Marketing strategy recommendation method and device based on big data analysis

Abstract

This application discloses a marketing strategy recommendation method and apparatus based on big data analysis, relating to the field of big data analysis technology. The method includes: collecting multi-source behavioral big data to construct a dynamic user profile; calculating behavioral deviation based on information entropy to quantify demand status and calculating marketing tolerance; constructing a strategy association network and using a graph neural network to retrieve and generate a candidate strategy set; constructing a simulation environment, mapping the dynamic user model to a simulation agent, and performing Monte Carlo simulations combined with competitive interference to generate predicted effect data; using marketing tolerance as a constraint, combining the predicted effect data to select target marketing strategies and updating the model in a closed loop. This invention solves the problems of inaccurate user demand judgment, insufficient quantification of promotion tolerance, and difficulty in predicting strategy effects in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a marketing strategy recommendation method and apparatus based on big data analytics. Background Technology

[0002] With the booming development of the digital cultural industry, cultural development companies have accumulated massive amounts of user behavior data, covering multi-source heterogeneous data such as online browsing, offline performance viewing, and member interaction. However, current cultural promotion strategies often rely on human experience or simple rule engines, such as sending mass push notifications based on the user's most recent ticket purchase time. This method has a significant lag and struggles to capture real-time changes in users' cultural consumption needs.

[0003] Existing big data-based marketing strategies primarily focus on product recommendations in the e-commerce sector, lacking adaptation to the specific characteristics of cultural consumption scenarios. Specifically: First, cultural consumption is characterized by being non-essential and highly experience-driven, making it difficult to accurately assess user needs and often overlooking the strong intent signals implied by sudden changes in user behavior patterns (such as frequent searches for specific shows). Second, cultural consumers are highly sensitive to psychological fatigue (i.e., tolerance) from promotional information, and existing frequency controls lack quantitative models for assessing user fatigue levels, easily leading to member unsubscriptions or churn. Third, cultural promotion strategies are mostly static combinations, lacking mechanisms for predicting effects in complex market competition environments (such as competition from other shows and holiday effects).

[0004] Therefore, there is an urgent need for a marketing strategy recommendation method that can accurately quantify users' cultural needs and tolerance levels, and can conduct effect simulations before strategy implementation. Summary of the Invention

[0005] This invention provides a marketing strategy recommendation method and apparatus based on big data analysis. This invention solves the problems of inaccurate judgment of user needs, insufficient quantification of promotion tolerance, and difficulty in predicting the effect of strategies in the existing technology.

[0006] In a first aspect, embodiments of the present invention provide a marketing strategy recommendation method based on big data analysis, the method comprising:

[0007] Collect multi-source behavioral big data of users, and construct dynamic user profiles that include time-series features and attribute features based on the multi-source behavioral big data;

[0008] Based on the dynamic user profile, the deviation between the user's current behavior pattern and historical behavior pattern is calculated to quantify the user's current needs and to calculate the user's marketing tolerance.

[0009] Construct a policy association network containing several policy meta-actions, and based on the dynamic user profile and the user's current demand status, retrieve matches in the policy association network to generate a candidate policy set;

[0010] A simulation environment is constructed, the dynamic user profile is mapped to the simulation agent of the simulation environment, the candidate strategy set is input into the simulation environment for simulation, and prediction effect data is generated.

[0011] Using the marketing tolerance as a constraint and combining the predicted effect data, a target marketing strategy is selected from the candidate strategy set, and the feedback data after the target marketing strategy is executed is sent back to the dynamic user profile for updating.

[0012] The technical solution provided in this application has at least the following beneficial effects:

[0013] Addressing the characteristics of cultural consumption as "non-essential and experience-driven," this approach constructs dynamic user profiles incorporating temporal and attribute features. By calculating the deviation between current and historical behavioral patterns based on information entropy, it accurately identifies abrupt changes in user behavior (such as a shift from casual browsing to strong interest in specific shows), thus distinguishing between "strong intent" and "wandering." This allows for the keen capture of users' immediate viewing impulses, resolving the issues of insufficient accuracy in judging user needs and ignoring strong intent signals in existing technologies. This enables a shift from "blindly pushing" to "acting on demand." To address the pain point of users' sensitivity to promotional information and their tendency to unsubscribe, a marketing tolerance index is introduced. This index comprehensively considers the time interval since the last marketing, the cumulative number of touchpoints, and the number of negative feedback behaviors such as complaints and unsubscriptions. Before the recommendation strategy is implemented, a "safety valve" assessment is conducted, forcing users in a highly fatigued state into a "silent period," effectively avoiding the negative impact of excessive marketing. This approach addresses employee dissatisfaction and churn, achieving a balance between commercial revenue and user experience, and resolving the lack of a quantitative model for psychological fatigue in existing frequency control methods. To address the intense competition in the cultural market (due to competing shows, holiday effects, etc.) and the prevalence of statically combined strategies, a simulation environment incorporating both simulated and competing agents was constructed. Using Monte Carlo simulation methods and incorporating competitor interference events, multi-dimensional effect data (such as predicted conversion rates and churn risk indices) after strategy execution were pre-rendered in a virtual environment. This allows strategy selection to be based not only on historical correlations but also on predictions of future complex market environments, significantly improving the scientific rigor and risk resistance of decision-making, and resolving the lack of an effect prediction mechanism in existing technologies. By transmitting feedback data after strategy execution (such as ticket purchase behavior, cancellation behavior, and browsing depth) back to the dynamic user profile, a closed-loop update and self-iteration of the dynamic user profile was achieved. This ensures that cultural marketing recommendations can dynamically adjust according to the evolution of users' artistic preferences, guaranteeing the continuous effectiveness and accuracy of cultural promotion services.

