E-commerce dynamic marketing resource intelligent allocation method and system based on user behavior characteristics

Abstract

The present application belongs to the technical field of e-commerce marketing resource scheduling, and particularly relates to an e-commerce dynamic marketing resource intelligent allocation method and system based on user behavior characteristics. The method acquires e-commerce user behavior data and pre-processes the same, screens high-confidence behavior records as benchmark samples, performs click behavior correction, jump-out behavior duration backtracking and stay duration offset correction on the remaining behavior records, forms a restored feature set, constructs a user feature vector, and uniformly converts different types of marketing resources into resource parameter groups with consistent field structures; completes user behavior label determination, user grouping and to-be-launched resource screening; continuously evaluates the marketing resources based on a rolling time window, suspends the launch of marketing resources whose continuous click conversion rate does not reach a preset threshold, and executes subsequent resource rearrangement and update in combination with feedback data. The present application can realize the unified screening, dynamic allocation and closed-loop adjustment of heterogeneous marketing resources.

Description

Technical Field

[0001] This invention belongs to the technical field of e-commerce marketing resource scheduling, specifically involving a method and system for intelligent allocation of dynamic e-commerce marketing resources based on user behavior characteristics. Background Technology

[0002] As the transaction volume of e-commerce platforms continues to expand, user behaviors such as accessing, browsing, adding to favorites, adding to cart, placing orders, reviewing, and sharing accumulate, forming user behavior data that is time-series, correlated, and diverse. Simultaneously, marketing resources have gradually expanded from the relatively simple advertising slots, coupons, and promotional recommendations of the early stages to various forms including internal and external traffic entry points, personalized display positions, promotional tools, content delivery channels, and merchant collaborative resources. How to dynamically allocate limited marketing resources based on user behavior characteristics formed at different touchpoints has become a crucial technical direction for improving resource utilization efficiency, enhancing user reach, and supporting the platform's refined operation.

[0003] Among the relevant existing technologies, typical technical approaches include: First, users are categorized and segmented based on their historical transaction records, click data, or basic tag information, and then marketing resources are delivered to different user groups according to preset rules. Secondly, based on real-time access behavior or periodic statistical results, resources such as ad slots, coupons, and recommended content are scheduled in a rule-based manner; Third, use machine learning or recommendation algorithms to predict user preferences and automatically distribute some marketing resources accordingly.

[0004] While the aforementioned technical approaches can achieve a certain degree of marketing automation in practical applications, they still have significant limitations overall: On the one hand, existing solutions typically focus on a single type of data or a small number of structured indicators, making it difficult to simultaneously integrate behavioral information such as user browsing patterns, interaction rhythms, content feedback, transaction tendencies, and terminal environments, resulting in an insufficiently comprehensive characterization of user features.

[0005] On the other hand, existing resource allocation methods are often based on static rules, fixed weights, or single prediction results, which are insufficient to respond to changes in user status, marketing scenario switching, and resource competition. This can easily lead to problems such as delayed resource allocation, imbalanced resource allocation, or insufficient utilization of high-value resources.

[0006] In addition, some existing technologies focus more on the execution of individual marketing actions, while giving insufficient consideration to the synergistic relationships between multiple types of marketing resources, priority conflicts, and overall matching in the dynamic allocation process, thus affecting the accuracy and flexibility of marketing resource allocation in e-commerce scenarios.

[0007] Therefore, it is necessary to provide a dynamic marketing resource allocation solution for e-commerce business scenarios to more effectively adapt to the actual needs of constantly changing user behavior characteristics and improve the targeting, timeliness and coordination of the marketing resource allocation process. Summary of the Invention

[0008] To address the above problems, the present invention aims to propose a method for intelligent allocation of dynamic marketing resources in e-commerce based on user behavior characteristics, comprising the following steps: S1. Obtain e-commerce user behavior data within a specified time period, and preprocess the user behavior data. The preprocessing includes: identifying and removing behavior records with consecutive behavior intervals of less than 500 milliseconds and consistent source pages as redundant clicks; marking behavior records with dwell time of less than 2 seconds or greater than 300 seconds as feature-missing records; performing time correction on behavior records with incorrect timestamps; and correcting user identity information based on the correspondence between user ID, device number, and network IP to form a standardized behavior sequence for benchmark sample screening. S2. Select high-confidence behavior records that meet preset conditions from the preprocessed user behavior data as benchmark samples, and perform click behavior correction, exit behavior duration backtracking and dwell time offset correction on the remaining behavior records based on the benchmark samples to form a restored feature set. S3. Construct user feature vectors based on the restored feature set, and obtain allocable marketing resources in the e-commerce platform. Convert different types of marketing resources into a resource parameter group that includes available time range, applicable user range, budget consumption rate, placement frequency limit and current number of bound users. S4. Based on the labeled user behavior records and the user feature vector, determine the behavior tags for the unlabeled user behavior records and complete user grouping. Match each user group with the resource parameter group and select the resource set to be deployed according to the preset strategy rules. S5. Based on the rolling time window, continuously evaluate the marketing resources in the resource set to be deployed. When the conversion rate of any marketing resource fails to reach the preset threshold for five consecutive clicks since deployment, suspend the deployment of the marketing resource in the current round and update the resource set to be deployed. S6. The updated set of resources to be deployed is rearranged according to budget priority and user preferences, and marketing resources are distributed to corresponding user groups. A correlation mapping matrix between user groups and marketing resources is established. Feedback data is generated after the resources are deployed, and the feedback data and user behavior data for the next period are input into the next round of resource evaluation process.

