A data processing method

Abstract

This disclosure provides a data processing method, comprising: acquiring interactive behavior data of a target user under reference indicators; determining a behavior representation vector based on the interactive behavior data; determining isolated anomaly-type representations and structural representations corresponding to the target user based on the behavior representation vectors; determining event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly-type representations and structural representations based on pre-configured custom rules and rule matching functions; determining the correlation strength attribute between the event chain attribute, sentiment evaluation attribute, and a pre-created graph structure and candidate essential issues; and determining the attribution analysis result of the target user based on the correlation strength attribute. The technical solution of this disclosure, based on the behavior representation vectors corresponding to interactive behavior data, determines the attribution analysis result associated with the target user, improving the comprehensiveness and accuracy of anomaly detection, and achieving a more accurate attribution analysis result for the target user.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a data processing method. Background Technology

[0002] With the development of technologies such as big data and machine learning, user attribution analysis has been widely used in the service optimization of telecom operators.

[0003] Currently, attribution analysis of user behavior largely relies on pre-set rule bases for matching and judgment. However, such methods are not adaptable enough to emerging or evolving behavioral patterns and are prone to problems such as recognition lag, misjudgment or omission. The overall flexibility and generalization ability need to be improved. Summary of the Invention

[0004] This disclosure provides a data processing method to improve the comprehensiveness and accuracy of anomaly detection, thereby achieving more accurate attribution analysis results for target users.

[0005] In a first aspect, embodiments of this disclosure provide a data processing method, the method comprising: Acquire interactive behavior data of target users under multiple reference indicators, and determine a behavior representation vector based on the interactive behavior data; Based on the behavior representation vector, the isolated anomaly class representation and the structural representation corresponding to the target user are determined. The isolated anomaly class representation is used to reflect the abnormal data determined by considering the interaction behavior data under different reference indicators individually. The structural representation is the abnormal data determined from the interaction behavior data by combining multiple reference indicators. Based on pre-configured custom rules and rule matching functions, the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structure class representation are determined. Based on the event chain attributes, sentiment assessment attributes, and a pre-created graph structure, the correlation strength attribute between the event chain and the candidate essential question is determined; wherein, the graph structure includes at least one node, and the node corresponds to the candidate essential question. Based on the association strength attribute, the attribution analysis results associated with the target user are determined.

[0006] Secondly, embodiments of the present invention also provide a data processing apparatus, the apparatus comprising: The behavior representation vector determination module is used to acquire the interaction behavior data of the target user under multiple reference indicators, so as to determine the behavior representation vector based on the interaction behavior data; The appearance determination module is used to determine the isolated abnormal appearance and the structural appearance corresponding to the target user based on the behavior representation vector. The isolated abnormal appearance is used to reflect the abnormal data determined by considering the interaction behavior data under different reference indicators individually. The structural appearance is the abnormal data determined from the interaction behavior data after combining multiple reference indicators. The event chain attribute determination module is used to determine the event chain attributes and sentiment evaluation attributes corresponding to the isolated abnormal class manifestations and the structural class manifestations based on pre-configured custom rules and rule matching functions. The association strength attribute determination module is used to determine the association strength attribute between the event chain attribute, the sentiment evaluation attribute, and the pre-created graph structure; wherein, the graph structure includes at least one node, and the node corresponds to the candidate essential question; The attribution analysis result determination module is used to determine the attribution analysis result associated with the target user based on the association strength attribute.

[0007] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the data processing method as described in any of the embodiments of the present invention.

[0008] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a data processing method as described in any of the embodiments of the present invention.

[0009] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program, characterized in that the computer program, when executed by a processor, implements the data processing method as described in any of the embodiments of the present invention.