[0014] In one optional implementation, multi-source behavioral big data of users is collected, and a dynamic user profile containing temporal and attribute features is constructed based on the multi-source behavioral big data, including:

[0015] Collect multi-source behavioral big data of users, including online behavior data, business transaction data, and interactive feedback data;

[0016] Multi-source behavioral big data is preprocessed to obtain preprocessed standard data;

[0017] Extract time-series and attribute features from the preprocessed standard data, and construct a dynamic user profile containing time-series and attribute features. The time-series features include current behavior sequences and historical behavior sequences.

[0018] In one optional implementation, based on the dynamic user profile, the deviation between the user's current behavior pattern and historical behavior pattern is calculated to quantify the user's current demand status, and the user's marketing tolerance is calculated, including:

[0019] Based on the current behavior sequence in the temporal features of the dynamic user profile, calculate the current behavior entropy corresponding to the user's current behavior pattern;

[0020] Based on the historical behavior sequence in the temporal features of the dynamic user profile, calculate the historical baseline entropy corresponding to the user's historical behavior pattern;

[0021] Based on the current behavioral entropy and the historical baseline entropy, the deviation between the user's current behavioral pattern and the historical behavioral pattern is calculated to quantify the user's current demand status.

[0022] Based on the attribute characteristics of the dynamic user profile, the user's marketing tolerance is calculated.

[0023] In one alternative implementation, the formula for the current behavior entropy is:

[0024]

[0025] In the formula, The entropy of the current behavior; Let j be the probability of the j-th action occurring; j, The behavior indicator is the set of behavior categories obtained by mapping the current behavior sequence; m is the total number of behaviors in the set of behavior categories; For the j-th behavior category in the set, Behavior The number of times it appears;

[0026] The formula for the demand state is:

[0027]

[0028] In the formula, For deviation degree; The historical baseline entropy; if If yes, the demand state is a strong intent state; otherwise, the demand state is a roaming state. The deviation threshold;

[0029] The formula for the marketing tolerance is:

[0030]

[0031] In the formula, For marketing tolerance; This refers to the time interval since the last marketing campaign; This represents the cumulative number of times the message has been received. This represents the number of negative feedback behaviors. This is the tolerance weighting coefficient.

[0032] In one optional implementation, a policy association network containing several policy meta-actions is constructed, and based on the dynamic user profile and the user's current demand state, matches are retrieved in the policy association network to generate a candidate policy set, including:

[0033] The marketing strategy is deconstructed into meta-action nodes corresponding to the strategy meta-actions, and a strategy association network is constructed based on the meta-action nodes. The edges in the strategy association network represent the association weights between meta-action nodes, and the node types of the meta-action nodes include channels, materials, rights and interests, and timing.

[0034] The implicit association vectors between several meta-action nodes in the strategy association network are extracted using a pre-built graph neural network.

[0035] Extract the user's preference feature vector from the dynamic user profile. In the policy association network, calculate the similarity between the preference feature vector and each implicit association vector. Retain the top-K matching meta-action nodes with the highest similarity scores to form an initial policy set constructed from the policy meta-actions corresponding to all matching meta-action nodes.

[0036] Based on the user's current needs, the initial strategy set is dynamically modified and expanded to generate a candidate strategy set.

[0037] In one optional implementation, a simulation environment is constructed, the dynamic user profile is mapped to a simulation agent of the simulation environment, the candidate strategy set is input into the simulation environment for simulation, and prediction effect data is generated, including:

[0038] Construct a simulation simulation environment, which includes a simulation agent that simulates the behavioral logic of a real user group and a competitive agent that simulates a market competition environment;

[0039] The dynamic user profile is mapped to a simulation agent in the simulation simulation environment. Based on the attribute characteristics of the dynamic user profile, the user's current demand state, and marketing tolerance, the initial state parameters of the simulation agent are set. The initial state parameters include the initial intent intensity parameter set according to the demand state, the initial fatigue sensitivity parameter set according to the marketing tolerance, and the initial decision preference vector set according to the attribute characteristics.

[0040] The candidate strategy set is input into a simulation environment for simulation and simulation to generate prediction effect data.

[0041] In one optional implementation, the candidate strategy set is input into a simulation environment for simulation and simulation to generate prediction effect data, including:

[0042] The candidate policy set is injected into the simulation agent of the simulation environment, and the competing agent of the simulation environment is used to generate interference events;

[0043] Using a simulation agent, the immediate utility is calculated based on the current state of the simulation agent, the set of candidate policies, and the interference events, using the following formula:

[0044]

[0045] In the formula, Let k be the k-th candidate policy in the candidate policy set at time t. The corresponding immediate utility; The current intent strength parameter of the simulation agent at time t represents the current state of the simulation agent. The current fatigue sensitivity parameter of the simulation agent at time t represents the current state of the simulation agent. The k-th candidate strategy at time t Chinese rights and interests actions Value score of rights and interests action nodes; The k-th candidate strategy at time t Chinese material element action The current decision preference vector of the simulated agent The material matching degree, the current decision preference vector belongs to the current state of the simulation agent; The k-th candidate strategy at time t Central Channel Yuan Action Channel node disturbance index; This is a disruptive event; This is the immediate utility weighting coefficient;

[0046] Using a simulation agent, behavioral decisions are made based on immediate utility and the current state of the simulation agent, generating user behavior at the current moment;

[0047] The Monte Carlo method is used to repeatedly make behavioral decisions, and the generated user behaviors are statistically analyzed to generate predictive performance data.

[0048] In one alternative implementation, the predicted performance data includes predicted conversion rate, predicted click-through rate, churn risk index, and predicted retention time.

[0049] In one optional implementation, using the marketing tolerance as a constraint and combining the predicted effect data, a target marketing strategy is selected from the candidate strategy set, and the feedback data after the execution of the target marketing strategy is sent back to the dynamic user profile for updating, including:

[0050] Using the aforementioned marketing tolerance as a constraint, if the marketing tolerance is less than the marketing tolerance threshold, it is determined that the user is in a state of high fatigue, the marketing strategy recommendation process is stopped, and the user waits for the next moment's marketing tolerance analysis.