[0009] As a preferred technical solution, the user behavior data comes from at least three data dimensions, including browsing history, click behavior, dwell time, jump behavior, adding to cart behavior, favorite behavior, purchase behavior, geolocation information, and device type.

[0010] As a preferred technical solution, the high-confidence behavior records in S2 simultaneously meet the following conditions: the time from entering the page to the first interaction is less than 10 seconds, the behavior segments containing continuous stay records for more than 30 seconds are included, and there are no interruption and reconnection events in the device switching path; only behavior records that simultaneously meet all of the above conditions are included in the benchmark sample set, and behavior records that do not meet all conditions are not included in the subsequent correction parameter generation.

[0011] As a preferred technical solution, in step S2, click behavior correction parameters, bounce behavior duration backtracking parameters, and dwell time offset correction parameters are generated based on the benchmark samples. The click behavior correction parameters are used to eliminate the impact of abnormal repeated clicks by the same user from the same page source. The bounce behavior duration backtracking parameters are used to fill in the time length of the continuous page dwell segment before the bounce. The dwell time offset correction parameters are used to offset and correct abnormal dwell times according to the dwell time distribution in the benchmark samples. The click behavior correction parameters, bounce behavior duration backtracking parameters, and dwell time offset correction parameters are then uniformly applied to non-benchmark sample behavior records to form the restored feature set.

[0012] As a preferred technical solution, the user feature vector constructed in S3 includes at least the browsing path length, single-category browsing depth, number of jumps per unit time, average single-item dwell time, dwell time fluctuation value, and continuous short-term browsing identifier. The continuous short-term browsing identifier is used to characterize whether the same user has three or more consecutive browsing paths with a single dwell time less than a preset duration threshold. The user feature vector serves as a unified input for user segmentation and resource matching.

[0013] As a preferred technical solution, the marketing resources in S3 include advertising exposure resources, discount coupon resources, and time-limited free shipping resources. Before entering the filtering process, different types of marketing resources are converted into resource parameter groups with consistent field structures. The field structure includes at least the available time range, applicable user range, budget consumption rate, maximum daily ad frequency, and current number of bound users to form a unified resource parameter pool, enabling different types of marketing resources to be compared, sorted, and called under the same filtering rules.

[0014] As a preferred technical solution, the preset strategy rules in S4 include a user interest priority threshold, a budget consumption ratio threshold, a user preference matching threshold, and an exclusivity condition. When the same marketing resource is bound to a mutually exclusive user group and the number of users in the mutually exclusive user group does not exceed the concurrent user limit, the marketing resource maintains a high priority in this round of screening. When the number of users in the mutually exclusive user group exceeds the concurrent user limit, the corresponding marketing resource is automatically reduced to the second priority, and the strategy hit source of the marketing resource is recorded.

[0015] As a preferred technical solution, the continuous evaluation in S5 includes at least calculating the reach rate, budget consumption ratio, and order conversion rate of each marketing resource, and forming a continuous click-conversion evaluation sequence in chronological order; when updating any rolling time window, the reach rate, budget consumption ratio, and order conversion rate of each marketing resource in the resource set to be deployed are recalculated, and the continuous click-conversion evaluation sequence is updated based on the recalculation results; when the click-conversion rate of any marketing resource fails to reach the preset threshold for five consecutive times since deployment, the deployment of that marketing resource in the current round is immediately suspended, and the deployment priority of the remaining marketing resources is updated simultaneously.

[0016] As a preferred technical solution, the association mapping matrix in S6 includes at least the resource allocation list, used budget, remaining budget, resource validity period, and remaining callable frequency corresponding to the user group; the feedback data includes at least unused resources, failed paths, time delays, and budget execution results; after each resource deployment, the association mapping matrix is ​​updated based on user behavior feedback, and the updated association mapping matrix is ​​used for the next round of resource evaluation and resource distribution.

[0017] This invention also provides a system for intelligent allocation of dynamic marketing resources in e-commerce based on user behavior characteristics, used to implement the method, including: The data acquisition and preprocessing module is used to acquire e-commerce user behavior data and form standardized behavior sequences. The benchmark sample screening and feature restoration module is used to screen high-confidence behavior records, generate click behavior correction parameters, bounce behavior duration backtracking parameters and dwell time offset correction parameters, and form a restored feature set; The feature construction and resource parameterization module is used to construct user feature vectors and convert different types of marketing resources into resource parameter groups with consistent field structures. The user segmentation and resource filtering module is used to determine behavior tags, segment users, and filter the set of resources to be deployed based on labeled user behavior records and user feature vectors. The dynamic evaluation and resource adjustment module is used to form a continuous click-conversion evaluation sequence based on a rolling time window. If the click-conversion rate of any marketing resource fails to reach the preset threshold after five consecutive clicks, the marketing resource will be suspended. The resource allocation and feedback update module is used to establish a mapping matrix between user groups and marketing resources, perform resource allocation, and generate feedback data.