[0010] The technical solution of this disclosure first acquires interactive behavior data of a target user under multiple reference indicators to determine a behavior representation vector based on the interactive behavior data. Then, based on the behavior representation vector, it determines the isolated anomaly class representation and the structural representation corresponding to the target user. The isolated anomaly class representation reflects the anomaly data determined by considering interactive behavior data under different reference indicators individually, while the structural representation is the anomaly data determined from the interactive behavior data after combining multiple reference indicators. Next, based on pre-configured custom rules and rule matching functions, it determines the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structural representation. Further, based on the event chain attributes, sentiment evaluation attributes, and a pre-created graph structure, it determines the association strength attribute with candidate essential questions. The graph structure includes at least one node, and each node corresponds to a candidate essential question. Finally, based on the association strength attribute, it determines the attribution analysis result associated with the target user. This solves the problem in the prior art where the identification of user behavior anomalies and the attribution of problems rely on a preset rule base for matching and judgment, resulting in insufficient adaptability to emerging or evolving behavioral patterns, leading to problems such as recognition lag, misjudgment, or missed judgment. This disclosure implements a method to determine the attribution analysis results associated with the target user based on the behavior representation vector corresponding to the interaction behavior data, thereby improving the comprehensiveness and accuracy of anomaly detection and achieving a more accurate attribution analysis result for the target user. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of exemplary embodiments of the present invention, the accompanying drawings used in describing the embodiments are briefly introduced below. Obviously, the accompanying drawings described are only a portion of the drawings of the embodiments to be described in this invention, and not all of the drawings. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0012] Figure 1 This is a schematic flowchart of a data processing method provided in an embodiment of this disclosure; Figure 2 This is a schematic flowchart of a data processing method provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of a data processing flow provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of the structure of a data processing apparatus provided in an embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0014] Before introducing the technical solutions provided in the embodiments of this disclosure, the application scenarios can be illustrated by examples. The technical solutions provided in the embodiments of this disclosure can be applied to scenarios where attribution analysis results associated with a target user are determined. Based on the technical solutions in the embodiments of this disclosure, attribution analysis results associated with a target user are determined based on the behavior representation vectors corresponding to interaction behavior data, improving the comprehensiveness and accuracy of anomaly detection, and achieving a more accurate effect in attribution analysis results for the target user.

[0015] Example 1 Figure 1 This is a flowchart illustrating a data processing method provided in an embodiment of this disclosure. This embodiment is applicable to situations where attribution analysis results associated with a target user are determined. The method can be executed by a data processing device, which can be implemented in the form of software and / or hardware. The hardware can be a mobile electronic device that can execute the data processing method provided in this technical solution.

[0016] like Figure 1 As shown, the method includes: S110. Obtain the interaction behavior data of the target user under multiple reference indicators, and determine the behavior representation vector based on the interaction behavior data.

[0017] In this context, "target user" refers to the individual user whose data needs to be analyzed and processed to determine the attribution analysis results. This can specifically correspond to an account, phone number, or user ID. "Reference metrics" refer to the dimensions or statistical methods used to collect the target user's interaction behavior data. Multiple reference metrics can include channel dimensions, business dimensions, time dimensions, and event attribute dimensions. Channel dimensions can include applications, SMS, IVR, and human customer service. Business dimensions can include pricing, processing, reach, and feedback. Time dimensions can include time windows of the past 7 days or 30 days. Event attribute dimensions can include event type, business line, and triggering reason.

[0018] It should be noted that interactive behavior data refers to the records of behavioral events left by target users during service usage. Interactive behavior data includes at least the following: tariff operation behavior data associated with the current terminal, service access data, business processing data, and user subjective feedback data. Tariff operation behavior data can include data on package changes or add-on subscriptions. Service access data can include system-triggered application notifications and SMS reminders. Business processing data can be data related to inquiries and processing initiated by target users through IVR or human customer service channels. User subjective feedback data can include complaint content, satisfaction ratings, and work order records from target users. Furthermore, when acquiring the target user's interactive behavior data, information such as the timestamp, channel, event type, event content primary key, and field values ​​of the corresponding event is also acquired.

[0019] It should also be noted that the behavior representation vector refers to the user behavior representation determined based on interaction behavior data, which is a fixed-dimensional numerical vector corresponding to the interaction behavior data.

[0020] Optionally, acquire interactive behavior data of the target user under multiple reference indicators; encode the interactive behavior data according to the event type corresponding to the interactive behavior data, and extract features from the interactive behavior data according to the first keyword information to obtain structured topic features; extract interactive behavior data according to the behavior occurrence time information in the same event type, and encode the extracted features through sequence hashing to obtain encoded features; and obtain the target user's behavior representation vector by concatenating the event type, topic features, and encoded features.

[0021] Event type refers to the category of behavioral events corresponding to the interactive behavior data, used to distinguish different interactive behavior events. Event types can include package changes, subscription / cancellation, receiving reminders, initiating inquiries, complaints, or evaluations, etc. Encoding interactive behavior data means mapping the interactive behavior data to standard category codes according to the event types corresponding to the interactive behavior data, that is, turning the interactive behavior data into a usable digital expression. Encoding interactive behavior data can align and unify different types of events at the semantic level. Primary keyword information refers to the set of keywords or key fields used to extract features from the event content. Primary keyword information can include package identifiers, product codes, reminder types, complaint reason categories, work order tags, and satisfaction levels, etc. Feature extraction refers to the process of converting interactive behavior data into structured, usable features based on primary keyword information. Feature extraction can include field mapping, etc. Thematic features refer to the result obtained by representing the content described by the event and the surrounding business points in a structured form. For example, structured thematic features can be: package change from A to B, abnormal deduction or overdue payment reminder, etc.