[0051] If the marketing tolerance is greater than or equal to the marketing tolerance threshold, then based on the predicted effect data, a predefined objective function is used to select the target marketing strategy with the largest target value from the candidate strategy set, as shown in the formula:

[0052]

[0053] In the formula, For the k-th candidate policy in the candidate policy set The predicted performance data includes predicted conversion rate, predicted click-through rate, churn risk index, and predicted retention time. For the k-th candidate policy in the candidate policy set The overall benefit, i.e., the target value; These are the weight coefficients of the objective function;

[0054] Distribute the target marketing strategy to the outreach channels and collect user feedback data in real time after the target marketing strategy is implemented;

[0055] Based on the feedback data, the user's dynamic user profile is updated to obtain an updated dynamic user profile.

[0056] Secondly, embodiments of the present invention provide a marketing strategy recommendation device based on big data analysis, used to implement a marketing strategy recommendation method, the device comprising:

[0057] The big data acquisition unit is used to collect multi-source behavioral big data of users and construct dynamic user profiles containing time-series features and attribute features based on the multi-source behavioral big data.

[0058] The status assessment unit is used to calculate the deviation between the user's current behavior pattern and historical behavior pattern based on the dynamic user profile, so as to quantify the user's current demand status and calculate the user's marketing tolerance.

[0059] The strategy construction unit is used to construct a strategy association network containing several strategy meta-actions, and based on the dynamic user profile and the user's current demand status, to search for matches in the strategy association network and generate a candidate strategy set.

[0060] The simulation and deduction unit is used to construct a simulation and deduction environment, map the dynamic user profile to a simulation agent of the simulation and deduction environment, input the candidate strategy set into the simulation and deduction environment for simulation and deduction, and generate prediction effect data;

[0061] The decision optimization unit is used to select a target marketing strategy from the candidate strategy set based on the marketing tolerance as a constraint and the predicted effect data, and to send the feedback data after the target marketing strategy is executed back to the dynamic user profile for updating.

[0062] A third aspect of this invention provides an electronic device, which includes:

[0063] At least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0064] The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0065] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention;

[0067] Figure 2 This is a flowchart illustrating the steps of a marketing strategy recommendation method based on big data analysis provided in an embodiment of the present invention.

[0068] Figure 3 This is a functional unit diagram of a marketing strategy recommendation device based on big data analysis provided in an embodiment of the present invention. Detailed Implementation

[0069] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0070] The present invention will be further described below with reference to the accompanying drawings.

[0071] Reference Figure 1 , Figure 1This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0072] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0073] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0074] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating device, a data storage module, a network communication module, a user interface module, and electronic programs.

[0075] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the marketing strategy recommendation device based on big data analysis stored in the memory 1005 through the processor 1001 and executes the marketing strategy recommendation method based on big data analysis provided in the embodiment of the present invention.

[0076] Reference Figure 2 The embodiments of the present invention provide a marketing strategy recommendation method based on big data analysis, the method comprising:

[0077] S201: Collect multi-source behavioral big data of users, and construct a dynamic user profile that includes time-series features and attribute features based on the multi-source behavioral big data;

[0078] S202: Based on the dynamic user profile, calculate the deviation between the user's current behavior pattern and historical behavior pattern to quantify the user's current demand status and calculate the user's marketing tolerance.

[0079] S203: Construct a policy association network containing several policy meta-actions, and based on the dynamic user profile and the user's current demand status, retrieve matches in the policy association network to generate a candidate policy set;

[0080] S204: Construct a simulation environment, map the dynamic user profile to a simulation agent of the simulation environment, input the candidate strategy set into the simulation environment for simulation, and generate prediction effect data;

[0081] S205: Using the marketing tolerance as a constraint and combining the predicted effect data, select the target marketing strategy from the candidate strategy set, and send the feedback data after the target marketing strategy is executed back to the dynamic user profile for updating.

[0082] The technical solution provided in this application has at least the following beneficial effects:

[0083] Addressing the characteristics of cultural consumption as "non-essential and experience-driven," this approach constructs dynamic user profiles incorporating temporal and attribute features. By calculating the deviation between current and historical behavioral patterns based on information entropy, it accurately identifies abrupt changes in user behavior (such as a shift from casual browsing to strong interest in specific shows), thus distinguishing between "strong intent" and "wandering." This allows for the keen capture of users' immediate viewing impulses, resolving the issues of insufficient accuracy in judging user needs and ignoring strong intent signals in existing technologies. This enables a shift from "blindly pushing" to "acting on demand." To address the pain point of users' sensitivity to promotional information and their tendency to unsubscribe, a marketing tolerance index is introduced. This index comprehensively considers the time interval since the last marketing, the cumulative number of touchpoints, and the number of negative feedback behaviors such as complaints and unsubscriptions. Before the recommendation strategy is implemented, a "safety valve" assessment is conducted, forcing users in a highly fatigued state into a "silent period," effectively avoiding the negative impact of excessive marketing. This approach addresses employee dissatisfaction and churn, achieving a balance between commercial revenue and user experience, and resolving the lack of a quantitative model for psychological fatigue in existing frequency control methods. To address the intense competition in the cultural market (due to competing shows, holiday effects, etc.) and the prevalence of statically combined strategies, a simulation environment incorporating both simulated and competing agents was constructed. Using Monte Carlo simulation methods and incorporating competitor interference events, multi-dimensional effect data (such as predicted conversion rates and churn risk indices) after strategy execution were pre-rendered in a virtual environment. This allows strategy selection to be based not only on historical correlations but also on predictions of future complex market environments, significantly improving the scientific rigor and risk resistance of decision-making, and resolving the lack of an effect prediction mechanism in existing technologies. By transmitting feedback data after strategy execution (such as ticket purchase behavior, cancellation behavior, and browsing depth) back to the dynamic user profile, a closed-loop update and self-iteration of the dynamic user profile was achieved. This ensures that cultural marketing recommendations can dynamically adjust according to the evolution of users' artistic preferences, guaranteeing the continuous effectiveness and accuracy of cultural promotion services.

