Customer acquisition heat prediction system and method for multi-channel marketing data fusion

Abstract

The application provides a customer acquisition heat prediction system and method for multi-channel marketing data fusion, relates to the technical field of market analysis and prediction, adopts an event abstraction model to fuse user behavior data of multiple marketing channels, forms cross-channel time sequence behavior sequences, integrates behavior records of users in different marketing channels, forms complete full-channel behavior path records, analyzes user behavior as a whole, guarantees comprehensiveness and depth, and utilizes a channel interaction weight lookup table to perform real-time weighted fusion processing on derived features, obtains a global user feature vector which can comprehensively reflect real interests and conversion tendencies of users and adapt to behavior changes, performs real-time matching of a customer acquisition heat value and derived features with preset strategy conditions, directly converts analysis results into executable instructions, performs automatic strategy matching and instruction generation, and realizes real-time and intelligent resource allocation.

Description

Technical Field

[0001] This invention relates to the field of market analysis and forecasting technology, specifically to a customer acquisition heat forecasting system and method that integrates multi-channel marketing data. Background Technology

[0002] In the era of digital marketing, businesses typically interact with potential customers through multiple marketing channels, such as official websites, apps, mini-programs, social media advertising, offline events, and telemarketing. However, existing technologies present the following technical challenges: Problem 1: The data from each marketing channel is independent, with different formats and a lack of effective integration mechanisms. Marketers can only see superficial indicators such as click-through rate and impressions of a single channel, and cannot gain a holistic understanding of a potential customer's complete behavioral path across all channels. The collection of customer data is singular, and the analysis of customer behavior is not in-depth enough, resulting in unreliable prediction results that cannot represent customer trends. The second problem is that existing customer value assessment models rely on a limited number of dimensions, such as spending amount, recent purchase time, or frequency of behavior on a single channel. These models cannot fully and dynamically reflect users' true interests and conversion tendencies. Furthermore, customer assessments rely on fixed rules or static weights, which cannot respond to dynamic customer behavior in real time. This leads to biases or even errors in predictive analysis results, resulting in low accuracy. Question 3: When issuing instructions for resource allocation and scheduling based on customer analysis results, the system mainly relies on manual experience or fixed strategies. It cannot generate response instructions intuitively and quickly based on customer analysis results, resulting in decision-making delays, low efficiency, and hindering customer analysis and execution. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: a customer acquisition heat prediction method based on multi-channel marketing data fusion, the method comprising: Acquire user behavior data from N marketing channels and convert the user behavior data into standardized event data; Generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on the time-series behavior sequences; The derived features are dynamically weighted and fused using a channel interaction weight lookup table to generate a global user feature vector. Calculate customer acquisition heat value based on global user feature vector; Based on customer acquisition heat value and derived characteristics, resource allocation instructions are automatically generated by matching condition rules through resource allocation strategy matrix; Execute resource allocation instructions and monitor the execution results, and periodically adjust the weight coefficients of the channel interaction weight lookup table based on the execution results.

[0004] Furthermore, the step of acquiring user behavior data from N marketing channels and converting the user behavior data into standardized event data includes: Acquire user behavior data from N marketing channels, including official websites, mobile applications, social media platforms, offline store systems, and third-party advertising platforms; User behavior data is transformed into standardized event data through an event abstraction model. The standardized event data includes user identifier, event type, time attribute, channel source, and timestamp.

[0005] Furthermore, the step of generating cross-channel time-series behavioral sequences based on standardized event data, and generating derived features based on the time-series behavioral sequences, includes: Using user identifiers as the primary key, standardized event data is sorted in ascending order by timestamp to generate a cross-channel time-series behavior sequence; Derivative features generated based on time-series behavioral sequences include path jump features, behavioral density fluctuation features, and channel depth distribution features; The path jump feature is obtained by statistically analyzing the number of times users complete key behavior conversions between different marketing channels within a preset time window. The behavior density fluctuation feature is obtained by calculating the coefficient of variation of the total amount of user behavior within a specified time range. The channel depth distribution feature is obtained by recording the maximum interaction depth value and distribution ratio of users in different marketing channels.

[0006] Furthermore, the step of dynamically weighting and fusing the derived features through a channel interaction weight lookup table to generate a global user feature vector includes: The derived features are input into the channel interaction weight lookup table, and the channel interaction weight lookup table stores the weight coefficients corresponding to different marketing channel combinations. Based on the time-series behavior sequence, match the channel interaction weight lookup table to find the channel combination and obtain the corresponding weight coefficient; Based on the obtained weight coefficients, the derived features of each marketing channel are weighted and summed to complete the dynamic weight fusion, and a global user feature vector that incorporates the dynamic weights is output.

[0007] Furthermore, the calculation of customer acquisition heat value based on global user feature vectors includes: Input the global user feature vector into the prediction model, calculate the probability of the user converting within a preset time window in the future, obtain the predicted probability value, normalize the predicted probability value to the range of [0,1], and obtain the customer acquisition heat value. The customer acquisition heat value is associated with the user identifier, and a heat visualization chart is generated based on the customer acquisition heat value. The heat visualization chart includes a geographic heat map and a channel conversion path map. The geographic heat map displays the regional heat distribution with color gradients, and the channel conversion path map displays the migration path of high-heat customers between marketing channels with flow animation.

[0008] Furthermore, the automatic generation of resource allocation instructions based on customer acquisition heat values ​​and derived characteristics, through resource allocation strategy matrix matching condition rules, includes: Establish a resource allocation strategy matrix, which includes M triggering conditions for resource allocation and a strategy rule base; The acquired customer acquisition heat value and derived features are matched with M trigger conditions. When the trigger conditions are met, the corresponding rules in the strategy rule base are automatically executed, and resource allocation instructions are generated based on the rules. The resource allocation instructions include channel budget adjustment, marketing content push and customer group classification operations.

[0009] Furthermore, the strategy rule base includes channel resource allocation rules, content resource allocation rules, and resource optimization rules.

[0010] Furthermore, the execution of resource allocation instructions and monitoring of execution results, and the periodic adjustment of weight coefficients in the channel interaction weight lookup table based on the execution results, include: Collect the actual conversion values ​​from the execution results, perform a difference analysis between the actual conversion values ​​and the predicted probability values, and calculate the prediction accuracy index. The actual conversion values ​​include user conversion rate, return on investment, and channel contribution index. Adjust the weight coefficients in the channel interaction weight lookup table based on the prediction accuracy index.

