User attribution method and system capable of meeting various configurations and conversions

Abstract

The invention provides a user attribution method and system capable of meeting various configurations and conversions, and belongs to the technical field of computer networks, and the method comprises the following steps: according to a unified identity mapping table, ordering behavior records on each device according to a time sequence, integrating the behavior records into a continuous behavior path data set, and determining a full-link view of a user conversion path; carrying out fragmentation storage and parallel analysis by adopting a distributed computing framework through the behavior path data set, and optimizing resource allocation to obtain a real-time processed user conversion path result set; according to a user conversion path result set processed in real time, a cache mechanism is constructed to store high-frequency access path data, and if a new query request is received, data is extracted from a cache preferentially, and a quick-response path analysis result is determined; through a path analysis result, an encryption technology is applied to protect a user identity identifier and behavior data, and in combination with an access control strategy, a safe and compliant attribution analysis data set is obtained. According to the method, the problem of cross-device user behavior tracking is solved, a full-link user conversion path view is provided, the path analysis requirement can be quickly responded, and data support is provided for precise marketing decision making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer networks, and particularly relates to a user attribution method and system that can meet various configurations and conversions. BACKGROUND

[0002] In the field of digital marketing, user attribution technology is an important bridge connecting advertising delivery and conversion effectiveness, and its accuracy directly affects the optimization of marketing strategies and the efficiency of resource allocation.

[0003] With the diversification of user behavior and the increasing demand for privacy protection, how to accurately track user conversion paths and protect data security has become a key issue that the industry needs to tackle.

[0004] Current user attribution methods have significant limitations in dealing with complex scenarios.

[0005] Many solutions rely too much on single device information or fixed tracking methods, making it difficult to adapt to user behavior patterns across devices and platforms.

[0006] The deeper problem is that these methods often ignore the balance between real-time and security when dealing with large-scale data, resulting in slow system response in high-concurrency environments or vulnerabilities in data protection, making it difficult to meet the dual needs of dynamic adaptation and privacy compliance in modern marketing.

[0007] The core challenge in this field is how to achieve accurate tracking and efficient processing in the face of diverse user behavior.

[0008] The first problem is the identification of user identity. When users frequently switch devices or change network environments, traditional identification methods can easily fail, causing the breakage of behavior links.

[0009] Further, this identification problem also brings a huge pressure on computing resources, as it requires processing massive amounts of data and completing multi-dimensional feature integration and analysis within a short period of time. If the computing process cannot be optimized, it will be difficult to meet the needs of real-time attribution.

[0010] These two problems are intertwined, requiring technology to be flexible and adaptable to different scenarios, and also requiring performance breakthroughs.

[0011] Therefore, how to ensure data security while integrating multi-dimensional features to achieve dynamic identification of user identity, and building an efficient computing framework to cope with high-concurrency scenarios, has become a key problem that needs to be solved in this research. SUMMARY

[0012] The present application provides a user attribution method that can meet various configurations and conversions, mainly including:

[0013] Obtain the interaction records of multiple data sources from a pre-established user behavior database, extract the original data set containing time stamp, access path and operation type for cross-device tracking requirements, and obtain a preliminary user behavior feature set;

[0014] According to the preliminary user behavior feature set, a multi-dimensional feature integration method is adopted to standardize the time, path and operation type, a unified feature vector matrix is constructed for data heterogeneity problem, and the user behavior correlation of cross-device is determined;

[0015] Through the feature vector matrix, the behavior characteristics between different devices are compared by applying a preset similarity calculation rule, if the similarity exceeds a preset threshold, it is judged as the same user identity, and a unified identity mapping table is obtained;

[0016] According to the unified identity mapping table, the behavior records on each device are sorted according to time sequence, and integrated into continuous behavior path data set, and the full link view of user conversion path is determined;

[0017] Through the behavior path data set, distributed computing framework is adopted for sharding storage and parallel analysis, resource allocation is optimized, and real-time processing user conversion path result set is obtained;

[0018] According to the real-time processing user conversion path result set, a cache mechanism is constructed to store high-frequency access path data, if a new query request is received, data is extracted from the cache first, and the path analysis result of fast response is determined;

[0019] Through the path analysis result, encryption technology is applied to protect user identity and behavior data, combined with access control strategy, safe and compliant attribution analysis data set is obtained.

[0020] The present application provides a user attribution system that can meet various configurations and conversions, mainly including:

[0021] The user behavior feature extraction module is used for obtaining the interaction records of multiple data sources from a pre-established user behavior database, extracting the original data set containing time stamp, access path and operation type for cross-device tracking requirements, and obtaining a preliminary user behavior feature set;

[0022] The multi-dimensional feature integration module is used for integrating the preliminary user behavior feature set, adopting a multi-dimensional feature integration method to standardize the time, path and operation type, constructing a unified feature vector matrix for data heterogeneity problem, and determining the user behavior correlation of cross-device;

[0023] The similarity calculation module is configured to apply a preset similarity calculation rule to compare the behavior characteristics between different devices through the feature vector matrix, and if the similarity exceeds a preset threshold, the same user identity is determined, and a unified identity mapping table is obtained.