[0084] In one optional implementation, multi-source behavioral big data of users is collected, and a dynamic user profile containing temporal and attribute features is constructed based on the multi-source behavioral big data, including:

[0085] S2011: Collect multi-source behavioral big data of users, including online behavioral data, business transaction data, business transaction data, and interactive feedback data;

[0086] In this embodiment, online behavior data (Web / APP logs) is collected using a Software Development Kit (SDK) to capture fine-grained user behavior on digital cultural platforms (such as WeChat official accounts, mini-programs, and ticketing apps). Specifically, this includes: the duration of browsing program details pages, the timestamp of clicking the "Want to Watch / Favorite" button, search keywords (such as "drama" and "symphony tickets"), browsing history of electronic program guides, and interactive clicks on online exhibitions.

[0087] Business transaction data: Connect to the ticketing system database to collect users' historical ticket purchase records (show type, price range, seat preference), membership card recharge and points change data, and cultural and creative product purchase records;

[0088] Offline scenario data: Combine IoT devices (such as venue WiFi probes, turnstiles) to collect user visit frequency, dwell heat maps (such as which exhibition hall they stay in the longest), and usage records through smart cultural terminals (such as reading booths, digital interactive screens);

[0089] Interactive feedback data: Collects user comments on the official WeChat account, completed post-performance surveys, and complaint or inquiry records from the customer service system;

[0090] S2012: Preprocess multi-source behavioral big data to obtain preprocessed standard data;

[0091] In this embodiment, data cleaning and preprocessing are performed as follows:

[0092] Denoising and Completion: The collected raw data is cleaned to remove noisy data such as crawler traffic and abnormal high-frequency clicks; for missing key fields (such as age and gender), mean imputation or prediction imputation algorithm based on random forest is used to complete them;

[0093] User identification: Using ID mapping technology, the user's OpenID, APP registered mobile phone number, membership card number and offline gate recognition FaceID are normalized and mapped to generate a unique "cultural member ID", breaking down the data silos between online browsing and offline performance;

[0094] Data standardization: Numerical features (such as consumption amount and length of stay) are normalized using Min-Max, and categorical features (such as province and device type) are encoded using One-Hot encoding to form preprocessed standard data;

[0095] S2013: Extract the temporal and attribute features from the preprocessed standard data and construct a dynamic user profile containing the temporal and attribute features, wherein the temporal features include the current behavior sequence and the historical behavior sequence;

[0096] In this embodiment, temporal feature extraction employs a sliding time window mechanism.

[0097] Current Behavior Sequence Window: Set to the current time 24 hours prior; for example: [Search_Modern Dance - Browse_Schedule Page - Click_Member Benefits], used to capture the user's immediate urge to watch a performance;

[0098] Historical Behavior Sequence Window: Set to advance 365 days; for example: [Ticket Purchase_Drama_3 times - Ticket Purchase_Concert_1 time - Pickup_Lecture_Invalid], used to depict users' long-term cultural consumption habits and art preferences;

[0099] Attribute feature extraction:

[0100] Static tags: age group, occupation (inferred), geographical location (distance from venue).

[0101] Cultural profile tags: Art preference score (e.g., drama 0.8, folk music 0.2), spending power level (average ticket price), membership level, price sensitivity (whether you often buy discounted tickets);

[0102] Model storage: Build dynamic user profiles and store them in the form of key-value pairs in the Redis cluster, supporting real-time updates.

[0103] In one optional implementation, based on the dynamic user profile, the deviation between the user's current behavior pattern and historical behavior pattern is calculated to quantify the user's current demand status, and the user's marketing tolerance is calculated, including:

[0104] S2021: Based on the current behavior sequence in the temporal features of the dynamic user profile, calculate the current behavior entropy corresponding to the user's current behavior pattern;

[0105] Demand State Quantification (Behavioral Deviation Calculation Based on Information Entropy):

[0106] Behavior category mapping: Mapping the collected fine-grained behaviors to a predefined set of cultural consumption behaviors, such as mapping to {browsing details, searching, selecting seats, purchasing tickets, claiming coupons, canceling, and refunding tickets};

[0107] The formula for the current behavior entropy is:

[0108]

[0109] In the formula, The entropy of the current behavior; Let j be the probability of the j-th action occurring; j, The behavior indicator is the set of behavior categories obtained by mapping the current behavior sequence; m is the total number of behaviors in the set of behavior categories; For the j-th behavior category in the set, Behavior The number of times it appears;

[0110] Calculate the degree of disorder in the user's behavior within the current window.

[0111] Example scenario: If the user's behavior sequence is [browse play A, browse play A, select seats, select seats], it indicates that the behavior is highly focused on a single goal. A lower value indicates a clear objective and may suggest a "strong intention to purchase tickets" state.

[0112] If the user behavior sequence is [browsing concerts, browsing exhibitions, clicking on cultural and creative products, exiting], the behavior is scattered. A higher level indicates a state of "cultural wandering" without a clear consumption goal;

[0113] S2022: Based on the historical behavior sequence in the time-series features of the dynamic user profile, calculate the historical baseline entropy corresponding to the user's historical behavior pattern. The calculation method is the same as the current behavior entropy. The baseline entropy is calculated based on the user's long-term behavior habits, reflecting the user's usual browsing style.

