pedestrian is output. To improve accuracy, additional post-processing, such as non-maximum suppression (NMS), can be performed to reduce redundant detections. Frame difference analysis is used for time-series analysis. Frame difference is a classic technique for counting objects based on differences between consecutive frames. By comparing images of adjacent frames in the same area, the change in the number of pedestrians is calculated, revealing the dynamic changes in pedestrian traffic at different time points. The pedestrian traffic data at each time point is combined with other market characteristics to generate time-series data on pedestrian traffic. This data, after processing, can provide important support for subsequent calculations of market popularity indices.

[0096] As described in step S303 above, based on the competitor features and the passenger flow change data, weighted fusion is performed to generate a comprehensive market heat index as the market feature. The market heat index can be regarded as a quantitative representation of the current state of the target market. It synthesizes the responses of users to different competitors and market dynamics, thereby providing a relevant market analysis basis for decision-makers. According to the extracted competitor features (such as price fluctuations, promotional activities, market responses) and passenger flow change data (such as traffic increase or decrease, passenger flow density), a mathematical model is constructed for weighted fusion. This model can be a simple weighted average, a regression model, or even a more complex machine learning model. The specific choice depends on the nature of the data and the analysis requirements. The setting of the weighting coefficients can be determined according to the importance of the features and their influence on the market heat. For example, if the discount intensity of a competitor has a greater impact during a specific period, the weight of the features of this competitor can be increased during the fusion process. At the same time, the weight of the passenger flow change should also take into account the particularity of different time periods. For example, during holidays and promotional seasons, the passenger flow change may better reflect the market heat. Finally, the generated comprehensive market heat index will be used as a key indicator of the market feature to guide the subsequent process of strategy generation.

[0097] In one embodiment, step S4 of converting the user features, market features, and competitor features into structured causal variables and constructing a dynamic causal graph based on the structured causal variables using a structural causal model includes:

[0098] S401: Extract the action type and intensity parameters in the competitor features and map them to treatment variables, extract the purchase willingness index in the user features and map it to the core outcome variable, and extract the market heat index in the market features and map it to the moderator variable;

[0099] S402: Construct a directed acyclic graph with the treatment variable as the cause, affected through the moderator variable, and acting on the core outcome variable to obtain the basic graph structure of the structural causal model;

[0100] S403: In the basic graph structure, introduce a preset time lag parameter for the causal path between the treatment variable and the core outcome variable to obtain the dynamic causal graph.

[0101] As described in step S401 above, the action type and intensity parameters from the competitor features are extracted and mapped as therapeutic variables; the purchase intention index from the user features is extracted and mapped as a core outcome variable; and the market popularity index from the market features is extracted and mapped as a moderating variable. Relevant action types and intensity parameters are identified from the competitor features, including the competitor's price reduction, promotional activity types (such as discounts, buy-one-get-one-free offers), and other market behaviors. In the structural causal model, these actions are considered "therapeutic variables," i.e., factors that can be directly intervened in, and their changes will affect subsequent user behavior. Next, the purchase intention index is extracted from the unified user profile. Purchase intention is the degree of interest a user has in a specific product, which is usually influenced by multiple factors, such as price sensitivity and brand loyalty. Mapping to a "core outcome variable" means that this variable is ultimately affected by the treatment variable, and its changes can reflect the effectiveness of market behavior. Finally, the market heat index is extracted from market characteristics. This is a quantitative indicator that reflects dynamic changes in the market and includes consumers' reactions to market activities. This feature is mapped to a "moderating variable," which records the moderating effect of market conditions on the core outcome variable (i.e., purchase intention). This mapping work lays the foundation for the subsequent construction of a causal model, enabling the model to intuitively reflect the causal relationships between variables and provide data support for subsequent inference.