[0018] The beneficial effects of this invention are as follows: First, this invention does not directly allocate marketing resources based on raw user behavior data. Instead, it first selects high-confidence behavior records from the user behavior data as benchmark samples, and then performs click behavior correction, bounce duration retrospection, and dwell time offset correction on the remaining behavior records, thereby forming a restored feature set that more closely approximates the true user intent. This improvement reduces the interference of abnormal clicks, false bounces, and dwell time deviations on subsequent judgment processes, allowing subsequent user segmentation and resource matching to be based on a more stable data foundation, and enhancing the consistency between resource allocation results and actual behavioral states.

[0019] Secondly, this invention unifies advertising exposure resources, discount coupon resources, and time-limited free shipping resources into a unified resource parameter group with a consistent field structure, and performs unified filtering and sorting based on user feature vectors. This improvement does not process individual marketing resources separately, but rather incorporates heterogeneous resources that were previously difficult to compare directly into the same allocation framework through a unified parameterization method. This allows different resources to complete matching, priority determination, and dynamic allocation under the same rules, which helps reduce the allocation fragmentation caused by differences in resource types and improves the organization and comparability of resource collaborative deployment.

[0020] Then, after completing the screening of resources to be deployed, this invention does not adopt a static deployment method. Instead, it continuously evaluates marketing resources based on a rolling time window, and pauses the current marketing resource if the click-through rate fails to reach a preset threshold for five consecutive times. Simultaneously, it incorporates feedback data and user behavior data from the next time period to enter the next round of resource evaluation. This improvement enables the resource deployment process to have continuous updating and timely adjustment capabilities, avoiding inefficient resources from continuously occupying budgets and deployment opportunities in the same round, and creating an iteratively optimized closed-loop relationship between user groups, marketing resources, and budget allocation. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0022] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention. Example

[0023] This embodiment provides a method for intelligent allocation of dynamic marketing resources in e-commerce based on user behavior characteristics, which can be applied to marketing scenarios on e-commerce platforms.

[0024] The method is deployed in a platform-side data processing environment, which is connected to the user behavior collection terminal, the marketing resource management terminal, and the order feedback terminal. This environment is used to collect, process, analyze, segment, filter, dynamically evaluate, and update user behavior data within a specified time period. It should be noted that this embodiment is only used to further illustrate the technical solution of this application and does not constitute a limitation on the scope of protection of this application.

[0025] In this embodiment, the e-commerce platform can be a comprehensive commodity trading platform or a vertical commodity sales platform. The platform provides registered users with services such as browsing, searching, clicking, adding to favorites, adding to cart, placing orders, making payments, and receiving discounts. The platform also has various available marketing resources, including advertising exposure resources, discount coupons, and time-limited free shipping resources. To avoid inconsistent participation in subsequent screening due to structural inconsistencies between different marketing resources, this embodiment first converts marketing resources from different sources and in different forms into resource parameter groups with a consistent field structure, and then dynamically allocates them based on the corrected user behavior characteristics. For example... Figure 1 As shown, it includes the following steps: S1. Obtain e-commerce user behavior data within a specified time period and perform preprocessing.

[0026] In this step, user behavior data is extracted from the platform's log system, event tracking system, and transaction record system over a specified seven-day period. This user behavior data includes at least several dimensions, such as browsing history, click behavior, dwell time, redirection behavior, adding to cart behavior, favorites behavior, purchase behavior, geolocation information, and device type. To ensure consistency in subsequent analysis, preprocessing is performed on the collected raw user behavior data.

[0027] Specifically, firstly, obviously invalid data records are deleted, such as records lacking both user and device identifiers, records with empty request paths, and records with unresolved timestamps. Secondly, click records from the same user on the same source page with consecutive actions less than 500 milliseconds apart are identified as redundant clicks and removed. Thirdly, the time field is standardized, converting all behavior records to a uniform time format, and the geographic location field is standardized to a uniform regional code. Behavior records with a dwell time of less than two seconds or more than three hundred seconds are not directly deleted but marked as feature-missing records for later restoration in conjunction with benchmark samples. Then, based on the correspondence between user ID, device number, and network IP, the potential identity splitting problem for the same user across devices and networks is corrected, ensuring that the original discrete behaviors can be attributed to the same actor as much as possible. After the above processing, a standardized behavior sequence is obtained for subsequent benchmark sample selection.

[0028] S2. Select high-confidence behavior records as the baseline sample, and perform behavior correction on the remaining behavior records.

[0029] In this step, behavioral records that can reliably reflect the browsing and interaction intentions of real users are further filtered from the standardized behavioral sequences as high-confidence behavioral records. Specific filtering criteria are: the time from page entry to the first interaction is less than ten seconds; it includes behavioral segments with continuous dwell times exceeding thirty seconds; and the behavioral path does not contain interruption and reconnection events during device switching. Only behavioral records that simultaneously meet all of the above conditions are included in the baseline sample set. Behavioral records that do not meet all conditions are not included in the generation of correction parameters but are still retained for subsequent correction and deployment analysis.