[0114] S2023: Calculate the deviation between the user's current behavior pattern and historical behavior pattern based on the current behavior entropy and historical baseline entropy, in order to quantify the user's current demand status;

[0115] The formula for the demand state is:

[0116]

[0117] In the formula, For deviation degree; The historical baseline entropy; if If the user is browsing casually but is focused on finding tickets today, the system should immediately intervene to assist in conversion. Otherwise, it should be judged as a "roaming state" (the system should focus on stimulating interest and displaying high-quality content, rather than aggressively promoting the product). The deviation threshold;

[0118] S2024: Calculate the user's marketing tolerance based on the attribute characteristics of the dynamic user profile;

[0119] The formula for the marketing tolerance is:

[0120]

[0121] In the formula, For marketing tolerance; The interval between the last marketing campaign and the current marketing campaign indicates that cultural consumption is relatively infrequent and has a long recovery period. If the interval is less than 3 days, this item will score very low. This represents the cumulative number of outreach messages. It includes SMS messages, app push notifications, and official account template messages. The number of negative feedback behaviors, such as clicking "unfollow," "complain," "unsubscribe from SMS," or "frequently close pop-ups," has a very high weight and severely lowers tolerance.

[0122] In one optional implementation, a policy association network containing several policy meta-actions is constructed, and based on the dynamic user profile and the user's current demand state, matches are retrieved in the policy association network to generate a candidate policy set, including:

[0123] S2031: Deconstruct the marketing strategy into meta-action nodes corresponding to the strategy meta-actions, and construct a strategy association network based on the meta-action nodes. The edges in the strategy association network represent the association weights between meta-action nodes. The node types of the meta-action nodes include channels, materials, rights and interests, and timing.

[0124] In this embodiment, the channel nodes are: official account template messages, mini-program pop-ups, SMS, ticketing platform homepage banners, venue on-site large screens, and member community messages;

[0125] Material elements: high-definition stills, trailer videos, director interview audio, reviews by famous writers (text and images), performance schedule, and countdown text for special offers.

[0126] Benefits include: early bird discounts, member-only discounts, points redemption, second ticket at half price, tickets to the lead actor's meet-and-greet, and limited-edition merchandise.

[0127] Timing: One week before the premiere (pre-show period), the day of the performance (last chance to snag a spot), weekend evenings (prime time), members' birthdays / holidays;

[0128] Edge weight construction: Based on historical delivery data, calculate the collaborative conversion rate between different nodes; for example, it was found that there are high-weight connections between "theater users" and "trailer videos" and "member discounts";

[0129] S2032: Use a pre-built graph neural network to extract implicit association vectors between several meta-action nodes in the strategy association network;

[0130] In this embodiment, a pre-built graph neural network (Graph Sampling and Aggregating (GraphSAGE) or Graph Attention Network (GAT)) is used for training, with the adjacency matrix and node feature vectors as input. By aggregating information from neighboring nodes, the graph neural network can learn the implicit deep associations between nodes. For example, it may discover implicit associations, such as the fact that "modern dance drama" is often implicitly related to "niche experimental video materials" and "limited-time discounts".

[0131] S2033: Extract the user's preference feature vector from the dynamic user profile. In the policy association network, calculate the similarity between the preference feature vector and each implicit association vector. Retain the Top-K matching meta-action nodes with the highest similarity scores to form an initial policy set constructed from the policy meta-actions corresponding to all matching meta-action nodes. For example, {Channel: APP Push, Material: Suspenseful Plot Video, Benefit: Member Points Deduction}.

[0132] S2034: Based on the user's current needs, dynamically modify and expand the initial strategy set to generate a candidate strategy set;

[0133] In this embodiment, dynamic correction is used:

[0134] Regarding "strong intent state":

[0135] Automatically enhance "benefit nodes," such as prioritizing the inclusion of "limited-time seat reservation" or "high discount" strategies to accelerate conversion;

[0136] Remove "exploratory" content (such as "activities you might be interested in") and replace it with "decision aid" content (such as "seats for this session are almost sold out").

[0137] Regarding "roaming status":

[0138] The focus is on "highly attractive content elements" (such as exquisite exhibition posters and immersive VR clips) and "low-threshold benefits" (such as free lecture tickets and small discounts on first orders), aiming to stimulate user interest rather than directly drive sales.

[0139] The initial strategy set generated by the final combination is dynamically modified to form a candidate strategy set containing multiple combination schemes.

[0140] In one optional implementation, a simulation environment is constructed, the dynamic user profile is mapped to a simulation agent of the simulation environment, the candidate strategy set is input into the simulation environment for simulation, and prediction effect data is generated, including:

[0141] S2041: Construct a simulation simulation environment, which includes a simulation agent that simulates the behavioral logic of a real user group and a competitive agent that simulates a market competition environment;

[0142] In this embodiment, the environment is defined as follows: A sandbox environment is constructed to simulate the dynamics of a real market;

[0143] Simulated agent construction: "cloning" the user profile of a real user into a simulated agent;

[0144] Initial intent strength: set to high (0.8) for strong intent state and low (0.2) for roaming state;

[0145] Initial fatigue sensitivity: Set as the reciprocal of marketing tolerance. Users with low tolerance are highly sensitive to interference.

[0146] Initial decision preferences: Inherit the user's artistic preferences (e.g., a preference for serious dramas and a rejection of vulgar comedies);

[0147] Competitive Agent: Introducing agents that simulate competitors to randomly generate "interference events" during the simulation process, such as introducing "competitive interference events": such as "a competitor's theater in the same city showing the same type of show", "a large music festival held on the same weekend to divert audiences", "a popular movie being released";

[0148] S2042: Map the dynamic user profile to a simulation agent in the simulation simulation environment, and set the initial state parameters of the simulation agent according to the attribute characteristics of the dynamic user profile, the user's current demand state, and marketing tolerance. The initial state parameters include an initial intent strength parameter set according to the demand state (determined by the demand state, with a high value initialized for a strong intent state and a low value initialized for a roaming state), an initial fatigue sensitivity parameter set according to the marketing tolerance (determined by the reciprocal of the marketing tolerance, with a higher fatigue sensitivity parameter for lower tolerance), and an initial decision preference vector set according to the attribute characteristics (inherited from the user's attribute characteristics, which determines the agent's sensitivity to different materials and benefits).