[0102] As described in step S402 above, a directed acyclic graph (DAG) is constructed, which takes the treatment variable as the cause, is influenced by the moderating variable, and acts on the core outcome variable, thus obtaining the basic graph structure of the structural causal model. A DAG is a graphical representation method used to show the causal relationships between variables, ensuring the logical clarity of the causal path and avoiding circular dependencies. First, when constructing the graph, the competitor feature mapped to the treatment variable is used as the starting node. This node represents a feature of the external market that directly affects user decisions. Next, according to the previously defined logic, directed edges are established between the competitor feature node and the purchase intention index (core outcome variable), indicating how the competitor's pricing or promotional behavior directly affects the user's purchase decision. Meanwhile, the market popularity index, as a moderating variable, plays a role in moderating the relationship between the therapeutic variable and the core outcome variable. To this end, in the directed acyclic graph, connections are established from the moderating variable node to the core outcome variable node. This indicates that market conditions (such as high or low popularity) may affect users' willingness to buy. For example, when market popularity is high, users' willingness to buy may be further stimulated, while when market popularity is low, users' interest may be weakened. In this way, the constructed causal graph provides a clear graphical structure for subsequent causal inference, making the relationship between different variables intuitively presented, and laying a solid foundation for subsequent counterfactual inference and strategic decision-making.

[0103] As described in step S403 above, a preset time lag parameter is introduced into the causal path between the treatment variable and the core outcome variable in the basic graph structure to obtain the dynamic causal graph. In complex market environments, the effects of interventions often do not appear immediately but are delayed over time. This phenomenon is the time lag effect in the causal path. First, by combining historical data and domain expertise, it is determined that the influence of the treatment variable (competitor characteristics) on the core outcome variable (user purchase intention) has a time lag characteristic. For example, after a user receives a discount promotion on a product, it may take some time to see its actual impact on sales and purchase intention. Therefore, the time lag parameter introduced into this causal path represents the delay between applying treatment (such as price reduction) and observing the effect (such as increased purchase intention). The introduction of the time lag parameter is achieved by explicitly identifying it in the representation of nodes and edges. This can be achieved by adding additional delay variables or adjusting path weights to reflect this time delay. In this way, the dynamic causal graph not only depicts the causal relationships between variables but also captures the temporal characteristics of these relationships, enabling the model to more accurately reflect the dynamic changes in the market.

[0104] In one embodiment, step S5, which involves performing counterfactual reasoning based on the dynamic causal graph to generate a causal reasoning result, includes:

[0105] S501: Based on the dynamic causal graph, construct a structural causal model calculation framework that includes the time delay parameters;

[0106] S502: Obtain the current market status and apply preset intervention measures to the treatment variables in the calculation framework based on the current market status;

[0107] S503: Based on the preset intervention measures, the probability distribution change of the core outcome variable within a set time window is deduced through the calculation framework;

[0108] S504: Output the expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures, as the result of the causal inference.

[0109] As described in step S501 above, a structural causal model computational framework incorporating the time-delay parameters is constructed based on the dynamic causal graph. The computational framework is designed to handle causal relationships between variables defined in the graph and to include time variables, particularly the aforementioned time-delay parameters. These time-delay parameters play a crucial role in causal inference because they characterize the time delay between interventions (such as price adjustments or promotional actions) and market feedback. When constructing the framework, a structural causal model (SCM) approach is used. This model not only considers immediate causal effects but also captures the dynamic characteristics of causal relationships by introducing time-delay variables. For example, after a competitor's promotion, user purchasing responses may take several days to materialize. The model equations must be cleverly designed to model this delayed effect. Through this dynamic causal model, researchers can establish equations to mathematically express the dependencies between variables. The construction of such a model requires not only precise mathematical descriptions but also full consideration of various factors influencing the results, ensuring that the comprehensive use of the model truly reflects the complexity of the market, thereby obtaining accurate and reliable results in the subsequent inference stage.