[0011] A customer acquisition heat prediction system that integrates multi-channel marketing data includes a unified data platform module, an intelligent prediction module, a dynamic decision engine module, and an adaptive optimization module.

[0012] Furthermore, the system includes: The unified data platform module is used to acquire user behavior data from N marketing channels, convert user behavior data into standardized event data, generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on time-series behavior sequences. The intelligent prediction module is used to dynamically fuse derived features through a channel interaction weight lookup table to generate a global user feature vector, and calculate the customer acquisition heat value based on the global user feature vector. The dynamic decision engine module automatically generates resource allocation instructions based on customer acquisition heat value and derived characteristics, through resource allocation strategy matrix matching condition rules; The adaptive optimization module executes resource allocation instructions and monitors the execution results, and periodically adjusts the weight coefficients of the channel interaction weight lookup table based on the execution results.

[0013] This invention provides a customer acquisition heat prediction system and method based on multi-channel marketing data fusion. It has the following beneficial effects: 1. This invention standardizes user behavior data collected from multiple marketing channels into a unified format, aggregates omnichannel events in chronological order using user identifiers as the primary key, and uses an event abstraction model to obtain cross-channel temporal behavior sequences for user behavior data fusion. This integrates user behavior records from different marketing channels to form a complete omnichannel behavior path record, enabling comprehensive and in-depth analysis of user behavior. It breaks down data silos, deeply integrates multi-channel information, and achieves in-depth and comprehensive insights into user behavior, thereby improving the credibility and representativeness of customer acquisition heat prediction results.

[0014] 2. This invention employs a method of fusing multi-dimensional derived features with dynamic weights. By calculating dynamic derived features such as path jumps, behavioral density fluctuations, and channel depth distribution, and using a channel interaction weight lookup table to perform real-time weighted fusion of derived features, a global user feature vector is generated that comprehensively reflects users' true interests and conversion tendencies and can adapt to behavioral changes. The weights of the weighted fusion are adaptively adjusted according to the user conversion rate, overcoming the limitations of fixed rules and static weights. This allows the prediction model to respond dynamically and accurately to user behavior, making the calculation of customer acquisition heat value more accurately reflect the user conversion probability, thereby reducing the error rate, improving the targeting of marketing strategies, and greatly improving the accuracy and real-time performance of customer acquisition heat prediction.

[0015] 3. This invention employs a method of automatically matching resource allocation strategy matrices and rules. By matching customer acquisition heat values ​​and derived features with preset strategy conditions in real time, it automatically triggers and generates specific resource allocation instructions. This results in an automated decision-making and execution closed loop that requires no manual intervention and can intuitively and quickly respond to prediction results. The analysis results are directly converted into executable instructions for automated strategy matching and instruction generation. This achieves real-time and intelligent resource allocation, completely changing the lagging decision-making mode that relies on human experience. Furthermore, the customer acquisition heat values ​​are intuitively displayed through heat visualization charts, facilitating quick understanding and execution, and improving the efficiency, speed, and accuracy of marketing resource allocation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the customer acquisition heat prediction method based on multi-channel marketing data fusion of the present invention. Figure 2 This is a data transmission flowchart of the customer acquisition heat prediction method based on multi-channel marketing data fusion of the present invention; Figure 3 This is an architecture diagram of the customer acquisition heat prediction system based on the fusion of multi-channel marketing data of the present invention. Detailed Implementation

[0017] 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. Example 1:

[0018] like Figures 1 to 3 As shown, the customer acquisition heat prediction method based on multi-channel marketing data fusion is characterized by the following: Step S100: Obtain user behavior data from N marketing channels and convert the user behavior data into standardized event data; Step S101: Obtain user behavior data from N marketing channels, including the official website, mobile application, social media platform, offline store system, and third-party advertising platform. User behavior data is obtained through raw records generated during the interaction between users and each marketing channel of the enterprise. The raw records are user behaviors based on the marketing channels, including website clicks, application logins, social media likes, scanning QR codes or clicking ads at offline stores, etc. When acquiring user behavior data, raw records are automatically collected through the tracking mechanisms of each marketing channel. For example, JavaScript code is deployed on the website to capture click and browsing events, SDKs are integrated into the mobile application to record user operation behaviors, such as login and download, likes and shares are extracted through API interfaces on social media platforms, QR code scanning behavior is recorded using scanning devices in offline stores, and ad interaction data is collected through click tracking systems on third-party advertising platforms. The data sets obtained from each marketing channel form raw records. The collected raw records are then stored and preliminarily cleaned in real time or in batches to form user behavior data, providing a foundation for the formation of standardized event data. When collecting user behavior data from multiple marketing channels, a multi-layered approach is adopted to protect user privacy, including data anonymization, encrypted transmission, and access control. First, during user behavior data collection, personal identification information (such as name and email) is processed using a hash function to generate anonymous user identifiers, removing sensitive data. Second, TLS / SSL encryption protocols are used during user behavior data transmission to ensure data security. Simultaneously, role-based access control policies are implemented, restricting access to user behavior data to authorized personnel, and regular security audits are conducted. Furthermore, adhering to the principle of data minimization, only necessary user behavior data is collected, and data anonymization techniques are applied during storage, such as generalizing precise geographical locations to regional levels. Finally, a user consent mechanism is integrated, informing users and obtaining explicit permission through a privacy policy before collecting user behavior data, and providing an opt-out option to allow users to control data usage, thereby comprehensively protecting user privacy. Step S102: Convert user behavior data into standardized event data using an event abstraction model. Standardized event data includes user identifier, event type, time attribute, channel source, and timestamp; where: The Event Abstraction Model (AIMM) is a data transformation framework used to eliminate the heterogeneity of multi-source data from different marketing channels, unify user behavior data into a structured format, and ensure the consistency of user behavior data. When processing user behavior data, firstly, the data is parsed to extract key fields such as user ID and timestamps. Secondly, data fields from different marketing channels are mapped to unified event model fields according to predefined mapping rules. These predefined mapping rules can be trained and set based on existing large-scale artificial intelligence models; for example, mapping "likes" on social media to the standard event type "like," and mapping numerical information to specific behavioral events. Finally, the unified event model fields are standardized in format, such as unifying the time format to ISO 8601 and encoding the channel source. Validation rules check data integrity and consistency, remove invalid or abnormal data, and output standardized event data. Standardized event data is uniformly formatted data processed by an event abstraction model, including fields such as user identifier, event type, time attribute, channel source, and timestamp. Standardized event data provides a consistent data foundation for generating cross-channel time-series behavioral sequences and derived features; among which: User identifier is a field that uniquely identifies a user, including a user ID or an anonymous identifier (such as a hash value), used to link the behavior of the same user across different marketing channels. For example, the behavior of user "user123" on the official website and mobile application is linked through this identifier. Event type describes the category of user behavior, including clicks, logins, likes, etc., used to classify and quantify user behavior interactions. For example, the event type "ad click" is used to analyze advertising effectiveness. Time attribute is information related to the time when the event occurred, including date, hour, etc., used to provide a basis for time series analysis. For example, the time attribute "2023-10-05" is used to calculate behavior density. Channel source is the marketing channel through which the event occurred, including N marketing channels such as the official website and social media, used to track the source of behavior. For example, the channel source "mobile application" is used to evaluate channel contribution. Timestamp is the precise time point when the event occurred, including Unix timestamps or ISO format, used to sort events and calculate intervals. For example, the timestamp "1696523400" is used to generate time series sequences.