[0030] After obtaining the baseline samples, three types of correction parameters are generated based on them. The first type is click behavior correction parameters, used to eliminate the behavior inflation problem caused by abnormal repeated clicks by the same user on the same page source within a short period of time. The second type is bounce behavior duration retrospective parameters, used to retrospectively fill in the time of the page stay segment before the bounce when there is a bounce but there is still a continuous browsing trajectory before the bounce. The third type is stay duration offset correction parameters, used to offset and correct the stay duration of records marked as missing features or obviously abnormal, based on the stay duration distribution in the baseline samples.

[0031] For example, if a user makes three consecutive clicks within one second on the same product details page, and the page source, device environment, and session identifier are all consistent, the three clicks can be merged into one valid click based on the click behavior correction parameters. Another example is if a user closes a product recommendation page immediately after browsing, resulting in a shorter recorded dwell time on the last page. In this case, the dwell time before the user exits the page is adjusted by combining the continuous page dwell characteristics of similar benchmark samples. Yet another example is if a behavior record shows a dwell time of zero but the user subsequently adds items to their shopping cart. In this case, the dwell time is adjusted by combining the user's browsing paths for similar products and the distribution of benchmark samples. After these processes, a feature set that more closely approximates the actual browsing intent and interaction process is formed.

[0032] S3. Construct user feature vectors based on the restored feature set and perform unified parameterization on marketing resources.

[0033] In this step, behavioral features that characterize user browsing depth, interest concentration, switching frequency, and dwell stability are first extracted from the restored feature set, and a user feature vector is constructed. The user feature vector includes at least browsing path length, single-category browsing depth, number of jumps per unit time, average dwell time per item, dwell time fluctuation value, and a continuous short-term browsing identifier. The continuous short-term browsing identifier indicates whether the same user has a browsing path with more than three consecutive occurrences and a single dwell time less than a preset duration threshold. Through these features, user browsing behavior within a specified time period can be transformed from a single event record into a unified feature representation that can be compared and filtered.

[0034] Subsequently, the platform's currently allocable marketing resources undergo unified parameterization. Specifically, the original attribute information of ad exposure resources, discount coupon resources, and time-limited free shipping resources is extracted and converted into resource parameter groups with a consistent field structure. These resource parameter groups include at least the available time range, applicable user range, budget consumption rate, maximum daily ad frequency, and number of currently bound users. For ad exposure resources, the effective ad slot time period, applicable category, audience restrictions, and budget deduction rules can be written into corresponding fields; for discount coupon resources, the coupon threshold, applicable category, validity period, and single-user redemption limit can be converted into unified fields; for time-limited free shipping resources, their effective time interval, applicable category range, and resource call frequency limit can be converted into unified fields. After unified conversion, a unified resource parameter pool is formed, providing a basis for subsequent comparison, sorting, and calling under the same filtering rules.

[0035] S4. Complete the identification of user behavior tags, user segmentation, and screening of resources to be deployed.

[0036] In this step, a batch of labeled user behavior records are selected as reference records. The tags in these labeled user behavior records may include high purchase tendency, price sensitivity tendency, activity response tendency, quick bounce tendency, and deep browsing without placing an order tendency. Based on the labeled user behavior records and user feature vectors, behavioral tags are determined for unlabeled user behavior records. This determination is not based solely on a single browsing count or click volume, but rather comprehensively considers browsing path length, dwell stability, jump frequency, continuous short browsing states, and path changes before and after adding items to the cart or favorites, thus making the tag determination process more consistent with actual behavioral patterns.

[0037] After tagging is completed, users are segmented. For example, users can be divided into high-interest conversion groups, price-sensitive promotion response groups, short-term browsing intervention groups, and repeat visit activation groups. Different groups correspond to different marketing resource adaptation directions. Then, each user segment is matched with resource parameter groups, and a set of resources to be deployed is selected according to preset strategy rules. These preset strategy rules include user interest priority thresholds, budget consumption percentage thresholds, user preference matching thresholds, and mutually exclusive deployment conditions. If a marketing resource is already bound to a mutually exclusive deployment user group and the number of users in that user group does not exceed the concurrent user limit, then that marketing resource maintains high priority in this round of selection; if it exceeds the concurrent user limit, it automatically drops to secondary priority, and the strategy hit source of that marketing resource is recorded. This method avoids excessive concentration of the same type of resources within the same time period and also avoids conflicting resources acting on the same user group simultaneously.

[0038] For example, for price-sensitive promotional response groups, priority can be given to matching discount coupons; for high-interest groups awaiting conversion, priority can be given to matching a combination of ad exposure and free shipping; for groups requiring short-term browsing intervention, priority can be given to matching low-budget, high-frequency, short-cycle notification-type resources. After screening, the set of resources to be deployed in the current round is obtained.

[0039] S5. Continuously evaluate the set of resources to be deployed based on a rolling time window, and suspend inefficient resources.

[0040] In this step, a rolling time window is constructed with a fixed duration, such as every 30 minutes as an evaluation window. This window can also be set to shorter or longer durations depending on the platform's business rhythm. For each marketing resource in the set of resources to be deployed, its reach rate, budget consumption ratio, and order conversion rate are calculated separately, and then arranged in chronological order to form a continuous click-conversion evaluation sequence. Specifically, the reach rate represents the proportion of the resource that actually reaches the target user group, the budget consumption ratio represents the extent to which the budget has been used, and the order conversion rate represents the resource's ability to generate orders after deployment.