[0110] As described in step S502 above, the current market state is obtained, and preset intervention measures are applied to the treatment variables in the calculation framework based on the current market state. The current market state includes the specific conditions of various dynamic factors in the market, such as changes in user preferences, competitor market actions, consumer purchase intentions, and changes in the overall market environment. This information can be obtained through a data acquisition system, sourced from multiple channels such as real-time monitoring data, market research results, and social media activities, ensuring that the information obtained accurately reflects the current market situation. After obtaining the current market state, the next step is to apply preset intervention measures to the treatment variables defined in the calculation framework. These intervention measures can be price adjustments, changes in promotional efforts, or changes in product distribution channels. Selecting appropriate preset intervention measures requires prior market analysis and causal reasoning to ensure that the selected measures can produce the expected results in the current market environment. When implementing interventions, the different impacts that different measures may have on core outcome variables (such as user purchase intentions) must be fully considered. By substituting these preset measures into the model, the system can prepare for deduction and analyze the effect of the current intervention in the current market state, thereby providing a scientific basis for strategy optimization.

[0111] As described in step S503 above, based on the preset intervention measures, the probability distribution changes of the core outcome variable within a set time window are deduced using the computational framework. The core outcome variable is the user's purchase intention or market reaction. By deducing the causal interactions between variables, it is possible to predict how specific intervention measures will affect the changing trend of the core outcome variable, thereby providing data support for subsequent decision-making. In the specific deduction process, the previously constructed structural causal model is run, taking into account time lag parameters, to simulate the changes of the core outcome variable after the implementation of the intervention measures under the current market conditions. The deduction process typically uses numerical methods for solving, such as Monte Carlo simulation or Bayesian updates, which can generate multiple possible predictions for the core outcome variable. At this point, the system will output the predicted probability distribution of the core outcome variable under different conditions within the set time window, providing decision-makers with different possible outcomes and risk expectations from the implementation of the strategy.

[0112] As described in step S504 above, the expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures are output as the causal inference result. The expected change typically refers to the difference between the predicted values ​​of the core outcome variable before and after the intervention. For example, if the user's purchase intention index is predicted to be 80 before the intervention and 90 after the intervention, the expected change is +10. This quantitative indicator helps decision-makers intuitively understand the expected effect of the intervention. Secondly, the confidence interval reflects the uncertainty of the expected change, which can be calculated using statistical methods, such as standard error and t-distribution. In practical applications, the confidence interval not only provides the center point of the expected change (e.g., the predicted change in most cases) but also gives the upper and lower limits. These parameters represent the range of variation of the change at a certain confidence level. Combining these two results forms a complete causal inference result, which can effectively enhance the scientific nature of decision-making and provide a basis for subsequent strategy adjustments.

[0113] In one embodiment, step S6, which involves inputting the causal reasoning result into a preset large-scale language model to generate a target market strategy for the target competitor, includes:

[0114] S601: Combine the expected change and confidence interval under the preset intervention measures in the causal reasoning results with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt;

[0115] S602: Input the strategy generation prompt into a preset large language model to generate multiple strategy options; wherein each strategy option includes an action plan, an expected quantitative effect indicator and a corresponding implementation risk assessment;

[0116] S603: Calculate a weighted score based on the expected quantitative performance indicators in each of the strategy options and the implementation risk assessment;

[0117] S604: Select the strategy option with the highest score as the target market strategy.

[0118] As described in step S601 above, the expected change and confidence interval under the preset intervention measures in the causal inference result are combined with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt. The expected change and its confidence interval under the preset intervention measures are obtained from the causal inference result. The expected change reflects the change in core outcome variables (such as purchase intention or sales revenue) after implementing a specific market strategy, while the confidence interval provides an assessment of the uncertainty of this change. Next, user characteristics and market characteristics are also the basis for constructing the prompt. User characteristics may include age, gender, purchase history, etc., while market characteristics include market popularity, competitive environment, etc. Integrating this information with the causal inference result can form a structured prompt, clearly informing the large language model which key factors should be considered in generating the strategy. Ultimately, the generated structured strategy prompt will contain comprehensive information about users, the market, and expected effects to optimize the generation capabilities of the large language model.