[0019] Step S200: Generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on the time-series behavior sequences; Step S201: Using the user identifier as the primary key, sort all standardized event data of the same user in ascending order by timestamp to generate a cross-channel time-series behavior sequence; Using the user identifier as the primary key signifies that the user identifier serves as a unique key value for associating and grouping all standardized event data for the same user, ensuring the accuracy of data retrieval and processing. With the user identifier as the primary key, standardized event data for the same user is aggregated through data queries or grouping operations (such as the GROUP BY statement in SQL), thereby constructing a complete behavioral profile for each user. Sort the standardized event data in ascending order by timestamp to maintain the temporal logic of user behavior, facilitating subsequent analysis of temporal behavior sequences and calculation of derived features. Sort by timestamp from smallest to largest to ensure that events are arranged in the actual order of occurrence, facilitating the rapid identification of user path jump characteristics, behavioral density fluctuation characteristics, and other derived features, while simultaneously improving the accuracy of the prediction model and avoiding biases caused by out-of-order data. A time-series behavior sequence is a sequence of user events arranged chronologically, including attributes such as event type, channel source, and timestamp. It is used to capture user behavior across channels, providing a data foundation for generating derived features and predicting user conversion rates. First, using the user identifier as the primary key, all events for the same user are filtered from standardized event data. Second, all events for the same user are sorted in ascending order by timestamp, forming a time-series behavior sequence for each user. Then, the time-series behavior sequence is stored in a structured data format (such as a list or data frame), where each event contains fields such as user identifier, event type, time attribute, channel source, and timestamp. Finally, the cross-channel time-series behavior sequence is output for further feature generation. Step S202: Derivative features generated based on time-series behavior sequences include path jump features, behavior density fluctuation features, and channel depth distribution features. These derived features are extracted from the time-series behavior sequences to quantify user behavior patterns. User activity, conversion potential, and channel preferences are assessed based on these features, thereby supporting customer acquisition heat prediction and resource allocation decisions. First, based on the time-series behavior sequences, calculation rules for path jump features, behavior density fluctuation features, and channel depth distribution features are defined, such as the number of conversions for path jump features and the coefficient of variation for behavior density fluctuation features. Second, the time-series behavior sequences of each user are traversed, and statistical and mathematical methods are applied to calculate the derived features for each user. For path jump features, the system iterates through the user's temporal behavior sequence and counts the number of events matching key behavior transition pairs within a preset time window. For behavior density fluctuation features, it calculates the coefficient of variation of the total number of daily behaviors within a specified time range. For channel depth distribution features, it records the maximum interaction depth achieved by the user on each marketing channel and calculates the distribution of the depth value of each marketing channel in the total depth value. For example, if a user jumps 5 times in 7 days, the behavior density coefficient of variation is 0.3, and the channel depth distribution is [Official Website: 0.6, Social: 0.4]. Finally, the feature values ​​of the derived features are integrated into a structured feature vector for subsequent dynamic weight fusion. Path jump feature measures the number of times a user completes key behavior conversions between different marketing channels within a preset time window. It assesses user activity and conversion potential by statistically analyzing conversion frequency. First, key behavior conversion pairs are defined and a time window (e.g., 24 hours) is set. Then, the user's temporal behavior sequence is traversed to detect whether adjacent events match key behavior conversion pairs within the time window. The number of matches is counted as the path jump feature value. Key behavior conversions refer to the conversion of a specific behavior from one marketing channel to another, such as "ad click (third-party advertising platform) to app download (mobile app)" or "social media like (social media platform) to website registration (official website)". Key behavior conversions are cross-marketing channel switching in user behavior that has conversion significance, including jumping from ad click to app download or jumping from social media interaction to website visit. When detecting whether adjacent events match key behavior conversion pairs, three criteria must be met: First, the event type and channel source must completely match the predefined key conversion pair; second, the timestamp difference between the two events must be within a preset time window (e.g., 24 hours); and third, the two events must be consecutive or adjacent in the time-series behavior sequence (without other key behaviors inserted in between). Adjacent events refer to two events in the user's time-series behavior sequence with consecutive timestamps or intervals within an allowed range. The number of matches is used as a path jump feature because the number of matches directly quantifies the frequency of user active conversions between marketing channels, effectively reflecting the user's exploration willingness and conversion potential. The number of matches is the total number of times the key behavior conversion pair was successfully detected under the preset time window and criteria. Behavioral density fluctuation characteristics measure the degree of fluctuation in the total amount of user behavior within a specified time period, including the coefficient of variation. The stability or anomaly of user behavior is identified by calculating the coefficient of variation of the total number of behavioral events. First, a time period (e.g., 7 days) is selected, and the total number of behavioral events each day is calculated. Then, the mean (μ) and standard deviation (σ) of the total number of user behavioral events within the time period are calculated. Finally, the coefficient of variation is used to... Through formula The behavior density fluctuation characteristic value is obtained. The level of the CV value is relative to the historical conversion rate and the threshold set by experience. Usually, a baseline CV value is set, generally 0.5. The CV value above the baseline is high CV, and the CV value below the baseline is low CV. A high CV value > 0.5 indicates that the user behavior fluctuates greatly and is unstable, which means that the user is in the decision-making period or has changeable interests, and the conversion uncertainty is high, but it may also contain high potential. A low CV value ≤ 0.5 indicates that the user behavior is stable and regular, which may indicate that the user loyalty is high or in a stable state, and the conversion path is predictable. For example, if the number of behaviors in 7 days is [5, 10, 15], then μ=10, σ≈5, CV≈0.5. The coefficient of variation is the ratio of the standard deviation to the mean. It is used to eliminate the influence of data units, measure relative volatility, and facilitate the comparison of the behavioral stability of different users or time periods. Channel depth distribution characteristics describe the maximum interaction depth values ​​achieved by users across different marketing channels and their distribution ratios. This includes the interaction depth levels (e.g., browsing, clicking, purchasing) and their distribution ratios for each marketing channel. User engagement is assessed by recording interaction depth values ​​and calculating distribution ratios. First, interaction depth levels are defined (e.g., level 1 for browsing, level 2 for clicking, and level 3 for purchasing). Then, the user's temporal behavior sequence is traversed, recording the maximum interaction depth value for each channel. Finally, the proportion of each marketing channel's interaction depth value to the total interaction depth value is calculated to guide resource allocation and personalized marketing. For example, if a user's maximum interaction depth value for channel A is 3 and their maximum interaction depth value for channel B is 2, then the total depth value is 5, and the distribution ratio is 60% for channel A and 40% for channel B. The interaction depth value is the highest level of engagement a user achieves on a particular channel, including numerical levels (e.g., levels 1 to 3), used to quantify the user's level of investment in that channel. The distribution ratio is the proportion of each marketing channel's depth value in the total depth value, including percentage values, used to identify channel preferences. Resource allocation and personalized marketing based on interaction depth include: First, define the depth levels, quantifying interaction depth into levels, such as Level 1 (browse / click), Level 2 (add to favorites / inquire), and Level 3 (purchase / register). Then, map user depth, determining the highest depth level a user reaches on each marketing channel based on their historical behavior. Finally, formulate an allocation strategy: allocate high-value resources, such as exclusive discounts and human customer service, to high-depth channels (e.g., depth ≥ 3); allocate nurturing resources, such as content pushes and light incentives, to medium-to-low-depth channels (e.g., depth = 1 or 2). For example, if user A has a depth of 3 (purchased) on the official website and a depth of 1 (liked only) on social media, the system automatically allocates "VIP customer exclusive discounts" to the official website channel and "product introduction videos" to the social media channel for nurturing. Identifying channel preferences through distribution ratios involves: first, calculating the distribution ratio, which is the proportion of a user's depth value across all marketing channels relative to the total depth. Then, setting a preference threshold (e.g., 30%), a user is considered to have a preference for a channel if its distribution ratio exceeds this threshold. Finally, applying the preference results, the preferred channel is prioritized for marketing outreach. For example, user B's total interaction depth is 10, with the "mobile app" channel having a depth value of 6 (60%) and the "offline store" channel having a depth value of 4 (40%). Since the mobile app's ratio exceeds 30% and is the highest, user B is determined to prefer the "mobile app" channel, and subsequent marketing pushes (such as new product launch notifications) will be prioritized through the mobile app.