[0022] It's important to note that the timing information of an event refers to its timestamp and related information. For example, derived information could include the interval, frequency, density, or most recent occurrence time of similar events. Sequence hashing can compress variable-length event sequence features into fixed-dimensional codes. Since the extracted features vary in length, sequence hashing yields fixed-dimensional encoded features, i.e., fixed-length feature vectors. After obtaining structured topic features, within the same event type, these features are encoded using sequence hashing to maintain a fixed dimension based on the temporal order, interval, frequency, or density of events, resulting in encoded features.

[0023] It should also be noted that concatenation refers to combining multiple feature vectors along different dimensions to form a longer vector. By concatenating event type, topic features, and encoding features along different dimensions, a fixed-dimensional numerical vector is obtained, which is the behavioral representation vector of the target user.

[0024] Specifically, first, the system acquires interactive behavior data of target users under multiple reference metrics. Then, it maps behavioral events in the data to standard category codes based on event type, and extracts structured thematic features by combining the first keyword information from the event content. Further, by analyzing the temporal sequence of behavioral events, features are extracted from the temporal arrangement of event types and the changes in time intervals and occurrence density between events, and these features are encoded using sequence hashing. Finally, the event types, thematic features, and the encoded features are concatenated to form a fixed-dimensional behavioral representation vector of the target user.

[0025] S120. Based on the behavior representation vector, determine the isolated anomaly class representation and structural representation corresponding to the target user.

[0026] Isolated anomaly representations are used to reflect anomalous data identified by considering interaction behavior data under different reference indicators individually. Structured representations are anomalous data identified from interaction behavior data by combining multiple reference indicators.

[0027] It's important to note that isolated anomalies refer to a target user's behavior significantly deviating from existing patterns within a certain timeframe. This manifests as abrupt path changes, abnormal event combinations, or an increased frequency of abnormal behaviors, potentially reflecting unidentified problem types or unusual experiences. In real-world business scenarios, this type of behavior might correspond to: a target user missing out on plan renewal due to delayed notifications, leading to frequent complaints; a target user frequently switching plans within a short period, reflecting dissatisfaction with current pricing; or a target user experiencing a surge in data usage without any proactive interaction, potentially indicating passive billing issues. Structural anomalies are composite structures formed by combining two or more existing behavioral anomalies in a specific way. For example, a target user might experience a combination of three behavioral types: failed plan renewal, delayed fee deduction notifications, and repeated complaints. Multiple users, within the same timeframe and similar business scenarios, might sequentially execute behavioral sequences such as network switching inquiries, sudden increases in data usage, and negative customer service feedback. These combined behaviors are difficult to determine through a single event or rule, but they frequently occur in scenarios affecting user satisfaction or potentially leading to network cancellations.

[0028] Specifically, after determining the behavior representation vector, the first step is to perform anomaly analysis on the target user's behavior based on the behavior representation vector. This identifies two types of anomalies for the target user: isolated anomalies and structural anomalies. Isolated anomalies correspond to anomalies identified solely based on interaction behavior data under a single reference indicator dimension, without considering other reference indicators. They reflect anomalies in a single dimension. Structural anomalies correspond to anomalies identified from interaction behavior data after jointly modeling information from multiple reference indicator dimensions. They reflect anomalies presented under cross-dimensional relationships or overall structural patterns.

[0029] Optionally, based on a preset aggregation time length and reference indicators, the behavior representation vector is aggregated to obtain a first behavior vector, and the first behavior vector is normalized to obtain a second behavior vector; based on the second behavior vector, the behavior state area where the target user's interaction behavior data is located is determined; based on the comprehensive state vector corresponding to the preset time selection slice, the behavior path evolution direction vector of the target user under the preset time selection slice is determined; based on the behavior path evolution direction vector corresponding to the same preset time selection slice, the behavior state area to which it belongs, and the comprehensive state vector, it is determined whether the target user's interaction behavior data is an isolated abnormal manifestation.