[0119] As described in step S602 above, the strategy generation prompt is input into a pre-set large-scale language model to generate multiple strategy options. Each strategy option includes an action plan, an expected quantitative performance indicator, and a corresponding implementation risk assessment. The pre-trained large-scale language model parses the input structured prompt, identifying key points such as user characteristics, market dynamics, expected results, and corresponding risks. By performing contextual analysis on this information, the model can infer the optimal response measures under different market scenarios. Then, the model generates multiple strategy options based on the input. Each strategy option should contain three main components: an action plan, an expected quantitative performance indicator, and an implementation risk assessment. The action plan describes the specific steps for implementing the market strategy, such as increasing advertising budgets or adjusting product pricing. The expected quantitative performance indicator aims to predict the possible effects after implementing the plan, such as a percentage increase in sales or an increase in market share. In addition, the implementation risk assessment considers factors such as market fluctuations and user acceptance, helping decision-makers to have leeway when choosing a strategy.

[0120] As described in step S603 above, a weighted score is calculated based on the expected quantitative effect indicator and the implementation risk assessment in each strategy option. Specifically, the comprehensive score = w1 × expected effect + w2 × (1 (Risk Level). A scoring standard needs to be established, which typically summarizes several key factors affecting expected results and risks. These factors may include the expected returns from the strategy, market share growth, rate of return, and implementation risk level. The expected effects of each strategy are then elaborated and quantitatively evaluated based on historical case studies, data analysis, and market sentiment. Secondly, since the importance of expected effects and risk factors varies across different strategies, the system should set appropriate weighting coefficients to reflect the influence of different factors. This process can employ data normalization techniques to ensure that all scores are on the same order of magnitude. Simultaneously, the weight of implementation risk can be dynamically adjusted based on industry practices or expert opinions. Through this weighted scoring process, the system will generate a comprehensive score for each strategy option, assessing its effectiveness.

[0121] As described in step S604 above, the strategy option with the highest score is selected as the target market strategy. The weighted scores of each strategy option are aggregated to determine the optimal strategy. The strategy with the highest score typically represents the best balance between expected results and implementation risks. From a business decision-making perspective, selecting a high-scoring strategy can minimize potential risks while ensuring high profitability. After selecting the highest-scoring strategy, a final market response plan is generated. This plan details the strategy's implementation steps, timeframe, and expected goals. For example, if the highest-scoring strategy is "increase advertising budget to increase market share," the subsequent plan will include budget allocation, timeline, and marketing channels. To further improve the rationality and operability of the decision, a monitoring plan for the implementation of the selected strategy may also be generated. This involves real-time evaluation of the implementation effects, ensuring that the company can flexibly respond to market changes and adjust strategies in a timely manner to optimize results.

[0122] In one embodiment, before step S6, which involves inputting the causal inference result into a preset large-scale language model to generate a target market strategy for the target competitor, the method further includes:

[0123] S511: Real-time acquisition of regional cultural activities related to the target market;

[0124] S512: Extract the regional cultural characteristics of the aforementioned regional cultural activities;

[0125] S513: Analyze the regional cultural characteristics to obtain regional cultural rules;

[0126] S514: Match the regional cultural rules with the preset general strategy parameter baseline, and calculate the parameter adjustment amount according to the rule content;

[0127] S515: Adjust the initial large-scale language model according to the parameter adjustment amount to obtain the preset large-scale language model.

[0128] As described in step S511 above, real-time acquisition of regional cultural events relevant to the target market is required. These cultural events include local festivals, traditions, important anniversaries, and other events that may influence consumer behavior. This information can be acquired in various ways, such as by accessing social media platforms, news websites, locally published event schedules, or through specific cultural event databases. In a rapidly changing market environment, understanding upcoming cultural events can help businesses be more forward-thinking in strategy development. For example, consumer shopping behavior, preferences, and needs often change before traditional holidays; the collected data needs to include the type, time, location, scale of the event, and user participation in related activities.

[0129] As described in step S512 above, the regional cultural characteristics of the regional cultural activities are extracted. Through natural language processing, the system analyzes text data collected from various data sources (such as social media posts, news reports, and announcements of local cultural activities). The goal is to identify and extract important features related to market behavior. For example, consumers in a certain region may be more sensitive to discounts during specific holidays than in other regions; this feature, after extraction, will become important target data. Secondly, the extraction process also needs to consider the background and social context of the cultural activities to avoid biased understanding and analysis.