[0020] Step S300: Dynamically fuse the derived features using a channel interaction weight lookup table to generate a global user feature vector; Step S301: Input the derived features into the channel interaction weight lookup table, and the channel interaction weight lookup table stores the weight coefficients corresponding to different marketing channel combinations; match the channel combinations in the channel interaction weight lookup table according to the time sequence behavior sequence to obtain the corresponding weight coefficients; The Channel Interaction Weight Lookup Table is a predefined, dynamically updated data structure used to store different marketing channel combinations and their corresponding weight coefficients. The table includes preset channel combinations (such as a channel list) and weight coefficients (such as numerical weights). These weights are derived from historical conversion rate analyses, business expert experience, or periodic training of machine learning models. For example, if historical conversion rate analysis reveals that the "social media + offline stores" combination contributes more to conversion rates, it is assigned a higher weight. The dynamic weight fusion of derived features through the Channel Interaction Weight Lookup Table more accurately reflects the impact of different marketing channel combinations on user conversion potential. The weight coefficients are allocated based on the historical conversion rates of the channel combinations. High-conversion-rate channel combinations are assigned higher weights (e.g., 0.8-1.0), medium-conversion-rate combinations are assigned medium weights (e.g., 0.4-0.7), and low-conversion-rate combinations are assigned lower weights (e.g., 0.1-0.3). For example, when the channel combination is "mobile app + official website," if historical conversion rates show a high conversion rate for this combination, the weight coefficient might be set to 0.9, while a single channel like "third-party advertising" might only have a weight of 0.3. When matching a time-series behavior sequence with a channel interaction weight lookup table, the process first extracts all channel sources involved in the user's behavior sequence to form the current user's channel combination. Then, the current user's channel combination is matched with preset channel combinations in the channel interaction weight lookup table. The judgment criteria are that the user's channel combination is completely consistent with the preset channel combination or has the highest similarity (e.g., using set intersection calculation). For example, if the user's channel combination includes "social media, offline stores", then the preset channel combination "social media & offline stores" in the channel interaction weight lookup table is matched, and the corresponding matching weight coefficient of the preset channel combination is obtained. A channel combination refers to the collection of multiple marketing channels that a user has interacted with within a certain period of time, including any two or more channels such as official websites, mobile applications, and social media platforms, such as "social media platform + offline store system" or "mobile application + official website + third-party advertising platform". These combinations are used to quantify cross-channel synergy effects. Step S302: Based on the obtained weight coefficients, perform a weighted summation of the derived features of each marketing channel. First, obtain the weight coefficients of the channel combination matching the current user from the channel interaction weight lookup table. Second, for each channel, extract the feature value of its corresponding derived feature, and multiply the feature value of each channel's derived feature by its corresponding weight coefficient to obtain a weighted feature value. Finally, sum the weighted feature values ​​of all channels to generate a comprehensive value, complete the dynamic weight fusion, and output the global user feature vector that incorporates the dynamic weights. The global user feature vector is a multi-dimensional numerical vector that integrates weighted derived features from all channels, including weighted path jump features, behavioral density fluctuation features, and channel depth distribution features. First, the comprehensive value of each derived feature is calculated by weighted summation. Then, the comprehensive values ​​are combined into a fixed-length vector, for example, in the form of [weighted path jump feature value, weighted behavioral density fluctuation feature value, weighted channel depth distribution feature value]. The global user feature vector serves as the input to the prediction model, comprehensively representing the user's cross-channel behavior patterns and conversion potential, and providing a data calculation basis for the calculation of customer acquisition heat value. Step S400: Calculate the customer acquisition heat value based on the global user feature vector; Step S401: Input the global user feature vector into the prediction model, calculate the probability that the user will convert within the future preset time window, obtain the predicted probability value, normalize the predicted probability value to the range of [0,1], and obtain the customer acquisition heat value. The predictive model is a machine learning model, such as logistic regression, random forest, or gradient boosting tree, used to predict the probability of user conversion based on a global user feature vector. First, the global user feature vector is input into the trained predictive model, which calculates the raw output score using an internal algorithm (such as the sigmoid function or tree set voting). Then, the raw output score is converted into a predicted probability value between 0 and 1 using an activation function (such as a logistic function). The predicted probability value represents the likelihood of a user converting within a preset time window. For example, if the preset model outputs a raw score of 1.2, after conversion using the sigmoid function, the predicted probability value is 0.77. The predicted probability value is a value between 0 and 1 output by the prediction model, representing the likelihood that a user will convert within a preset time window (such as 7 days). It is used to quantify user conversion potential and serves as the basis for customer acquisition heat value. Customer acquisition heat value is a normalized result of predicted probability values, ranging from [0,1], including heat value score and level classification (such as high, medium, low). It is obtained by linearly or non-linearly scaling the predicted probability values ​​to the range of [0,1], for example using the Min-Max normalization formula: Customer acquisition heat value = (probability value - minimum probability) / (maximum probability - minimum probability). It is used to intuitively evaluate user customer acquisition priorities and guide resource allocation and marketing strategies. The heat score of customer acquisition heat value is a normalized value of the predicted probability value, ranging from [0,1]. It is used to quantify the conversion potential of users and provide data support for precision marketing strategies. The higher the heat score, the greater the probability that the user will convert within a preset time window in the future. The marketing team can prioritize the allocation of resources accordingly. For example, users with a heat score of 0.85 are more likely to complete a purchase than users with a heat score of 0.3, so they should receive more immediate marketing outreach. The customer acquisition heat value classification is defined based on the heat value threshold set by business needs. It is usually divided into three equal parts or divided by business experience. Specifically, the classification is as follows: low heat value (0-0.3) represents low conversion potential, medium heat value (0.3-0.7) represents medium potential, and high heat value (0.7-1) represents high conversion potential. For example, an e-commerce company sets the customer acquisition heat value threshold corresponding to the top 20% of users in the history of conversion rate as 0.7 as the high heat value dividing line. When prioritizing user acquisition based on customer acquisition heat value, the user list is first sorted in descending order of heat value score. Then, the order is dynamically adjusted based on channel cost efficiency. High-heat users are given priority in high-cost channel resources, while medium- and low-heat users are allocated standard or nurturing outreach in descending order of their scores. For example, the system automatically generates a priority queue: customer acquisition heat value 0.9 → 0.85 → 0.8..., and simultaneously calculates the expected return on investment (ROI) of each marketing channel for fine-tuning. When allocating resources and developing marketing strategies based on customer acquisition heat value, high-heat users are assigned dedicated customer service and high-budget channels (such as search advertising), medium-heat users are assigned automated marketing processes (such as email sequences), and low-heat users are assigned low-cost channels (such as organic traffic from social media). Limited-time offers and personalized product recommendations are pushed to users with high customer acquisition heat value, educational content is pushed to users with medium customer acquisition heat value, and brand exposure is provided to users with low customer acquisition heat value. For example: if a user's