[0041] During any rolling time window update, the aforementioned metrics for each marketing resource in the resource set to be deployed are recalculated, and the continuous click-to-conversion evaluation sequence is updated accordingly. If a marketing resource fails to reach the preset threshold for click-to-conversion rate for five consecutive times since its launch, its deployment in the current round is immediately paused. After pause, the resource is removed from the current resource set to be deployed or downgraded to a pending observation status, and the deployment priority of the remaining marketing resources is updated simultaneously. This prevents continuously inefficient resources from occupying budget and deployment opportunities for extended periods, improving the dynamic adaptability of resource allocation within the current round.

[0042] For example, if a discount coupon is clicked in five consecutive evaluation windows but no valid orders are generated, and its click-through rate remains below a preset threshold, it can be determined that it is no longer suitable as a primary investment source for the current user group. The system will then suspend further deployment of the discount coupon and allocate more budget to free shipping or exposure resources that are more suitable for the same user group. This continuous evaluation and suspension mechanism allows resource deployment to be adjusted in a timely manner during the deployment process, rather than relying on a unified adjustment after a fixed period ends.

[0043] S6. Rearrange the updated set of resources to be deployed and perform resource distribution, while forming a closed-loop update based on feedback data.

[0044] In this step, after pausing inefficient resources and updating the resource set to be deployed, the updated resource set is rearranged according to budget priority and user preferences. Specifically, this considers both the current remaining budget, budget consumption rate, and available timeframe of the resources, as well as the degree of preference matching among user groups, recent behavioral tendencies, and historical response data of similar resources. After rearrangement, the corresponding marketing resources are distributed to the corresponding user groups, and a mapping matrix between user groups and marketing resources is established.

[0045] The association mapping matrix includes at least the resource allocation list, used budget, remaining budget, resource validity period, and remaining callable frequency for each user group. After resource deployment, the system further generates feedback data. This feedback data includes at least unused resources, failed paths, time delays, and budget execution results. Failed paths can be recorded as user reach but not click, click but not claim, claim but not place an order, or place an order but not complete payment. The time delay records the time interval between resource reach and response. After each deployment, the association mapping matrix is ​​updated based on user behavior feedback, and the updated matrix, along with user behavior data for the next time period, is input into the next round of resource evaluation.

[0046] For example, if a user group is allocated ad impressions and free shipping in the current round, and after the campaign is completed, it is found that ad impressions have a high reach rate but a low order conversion rate, while free shipping has a moderate reach rate but a high payment completion rate, then in the next round of resource reordering, the priority of free shipping resources can be increased, and the budget percentage of similar impression resources can be reduced. By continuously introducing new behavioral data and feedback data from the previous round of campaigns, this embodiment achieves a closed-loop resource allocation process from behavior collection, behavior correction, feature construction, user segmentation, resource selection, dynamic evaluation to feedback updates.

[0047] In summary, the method in this embodiment does not directly allocate marketing resources based on the original behavior logs. Instead, it first establishes a benchmark sample through high-confidence behavior records, and then uses the benchmark sample to correct click behavior, bounce duration, and dwell time offset of the remaining behavior records, thereby forming a restored feature set that is closer to the real user intent. On this basis, heterogeneous marketing resources are uniformly converted into a resource parameter group with a consistent structure, and unified screening and dynamic allocation are completed by combining user feature vectors. At the same time, the resource delivery status is continuously evaluated through a rolling time window, and the correlation mapping matrix is ​​continuously updated in combination with feedback data, so that the marketing resources form an iteratively adjustable closed-loop allocation process among different user groups, thereby completing the intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics. Example

[0048] This embodiment provides an intelligent allocation system for dynamic marketing resources in e-commerce based on user behavior characteristics. The system is used to implement the method described above.

[0049] The system can be deployed on the e-commerce platform's business server, marketing scheduling server, cloud data processing node, or a distributed processing environment consisting of multiple service nodes, to realize the dynamic allocation of marketing resources to different user groups.

[0050] In this embodiment, as Figure 2 As shown, the system includes: a data acquisition and preprocessing module, a benchmark sample screening and feature restoration module, a feature construction and resource parameterization module, a user segmentation and resource screening module, a dynamic evaluation and resource adjustment module, and a resource allocation and feedback update module.

[0051] The modules are connected according to a preset data flow sequence, or they can interact through a shared data pool, message queue, task scheduling bus, or unified scheduling interface. These modules can be deployed centrally or distributed, but their input-output relationships remain consistent.

[0052] The data acquisition and preprocessing module is used to acquire e-commerce user behavior data and form standardized behavior sequences.

[0053] Specifically, the data acquisition and preprocessing module connects to the platform's data collection terminal, page access log terminal, order transaction terminal, and user terminal identification terminal, and collects user browsing history, click behavior, dwell time, jump behavior, add to cart behavior, favorite behavior, purchase behavior, geographical location information, and device type and other behavioral data within a specified time period.