[0024] The behavior path integration module is configured to sort the behavior records on each device in a time sequence according to the unified identity mapping table, integrate the behavior records into a continuous behavior path data set, and determine a full-link view of the user conversion path.

[0025] The distributed computing module is configured to perform sharding storage and parallel analysis by using a distributed computing framework through the behavior path data set, optimize resource allocation, and obtain a real-time processed user conversion path result set.

[0026] The cache optimization module is configured to construct a cache mechanism to store high-frequency access path data according to the real-time processed user conversion path result set, and if a new query request is received, the data is extracted from the cache to determine a fast response path analysis result.

[0027] The security compliance module is configured to apply encryption technology to protect the user identity and behavior data through the path analysis result, combine an access control strategy, and obtain a security-compliant attribution analysis data set.

[0028] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0029] The application discloses a user attribution system and method capable of meeting multiple configurations and conversions, and realizes cross-device user behavior tracking and conversion path analysis. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The flowchart of the user attribution method capable of meeting multiple configurations and conversions of the application.

[0031] Figure 2 The schematic diagram of the user attribution method capable of meeting multiple configurations and conversions of the application.

[0032] Figure 3Another schematic diagram of the user attribution method that can meet various configurations and transformations of the present application.

[0033] Figure 4 A structural schematic diagram of the user attribution system that can meet various configurations and transformations of the present application. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be described clearly and in detail below with reference to the drawings in the embodiments of the present application. The described embodiments are only some of the embodiments of the present application.

[0035] As shown in Figure 1 , Figure 2 and Figure 3 , the present embodiment can meet various configurations and transformations of the user attribution method, which can specifically include:

[0036] Step S101, obtaining the interaction records of multiple data sources from a pre-established user behavior database, extracting the original data set containing the time stamp, access path and operation type for the cross-device tracking demand, and obtaining the preliminary user behavior feature set.

[0037] Obtain the user interaction records stored in multiple data sources, read the original interaction data containing the device identifier, time stamp, access path through the data interface, and obtain the complete cross-device behavior data set. According to the time stamp information in the interaction record, the time window segmentation method is used to divide the user behavior data in time sequence, if the time interval exceeds the preset threshold, it is judged as different sessions, and the behavior fragments organized in time sequence are obtained. By analyzing the access path features in the behavior fragments, the page jump sequence, dwell time, click position and other path attributes are extracted, and the browsing track mode of the user on different devices is determined. The clustering algorithm is used to group process the browsing track mode, the behavior correlation degree between devices is calculated according to the path similarity and operation type similarity, and the potential same user cross-device behavior cluster is obtained. If the same access path sequence exists in the behavior cluster and the time stamp is continuous, the device ownership relationship is determined through the user identity association rule, and the cross-device user identity matching result is determined. According to the identity matching result, the sequence pattern mining algorithm is used to extract the frequent behavior mode from the user behavior feature set, and the user cross-device operation habit and preference feature is obtained. By fusing the operation habit and preference feature and the device use time distribution, the user cross-device behavior profile is constructed, and the complete behavior track and mode recognition label of each user on different devices is determined.

[0038] Illustratively, the system first extracts user interaction records from seven different data sources such as mobile applications, web pages, smart TVs, etc. through a distributed data collection interface, uses a user identifier-based association algorithm to aggregate the behavior data of the same user on different devices, and establishes a cross-device user portrait through device fingerprint technology and Cookie mapping. In the data preprocessing stage, the time series alignment algorithm is used to convert the time stamps from different time zones into UTC format, ensuring the consistency of the time dimension, and the data deduplication algorithm is used to eliminate duplicate records, finally obtaining an original data set containing 12,000 valid interaction records. In the feature extraction process, the system uses the N-gram algorithm to analyze the user access path sequence, converts the page jump path into a path segment with a length of 3, calculates the path weight through the TF-IDF algorithm, identifies the user's common navigation mode, and divides the user behavior into four categories: browsing, clicking, purchasing, and sharing based on the operation type classifier. Each type of behavior is assigned a different weight coefficient, which is 0.1, 0.3, 0.8, and 0.5 respectively. Similar behavior patterns are classified through clustering algorithms, and the Pearson correlation coefficient is used to calculate the correlation between different features, finally constructing a user behavior feature vector containing 15 feature variables in three dimensions: time preference, path preference, and operation preference, providing a data basis for subsequent personalized recommendation and user clustering analysis.

[0039] Step S102, according to the preliminary user behavior feature set, using multi-dimensional feature integration method to standardize the time, path and operation type, constructing a unified feature vector matrix for data heterogeneity problem, determining the cross-device user behavior association.

[0040] By processing time information, path records and operation categories, the feature set is cleaned and format-converted using a data preprocessing tool to obtain a standardized behavior data set. According to the standardized behavior data set, a multi-dimensional feature extraction method is used to hierarchically map the time information and path records, construct a preliminary feature vector, and determine the distribution pattern of cross-device behavior. If the distribution pattern of the feature vector has significant differences between different devices, the matching degree of the operation category and the path record is compared to determine whether there is behavior correlation, and a preliminary correlation result is obtained. For the preliminary correlation result, a clustering algorithm is used to group the feature vectors, and the cross-device nature and the similarity of user behavior are classified to obtain a hierarchical behavior correlation cluster. According to the behavior correlation cluster, the continuity features of the time information and the operation category are extracted, a unified vector matrix is constructed, and the deep connection of cross-device user behavior is determined. Through the vector matrix, a standardization processing method is used to smooth the adjustment of the distribution characteristics of cross-device behavior, and a final user behavior correlation mapping is obtained. If there is incomplete behavior data of some devices in the user behavior correlation mapping, an integration method is used to interpolate and supplement the missing part to determine the complete cross-device behavior correlation structure.