[0030] The normalization process is performed by incorporating the first behavior vectors of all users. Different behavior state zones correspond to different indicator thresholds, and these thresholds are dynamically adjusted based on a smoothing factor.

[0031] The preset aggregation time length refers to the fixed time window or granularity used when statistically summarizing the behavioral representation vectors of the target user. Using the preset aggregation time length, multiple behavioral representation vectors within the same window can be aggregated into a first behavioral vector at a given moment or over a given time period.

[0032] It should be noted that the aggregation of behavioral representation vectors to obtain the first behavioral vector refers to summarizing and calculating the original behavioral events or features within a preset aggregation time period to form the first behavioral vector for that time period, so that a single first behavioral vector can represent the overall behavioral status during that time period. For example, behavioral representation vectors can be aggregated by month, week, or a rolling window of the past 30 days. Reference indicators can correspond to dimensions such as monthly average consumption index, complaint density, and service usage rate. The first behavioral vector can be represented by the following structure: ; in, This can represent the target user at the time of aggregation. The average monthly consumption index; This can represent the target user at the time of aggregation. The density of complaints. For example, This can represent the target user at the time of aggregation. The average number of complaints per month; This can represent the target user at the time of aggregation. The service usage rate. For example, This can represent the target user at the time of aggregation. The percentage of voice or data usage. For example, a target user's phone bill in June was... If a target user has no complaints and their data usage is within a normal range, their first behavioral vector will reflect a low complaint rate and balanced usage. However, if this target user experiences a sudden increase in data usage and a sharp rise in complaints in July, their first behavioral vector will deviate significantly from the previous period.

[0033] It should be noted that the formula for normalizing the first row of vectors to obtain the second row of vectors can be: ; in, This refers to the second row being a vector; This refers to the mean vector; This refers to the standard deviation vector. The mean vector used when normalizing the first row of vectors. with standard deviation vector This is obtained by statistically analyzing the first behavior vectors of all users, avoiding inconsistencies in the behavioral dimension scale between different users, ensuring that different users are in the same scale space, and enhancing the comparability of comparisons between different users. Furthermore, and It can be updated monthly based on a scrolling window.

[0034] It should be noted that behavioral state zones refer to several predefined behavioral paradigm regions in the standardized behavioral vector space, used to express the business risk or service sensitivity state of the target user. For example, behavioral state zones may include a stable zone with few complaints and stable consumption, a high-risk zone with a surge in complaints or abnormal usage, and a borderline zone that is not exceeded but shows abnormal trends. Each behavioral state zone can be dynamically defined through indicator thresholds. The different indicator thresholds corresponding to different behavioral state zones mean that the judgment boundary range of each behavioral state zone is different, thus different behavioral state zones correspond to different risk levels and judgment criteria. For example, high-risk zones have stricter or higher thresholds for complaint density or abnormal usage. Each behavioral state zone is dynamically defined through indicator thresholds. Based on the second behavioral vector, the formula for determining the behavioral state zone where the target user's interaction behavior data belongs can be: ; in, This refers to the behavior state area where the target user's interaction behavior data is located; Here, the second row represents a vector; This represents the selected metric function. For example, It can be a weighted combination of consumption density or complaint rate, etc. For the first The indicator thresholds corresponding to each behavioral state zone are used to determine the behavioral state zone in which the target user's interaction behavior data falls, by substituting the target user's second behavior vector for that time period into the formula. This determines whether the target user falls into a stable zone, a marginal zone, or a high-risk zone.

[0035] It should be noted that the thresholds for behavioral status indicators are updated dynamically based on changes in business cycles and overall user distribution. A smoothing factor is used to weight and fuse the current and historical thresholds. The indicator thresholds are dynamically adjusted based on this smoothing factor. ; in, It refers to the first The new indicator thresholds corresponding to each behavioral state zone. This refers to the smoothing factor, which controls the weight balance between the current state and historical states. A larger value indicates a greater reliance on recent data. This refers to recent indicator thresholds; This refers to historical indicator thresholds. The default value can be 0.7.