[0149] S2043: Input the candidate strategy set into the simulation environment for simulation and generate prediction effect data.

[0150] In one optional implementation, the candidate strategy set is input into a simulation environment for simulation and simulation to generate prediction effect data, including:

[0151] S20431: Inject the candidate strategy set into the simulation agent of the simulation environment, and use the competing agent of the simulation environment to generate interference events;

[0152] In this embodiment, the strategies in the candidate strategy set are injected into the simulation environment one by one, and at the same time, the competing agent triggers interference events according to a certain probability distribution.

[0153] S20432: Using a simulation agent, calculate the immediate utility based on the current state of the simulation agent, the candidate policy set, and the interference events, using the following formula:

[0154]

[0155] In the formula, Let k be the k-th candidate policy in the candidate policy set at time t. The corresponding immediate utility; The current intent strength parameter of the simulation agent at time t represents the current state of the simulation agent. The current fatigue sensitivity parameter of the simulation agent at time t represents the current state of the simulation agent. The k-th candidate strategy at time t Chinese rights and interests actions Value score of rights and interests action nodes; The k-th candidate strategy at time t Chinese material element action The current decision preference vector of the simulated agent The material matching degree, the current decision preference vector belongs to the current state of the simulation agent; The k-th candidate strategy at time t Central Channel Yuan Action Channel node disturbance index; This is a disruptive event; This is the immediate utility weighting coefficient;

[0156] Positive incentives: The content pushed is the user's favorite "suspense drama" (high material matching degree) and "member exclusive discount" (high value of benefits).

[0157] Negative losses: Users are currently fatigued and highly sensitive, but the push channel is "SMS" (high disturbance index), or at this time, competitors launch "half-price tickets" (strong external interference).

[0158] S20433: Using a simulation agent, make behavioral decisions based on immediate utility and the current state of the simulation agent, and generate user behavior at the current moment;

[0159] In this embodiment, the behavior probability is converted using the Softmax function. For example, if Extremely high probability that the simulated agent will execute "click" or "purchase / registration" actions; if If the value is negative, the simulation agent is highly likely to execute an "ignore" or "complain" action.

[0160] S20434: The Monte Carlo method is used to repeatedly make behavioral decisions several times, and the generated user behaviors are statistically analyzed to generate predictive performance data, which includes predicted conversion rate, predicted click-through rate, churn risk index and predicted retention time.

[0161] In this embodiment, due to the randomness of a single decision, the system uses the Monte Carlo method to run the same strategy repeatedly (e.g., 1000 times). The frequency of "conversion" in these 1000 simulations is the predicted conversion rate. The weighted frequency of "complaints / blocking" is the churn risk index. The predicted click-through rate is obtained by counting the number of times the simulation agent performs a "click / view" action (rather than directly ignoring or closing) in the initial stage after receiving the strategy notification, divided by 1000. Retention time is a continuous numerical indicator, not a discrete event frequency. During the simulation, the internal state of the simulation agent includes an "interest decay factor." In each simulation, the agent will adjust its behavior based on immediate utility. The agent continuously browses or interacts in the simulation environment until the utility value falls below the preset "exit threshold" or an external interference event is triggered, causing a forced exit. The time step from entering to exiting the environment is recorded in each simulation (e.g., the time step in the simulation can be mapped to seconds or minutes in reality), and the predicted retention time is the average of 100 simulations; thus, multi-dimensional prediction effect data is generated.

[0162] In one optional implementation, using the marketing tolerance as a constraint and combining the predicted effect data, a target marketing strategy is selected from the candidate strategy set, and the feedback data after the execution of the target marketing strategy is sent back to the dynamic user profile for updating, including:

[0163] S2051: Using the marketing tolerance as a constraint, if the marketing tolerance is less than the marketing tolerance threshold, it is determined that the user is in a state of high fatigue, the marketing strategy recommendation process is stopped, and the marketing tolerance analysis is awaited at the next moment.

[0164] In this embodiment, the tolerance is subject to a hard constraint:

[0165] This is a "safety valve"; before screening, the user's marketing tolerance is checked again.

[0166] If the marketing tolerance is less than the marketing tolerance threshold, it means that the user is in a period of extreme fatigue or aversion. At this time, no matter how good the strategy prediction effect is, the recommendation process should be forcibly terminated and enter a "quiet period" to wait for the user to self-repair, so as to protect the user experience to the greatest extent.

[0167] S2052: If the marketing tolerance is greater than or equal to the marketing tolerance threshold, then based on the predicted effect data, using a predefined objective function, the target marketing strategy with the largest target value is selected from the candidate strategy set, as shown in the formula:

[0168]

[0169] In the formula, For the k-th candidate policy in the candidate policy set The predicted performance data includes predicted conversion rate, predicted click-through rate, churn risk index, and predicted retention time. For the k-th candidate policy in the candidate policy set The overall benefit, i.e., the target value; The weight coefficients of the objective function; if a strategy predicts a high conversion rate, but also predicts a very high risk of user churn, its overall score will be reduced, or the strategy may even be abandoned, thus achieving a balance between commercial interests and user experience.