[0041] For example, for the initial set of user behavior features, the system first standardizes time, path, and operation type, uses a Z-score-based standardization algorithm to adjust the data range of the time dimension to a distribution with a mean of 0 and a standard deviation of 1, for example, converts the user's daily active time from the original 0 to 24 hour interval to a standardized value, where the average active time of a certain user is 6.5 hours, and the standardized value is 0.8 after calculation. At the same time, the path data is normalized by length, and the average number of jumps in the access path is limited to within 5 times. If a user's path jump number is 8 times, the first 5 jump records are retained by the truncation algorithm, and the operation type is converted to a numerical vector by encoding, for example, the "search" operation is encoded as 1.0, and the "collection" operation is encoded as 2.0, to ensure the consistency of the dimensions of different types of data. Subsequently, to address the data heterogeneity problem, the system constructs a unified feature vector matrix and uses principal component analysis (PCA) to reduce the dimensionality of the multi-dimensional features, for example, extracts the top 95% of the main components from the initial 20 feature dimensions, and finally retains 8 dimensions to generate an 8-column feature matrix. Each user's feature vector contains time activity, path complexity, and other indicators. The vector value of a certain user in the matrix is [0.75, 1.23, 0.45, 2.10, 0.88, 1.15, 0.32, 0.67]. Finally, to determine the correlation of user behavior across devices, the system introduces a cosine similarity-based calculation method to analyze the similarity of feature vectors between different devices, for example, the similarity between user A's feature vector on a mobile device and the vector on a tablet device is 0.92, indicating a high correlation between behaviors. If the similarity is less than 0.5, a secondary verification is performed by supplementing business data such as device login frequency (e.g., 3 daily logins) to ensure the accuracy of the correlation judgment, forming a complete logical chain from data standardization to correlation analysis.

[0042] Step S103, by the feature vector matrix, apply a pre-set similarity calculation rule to compare the behavior features between different devices. If the similarity exceeds a pre-set threshold, it is determined to be the same user identity, and a unified identity mapping table is obtained.

[0043] The feature vector matrix is used to compare the device behaviors by using a preset evaluation rule, calculate the similarity values between different devices, and obtain a preliminary similarity distribution result. According to the similarity distribution result, for the device behaviors with a similarity value higher than a preset threshold, a correlation judgment is performed to determine a device combination that may belong to the same user identity. A clustering algorithm is used to group the device combination, and a classified device correlation group is obtained according to the similarity values of the device behaviors and the distribution characteristics of the feature vectors. Through the device correlation group, a common mode of the device behaviors is extracted, a unified identity mapping framework is constructed, and a cross-device user identity correspondence relationship is determined. For the identity mapping framework, a data verification tool is used to analyze the device differences. If it is found that part of the device behavior data is inconsistent, a data fusion method is used for adjustment to obtain a revised identity mapping table. According to the revised identity mapping table, the correlation information of the device behaviors is integrated, a complete cross-device user identity profile is constructed, and a final unified identification structure is determined. Through the unified identification structure, the feature vectors of all device behaviors are standardized, a cross-device user behavior consistency record is generated, and the final integration of identity correlation is completed.

[0044] For example, based on the feature vector matrix, the system first compares and analyzes the behavior characteristics between different devices, calculates the distance difference between the feature vectors of the devices by using the Euclidean distance algorithm, for example, the feature vector of a user on a notebook computer is [1.25, 0.88, 2.34, 1.67], and the feature vector of the user on a smart watch is [1.30, 0.85, 2.28, 1.72]. The Euclidean distance between the two vectors is calculated to be 0.09, indicating that the behavior characteristics of the two devices are very close. Then, the system compares the calculated distance value with a preset threshold value 0.15. If the distance is less than the threshold value, it is preliminarily determined that it is the same user identity. Then, the system introduces a weighted scoring mechanism for further verification, assigns a higher weight to the key dimensions in the behavior characteristics, such as operation frequency, for example, the operation frequency weight is set to 0.4, and the weight of other dimensions is 0.2. The weighted comprehensive similarity score is calculated. If the score is 0.87 and exceeds the preset threshold value 0.8, it is confirmed that the behaviors of the devices are consistent. Finally, the system generates a unified identity mapping table, maps the feature data corresponding to different devices to the same user ID, for example, user ID_001 is associated with notebook computer device number NB_123 and watch device number SW_456, forms a mapping record, and performs logical verification in combination with auxiliary data such as device registration time (for example, the registration time interval is less than 7 days) to ensure that the mapping table is accurate and correct. The whole process is automatically completed by the algorithm, and a complete chain from feature comparison to identity unification is constructed.

[0045] Step S104, according to the unified identity mapping table, the behavior records on each device are sorted in time sequence, integrated into continuous behavior path data set, and the full link view of user conversion path is determined.