[0054] Since the raw data often comes from different sources, has inconsistent formats, and contains abnormal records, the data acquisition and preprocessing module further performs preprocessing operations. These preprocessing operations include removing invalid data, eliminating duplicate behaviors, standardizing time formats, normalizing geographic location information, and correcting user identity information. Specifically, removing invalid data deletes data with severely missing fields, invalid page sources, or abnormal time records; eliminating duplicate behaviors deletes redundant clicks with excessively short intervals between consecutive behaviors and consistent source pages; standardizing time formats unifies time fields from different sources into the same format; normalizing geographic location information converts location descriptions of different granularities into a unified geographic identifier; and correcting user identity information merges behavioral data generated by the same user on different devices and under different network environments based on the correspondence between user ID, device number, and network IP. After the above processing, a standardized behavioral sequence is output.

[0055] The benchmark sample screening and feature restoration module is used to screen high-confidence behavior records, generate correction parameters, and form a restored feature set.

[0056] After receiving the standardized behavior sequence, the benchmark sample screening and feature restoration module first selects high-confidence behavior records as benchmark samples according to preset conditions.

[0057] The preset conditions include: the time from page entry to the first interaction is less than a set value, the behavior path contains behavior segments with continuous dwell time exceeding a set duration, and the device switching path does not contain interruption and reconnection events. Only behavior records that simultaneously meet all conditions are included in the benchmark sample set. Subsequently, click behavior correction parameters, exit behavior duration backtracking parameters, and dwell time offset correction parameters are generated based on the benchmark samples.

[0058] The click behavior correction parameter is used to eliminate the impact of abnormal repeated clicks by the same user on the same page source; the bounce behavior duration backtracking parameter is used to fill in the time length of the continuous page stay segment before the bounce; the stay duration offset correction parameter is used to offset and correct the abnormal stay duration according to the stay duration distribution in the benchmark sample.

[0059] Subsequently, the benchmark sample screening and feature restoration module applies the three types of correction parameters to non-benchmark sample behavior records to form a restored feature set. By setting this module, the system does not directly send the original behavior logs into subsequent analysis, but first uses high-confidence behavior records to establish correction criteria, and then restores the remaining behaviors, thereby making the basis of subsequent analysis closer to the real user intent.

[0060] The feature construction and resource parameterization module is used to construct user feature vectors and convert different types of marketing resources into resource parameter groups with consistent field structures.

[0061] Specifically, the feature construction and resource parameterization module extracts features such as browsing path length, single-category browsing depth, number of jumps per unit time, average single-item dwell time, dwell time fluctuation value, and continuous short-term browsing identifiers from the restored feature set, and combines them to form a user feature vector.

[0062] The continuous short-duration browsing identifier is used to characterize whether the same user has a browsing path that is repeated more than three times with each visit lasting less than a preset duration threshold. Through this user feature vector, the originally discrete user behavior records are transformed into a unified input that can be compared, matched, and grouped.

[0063] On the other hand, the feature construction and resource parameterization module also connects to the marketing resource management terminal to read the currently allocable marketing resources on the platform. These marketing resources include advertising exposure resources, discount coupons, and time-limited free shipping resources. Because different marketing resources differ in their sources, fields, and constraints, this module performs unified parameterization processing on different marketing resources, converting them into resource parameter groups with consistent field structures. These resource parameter groups at least include the available time range, applicable user range, budget consumption rate, maximum daily ad frequency, and currently bound user count. After unified parameterization, different types of marketing resources can be compared, sorted, and invoked under the same filtering logic.

[0064] The user segmentation and resource filtering module is used to determine behavioral tags, segment users, and filter the set of resources to be deployed.

[0065] Specifically, the user segmentation and resource filtering module has a pre-set batch of labeled user behavior records. The tags in the labeled user behavior records may include deep browsing tendency, price sensitivity tendency, activity response tendency, quick exit tendency, and conversion tendency.

[0066] After receiving user feature vectors, the user segmentation and resource filtering module determines behavior tags for unlabeled user behavior records based on labeled user behavior records and completes user segmentation based on the behavior tag results. After segmentation, each user segment is matched with resource parameter groups, and a set of resources to be deployed is selected according to preset strategy rules. The preset strategy rules include user interest priority threshold, budget consumption ratio threshold, user preference matching threshold, and mutually exclusive deployment conditions.

[0067] For example, if a marketing resource is bound to a mutually exclusive user group and the number of users in that user group does not exceed the concurrent user limit, then that marketing resource will maintain high priority in this round of screening; if it exceeds the concurrent user limit, it will automatically be downgraded to the second priority, and the corresponding strategy hit source will be recorded. By setting this module, the system can incorporate user group characteristics, resource constraints, and ad exclusion relationships into the screening process simultaneously, rather than simply allocating based on a single click behavior or a single budget condition.

[0068] The dynamic evaluation and resource adjustment module is used to form a continuous click conversion evaluation sequence based on a rolling time window, and to pause the marketing resource delivery when the marketing resource is continuously below a preset threshold.

[0069] Specifically, after receiving the set of resources to be deployed, the dynamic evaluation and resource adjustment module continuously calculates the reach rate, budget consumption ratio and order conversion rate of each marketing resource using a rolling time window as the evaluation unit, and forms a continuous click conversion evaluation sequence in chronological order.