[0036] It should be noted that a preset time slice refers to a discrete time period sequence used to observe behavioral evolution; for example, a preset time slice could be three consecutive months. State vectors can be constructed on these preset time slices, and changes in adjacent slices can be compared. The comprehensive state vector refers to the target user's behavioral vector on the preset time slice, comprehensively reflecting the overall level of multiple indicators such as consumption, complaints, and usage within the preset time slice. The comprehensive state vector corresponding to a preset time slice may include: , as well as When the preset time period is selected as three consecutive months, This refers to the vector corresponding to the second row of the first month. This refers to the second row of the vector corresponding to the second month. This refers to the second behavior vector corresponding to the third month. The behavior path evolution direction vector is a vector that characterizes the direction and magnitude of the target user's behavior change from one time slice to the next. It reflects the trend of the target user's behavior and whether the change is abrupt. To determine the target user's behavior path evolution direction vector within a preset time slice, based on the comprehensive state vector corresponding to that slice, the difference between adjacent comprehensive state vectors within the preset time slice can be used to obtain the behavior path evolution direction vector describing the direction of behavior change. The formula for determining the target user's behavior path evolution direction vector within a preset time slice can be: , ; in, Indicates the target user within a time period The second vector is a multi-vector normalized from features such as the consumption index, complaint density, and service usage rate. The early evolution direction vector represents the time period. arrive The direction of behavioral change; The recent evolution direction vector represents the direction of evolution from the time period. arrive The direction of behavioral change.

[0037] In this embodiment, based on the comprehensive state vector and the behavior state region to which the comprehensive state vector belongs, it is determined whether a transition has occurred in the behavior state partition corresponding to the preset time selection slice; if a transition occurs, it is determined that the target user's interactive behavior data has an isolated abnormality; and / or, based on the behavior path evolution direction vector, behavior direction mutation information is determined; when the behavior direction mutation information meets preset conditions, it is determined that the target user's interactive behavior data has an isolated abnormality.

[0038] It should be noted that whether a transition occurs in the behavioral state partition corresponding to the preset time selection slice refers to comparing whether the target user's behavioral state partition differs across different slices within the preset time selection slice. The formula for region transition detection can be: ; in, and This refers to the behavioral state region where the target user resides. Each behavioral state region represents a stable behavioral state paradigm. When the target user's second behavioral vector crosses the region boundary... ,Right now When the value is 1, it is considered that a transition has occurred in the behavior state partition corresponding to the preset time selection slice. Otherwise, then... A value of 0 indicates that no transition has occurred in the behavioral state partition corresponding to the preset time selection slice. Once a transition has been determined in the behavioral state partition corresponding to the preset time selection slice, it is determined that the target user's interaction behavior data exhibits an isolated anomaly. This means that when the target user's behavioral state partition jumps from one region to another, the change is considered significant enough to constitute an isolated anomaly on its own.

[0039] It should be noted that behavioral direction change information refers to information used to determine whether a sharp turn occurs in the trajectory. The formula for determining behavioral direction change information based on the behavioral path evolution direction vector can be: ; in, This refers to information about behavioral direction mutations. If the behavioral direction mutation information is greater than the mutation angle threshold... In other words, if preset conditions are met, it indicates a sudden change in the direction of the user's behavior trajectory, confirming the existence of isolated anomalies in the target user's interaction behavior data. When the abrupt change in the target user's behavior direction is greater than a change angle threshold, it is considered to pose a risk of path deviation. The interaction behavior data confirmed to contain isolated anomalies is then added to the candidate new phenomenon pool.

[0040] For example, this illustrates how target user interaction behavior data can be labeled as isolated anomaly-type representations: Suppose a target user's second behavior vectors over the past 3 months are as follows: , as well as Their corresponding behavioral state areas are respectively , as well as If the target user is It remains in a stable behavioral state at all times. But at the current moment Leap to the region ,and Not equal to If the preset time slice corresponds to a behavioral state partition, a transition occurs. In another scenario, assuming the target user remains within the same behavioral state partition, but their behavioral direction abruptly reverses, the behavioral direction abrupt change information is obtained through vector calculation. If the angle of change exceeds the threshold, the behavioral direction abrupt change condition is met, thus confirming the existence of isolated anomaly-type manifestations in the target user's interaction behavior data. In any of the above cases, the system will determine that the target user's current behavioral path has a significant abnormal deviation, automatically generate isolated anomaly-type manifestations, and add them to the candidate new manifestation pool.

[0041] Optionally, a central trajectory is selected from multiple sets of representation clusters associated with existing representation recognition results to determine a representative path set; time splicing is performed on the representation paths of the target user in multiple time slices to determine a composite behavior vector, and multiple users' frequently occurring behavior sequences under the same reference indicators are determined, as well as high-frequency event sequences based on a sliding time window; based on the composite behavior vector, behavior sequence, high-frequency event sequence, and representative path set, coverage assessment results, business label offset information and business indicator offset information involved in the target user are determined; when at least two of the coverage assessment results, business label offset information and business indicator offset information meet preset conditions, the target user's interaction behavior data, composite behavior vector, behavior sequence and high-frequency event sequence are determined as a structured representation.