[0046] Through the identity and mapping table, the device records are preliminarily classified and arranged in sequence according to time sequence, and the device behavior data set sorted by time is obtained. According to the device behavior data set sorted by time, the key time nodes of user behavior are extracted by using time series analysis method, a preliminary behavior path framework is constructed, and the initial structure of the behavior path is determined. According to the initial structure of the behavior path, the behavior continuity in the device records is verified in combination with the characteristics of the integrated set. If there is a time breakpoint or data loss in the behavior path, the data interpolation method is used for supplement, and the complete path data set is obtained. Through the complete path data set, the correlation between user behavior and conversion path is analyzed, the clustering algorithm is used to group the behavior path, and the classification set of different conversion paths is obtained. According to the classification set, the common features in each group of conversion paths are extracted, the framework model of the full link view is constructed, if the path features of a group do not match the overall view, the feature adjustment tool is used for correction, and the unified path view structure is determined. According to the unified path view structure, the user behavior and behavior continuity data are integrated, the cross-device behavior path panoramic record is generated, and the final conversion path full link view is determined. Through the conversion path full link view, the key nodes in the behavior path are labeled, the data storage tool is used to save the structure information of the full link view, and the persistent path view data is obtained.

[0047] Exemplarily, based on the unified identity mapping table, the system first integrates the user's behavior records on different devices, automatically extracts the behavior data of user ID_002 on mobile device MP_789, such as browsing record timestamp 2023-10-01 10:15:23, click time of a certain application page for 120 seconds, and behavior data on tablet device TB_456, such as login timestamp 2023-10-01 11:20:45, search keyword frequency 5 times / minute, and sorts these records by timestamp field to form an early to late time series data set. Subsequently, the system utilizes the time window algorithm to classify the behavior records with a time interval less than 30 minutes as the same session, for example, the behavior time difference of the above mobile phone and tablet is 65 minutes, which exceeds the threshold and is divided into two sessions, while the time interval of three consecutive clicks in the mobile phone is less than 10 minutes, which is classified as one session, generating session ID_S001. Next, the system connects the behavior records in each session in chronological order through the behavior path analysis algorithm to construct the user conversion path, for example, the user jumps from mobile phone browsing page to tablet search keyword in session ID_S001, and the path weight is calculated as 0.75 according to the behavior frequency, which exceeds the preset threshold 0.5 and is marked as a high-frequency path. Finally, the system integrates all session paths to generate a full-link view data set, for example, the full path of user ID_002 is displayed as "mobile phone browsing-tablet searching-mobile phone payment", and combined with business data such as payment amount 500 yuan for correlation analysis, the payment behavior is automatically marked as the core node to form a complete logical chain from behavior record to path view, which assists subsequent business optimization.

[0048] Step S105, through the behavior path data set, a distributed computing framework is used for sharded storage and parallel analysis to optimize resource allocation, and a real-time processed user conversion path result set is obtained.

[0049] The behavior path data set is obtained, and the data is partitioned by user identification using a hash fragmentation algorithm. If the data volume exceeds a preset threshold, a multi-level fragmentation mechanism is started to obtain a uniformly distributed data fragment set. For the data fragment set, parallel processing is performed by computing nodes in a distributed architecture, and the processing tasks are dynamically allocated according to the computing capacity of each node to obtain a node-level processing task allocation table. According to the processing task allocation table, each computing node synchronously performs a path analysis operation, and a load balancing strategy is used to monitor the node running state. If the processing speed of a certain node is lower than the average level, the task allocation weight is readjusted to determine the optimized computing resource configuration. Through the optimized computing resource configuration, the intermediate results generated by each node are aggregated in real time, and the fragment calculation results are combined using a streaming processing technology to obtain unified path analysis intermediate data. For the path analysis intermediate data, a memory caching mechanism is used to improve data access speed, and the results are sorted according to the time characteristics of the conversion path to obtain a time-sequenced conversion path candidate set. According to the conversion path candidate set, a clustering algorithm is used to identify similar path patterns. If the path similarity is higher than a set standard, they are merged into the same category to determine the final conversion path type grouping. Through the conversion path type grouping, the statistical characteristics and behavior patterns of each group are integrated, and a data compression technology is used to optimize the storage space to obtain a complete real-time conversion path analysis result set.

[0050] For example, the system uses the Apache Spark distributed computing framework to perform fragmentation processing on the user conversion path data set, and the behavior path data containing user ID_005 is distributed to 8 computing nodes according to the hash algorithm. Each node processes about 125,000 records, and the memory-level computing speed is accelerated through the RDD elastic distributed data set. The system configures the memory cache of each fragment to be 2GB, and the CPU core number is set to 4 cores to ensure that the processing delay of a single fragment is controlled within 150 milliseconds. Based on the MapReduce parallel computing model, the system performs real-time analysis on the cross-device conversion behavior of user ID_005 between desktop device DT_321 and mobile device MO_654, updates the calculation results every 5 seconds using a sliding window mechanism, and sets the window size to 300 seconds. The system uses the Kafka message queue to receive real-time behavior data streams, processes up to 5,000 messages per second, and routes the behavior data of different users to the corresponding processing nodes through a partition strategy. At the resource scheduling level, the system uses a dynamic load balancing algorithm. When the CPU usage of a certain node exceeds 80%, the system automatically triggers task migration and reallocates part of the computing tasks to nodes with lower load to ensure that the overall processing performance is stable at an average response time of less than 200 milliseconds. Finally, the system outputs a real-time result set containing user conversion probability, path completion degree, and behavior prediction value.