[0070] Each time the rolling time window updates, the reach rate, budget consumption ratio, and order conversion rate of each marketing resource in the resource set to be deployed are recalculated, and the continuous click-to-conversion evaluation sequence is updated accordingly. If the click-to-conversion rate of any marketing resource fails to reach the preset threshold for five consecutive times since its launch, the dynamic evaluation and resource adjustment module immediately suspends the deployment of that marketing resource in the current round and simultaneously updates the deployment priority of the remaining marketing resources. Through this structure, the system can promptly identify inefficient resources before the resource deployment ends, preventing them from continuously consuming budget and deployment opportunities, thereby improving the real-time adjustment capability during resource scheduling.

[0071] The resource allocation and feedback update module is used to establish a correlation mapping matrix between user groups and marketing resources, perform resource allocation, and generate feedback data.

[0072] Specifically, the resource allocation and feedback update module receives the updated set of resources to be deployed, rearranges the resources according to budget priority and user preferences, and distributes the resources to the corresponding user groups. During the resource distribution process, a correlation mapping matrix is ​​established between user groups and marketing resources. The correlation mapping matrix includes at least the resource allocation list, used budget, remaining budget, resource validity period, and remaining callable frequency for each user group. After the resource deployment is completed, the resource allocation and feedback update module further generates feedback data.

[0073] The feedback data includes at least unused resources, failure paths, time delays, and budget execution results. Failure paths can be used to describe status paths such as "reached but not clicked," "clicked but not claimed," "claimed but not ordered," or "ordered but not paid." Time delays can be used to record the time interval between resource reach and actual user feedback. Budget execution results can be used to record the budget usage of resources in the current round.

[0074] Subsequently, the resource allocation and feedback update module updates the association mapping matrix based on the user's behavior feedback again, and sends the updated association mapping matrix and the user behavior data of the next time period into the next round of resource evaluation process to form a closed-loop update mechanism.

[0075] In a specific operational scenario, the e-commerce platform activates the system described in this embodiment during the evening peak marketing period. The data acquisition and preprocessing module first extracts user browsing, clicking, favorites, adding to cart, and order placement behavior data within a specified time period from the log system and the order system, and forms a standardized behavior sequence.

[0076] Subsequently, the benchmark sample selection and feature restoration module selects high-confidence behavior records as benchmark samples, corrects for abnormal clicks, missing pre-bounce dwell times, and abnormal dwell durations, and outputs a restored feature set. The feature construction and resource parameterization module then constructs user feature vectors based on the restored feature set and uniformly converts ad exposure resources, discount coupon resources, and time-limited free shipping resources into resource parameter groups.

[0077] The user segmentation and resource filtering module uses this information to determine behavioral tags and segment users, and then filters out the set of resources to be deployed based on preset strategy rules.

[0078] The dynamic evaluation and resource adjustment module continuously generates a continuous click-to-conversion evaluation sequence during resource deployment. Resources that fail to reach a preset threshold for conversion rate after five consecutive clicks are paused. The resource allocation and feedback update module distributes resources according to the updated deployment priority and updates the associated mapping matrix and feedback data after deployment to support the next round of dynamic evaluation.

[0079] In summary, the system provided in this embodiment is not simply a patchwork of user behavior analysis and marketing delivery modules. Instead, its system structure incorporates a "benchmark sample screening and behavior correction" processing link, a dual-unified structure of "user feature vectors and unified resource parameter groups," a dynamic judgment structure of "continuous click conversion evaluation under a rolling time window," and a closed-loop update structure of "association mapping matrix and feedback data updates." This creates a continuous linkage between user behavior analysis, resource screening, resource delivery, and delivery feedback. Through this modular system structure, intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics can be achieved.

[0080] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent allocation of dynamic marketing resources in e-commerce based on user behavior characteristics, characterized in that, Includes the following steps: S1. Obtain e-commerce user behavior data within a specified time period, and preprocess the user behavior data. The preprocessing includes: determining and removing behavior records with consecutive behavior intervals of less than 500 milliseconds and consistent source pages as redundant clicks; unifying the time field to convert all behavior records into a unified time format, and standardizing the geographic location field to a unified regional code. Records of behaviors with a dwell time of less than 2 seconds or more than 300 seconds are marked as records with missing features; time correction is performed on behavior records with incorrect timestamps; Based on the correspondence between user ID, device number, and network IP, user identity information is corrected to form a standardized behavioral sequence for benchmark sample screening; S2. Select high-confidence behavior records from the preprocessed user behavior data as benchmark samples, and perform click behavior correction, exit behavior duration backtracking and dwell time offset correction on the remaining behavior records based on the benchmark samples to form a restored feature set; S3. Construct a user feature vector based on the restored feature set, obtain the allocatable marketing resources in the e-commerce platform, and convert different types of marketing resources into a resource parameter group that includes available time range, applicable user range, budget consumption rate, placement frequency limit and current number of bound users. S4. Based on the labeled user behavior records and the user feature vector, determine the behavior tags for the unlabeled user behavior records and complete user grouping. Match each user group with the resource parameter group and select the resource set to be deployed according to the preset strategy rules. S5. Based on the rolling time window, continuously evaluate the marketing resources in the resource set to be deployed. When the conversion rate of any marketing resource fails to reach the preset threshold for five consecutive clicks since deployment, suspend the deployment of the marketing resource in the current round and update the resource set to be deployed. S6. The updated set of resources to be deployed is rearranged according to budget priority and user preferences, and marketing resources are distributed to corresponding user groups. A correlation mapping matrix between user groups and marketing resources is established. Feedback data is generated after the resources are deployed, and the feedback data and user behavior data for the next period are input into the next round of resource evaluation process.

2. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The high-confidence behavior records in S2 simultaneously meet the following conditions: T1: The time from entering the page to the first interaction is less than 10 seconds; T2: Includes behavioral segments with continuous dwell time exceeding 30 seconds; T3: and does not include interruption and reconnection events in the device switching path; Only behavioral records that simultaneously meet all of the above conditions are included in the baseline sample set; behavioral records that do not meet any of the conditions are not included in the subsequent generation of correction parameters.

3. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: In step S2, click behavior correction parameters, bounce behavior duration backtracking parameters, and dwell time offset correction parameters are generated based on the benchmark sample. The click behavior correction parameter is used to eliminate the impact of abnormal repeated clicks by the same user on the same page source; the bounce behavior duration backtracking parameter is used to fill in the time length of the continuous page stay segment before the bounce; and the stay duration offset correction parameter is used to offset and correct the abnormal stay duration according to the stay duration distribution in the benchmark sample. The click behavior correction parameters, the bounce behavior duration backtracking parameters, and the dwell time offset correction parameters are uniformly applied to the non-benchmark sample behavior records to form the restored feature set.

4. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The user feature vector constructed in S3 includes at least the browsing path length, single-category browsing depth, number of jumps per unit time, average single-item dwell time, dwell time fluctuation value, and continuous short-term browsing identifier. The continuous short-term browsing identifier is used to characterize whether the same user has three or more consecutive browsing paths with a single dwell time of less than a preset duration threshold. The user feature vector serves as a unified input for user segmentation and resource matching.

5. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The marketing resources in S3 include advertising exposure resources, discount coupons, and time-limited free shipping resources. Different types of marketing resources are converted into resource parameter groups with consistent field structures before entering the filtering process. The field structure includes at least the available time range, applicable user range, budget consumption rate, maximum daily ad frequency, and current number of bound users, forming a unified resource parameter pool that allows different types of marketing resources to be compared, sorted, and invoked under the same filtering rules.

6. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The preset strategy rules in S4 include: user interest priority threshold, budget consumption ratio threshold, user preference matching threshold, and exclusion condition; When the same marketing resource is bound to a mutually exclusive user group and the number of users in that mutually exclusive user group does not exceed the concurrent user limit, the marketing resource will maintain high priority in this round of screening. When the number of users in the mutually exclusive user group exceeds the concurrent user limit, the corresponding marketing resource is automatically downgraded to the second priority, and the source of the marketing resource's strategy hit is recorded.

7. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The continuous evaluation in S5 includes at least the following: calculating the reach rate, budget consumption ratio and order conversion rate of each marketing resource, and forming a continuous click-conversion evaluation sequence in chronological order; When updating at any rolling time window, recalculate the reach rate, budget consumption ratio, and order conversion rate of each marketing resource in the resource set to be deployed, and update the continuous click conversion evaluation sequence based on the recalculation results; when the click conversion rate of any marketing resource fails to reach the preset threshold for five consecutive times since deployment, immediately suspend the deployment of the marketing resource in the current round, and simultaneously update the deployment priority of the remaining marketing resources.

8. The method for intelligent allocation of e-commerce dynamic marketing resources based on user behavior characteristics according to claim 1, characterized in that: The association mapping matrix in S6 includes at least: the resource allocation list, used budget, remaining budget, resource validity period, and remaining callable frequency corresponding to the user group; The feedback data includes at least unused resources, failed paths, time delays, and budget execution results. After each resource deployment, the association mapping matrix is ​​updated based on user behavior feedback, and the updated association mapping matrix is ​​used for the next round of resource evaluation and resource distribution.

9. A system for intelligent allocation of dynamic marketing resources in e-commerce based on user behavior characteristics, used to implement the method as described in any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to acquire e-commerce user behavior data and form standardized behavior sequences. The benchmark sample screening and feature restoration module is used to screen high-confidence behavior records, generate click behavior correction parameters, bounce behavior duration backtracking parameters and dwell time offset correction parameters, and form a restored feature set; The feature construction and resource parameterization module is used to construct user feature vectors and convert different types of marketing resources into resource parameter groups with consistent field structures. The user segmentation and resource filtering module is used to determine behavior tags, segment users, and filter the set of resources to be deployed based on labeled user behavior records and user feature vectors. The dynamic evaluation and resource adjustment module is used to form a continuous click-conversion evaluation sequence based on a rolling time window. If the click-conversion rate of any marketing resource fails to reach the preset threshold after five consecutive clicks, the marketing resource will be suspended. The resource allocation and feedback update module is used to establish a mapping matrix between user groups and marketing resources, perform resource allocation, and generate feedback data.

Images