customer acquisition heat value is 0.9 and their channel preference is mobile applications, the system immediately allocates 50 yuan of advertising budget to in-app push notifications and triggers a "100 yuan discount on first order" coupon; if a user's customer acquisition heat value is 0.4, 5 yuan of budget is allocated to social media for content marketing. Step S402: First, associate the customer acquisition heat value with the user identifier, using the user identifier as the primary key. Bind the calculated customer acquisition heat value to the corresponding user identifier through database operations (such as SQL UPDATE or data frame merging) to form key-value pairs. This ensures the traceability and personalized application of each user's customer acquisition heat value, facilitating the subsequent generation of visualization charts and execution of resource allocation. Complete the association between the customer acquisition heat value and the user identifier, and organize the key-value pairs in table or JSON format to generate structured data. The structured data includes user identifiers, customer acquisition heat values, derived features, etc., to provide a standardized data foundation and support efficient querying, analysis, and visualization. Then, heatmaps are generated based on structured data containing customer acquisition heat values. These heatmaps include geographic heatmaps and channel conversion path maps. They visually display the distribution of customer acquisition heat values ​​and user behavior paths, helping decision-makers quickly identify high-potential regions and channels. Relevant fields (such as geographic location or channel sequence) are extracted from the structured data, and visualization tools (such as Tableau or D3.js) are used to create charts. For example, a geographic heatmap can be generated by aggregating geographic coordinates and customer acquisition heat values, or a channel conversion path map can be generated through sequence analysis. The geographic heatmap displays the regional heat distribution using color gradients, including geographic coordinates (latitude and longitude), customer acquisition heat value intensity, and color gradients. The heatmap is used to identify high-potential customer acquisition areas. It collects users' geographic attributes (such as location resolved from IP addresses) and customer acquisition heat values, aggregates them into geographic regions, and uses heat rendering algorithms (such as kernel density estimation) to map the customer acquisition heat values ​​into color gradients and visualize them on a map to form a geographic heatmap. The channel conversion path map uses flow animation to show the migration path of high-heat customers between marketing channels, including channel nodes, flow arrows, and path frequency. It is used to analyze users' cross-channel behavior patterns, filter channel conversion data of users with high customer acquisition heat values ​​from time-series behavior sequences, construct a directed graph (nodes are channels, edges are conversion times), use force-directed layout algorithms to visualize the path, and show the flow through animation effects. Step S500: Based on customer acquisition heat value and derived characteristics, automatically generate resource allocation instructions by matching condition rules through resource allocation strategy matrix; Step S501: Establish a resource allocation strategy matrix, which includes M trigger conditions for resource allocation and a strategy rule base. The resource allocation strategy matrix is ​​set based on historical conversion rate analysis and business goal optimization. For example, if historical conversion rate analysis reveals that increasing the budget can improve the conversion rate by 20% in scenarios where "customer acquisition heat value > 0.8 and the main channels are clearly defined," this is set as a condition. Wherein: The trigger conditions for M include three core categories: first, customer acquisition heat value conditions (such as continuous growth or consistently below a threshold); second, time conditions (such as preset duration or specific period); and third, behavioral characteristic conditions (such as channel clarity or insufficient interaction depth). For example, specific conditions are defined as "customer acquisition heat value increases by ≥0.1 for 3 consecutive days", "the proportion of main source channels is >60%", and "the interaction depth of key channels is <2 for 5 consecutive days". The strategy rule base includes channel resource allocation rules, content resource allocation rules, and resource optimization rules; Channel resource allocation rules: When the customer acquisition heat value in a specific geographical area continues to increase and the main source channels are clear, the budget allocation for the corresponding channels in that area will be automatically increased, and marketing content for that area will be pushed. Content resource allocation rules: When customer acquisition heat value is high and interest characteristics match specific products, but the interaction depth on key channels is insufficient, expert consultation resources are automatically allocated and reached through the most active channels; Resource optimization rules automatically remove customers from the list of high-cost marketing channels (such as paid advertising) and transfer them to low-cost nurturing channels (such as email marketing) when the customer acquisition heat value is continuously lower than the heat value threshold (such as 0.3) and exceeds the preset time (such as 7 days). Historical conversion rate refers to the percentage of users who actually converted within a specific historical period, including overall conversion rate, channel-specific conversion rate, and time-specific conversion rate. It is calculated by dividing the number of users who converted within the statistical period by the total number of users reached. For example, the conversion rate of the official website channel in the last 30 days = (200 users who completed a purchase) / (5000 users who visited the official website) = 4%. This data is obtained by integrating data from enterprise CRM systems, website analytics tools, and advertising platforms, and is used to provide a benchmark reference for resource allocation strategy matrices, helping to set reasonable heat thresholds and channel weights. Step S502: Match the acquired customer acquisition heat value and derived characteristics with M trigger conditions. Periodically scan each user's customer acquisition heat value and derived characteristics against each of the M trigger conditions in the strategy matrix. The matching process uses conditional judgment logic (such as if-then rules). The judgment criteria include numerical comparison (such as customer acquisition heat value ≥ heat value threshold), trend analysis (such as continuous growth), and logical operations (such as AND or OR relationships of multiple conditions). Based on preset business logic and data analysis results, for example, when the three sub-conditions "customer acquisition heat value > 0.7", "main channel clarity > 0.6", and "duration ≥ 3 days" are all true, the trigger condition is determined to be met. The basis is that historical conversion rates prove that the return on resource investment under this combination of conditions is the highest. When the trigger condition is met, the corresponding rule in the conditional strategy rule base is automatically executed to generate a resource allocation instruction; for example: If the triggering conditions are met, such as when a user's customer acquisition heat value is >0.7 for 3 consecutive days and the proportion of main channels is >60%, the system fully matches the "high heat value clearly defined channel" condition and triggers the channel budget increase instruction; If the triggering conditions are not met, such as when the user acquisition heat value is consistently <0.2 and the channels are scattered, it matches the "high potential user" condition 0 and no resource allocation is triggered; If the triggering conditions are partially met, such as a user acquisition heat value of 0.6 (meeting the threshold of 0.5) but the main channel accounts for only 40% (not meeting 60%), the system will perform some actions according to the strategy rule base, such as only pushing general marketing content without adjusting the budget; Resource allocation instructions are execution commands automatically generated based on the strategy rule base after the triggering conditions are met. They include three core components: channel budget adjustment, marketing content push, and customer group classification. Channel budget adjustment specifically refers to dynamically modifying the budget allocation ratio of each marketing channel; marketing content push includes selecting specific content templates and sending them through designated channels, matching templates from the content library based on user interest characteristics, and sending them through the marketing automation platform; customer group classification divides users into different marketing groups according to rules; resource allocation instructions are sent to various execution systems (such as advertising systems and CRM systems) via API interfaces, and execution logs are recorded for subsequent optimization. Step S600: Execute resource allocation instructions and monitor the execution results, and periodically adjust the weight coefficients of the channel interaction weight lookup table based on the execution results.