[0130] As described in step S513 above, the regional cultural characteristics are analyzed to obtain regional cultural rules. The goal of this step is to further analyze the extracted regional cultural characteristics to generate specific regional cultural rules. Regional cultural rules are a more systematic and structured form of knowledge that allows companies to clarify the impact of cultural characteristics on consumer behavior when formulating market strategies. For example, a region may have early promotional activities on specific holidays or a preference for specific products. By aggregating and analyzing the previously extracted regional cultural characteristics, common patterns and regularities can be found. Statistical analysis methods (such as association rule mining) can be used to discover the connections between these characteristics. The results of these analyses will constitute "regional cultural rules," specifically expressed as "If a certain cultural activity occurs, consumers' demand for the corresponding products will increase" or "On a certain holiday, users' sensitivity to discounts increases to a certain level."

[0131] As described in step S514 above, the regional cultural rules are matched with a preset general strategy parameter baseline, and the parameter adjustment amount is calculated according to the rule content. The general strategy parameter baseline represents the strategy parameters that enterprises typically adopt in the absence of a special cultural background, such as standard discount levels and promotion duration. By matching regional cultural rules, the system can identify which parameters need to be adjusted to adapt to a specific cultural environment. For example, a certain region may tend to focus more on discount intensity than promotion duration during major holidays. The system needs to identify these differences and calculate the corresponding "parameter adjustment amount". This process usually relies on mathematical models and rule engines, combined with business logic for correction. During the calculation process, the system quantifies and adjusts the parameters according to the specific requirements of the regional cultural rules.

[0132] As described in step S515 above, the initial large-scale language model is adjusted according to the parameter adjustment amount to obtain the preset large-scale language model. Large-scale language models (such as the GPT series) typically have powerful text generation and understanding capabilities, but in actual market strategy generation, if they cannot adapt to the cultural rules of a specific market, their output may lack targeting and effectiveness. First, the initial model needs to be fine-tuned or specific training data needs to be added. This process may include introducing new cultural feature sets to ensure that the large-scale language model can effectively consider these cultural factors when generating strategies. In addition, the model's adaptability to new inputs can be enhanced by changing the model's hyperparameters, such as the learning rate or the number of training epochs. Through this effective adjustment, enterprises can use large-scale language models to generate various targeted market strategies, ensuring the localization and cultural adaptability of the strategies, enabling enterprises to gain an advantage in a highly competitive environment.

[0133] Reference Figure 3 The present invention also provides a competitor market strategy generation device, the device comprising:

[0134] The acquisition module 902 is used to acquire real-time behavioral data of multiple users on various platforms, as well as market data of the target market and competitor data of the target competitors;

[0135] The comparison module 904 is used to perform cross-domain semantic alignment and fusion of multiple real-time behavioral data through a comparison learning algorithm to obtain a unified user profile for each user.

[0136] Extraction module 906 is used to extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0137] The conversion module 908 is used to convert the user characteristics, market characteristics and competitor characteristics into structured causal variables, and to construct a dynamic causal graph based on the structured causal variables using a structured causal model.

[0138] The reasoning module 910 is used to perform counterfactual reasoning based on the dynamic causal graph and generate causal reasoning results;

[0139] The generation module 912 is used to input the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor.

[0140] In one embodiment, the comparison module 904 includes:

[0141] The user information acquisition submodule is used to acquire user information for each user and to mark each real-time behavior data based on the user information.

[0142] The contrastive learning model construction submodule is used to construct an encoder-based contrastive learning model, which uses real-time behavior data from the same user but different platforms as positive sample pairs for semantic convergence, and uses real-time behavior data from different users as negative sample pairs for semantic distance, in order to train the encoder.

[0143] The mapping submodule is used to input multiple real-time behavioral data of each user into the trained encoder, map them into a unified vector space, and generate a multi-dimensional user feature vector for the user, which serves as the unified user profile for the user.