[0170] S2053: Distribute the target marketing strategy to the outreach channels (SMS gateway, Push notification service, advertising system, etc.) and collect user feedback data in real time after the target marketing strategy is implemented;

[0171] In this embodiment, the feedback data types include:

[0172] Positive feedback: The user clicked the push notification, claimed the coupon, and completed the ticket purchase;

[0173] Negative feedback: Users delete messages, unsubscribe from the official account, and repeatedly click "back" in the background;

[0174] Time-series feedback: Users accessed the venue's transportation guide multiple times within 3 days before the event started after purchasing tickets;

[0175] S2054: Based on the feedback data, update the user's dynamic user profile to obtain an updated dynamic user profile;

[0176] In this embodiment, the timing feature is updated by appending the current "click / purchase ticket" action to the user's current behavior sequence, updating the user profile, and using it for the next round of recommendations.

[0177] Update attribute characteristics:

[0178] If a user has never seen a dance drama before but has purchased tickets this time, the system will increase the weight of their "dance" interest tag;

[0179] If a user is price-sensitive and uses large coupons, the system will label them as a "price-driven" user.

[0180] Update policy related network:

[0181] If the strategy is successfully transformed, the connection weight between the "meta-action nodes" in the strategy will be enhanced (such as increasing the correlation between "preview video" and "stage play").

[0182] If a strategy leads to users unsubscribing, the weight of that strategy combination will be significantly reduced, and it will also be demoted in subsequent recommendations to that user group.

[0183] This invention also provides a marketing strategy recommendation device 300 based on big data analysis, referring to... Figure 3 The device may include the following units:

[0184] The big data acquisition unit 301 is used to collect multi-source behavioral big data of users and construct a dynamic user profile containing time-series features and attribute features based on the multi-source behavioral big data.

[0185] The status assessment unit 302 is used to calculate the deviation between the user's current behavior pattern and historical behavior pattern based on the dynamic user profile, so as to quantify the user's current demand status and calculate the user's marketing tolerance.

[0186] The strategy construction unit 303 is used to construct a strategy association network containing several strategy meta-actions, and based on the dynamic user profile and the user's current demand status, to search for matches in the strategy association network and generate a candidate strategy set.

[0187] The simulation simulation unit 304 is used to construct a simulation simulation environment, map the dynamic user profile to a simulation agent of the simulation simulation environment, input the candidate strategy set into the simulation simulation environment for simulation simulation, and generate prediction effect data.

[0188] The decision optimization unit 305 is used to select a target marketing strategy from the candidate strategy set based on the marketing tolerance as a constraint and the predicted effect data, and to send the feedback data after the target marketing strategy is executed back to the dynamic user profile for updating.

[0189] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0190] Memory, used to store computer programs;

[0191] When a processor executes a program stored in memory, it implements the marketing strategy recommendation method based on big data analysis of the present invention.

[0192] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0193] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0194] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the marketing strategy recommendation method based on big data analysis according to embodiments of the present invention.

[0195] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable vehicles (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0196] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0197] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0198] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0199] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0200] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A marketing strategy recommendation method based on big data analysis, characterized in that, The method includes: Collect multi-source behavioral big data of users, and construct dynamic user profiles that include time-series features and attribute features based on the multi-source behavioral big data; Based on the dynamic user profile, the deviation between the user's current behavior pattern and historical behavior pattern is calculated to quantify the user's current needs and to calculate the user's marketing tolerance. Construct a policy association network containing several policy meta-actions, and based on the dynamic user profile and the user's current demand status, retrieve matches in the policy association network to generate a candidate policy set; A simulation environment is constructed, the dynamic user profile is mapped to the simulation agent of the simulation environment, the candidate strategy set is input into the simulation environment for simulation, and prediction effect data is generated. Using the marketing tolerance as a constraint and combining the predicted effect data, a target marketing strategy is selected from the candidate strategy set, and the feedback data after the target marketing strategy is executed is sent back to the dynamic user profile for updating.

2. The marketing strategy recommendation method based on big data analysis according to claim 1, characterized in that, Collect multi-source behavioral big data of users, and construct dynamic user profiles containing time-series features and attribute features based on the multi-source behavioral big data, including: Collect multi-source behavioral big data of users, including online behavior data, business transaction data, and interactive feedback data; Multi-source behavioral big data is preprocessed to obtain preprocessed standard data; Extract time-series and attribute features from the preprocessed standard data, and construct a dynamic user profile containing time-series and attribute features. The time-series features include current behavior sequences and historical behavior sequences.

3. The marketing strategy recommendation method based on big data analysis according to claim 2, characterized in that, Based on the dynamic user profile, the deviation between the user's current behavior pattern and historical behavior pattern is calculated to quantify the user's current needs and to calculate the user's marketing tolerance, including: Based on the current behavior sequence in the temporal features of the dynamic user profile, calculate the current behavior entropy corresponding to the user's current behavior pattern; Based on the historical behavior sequence in the temporal features of the dynamic user profile, calculate the historical baseline entropy corresponding to the user's historical behavior pattern; Based on the current behavioral entropy and the historical baseline entropy, the deviation between the user's current behavioral pattern and the historical behavioral pattern is calculated to quantify the user's current demand status. Based on the attribute characteristics of the dynamic user profile, the user's marketing tolerance is calculated.

4. The marketing strategy recommendation method based on big data analysis according to claim 3, characterized in that, The formula for the current behavior entropy is: In the formula, The entropy of the current behavior; Let j be the probability of the j-th action occurring; j, A behavior indicator for the set of behavior categories obtained by mapping the current behavior sequence; m is the total number of behaviors in the set of behavior categories; For the j-th behavior category in the set, Behavior The number of times it appears; The formula for the demand state is: In the formula, For deviation degree; The historical baseline entropy; if If the demand is strong, the demand state is a strong intention state; otherwise, the demand state is a roaming state. The deviation threshold; The formula for the marketing tolerance is: In the formula, For marketing tolerance; This refers to the time interval since the last marketing campaign; This represents the cumulative number of times the message has been received. This represents the number of negative feedback behaviors. This is the tolerance weighting coefficient.