[0021] Step S601: Collect the actual conversion values ​​from the execution results, perform a difference analysis between the actual conversion values ​​and the predicted probability values, and calculate the prediction accuracy index; the actual conversion values ​​include user conversion rate, return on investment, and channel contribution index. The results include user conversion rate, ROI, and channel contribution. These results are obtained by integrating feedback data from various marketing systems and are acquired in real time from data sources such as the official website conversion tracking system, mobile application analytics platform, CRM system, and advertising platform via API interfaces. User conversion rate refers to the proportion of users who actually convert, calculated by dividing the number of users who convert within a preset time window by the total number of users. ROI is the ratio of marketing investment to the revenue generated, calculated by dividing the total revenue by the total investment cost. Channel contribution is the proportion of each marketing channel's contribution to the conversion, and the conversion credit is allocated to each contact channel through attribution analysis models. Discrepancy analysis involves comparing the deviations between predicted probability values ​​and actual conversion values. It compares the predicted probability value for each user with the actual conversion status. The calculation of discrepancy analysis includes calculating the mean absolute deviation (MAE), mean absolute percentage error (MAPE), and model discrimination metrics (such as the fluctuation of AUC-ROC), including: Each user's predicted probability value is mapped one-to-one with its actual conversion status (0 indicates no conversion, 1 indicates conversion) in the subsequent time window, forming a data list containing both predicted and actual values. Calculate the absolute error by iterating through the data list and calculating the deviation between the predicted probability value and the actual conversion value for each user, i.e., |predicted value - actual value|. To calculate the mean absolute deviation (MAE), sum the absolute errors of all users and then divide by the total number of users. That is, MAE = Σ|predicted value - actual value| / N; Calculate the mean absolute percentage error (MAPE). For users who actually converted (i.e., the actual value is 1), calculate the percentage of their absolute error relative to the actual value, which is |predicted value - 1| / 1 * 100%. Then, calculate the average of all these percentages. The model discrimination index (AUC-ROC) is a core metric for evaluating the discriminative power of binary classification models. It measures the model's ability to correctly distinguish between "converted users" and "non-converted users." A higher AUC value (closer to 1) indicates better discrimination, meaning it can predict higher probabilities for converted users. This includes: Sort all users according to their predicted probability values ​​from highest to lowest, set a probability threshold, and calculate the true positive rate (TPR) and false positive rate (FPR). Starting with the highest probability, for each user's predicted probability value, users whose predicted probability value is greater than or equal to the probability threshold are predicted to "convert," while those whose predicted probability value is less than or equal to the threshold are predicted to "not convert." Based on the prediction results and the actual results, calculate the true positive rate (TPR = number of true positives / total number of users who actually converted) and the false positive rate (FPR = number of false positives / total number of users who did not actually convert). Plot the ROC curve and calculate the AUC. Plot all threshold points with FPR as the x-axis and TPR as the y-axis and connect them to form the ROC curve. The AUC is the area under this curve, which is usually calculated using the trapezoidal integral method.