[0144] In one embodiment, the extraction module 906 includes:

[0145] The parsing submodule is used to parse multilingual competitor announcement text from the target market in real time using a pre-trained multilingual BERT model, identify key action types and quantification parameters in the announcement, and generate competitor features.

[0146] The customer flow change data extraction submodule is used to perform time-series analysis on satellite images of the relevant areas of the target market using a computer vision model to extract customer flow change data of offline stores;

[0147] The weighted fusion submodule is used to perform weighted fusion based on the competitor characteristics and the customer flow change data to generate a comprehensive market popularity index as the market characteristic.

[0148] In one embodiment, the conversion module 908 includes:

[0149] The market popularity index extraction submodule is used to extract the action type and intensity parameters from the competitor features and map them as treatment variables, extract the purchase intention index from the user features and map it as a core outcome variable, and extract the market popularity index from the market features and map it as a moderating variable.

[0150] The basic graph structure acquisition submodule is used to construct a directed acyclic graph that takes the treatment variable as the cause, is influenced by the moderating variable, and acts on the core outcome variable, thereby obtaining the basic graph structure of the structural causal model.

[0151] The dynamic causal graph acquisition submodule is used to introduce a preset time delay parameter into the causal path between the treatment variable and the core outcome variable in the basic graph structure, thereby obtaining the dynamic causal graph.

[0152] In one embodiment, the inference module 910 includes:

[0153] The structural causal model computation framework construction submodule is used to construct a structural causal model computation framework containing the time delay parameters based on the dynamic causal graph.

[0154] The market status acquisition submodule is used to acquire the current market status and apply preset intervention measures to the treatment variables in the calculation framework based on the current market status.

[0155] The deduction submodule is used to deduce the probability distribution change of the core outcome variable within a set time window based on the preset intervention measures and through the calculation framework.

[0156] The output submodule is used to output the expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures, as the result of the causal inference.

[0157] In one embodiment, the generation module 912 includes:

[0158] The strategy generation prompt construction submodule is used to combine the expected change and confidence interval under the preset intervention measures in the causal inference results with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt.

[0159] The strategy option generation submodule is used to input the strategy generation prompt into a preset large language model to generate multiple strategy options; wherein each strategy option includes an action plan, an expected quantitative effect indicator, and a corresponding implementation risk assessment.

[0160] The weighted scoring submodule is used to perform weighted scoring based on the expected quantitative effect indicators and the implementation risk assessment in each of the strategy options;

[0161] The selection submodule is used to select the strategy option with the highest score as the target market strategy.

[0162] In one embodiment, the competitor's market strategy generation device further includes:

[0163] The regional cultural activities acquisition module is used to acquire regional cultural activities related to the target market in real time.

[0164] A regional cultural feature extraction module is used to extract the regional cultural features of the said regional cultural activities;

[0165] A regional cultural rules acquisition module is used to parse the regional cultural characteristics to obtain regional cultural rules;

[0166] The parameter adjustment calculation module is used to match the regional cultural rules with the preset general strategy parameter baseline, and calculate the parameter adjustment amount according to the rule content.

[0167] The large language model acquisition module is used to adjust the initial large language model according to the parameter adjustment amount to obtain the preset large language model.

[0168] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a method for generating market strategies for competitors. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to execute a method for generating market strategies for competitors. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0169] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps:

[0170] Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products;

[0171] By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user.

[0172] Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0173] The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model.

[0174] Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results;

[0175] The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor.

[0176] By integrating real-time data from multiple platforms and constructing dynamic user profiles, the system can accurately capture the impact of market changes on user behavior, effectively overcome the limitations of traditional prediction models, and significantly improve prediction accuracy. By adopting a method that combines causal reasoning and dynamic modeling, it breaks through the rigid processing mode of traditional rule engines, enabling intelligent analysis of market competition and automatic generation of optimal response strategies, greatly shortening response time. Finally, through an intelligent strategy generation mechanism, it achieves high efficiency and precision in the decision-making process, not only significantly improving strategy generation efficiency but also ensuring the scientific nature and applicability of strategy solutions, providing reliable technical support for enterprises to maintain a competitive advantage in a dynamic market environment.