[0051] Step S106, according to the real-time processing of the user conversion path result set, build cache mechanism storage high frequency access path data, if receiving new query request, then priority from cache extraction data, determine the path analysis results of fast response.

[0052] Obtain the real-time conversion path result set, according to the historical query record statistics of each path access frequency, if the access frequency exceeds the preset reference value, then mark as high frequency path data, establish high frequency path data identification table. For the high frequency path data identification table, using bloom filter algorithm to build data index structure, according to the path characteristic value to generate hash index key, obtain the fast retrieval index mapping relationship of path data. According to the index mapping relationship, design three layer cache storage architecture, if the path data access frequency belongs to the highest level, then store to the first level cache, if belongs to high level, then store to the second level cache, determine the hierarchical cache data distribution scheme. Through the hierarchical cache data distribution scheme, monitor the storage capacity usage of each cache level, if the storage capacity of a certain level reaches the capacity limit, then start data elimination mechanism to clean up low frequency access data, get the optimized cache space configuration. For the optimized cache space configuration, when the system receives path query request, parse the query condition to generate query characteristic code, according to the query characteristic code matching cache index, judge the cache location of target path data. According to the cache location judgment result, if the target data is located in the first level cache, then directly return the query result, if located in the second or third level cache, then extract data and update to the upper cache, obtain path query response data. Through the path query response data, record the response time and data source level of this query, update the path access frequency statistics information, get the path query complete result set containing performance index.

[0053] For example, the system deploys a Redis cluster as a distributed cache layer, configures a master-slave replication mode containing 3 master nodes and 6 slave nodes, and allocates 16 GB of memory space to each master node to store hot data of user conversion paths. The cache strategy uses the LRU (Least Recently Used) algorithm, and when the memory usage reaches 85%, it automatically cleans up path records with access frequency less than 10 times per hour, and sets the TTL (Time To Live) expiration time to 1800 seconds to ensure data timeliness. The system uses a Bloom filter pre-judgment mechanism, uses 3 hash functions to quickly filter query requests, and controls the false positive rate within 0.1%, effectively reducing invalid cache queries. When receiving a path query request of user ID_078, the system first calculates the MD5 hash value of the user behavior sequence as the cache key, and then locates the corresponding cache shard node through a consistent hashing algorithm. Cache hit detection uses a two-level index structure, the first level index performs coarse-grained matching based on user identification, and the second level index performs precise positioning based on timestamp and device type, with an average retrieval time controlled within 5 milliseconds. If the cache is not hit, the system triggers a back-to-source query mechanism to obtain complete path data from the persistent storage layer, and asynchronously updates the cache content while recording the access weight value of the path increased by 0.2, providing a reference basis for subsequent cache eviction strategies.

[0054] Step S107, through the path analysis result, the user identity and behavior data are protected by encryption technology, and combined with the access control policy, the safe and compliant attribution analysis data set is obtained.

[0055] Through the path analysis result, the original record of user identity and behavior data is obtained, and the original record is processed by encryption technology to generate an encrypted user data set to determine the protected data basis. According to the encrypted user data set, combined with the pre-established access control policy, the data access permission is classified and divided, if the access request is detected not to meet the permission requirements, the access is refused and the abnormal log is recorded, the data access range after permission verification is obtained. For the data access range after permission verification, the key features of behavior data are obtained, and the features are classified and processed by using a logistic regression model to judge the correlation between behavior data and path analysis result, and a classified data correlation mapping is constructed. Through the classified data correlation mapping, the core data fields related to attribution analysis are extracted, if the integrity of the core data fields is lower than the preset threshold, the data completion process is triggered to obtain the complete data combination after supplement. According to the complete data combination after supplement, a safe and compliant data processing framework is constructed, the user identity is processed by secondary encryption, and the data protection level that meets the compliance requirements is determined. For the data protection level that meets the compliance requirements, the path analysis result and the attribution analysis demand are integrated, the data is stored in a standardized format, and the final attribution analysis data set is obtained.