[0042] The central trajectory is used to represent the representative path of the corresponding representation cluster set. The representative path set includes multiple representative paths, which reflect the occurrence chain of behavioral interaction data.

[0043] It should be noted that existing appearance recognition results refer to the appearance detection and clustering outputs already completed on historical data, which include the user behavior trajectories corresponding to various appearances and their respective appearance clusters. An appearance cluster set refers to a collection of multiple clusters formed by aggregating similar appearance trajectories according to the similarity of their behavioral structures. This appearance cluster set can be written as... Each of them This corresponds to a known abnormal behavior pattern or business problem pattern. The central trajectory refers to the most representative behavioral path within a set of representation clusters, used to summarize the typical behavioral links and structural characteristics of that cluster, and used as a template or component unit in subsequent analysis. Selecting the central trajectory from multiple representation cluster sets means selecting the most representative typical trajectory or typical sample of a high-density area within each cluster, representing the behavior pattern of that cluster. The representative path set is the path set formed by summarizing the central trajectories selected from each representation cluster. The representative path set includes multiple representative paths, meaning that it does not output only one template path, but rather outputs representative paths for multiple representation clusters, thus covering the typical behavioral patterns of different known representation types. Each representative path is represented using a unified vector encoding method and can be used for distance calculation and coverage evaluation. These paths can be represented using a unified vector encoding method as follows: .in, The second row is a vector after normalization. This represents the encoded behavioral characteristic dimensions, such as event type, channel distribution, rhythm characteristics, and business tags. A representative path refers to a typical behavioral sequence template used to reflect the chain of behavioral interaction data occurrences. For example, a representative path could be presented as: reminder failure → package misoperation → complaint submission or order → query → cancellation → reorder, etc.

[0044] It should be noted that the apparent paths of a target user across multiple time slices refer to the identified apparent trajectories of the target user appearing in different time periods. These trajectories may correspond to different abnormal appearances or behavioral patterns at different stages. For example, the apparent paths of a target user across multiple time slices may refer to the identified apparent trajectories of the target user appearing in different months or different sliding time windows. After determining the apparent paths of a target user across multiple time slices, the multiple apparent trajectories of the same user appearing in different time periods can be sequentially concatenated in chronological order to characterize the cross-apparent coupling structure in the user's behavioral evolution process. The composite behavior vector refers to the vectorized representation of the concatenated cross-time period apparent trajectories. The composite behavior vector can be represented as: ,in, This refers to a composite behavior vector. and These represent the user's apparent path at two different times. (The image is then concatenated.) It can capture newly emerging coupling structures during the evolution of individual behavior. For example, if a user complains about a failed plan change in March and then frequently exceeds their data limit in April, the apparent paths occurring in multiple time slices can be spliced ​​together to form a composite behavioral vector of failed change + abnormal over-limit usage, reflecting that the user may have been misled about pricing.

[0045] Among them, a behavioral sequence refers to a sequence composed of multiple representational vectors arranged in chronological order, used to represent the continuous behavioral chain of a user group over a period of time. A behavioral sequence in which multiple users frequently occur under the same reference indicators refers to: statistically analyzing combinations of frequently occurring behavioral sequences among multiple users at the group level, under the same business criteria or the same reference indicators, to abstract structurally abnormal combinations common to the scenario. This can be used to represent the representational path. and The splicing together constitutes a high-frequency behavioral sequence, that is, a high-frequency behavioral sequence can be represented as: For example, if multiple users experience a combination of delayed reminders, failed renewals, and complaints when a certain service plan expires, the system can abstract this as a high-frequency behavioral sequence. A high-frequency event sequence refers to an event chain sequence that occurs frequently and has a stable co-occurrence structure within a sliding time window, used to characterize the underlying combinations of potential structured representations. When determining high-frequency event sequences based on the sliding time window, sequence pattern mining can be performed on the event-level sequences within the sliding time window to extract frequently co-occurring event combinations within the window. A high-frequency event sequence can be represented as: Package changes → delayed notifications → complaint feedback. Among these, sequence mining algorithms such as PrefixSpan can be used to identify high-frequency event sequences based on a sliding time window.