[0177] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps:

[0178] Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products;

[0179] By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user.

[0180] Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively.

[0181] The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model.

[0182] Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results;

[0183] The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor.

[0184] By integrating real-time data from multiple platforms and constructing dynamic user profiles, the system can accurately capture the impact of market changes on user behavior, effectively overcome the limitations of traditional prediction models, and significantly improve prediction accuracy. By adopting a method that combines causal reasoning and dynamic modeling, it breaks through the rigid processing mode of traditional rule engines, enabling intelligent analysis of market competition and automatic generation of optimal response strategies, greatly shortening response time. Finally, through an intelligent strategy generation mechanism, it achieves high efficiency and precision in the decision-making process, not only significantly improving strategy generation efficiency but also ensuring the scientific nature and applicability of strategy solutions, providing reliable technical support for enterprises to maintain a competitive advantage in a dynamic market environment.

[0185] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0186] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0187] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating a market strategy for competing products, characterized in that, The method includes: Acquire real-time behavioral data of multiple users across various platforms, as well as market data for the target market and competitor data for the target products; By using a contrastive learning algorithm, multiple real-time behavioral data are cross-domain semantically aligned and fused to obtain a unified user profile for each user. Extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively. The user characteristics, market characteristics, and competitor characteristics are converted into structured causal variables, and a dynamic causal graph is constructed based on the structured causal variables using a structured causal model. Based on the dynamic causal graph, counterfactual reasoning is performed to generate causal reasoning results; The causal reasoning results are input into a pre-set large-scale language model to generate a target market strategy for the target competitor. The step of converting the user characteristics, market characteristics, and competitor characteristics into structured causal variables, and constructing a dynamic causal graph based on the structured causal variables using a structured causal model, includes: Extract the action type and intensity parameters from the competitor features and map them as treatment variables; extract the purchase intention index from the user features and map it as a core outcome variable; extract the market popularity index from the market features and map it as a moderating variable. Construct a directed acyclic graph that takes the treatment variable as the cause, is influenced by the moderating variable, and acts on the core outcome variable, to obtain the basic graph structure of the structural causal model; In the basic graph structure, a preset time lag parameter is introduced for the causal path between the treatment variable and the core outcome variable, thereby obtaining the dynamic causal graph; The step of performing counterfactual reasoning based on the dynamic causal graph to generate causal reasoning results includes: Based on the dynamic causal graph, a structural causal model computation framework including the time delay parameters is constructed; Obtain the current market status and apply preset intervention measures to the treatment variables in the calculation framework based on the current market status; Based on the preset intervention measures, the probability distribution changes of the core outcome variables within a set time window are deduced through the computational framework. The expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures are output as the causal inference result. The step of inputting the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor includes: The expected change and confidence interval under the preset intervention measures in the causal reasoning results are combined with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt. The strategy generation prompts are input into a preset large language model to generate multiple strategy options; each strategy option includes an action plan, expected quantitative performance indicators, and corresponding implementation risk assessment. A weighted score is assigned based on the expected quantitative performance indicators in each of the strategy options and the implementation risk assessment. Select the strategy option with the highest score as the target market strategy.

2. The method for generating market strategies for competitors according to claim 1, characterized in that, The step of performing cross-domain semantic alignment and fusion of multiple real-time behavioral data using a contrastive learning algorithm to obtain a unified user profile for each user includes: Obtain user information for each user, and label each real-time behavior data based on the user information; A contrastive learning model based on an encoder is constructed, in which real-time behavior data from the same user but different platforms are used as positive sample pairs for semantic convergence, and real-time behavior data from different users are used as negative sample pairs for semantic distance, in order to train the encoder. Multiple real-time behavioral data of each user are input into the trained encoder and mapped into a unified vector space to generate a multi-dimensional user feature vector for that user, which serves as the unified user profile for that user.