[0022] In user conversion rate calculation, converted users refer to users who complete the ultimate target behavior defined by the enterprise within a preset time window. Converted users specifically include, but are not limited to: users who have completed a purchase transaction, users who have completed membership registration, users who have submitted high-quality sales leads (such as booking a test drive), and users who have downloaded and activated the core application. The data is mainly obtained by integrating with the enterprise's business systems. First, a list of purchasing user IDs is obtained from the transaction system (such as the order database of an e-commerce platform). Second, a list of users who have completed registration or filled in information is obtained from the user management system (such as CRM or CDP). Finally, these scattered user identifiers are associated with and deduplicated with the unified user identifiers in the marketing data platform through data pipelines, thereby calculating the total number of converted users within the period.

[0023] Prediction accuracy metrics include AUC-ROC, precision, recall, and F1 score. For example, AUC-ROC is calculated by plotting the true positive rate versus false positive rate curve and calculating the area under the curve. The specific steps are: first, sort users by prediction probability; calculate TPR and FPR at different thresholds; then, calculate the AUC value using the trapezoidal rule. The closer the value is to 1, the more accurate the prediction. Prediction accuracy metrics are a comprehensive evaluation system based on the results of difference analysis and model evaluation. All calculated metric values ​​from the difference analysis are compiled into a structured report to form prediction accuracy metrics, which guide model optimization decisions. Step S602: Based on the prediction accuracy index, the gradient descent algorithm is used to adjust the weight coefficients in the channel interaction weight lookup table.

[0024] When adjusting weight coefficients based on prediction accuracy metrics, firstly, the direction and magnitude of weight adjustment are determined based on the prediction accuracy metrics calculated from difference analysis (e.g., a decrease in AUC or an increase in MAE). Then, optimization algorithms such as gradient descent are used to calculate the gradient of the weight coefficients relative to the accuracy metrics, with the goal of improving accuracy. The weights are then updated along the negative gradient direction. The detailed adjustment process involves defining a loss function L (e.g., cross-entropy loss), calculating the loss function for each weight coefficient... partial derivatives Then follow the formula Update the weights, where It's the learning rate. The updated weight coefficients are as follows: For example, suppose the initial weights of channels A and B are 0.6 and 0.4, respectively. After a monitoring period, it is found that the predicted AUC of this combination drops from 0.8 to 0.75. The gradient calculation shows that the weight allocation is unreasonable. After adjusting it with a learning rate of 0.01 using the gradient descent method, the new weights become A: 0.58 and B: 0.42, which improves the accuracy of subsequent predictions. The whole process is executed periodically by an automated script to ensure that the weight coefficients are continuously optimized.

[0025] In this embodiment, user behavior data collected from multiple marketing channels is standardized into a unified format, and the processing of omnichannel events is aggregated in chronological order using user identifiers as the primary key. An event abstraction model is used to obtain cross-channel time-series behavior sequences for user behavior data fusion. This can integrate user behavior records from different marketing channels to form a complete omnichannel behavior path record, analyze user behavior as a whole, ensure comprehensiveness and depth, break down data silos, deeply integrate multi-channel information, thereby achieving in-depth and comprehensive insights into user behavior and improving the credibility and representativeness of customer acquisition heat prediction results. This method employs a fusion approach combining multi-dimensional derived features and dynamic weights. By calculating dynamic derived features such as path jumps, behavioral density fluctuations, and channel depth distribution, and utilizing a channel interaction weight lookup table to perform real-time weighted fusion of derived features, a global user feature vector is generated that comprehensively reflects users' true interests and conversion tendencies and can adapt to behavioral changes. The weights of the fusion are adaptively adjusted based on user conversion rates, overcoming the limitations of fixed rules and static weights. This enables the prediction model to respond dynamically and accurately to user behavior, making the calculation of customer acquisition heat value more accurately reflect the user conversion probability, thereby reducing the error rate, improving the targeting of marketing strategies, and greatly enhancing the accuracy and real-time performance of customer acquisition heat prediction. By employing a resource allocation strategy matrix and rule-based automatic matching method, customer acquisition heat values ​​and derived features are matched in real time with preset strategy conditions. This automatically triggers and generates specific resource allocation instructions, resulting in an automated decision-making and execution closed loop that requires no manual intervention and can intuitively and quickly respond to prediction results. The analysis results are directly converted into executable instructions for automated strategy matching and instruction generation, achieving real-time and intelligent resource allocation. This completely changes the lagging decision-making model that relies on human experience. Furthermore, customer acquisition heat values ​​are intuitively displayed through heat visualization charts, facilitating quick understanding and execution, and improving the efficiency, speed, and accuracy of marketing resource allocation. Example 2:

[0026] like Figures 1 to 3 As shown, the customer acquisition heat prediction system integrates multi-channel marketing data. The system includes a unified data platform module, an intelligent prediction module, a dynamic decision engine module, and an adaptive optimization module. The system comprises: The unified data platform module is used to acquire user behavior data from N marketing channels, convert user behavior data into standardized event data, generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on time-series behavior sequences. The intelligent prediction module is used to dynamically fuse derived features through a channel interaction weight lookup table to generate a global user feature vector, and calculate the customer acquisition heat value based on the global user feature vector. The dynamic decision engine module automatically generates resource allocation instructions based on customer acquisition heat value and derived characteristics, through resource allocation strategy matrix matching condition rules; The adaptive optimization module executes resource allocation instructions and monitors the execution results, periodically adjusting the weight coefficients of the channel interaction weight lookup table based on the execution results. Example 3:

[0027] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code that, when executed by the one or more processors, can perform the customer acquisition heat prediction system and method for multi-channel marketing data fusion as described above.