[0046] It should be noted that the coverage evaluation result is used to measure whether a composite behavior vector can be covered by the existing set of representation clusters. When determining the coverage evaluation result, a distance metric can be used between the composite behavior vector and the existing set of representation clusters. The minimum distance between the composite behavior vector and the existing set of representation clusters is used to determine whether the composite behavior vector is covered by the existing set of representation clusters. The minimum distance between the composite behavior vector and the existing set of representation clusters is... If the minimum distance between the composite behavior vector and the existing representation cluster set is greater than a threshold... If the preset conditions are met, it means that the existing set of representational clusters does not cover it sufficiently, thus the composite behavior vector has the potential to add new structures.

[0047] Among them, business tags refer to a set of tags used to characterize the business attributes or essential performance of target users. For example, business tags may include satisfaction level, essential problem type, complaint frequency, and processing success rate, used to determine whether the composite behavior vector corresponds to user group characteristics different from existing appearances. Business tag offset information refers to information indicating a significant change in the distribution of the user group involved in the composite behavior vector on the essential business tags relative to the baseline cluster. This can be expressed as... Divergence measures the new distribution Compared with the baseline distribution The differences. Based on The formula for determining the business label offset information of the target user based on divergence can be: ; in, This refers to the business label offset information; This indicates that the user group in the composite behavior vector is represented by the business label. The probability distribution on; This indicates the user group in the benchmark cluster in terms of business tags. The probability distribution on; This indicates a specific category or value for the label variable. For example, a level among high / medium / low, or a category among abnormal deduction / signal problem / service processing failure. Indicates business tags This random variable takes on a specific value The event. If the business label offset information is greater than the significance threshold for label distribution offset. If the preset conditions are met, it means that the composite behavior vector has the potential to add new structures.

[0048] It should be noted that business indicator offset information refers to the significant differences in key business indicators, such as complaint rate and feedback satisfaction, between the user group involved in the composite behavior vector and the baseline cluster. For example, business indicator offset information can be verified through statistical tests such as the mean difference test. When the difference in business indicator offset information is significant, it proves that the preset conditions are met, indicating that the composite behavior vector has the potential to add new structures. Furthermore, when at least two of the coverage assessment results, business tag offset information, and business indicator offset information meet the preset conditions, the corresponding composite behavior vector is identified as a structured new representation and output to the structured new representation pool.

[0049] Specifically, based on historically identified representation clusters, the central trajectory representing each representation cluster is extracted as a representative path set. Then, the representation paths of the target user across multiple time slices are concatenated in chronological order to form a composite behavior vector. At the group level, high-frequency behavior sequences under the same reference indicators and high-frequency event sequences under sliding time windows are mined. Finally, a comprehensive judgment is made by combining evaluation results such as coverage, business label distribution offset, and significant differences in business indicators. When at least two of the threshold conditions are met, the interaction behavior data corresponding to the target user and its derived composite behavior vector, behavior sequence, and high-frequency event sequence are identified as structural representations.

[0050] S130. Based on pre-configured custom rules and rule matching functions, determine the event chain attributes and sentiment assessment attributes corresponding to isolated abnormal class representations and structural class representations.

[0051] Among them, pre-configured custom rules refer to a set of rules that are pre-written by business personnel or users based on business experience and can be dynamically expanded. A rule matching function is a function that determines whether a rule is matched. Through rule matching functions, the pattern of whether an isolated abnormal or structural phenomenon exhibits a certain event chain can be transformed into a computable judgment result.

[0052] It should be noted that the event chain attribute refers to the attribute result used to characterize whether a specific operation chain and its strength exist in the interaction behavior data corresponding to isolated anomalous representations and structured representations. The sentiment assessment attribute refers to the sentiment assessment conclusion obtained corresponding to isolated anomalous representations or structured representations.

[0053] Specifically, for the identified isolated anomalous and structural representations, custom rules can be used to match the corresponding interaction behavior data to identify whether typical anomalous triggering operation chains exist, thereby obtaining event chain attributes. Furthermore, text sentiment can be calibrated to determine the sentiment assessment attributes corresponding to the isolated anomalous and structural representations.

[0054] S140. Based on the event chain attributes, sentiment assessment attributes, and the pre-created graph structure, determine the correlation strength attributes between the candidate essential questions and the relevant issues.