5. The marketing strategy recommendation method based on big data analysis according to claim 4, characterized in that, Construct a policy association network containing several policy meta-actions, and based on the dynamic user profile and the user's current demand state, retrieve matches in the policy association network to generate a candidate policy set, including: The marketing strategy is deconstructed into meta-action nodes corresponding to the strategy meta-actions, and a strategy association network is constructed based on the meta-action nodes. The edges in the strategy association network represent the association weights between meta-action nodes, and the node types of the meta-action nodes include channels, materials, rights and interests, and timing. The implicit association vectors between several meta-action nodes in the strategy association network are extracted using a pre-built graph neural network. Extract the user's preference feature vector from the dynamic user profile. In the policy association network, calculate the similarity between the preference feature vector and each implicit association vector. Retain the top-K matching meta-action nodes with the highest similarity scores to form an initial policy set constructed from the policy meta-actions corresponding to all matching meta-action nodes. Based on the user's current needs, the initial strategy set is dynamically modified and expanded to generate a candidate strategy set.

6. The marketing strategy recommendation method based on big data analysis according to claim 5, characterized in that, A simulation environment is constructed, the dynamic user profile is mapped to a simulation agent within the simulation environment, the candidate strategy set is input into the simulation environment for simulation, and prediction effect data is generated, including: Construct a simulation simulation environment, which includes a simulation agent that simulates the behavioral logic of a real user group and a competitive agent that simulates a market competition environment; The dynamic user profile is mapped to a simulation agent in the simulation simulation environment. Based on the attribute characteristics of the dynamic user profile, the user's current demand state, and marketing tolerance, the initial state parameters of the simulation agent are set. The initial state parameters include the initial intent intensity parameter set according to the demand state, the initial fatigue sensitivity parameter set according to the marketing tolerance, and the initial decision preference vector set according to the attribute characteristics. The candidate strategy set is input into a simulation environment for simulation and simulation to generate prediction effect data.

7. The marketing strategy recommendation method based on big data analysis according to claim 6, characterized in that, The candidate strategy set is input into a simulation environment for simulation and simulation to generate prediction effect data, including: The candidate policy set is injected into the simulation agent of the simulation environment, and the competing agent of the simulation environment is used to generate interference events; Using a simulation agent, the immediate utility is calculated based on the current state of the simulation agent, the set of candidate policies, and the interference events, using the following formula: In the formula, Let k be the k-th candidate policy in the candidate policy set at time t. The corresponding immediate utility; The current intent strength parameter of the simulation agent at time t represents the current state of the simulation agent. The current fatigue sensitivity parameter of the simulation agent at time t represents the current state of the simulation agent. The k-th candidate strategy at time t Chinese rights and interests actions Value score of rights and interests action nodes; The k-th candidate strategy at time t Chinese material element action The current decision preference vector of the simulated agent The material matching degree, the current decision preference vector belongs to the current state of the simulation agent; The k-th candidate strategy at time t Central Channel Yuan Action Channel node disturbance index; This is a disruptive event; This is the immediate utility weighting coefficient; Using a simulation agent, behavioral decisions are made based on immediate utility and the current state of the simulation agent, generating user behavior at the current moment; The Monte Carlo method is used to repeatedly make behavioral decisions, and the generated user behaviors are statistically analyzed to generate predictive performance data.

8. The marketing strategy recommendation method based on big data analysis according to claim 7, characterized in that, The predicted performance data includes predicted conversion rate, predicted click-through rate, churn risk index, and predicted retention time.

9. The marketing strategy recommendation method based on big data analysis according to claim 8, characterized in that, Using the marketing tolerance as a constraint and combining the predicted effect data, a target marketing strategy is selected from the candidate strategy set, and the feedback data after the execution of the target marketing strategy is sent back to the dynamic user profile for updating, including: Using the aforementioned marketing tolerance as a constraint, if the marketing tolerance is less than the marketing tolerance threshold, it is determined that the user is in a state of high fatigue, the marketing strategy recommendation process is stopped, and the user waits for the next moment's marketing tolerance analysis. If the marketing tolerance is greater than or equal to the marketing tolerance threshold, then based on the predicted effect data, a predefined objective function is used to select the target marketing strategy with the largest target value from the candidate strategy set, as shown in the formula: In the formula, For the k-th candidate policy in the candidate policy set The predicted performance data includes predicted conversion rate, predicted click-through rate, churn risk index, and predicted retention time. For the k-th candidate policy in the candidate policy set The overall benefit, i.e., the target value; These are the weight coefficients of the objective function; Distribute the target marketing strategy to the outreach channels and collect user feedback data in real time after the target marketing strategy is implemented; Based on the feedback data, the user's dynamic user profile is updated to obtain an updated dynamic user profile.

10. A marketing strategy recommendation device based on big data analysis, used to implement the marketing strategy recommendation method as described in any one of claims 1-9, characterized in that, The device includes: The big data acquisition unit is used to collect multi-source behavioral big data of users and construct dynamic user profiles containing time-series features and attribute features based on the multi-source behavioral big data. The status assessment unit is used to calculate the deviation between the user's current behavior pattern and historical behavior pattern based on the dynamic user profile, so as to quantify the user's current demand status and calculate the user's marketing tolerance. The strategy construction unit is used to construct a strategy association network containing several strategy meta-actions, and based on the dynamic user profile and the user's current demand status, to search for matches in the strategy association network and generate a candidate strategy set. The simulation and deduction unit is used to construct a simulation and deduction environment, map the dynamic user profile to a simulation agent of the simulation and deduction environment, input the candidate strategy set into the simulation and deduction environment for simulation and deduction, and generate prediction effect data. The decision optimization unit is used to select a target marketing strategy from the candidate strategy set based on the marketing tolerance as a constraint and the predicted effect data, and to send the feedback data after the target marketing strategy is executed back to the dynamic user profile for updating.

Images