[0028] The methods or systems according to embodiments of this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. Storage devices in the electronic device, such as ROM or hard disk, may store the multi-channel marketing data fusion customer acquisition heat prediction system and method provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components in the electronic device shown in this application may be omitted according to actual needs. Example 4:

[0029] One embodiment of this application discloses a computer-readable storage medium. The computer-readable storage medium stores computer-readable instructions. When executed by a processor, the computer-readable instructions can perform the multi-channel marketing data fusion customer acquisition heat prediction system and method according to the embodiment of this application, as described with reference to the above figures. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0030] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowcharts can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as a customer acquisition heat prediction system and method based on multi-channel marketing data fusion. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0031] The multi-channel marketing data fusion customer acquisition heat prediction system and method includes a processor, a machine-readable storage medium, and the machine-readable storage medium and the processor are connected. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above method.

[0032] It should be noted that, in this document, relational terms such as "first" and "second" are used merely 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. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0033] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A customer acquisition heat prediction method based on multi-channel marketing data fusion, characterized in that, The method includes: Acquire user behavior data from N marketing channels and convert the user behavior data into standardized event data; Generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on the time-series behavior sequences; The derived features are dynamically weighted and fused using a channel interaction weight lookup table to generate a global user feature vector. Calculate customer acquisition heat value based on global user feature vector; Based on customer acquisition heat value and derived characteristics, resource allocation instructions are automatically generated by matching condition rules through resource allocation strategy matrix; Execute resource allocation instructions and monitor the execution results, and periodically adjust the weight coefficients of the channel interaction weight lookup table based on the execution results.

2. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 1, characterized in that, The process of acquiring user behavior data from N marketing channels and converting the user behavior data into standardized event data includes: Acquire user behavior data from N marketing channels, including official websites, mobile applications, social media platforms, offline store systems, and third-party advertising platforms; User behavior data is transformed into standardized event data through an event abstraction model. The standardized event data includes user identifier, event type, time attribute, channel source, and timestamp.

3. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 2, characterized in that, The process of generating cross-channel time-series behavior sequences based on standardized event data, and generating derived features based on these time-series behavior sequences, includes: Using user identifiers as the primary key, standardized event data is sorted in ascending order by timestamp to generate a cross-channel time-series behavior sequence; Derivative features generated based on time-series behavioral sequences include path jump features, behavioral density fluctuation features, and channel depth distribution features.

4. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 3, characterized in that, The step of dynamically weighting and fusing derived features through a channel interaction weight lookup table to generate a global user feature vector includes: The derived features are input into the channel interaction weight lookup table, and the channel interaction weight lookup table stores the weight coefficients corresponding to different marketing channel combinations. Based on the time-series behavior sequence, match the channel interaction weight lookup table to find the channel combination and obtain the corresponding weight coefficient; Based on the obtained weight coefficients, the derived features of each marketing channel are weighted and summed to complete the dynamic weight fusion, and a global user feature vector that incorporates the dynamic weights is output.

5. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 4, characterized in that, The calculation of customer acquisition heat value based on global user feature vectors includes: Input the global user feature vector into the prediction model, calculate the probability of the user converting within a preset time window in the future, obtain the predicted probability value, normalize the predicted probability value to the range of [0,1], and obtain the customer acquisition heat value. The customer acquisition heat value is associated with the user identifier, and a heat visualization chart is generated based on the customer acquisition heat value. The heat visualization chart includes a geographic heat map and a channel conversion path map. The geographic heat map displays the regional heat distribution with color gradients, and the channel conversion path map displays the migration path of high-heat customers between marketing channels with flow animation.

6. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 5, characterized in that, The method of automatically generating resource allocation instructions based on customer acquisition heat values ​​and derived characteristics, through resource allocation strategy matrix matching condition rules, includes: Establish a resource allocation strategy matrix, which includes M triggering conditions for resource allocation and a strategy rule base; The acquired customer acquisition heat value and derived features are matched with M trigger conditions. When the trigger conditions are met, the corresponding rules in the strategy rule base are automatically executed, and resource allocation instructions are generated based on the rules. The resource allocation instructions include channel budget adjustment, marketing content push and customer group classification operations.

7. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 6, characterized in that, The strategy rule base includes channel resource allocation rules, content resource allocation rules, and resource optimization rules.

8. The customer acquisition heat prediction method based on multi-channel marketing data fusion according to claim 7, characterized in that, The execution of resource allocation instructions and monitoring of execution results, and the periodic adjustment of weight coefficients in the channel interaction weight lookup table based on the execution results, include: Collect the actual conversion values ​​from the execution results, perform a difference analysis between the actual conversion values ​​and the predicted probability values, and calculate the prediction accuracy index. The actual conversion values ​​include user conversion rate, return on investment, and channel contribution index. Adjust the weight coefficients in the channel interaction weight lookup table based on the prediction accuracy index.

9. A customer acquisition heat prediction system integrating multi-channel marketing data, characterized in that, The system includes a unified data chassis module, an intelligent prediction module, a dynamic decision engine module, and an adaptive optimization module.

10. The customer acquisition heat prediction system based on multi-channel marketing data fusion according to claim 9, characterized in that, The system includes: The unified data platform module is used to acquire user behavior data from N marketing channels, convert user behavior data into standardized event data, generate cross-channel time-series behavior sequences based on standardized event data, and generate derived features based on time-series behavior sequences. The intelligent prediction module is used to dynamically fuse derived features through a channel interaction weight lookup table to generate a global user feature vector, and calculate the customer acquisition heat value based on the global user feature vector. The dynamic decision engine module automatically generates resource allocation instructions based on customer acquisition heat value and derived characteristics, through resource allocation strategy matrix matching condition rules; The adaptive optimization module executes resource allocation instructions and monitors the execution results, and periodically adjusts the weight coefficients of the channel interaction weight lookup table based on the execution results.

Images