3. The method for generating market strategies for competitors according to claim 1, characterized in that, In the step of extracting feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features respectively, the step of extracting market features from the market data includes: Using a pre-trained multilingual BERT model, multilingual competitor announcement texts from the target market are parsed in real time to identify key action types and quantification parameters in the announcements, thereby generating competitor features. By using computer vision models to perform time-series analysis on satellite images of relevant areas of the target market, data on changes in customer flow at offline stores can be extracted. Based on the competitor characteristics and the customer flow change data, a weighted fusion is performed to generate a comprehensive market popularity index, which serves as the market characteristic.

4. The method for generating market strategies for competitors according to claim 1, characterized in that, Before the step of inputting the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor, the method further includes: Real-time acquisition of regional cultural activities related to the target market; Extract the regional cultural characteristics of the aforementioned regional cultural activities; The regional cultural characteristics are analyzed to obtain regional cultural rules; The regional cultural rules are matched with a preset general strategy parameter baseline, and the parameter adjustment amount is calculated based on the rule content. The initial large-scale language model is adjusted according to the parameter adjustment amount to obtain the preset large-scale language model.

5. A competitor's market strategy generation device, characterized in that, The device includes: The acquisition module is used to acquire real-time behavioral data of multiple users on various platforms, as well as market data of the target market and competitor data of the target competitors; The comparison module is used to perform cross-domain semantic alignment and fusion of multiple real-time behavioral data through a comparison learning algorithm to obtain a unified user profile for each user. The extraction module is used to extract feature data from the unified user profile, the market data, and the competitor data to obtain user features, market features, and competitor features, respectively. The conversion module is used to convert the user characteristics, market characteristics, and competitor characteristics into structured causal variables, and to construct a dynamic causal graph based on the structured causal variables using a structured causal model. The reasoning module is used to perform counterfactual reasoning based on the dynamic causal graph and generate causal reasoning results. The generation module is used to input the causal reasoning results into a preset large-scale language model to generate a target market strategy for the target competitor. The conversion module includes: The market popularity index extraction submodule is used to extract the action type and intensity parameters from the competitor features and map them as treatment variables, extract the purchase intention index from the user features and map it as a core result variable, and extract the market popularity index from the market features and map it as a moderating variable. The basic graph structure acquisition submodule is used to construct a directed acyclic graph that takes the treatment variable as the cause, is influenced by the regulating variable, and acts on the core outcome variable, thereby obtaining the basic graph structure of the structural causal model. The dynamic causal graph acquisition submodule is used to introduce a preset time delay parameter into the causal path between the treatment variable and the core outcome variable in the basic graph structure, thereby obtaining the dynamic causal graph; The reasoning module includes: The structural causal model computation framework construction submodule is used to construct a structural causal model computation framework containing the time delay parameters based on the dynamic causal graph. The market status acquisition submodule is used to acquire the current market status and apply preset intervention measures to the treatment variables in the calculation framework based on the current market status. The deduction submodule is used to deduce the probability distribution change of the core outcome variable within a set time window based on the preset intervention measures and through the calculation framework. The output submodule is used to output the expected change of the core outcome variable and its corresponding confidence interval under the preset intervention measures, as the result of the causal inference. The generation module includes: The strategy generation prompt construction submodule is used to combine the expected change and confidence interval under the preset intervention measures in the causal inference results with the corresponding user characteristics and market characteristics to construct a structured strategy generation prompt. The strategy option generation submodule is used to input the strategy generation prompt into a preset large language model to generate multiple strategy options; wherein each strategy option includes an action plan, expected quantitative performance indicators and corresponding implementation risk assessment. The weighted scoring submodule is used to perform weighted scoring based on the expected quantitative effect indicators and the implementation risk assessment in each of the strategy options; The selection submodule is used to select the strategy option with the highest score as the target market strategy.

6. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, causes the processor to perform the steps of the competitor's market strategy generation method as described in any one of claims 1 to 4.

7. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the competitor's market strategy generation method as described in any one of claims 1 to 4.

Images