[0056] For example, on the basis of the path analysis result, the system first encrypts the user identity and behavior data, uses the AES-256 algorithm to encrypt the user ID and behavior log, the key length is 256 bits, ensures the security of the data in the transmission and storage process, the encryption process takes about 200 milliseconds to process 1000 records, then the system generates a 32-bit random salt value to bind with the encrypted data, further improves the cracking difficulty. Next, the system combines the access control strategy, divides the data access permission into three levels through the role-based permission management mechanism, namely data analyst, business administrator and system operation and maintenance, each level of permission corresponds to different data field access range, for example, data analysts can only access the desensitized behavior data, while business administrators can access the complete path analysis result, the permission verification process is realized through the JWT token, and the verification time is controlled within 50 milliseconds. In order to form a safe and compliant attribution analysis data set, the system performs hierarchical processing on the encrypted data, anonymizes the user behavior data using the K-anonymity algorithm, ensures that each group of data contains at least 5 similar user records, reduces the risk of individual identification, and the anonymization processing of each batch of 10000 data takes about 1.5 seconds. At the same time, the system performs association analysis on the path data, uses the Apriori algorithm to mine the association rules between user behaviors, sets the support threshold to 0.3 and the confidence threshold to 0.7, analyzes the high-frequency behavior combination, for example, the association probability of page browsing and purchase behavior is 0.82, and finally stores the analysis result as a compliant data set, the storage format is Parquet, and the file size is limited to within 500MB, ensuring efficient reading and compliance of the data. To supplement the business association, the system also introduces time dimension analysis, groups user behaviors by 24-hour period, calculates the behavior distribution proportion of each time period, for example, the behavior proportion of 0-6am is 12.5%, to optimize the accuracy of the attribution model. The overall process is realized through automatic scripts without human intervention.

[0057] As Figure 4 shown, the present application provides a user attribution system that can meet various configurations and transformations, mainly including:

[0058] A user behavior feature extraction module is used to obtain interaction records of multiple data sources from a pre-established user behavior database, and to extract original data sets containing timestamps, access paths and operation types for cross-device tracking requirements, to obtain a preliminary user behavior feature set;

[0059] A multi-dimensional feature integration module is used to standardize the time, path and operation type according to the preliminary user behavior feature set, and to construct a unified feature vector matrix for data heterogeneity problem and determine the cross-device user behavior association;

[0060] The similarity calculation module is configured to apply a preset similarity calculation rule to compare the behavior features between different devices through the feature vector matrix, and if the similarity exceeds a preset threshold, the same user identity is determined, and a unified identity mapping table is obtained.

[0061] The behavior path integration module is configured to sort the behavior records on each device according to a time sequence, integrate the behavior records into a continuous behavior path data set, and determine a full-link view of a user conversion path according to the unified identity mapping table.

[0062] The distributed computing module is configured to perform sharding storage and parallel analysis by using a distributed computing framework, optimize resource allocation, and obtain a real-time processed user conversion path result set through the behavior path data set.

[0063] The cache optimization module is configured to construct a cache mechanism to store high-frequency access path data according to the real-time processed user conversion path result set, and if a new query request is received, the cache is preferentially used to extract data and determine a fast response path analysis result.

[0064] The security compliance module is configured to apply encryption technology to protect user identity and behavior data through the path analysis result, combine an access control strategy, and obtain a security-compliant attribution analysis data set.

[0065] In addition, it should be noted that various specific technical features described in the above specific embodiments can be combined in any appropriate manner without contradiction, and to avoid unnecessary repetition, various possible combinations are not described again by the present application. In addition, various different embodiments of the present application can also be combined in any manner, as long as they do not violate the spirit of the present application, and they should also be considered as disclosed by the present application.

Claims

1. A user attribution method that can satisfy various configurations and transformations, characterized in that, The method includes: Interaction records from multiple data sources are obtained from a pre-established user behavior database. To meet the cross-device tracking requirements, the raw dataset containing timestamps, access paths, and operation types is extracted to obtain a preliminary set of user behavior features. Based on the preliminary set of user behavior features, time, path and operation type are standardized by multi-dimensional feature integration. A unified feature vector matrix is ​​constructed to address the data heterogeneity problem and determine the cross-device user behavior correlation. Using the feature vector matrix, the behavioral features of different devices are compared using a preset similarity calculation rule. If the similarity exceeds a preset threshold, the same user identity is determined, and a unified identity mapping table is obtained. Based on the unified identity mapping table, the behavior records on each device are sorted by time series and integrated into a continuous behavior path dataset to determine the full-link view of the user conversion path; Using the aforementioned behavioral path dataset, a distributed computing framework is employed for sharded storage and parallel analysis to optimize resource allocation and obtain a real-time user conversion path result set. Based on the real-time processed user conversion path result set, a caching mechanism is constructed to store high-frequency access path data. If a new query request is received, data is retrieved from the cache first to determine the path analysis result for a fast response. Based on the path analysis results, encryption technology is applied to protect user identity and behavioral data. Combined with access control policies, a secure and compliant attribution analysis dataset is obtained.

2. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The process involves retrieving interaction records from multiple data sources from a pre-established user behavior database. For cross-device tracking requirements, a raw dataset containing timestamps, access paths, and operation types is extracted to obtain a preliminary set of user behavior features, including: Obtain user interaction records stored in multiple data sources, and read the raw interaction data containing device identifiers, timestamps, and access paths through the data interface to obtain a complete cross-device behavior dataset; Based on the timestamp information in the interaction record, the user behavior data is divided into time sequences using a time window segmentation method. If the time interval exceeds a preset threshold, it is determined to be different sessions, and behavior segments organized by time sequence are obtained. By analyzing the access path features in the behavioral segments, path attributes such as page jump sequence, dwell time and click position are extracted to determine the user's browsing trajectory pattern on different devices; Clustering algorithms are used to group the browsing trajectory patterns, and the behavioral correlation between devices is calculated based on path similarity and operation type similarity to obtain potential cross-device behavior clusters of the same user; If there are identical access path sequences and consecutive timestamps in the behavior cluster, the device affiliation is determined by the user identity association rules, and the cross-device user identity matching result is determined. Based on the identity matching results, a sequence pattern mining algorithm is used to extract frequent behavior patterns from the user behavior feature set to obtain the user's cross-device operation habits and preference features. By integrating the aforementioned operating habits and preference characteristics with device usage time distribution, a user cross-device behavior profile is constructed to determine each user's complete behavioral trajectory and pattern recognition labels on different devices.

3. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, Based on the preliminary user behavior feature set, a multi-dimensional feature integration method is used to standardize time, path, and operation type, construct a unified feature vector matrix to address data heterogeneity issues, and determine cross-device user behavior correlations, including: By processing time information, path records, and operation categories, and using data preprocessing tools to clean and convert the feature set, a standardized behavioral data set is obtained. Based on the standardized behavioral data set, a multi-dimensional feature extraction method is used to perform hierarchical mapping of time information and path records, construct a preliminary feature vector, and determine the distribution pattern of cross-device behavior. If the distribution pattern of the feature vectors differs significantly across different devices, the matching degree between the operation category and the path record is compared to determine whether there is a behavioral correlation, thus obtaining a preliminary correlation result. Based on the preliminary correlation results, a clustering algorithm is used to group the feature vectors and classify them according to cross-device characteristics and user behavior similarity to obtain hierarchical behavior correlation clusters; Based on the behavior association clusters, extract the continuity features of time information and operation categories, construct a unified vector matrix, and determine the deep connections between user behaviors across devices. Using the vector matrix, a standardization method is employed to smooth the distribution characteristics of cross-device behavior, resulting in the final user behavior correlation mapping. If the user behavior association mapping contains incomplete behavior data for some devices, the missing parts are interpolated and supplemented using an integration method to determine the complete cross-device behavior association structure.

4. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The process involves comparing behavioral features between different devices using the feature vector matrix and applying preset similarity calculation rules. If the similarity exceeds a preset threshold, the devices are identified as belonging to the same user, resulting in a unified identity mapping table. This includes: Using the feature vector matrix, the device behaviors are compared using preset evaluation rules to calculate the similarity values ​​between different devices and obtain preliminary similarity distribution results. Based on the similarity distribution results, for device behaviors with similarity values ​​higher than a preset threshold, an association judgment is performed to determine device combinations that may belong to the same user identity; Clustering algorithms are used to group the device combinations, and the classified device association groups are obtained based on the similarity values ​​of device behaviors and the distribution characteristics of feature vectors. By extracting common patterns of device behavior from the device association groups, a unified identity mapping framework is constructed to determine the cross-device user identity correspondence. For the aforementioned identity mapping framework, a data verification tool is used to analyze device differences. If inconsistencies are found in the behavioral data of some devices, adjustments are made through data fusion methods to obtain a corrected identity mapping table. Based on the revised identity mapping table, the associated information of device behavior is integrated to construct a complete cross-device user identity profile and determine the final unified identifier structure; Through the unified identifier structure, the feature vectors of all device behaviors are standardized to generate a consistent record of user behavior across devices, thus completing the final integration of identity association.

5. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The step of sorting and integrating the behavior records on each device according to the time series based on the unified identity mapping table into a continuous behavior path dataset to determine the full-link view of the user conversion path includes: Using the identification and mapping table, the records of each device are initially classified and organized, and then arranged sequentially according to the time series to obtain a device behavior dataset sorted by time. Based on the device behavior dataset sorted by time, a time series analysis method is used to extract key time nodes of user behavior, construct a preliminary behavior path framework, and determine the initial structure of the behavior path. Based on the initial structure of the behavior path, and taking into account the characteristics of the integrated set, the continuity of behavior in the device records is verified. If time breakpoints or missing data are found in the behavior path, they are supplemented by data interpolation methods to obtain a complete path dataset. Using the complete path dataset, the correlation between user behavior and conversion paths is analyzed, and clustering algorithms are used to group the behavior paths to obtain a set of categories for different conversion paths. Based on the classification set, common features in each group of conversion paths are extracted to construct a framework model of the full-link view. If the features of a certain group of paths do not match the overall view, they are corrected by feature adjustment tools to determine a unified path view structure. For the unified path view structure, user behavior and behavior consistency data are integrated to generate a panoramic record of behavior paths across devices, and the final conversion path full-link view is determined. By using the full-link view of the transformation path, key nodes in the behavior path are marked, and data storage tools are used to save the structural information of the full-link view, resulting in persistent path view data.

6. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The process involves using the behavioral path dataset, employing a distributed computing framework for sharded storage and parallel analysis, optimizing resource allocation, and obtaining a real-time user conversion path result set, including: The behavior path dataset is obtained, and the data is partitioned according to user identifier using a hash partitioning algorithm. If the data volume exceeds a preset threshold, a multi-level partitioning mechanism is activated to obtain a uniformly distributed set of data partitions. For the data shard set, parallel processing is performed through computing nodes in a distributed architecture. Processing tasks are dynamically allocated according to the computing capabilities of each node to obtain a node-level processing task allocation table. According to the processing task allocation table, each computing node synchronously performs path analysis operations and uses a load balancing strategy to monitor the node's operating status. If the processing speed of a node is lower than the average level, the task allocation weights are readjusted to determine the optimized computing resource configuration. With the optimized computing resource configuration, the intermediate results generated by each node are aggregated in real time, and the fragmented computing results are merged using streaming processing technology to obtain unified path analysis intermediate data. For the intermediate data of the path analysis, a memory caching mechanism is used to improve data access speed, and the results are sorted according to the time characteristics of the conversion path to obtain a candidate set of conversion paths arranged in time sequence; Based on the candidate set of conversion paths, similar path patterns are identified through clustering algorithms. If the path similarity is higher than the set standard, they are merged into the same category to determine the final conversion path type grouping. By grouping the conversion path types, integrating the statistical characteristics and behavioral patterns of each group, and using data compression technology to optimize storage space, a complete set of real-time conversion path analysis results is obtained.

7. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The step involves constructing a caching mechanism to store frequently accessed path data based on the real-time processed user conversion path result set. If a new query request is received, data is preferentially retrieved from the cache to determine the path analysis results for a fast response, including: Obtain the real-time conversion path result set, and count the access frequency of each path according to the historical query records. If the access frequency exceeds the preset benchmark value, it is marked as high-frequency path data, and a high-frequency path data identification table is established. For the high-frequency path data identifier table, a Bloom filter algorithm is used to construct a data index structure, and a hash index key is generated based on the path feature value to obtain a fast retrieval index mapping relationship for path data; Based on the index mapping relationship, a three-layer cache storage architecture is designed. If the path data access frequency is extremely high, it is stored in the first-level cache; if it is high, it is stored in the second-level cache. The hierarchical cache data distribution scheme is determined. The hierarchical cache data distribution scheme monitors the storage capacity usage of each cache level. If the storage capacity of a certain level reaches its upper limit, a data eviction mechanism is activated to clean up low-frequency access data, resulting in an optimized cache space configuration. For the optimized cache space configuration, when the system receives a path query request, it parses the query conditions to generate a query feature code, matches the cache index according to the query feature code, and determines the cache location of the target path data. Based on the cache location determination result, if the target data is located in the first-level cache, the query result is returned directly; if it is located in the second-level or third-level cache, the data is extracted and synchronously updated to the upper-level cache to obtain the path query response data. By querying the response data through the path, the response time and data source level of this query are recorded, the path access frequency statistics are updated, and a complete result set of path queries containing performance indicators is obtained.

8. The user attribution method according to claim 1, which can satisfy multiple configurations and transformations, is characterized in that, The path analysis results are used to protect user identity and behavioral data using encryption technology, and combined with access control policies to obtain a secure and compliant attribution analysis dataset, including: Based on the path analysis results, the original records of user identity and behavior data are obtained. These original records are then processed using encryption technology to generate an encrypted user data set, thus determining the protected data foundation. Based on the encrypted user data set and the pre-established access control policy, data access permissions are classified into levels. If an access request is detected that does not meet the permission requirements, access is denied and an exception log is recorded to obtain the data access range after permission verification. Based on the data access scope after permission verification, key features of behavioral data are obtained, and a logistic regression model is used to classify the features. The correlation between behavioral data and path analysis results is judged, and a data association mapping is constructed after classification. By mapping the classified data, the core data fields related to attribution analysis are extracted. If the completeness of the core data fields is lower than the preset threshold, the data completion process is triggered to obtain the complete data combination after completion. Based on the complete supplementary data combination, a secure and compliant data processing framework is constructed, and secondary encryption is performed on user identity identifiers to determine a data protection level that meets compliance requirements. For data protection levels that meet compliance requirements, the path analysis results and attribution analysis needs are integrated, and the data is stored in a structured format using a standardized format to obtain the final attribution analysis dataset.

9. A user attribution system capable of meeting various configurations and transformations, characterized in that: The system includes: The user behavior feature extraction module is used to obtain interaction records from multiple data sources from a pre-established user behavior database. For cross-device tracking needs, it extracts the raw dataset containing timestamps, access paths, and operation types to obtain a preliminary set of user behavior features. The multi-dimensional feature integration module is used to standardize time, path and operation type based on the preliminary user behavior feature set by using a multi-dimensional feature integration method, construct a unified feature vector matrix to address the data heterogeneity problem, and determine the cross-device user behavior correlation. The similarity calculation module is used to compare the behavioral features between different devices by applying preset similarity calculation rules through the feature vector matrix. If the similarity exceeds the preset threshold, it is determined to be the same user identity, and a unified identity mapping table is obtained. The behavior path integration module is used to sort the behavior records on each device by time sequence according to the unified identity mapping table, integrate them into a continuous behavior path dataset, and determine the full-link view of the user conversion path. The distributed computing module is used to perform sharded storage and parallel analysis using the behavior path dataset through a distributed computing framework, optimize resource allocation, and obtain a real-time user conversion path result set. The cache optimization module is used to build a cache mechanism to store high-frequency access path data based on the real-time processed user conversion path result set. If a new query request is received, data is retrieved from the cache first to determine the path analysis result for a fast response. The security and compliance module is used to protect user identity and behavioral data using encryption technology based on the path analysis results, and combined with access control policies to obtain a security and compliance attribution analysis dataset.

Images