[0055] The graph structure includes at least one node, which corresponds to a candidate essential problem. A candidate essential problem refers to a possible root cause or problem type. For example, candidate essential problems could include billing anomalies, incompatible service plans, harassing inquiries, poor service attitude, poor channel experience, and misleading marketing. The graph structure is a mechanism that uses nodes and edges to express causal relationships, establishing a mapping between manifestation, touchpoint, and problem. A manifestation is the outward phenomenon derived from the attributes of an event chain. For example: continuous failures → repeated inquiries → complaints or frequent unsubscriptions → negative wording, etc. A touchpoint refers to the business process, channel, or product that enables the manifestation. A problem refers to the essential issue that must ultimately be attributed to. Each node in the graph structure must contain at least a problem node; that is, each node corresponds to a candidate essential problem. Edges or paths in the graph structure represent how a certain manifestation, through a certain touchpoint, may lead to a certain problem.

[0056] It should be noted that the edges or paths in the graph structure are used to represent the causal relationship between the appearance and the candidate essential problem. The causal relationship can be automatically generated based on the contribution statistics of historical interaction behavior data, and business personnel are allowed to supplement the causal relationship or adjust the path weight to realize the dynamic evolution and personalized configuration of the causal graph.

[0057] Optionally, based on the objective function and auxiliary function, the event chain attributes and sentiment evaluation attributes are processed in combination with the graph structure to obtain the correlation strength attribute of the candidate essential question relative to the interaction behavior data.

[0058] It should be noted that for each candidate path corresponding to isolated anomaly representations and structural representations, the correlation strength attribute between the candidate essential problem and the representation corresponding to the interaction behavior data can be determined based on the objective function. The objective function can be: ; in, This refers to the essential problem of candidates. Compared to the appearance of interaction behavior data The correlation strength attribute. The higher the value of the correlation strength attribute, the more likely the current phenomenon is to correspond to the candidate essential problem. . This refers to statistical analysis of historical data. When it occurs, the problem The probability of it also occurring. This refers to fixed causal weights that are set manually. Indicates the first A user-defined helper function to enhance context relevance or semantic interpretability. , as well as These are the weighting coefficients for the three types of correlation factors. The larger the scale, the more it relies on historical statistics; The larger the scale, the greater the emphasis on prior knowledge. The larger the value, the more emphasis is placed on contextual enhancement.

[0059] It should also be noted that the auxiliary function can be: ; Among them, among them, Indicates from behavioral appearance The time interval between the occurrence of an issue and the feedback of the problem measures the timeliness of the events; the shorter the interval, the higher the correlation. This indicates the degree of interaction and coupling between the touchpoints and problem types involved in the manifestation and the business process, such as whether they are bound to the same package or the same contact channel; This indicates the warning level or business impact level of the current scenario or problem type. For example, whether it falls under the category of high-risk regulatory complaints.

[0060] Specifically, a causal path expression mechanism represented by a graph structure is introduced. An explicit mapping relationship between appearances, touchpoints, and issues is established in the graph structure. Event chain attributes and sentiment evaluation attributes are used as input information on the appearance side and the sentiment side, respectively. The information is propagated or aggregated along the path in the graph structure to calculate the comprehensive correlation between the appearances corresponding to the interactive behavior data and each candidate essential issue, which is used as the correlation strength attribute.

[0061] S150. Based on the association strength attribute, determine the attribution analysis results associated with the target user.

[0062] The attribution analysis result refers to the final determination of the essential problem output. The attribution analysis result typically includes: the selected problem type, the corresponding correlation strength attribute, and necessary explanatory information such as the hit manifestations, touchpoints, and timelines.

[0063] Optionally, based on preset output attributes and context adjustment functions, determine the judgment threshold corresponding to the candidate essential question; based on the judgment threshold and association strength attribute corresponding to each candidate essential question, determine the usable candidate essential question; based on the usable candidate essential question and the target strategy, determine the target candidate essential question; and based on the target candidate essential question, determine the attribution analysis result.

[0064] The target strategy is used to reflect weighted information across multiple preset dimensions.

[0065] It should be noted that preset output attributes are a set of output control parameters predefined by the system. For example, preset output attributes could be a default output threshold, the maximum number of issues to output, whether multiple issues are allowed to be listed concurrently, or output strategies under different risk levels. These preset output attributes serve as the initial threshold for determining whether to include an issue in the output. To enhance the system's adaptability to different business scenarios, a set of context adjustment functions is introduced. Context-adjusting functions can automatically and dynamically adjust output criteria based on event background, user attributes, or unexpected situations to form a final judgment threshold. For example, they can automatically adjust output criteria based on whether it is a high-risk scenario, whether the user is high-value, whether it is a holiday or other special period, or whether it is a system failure period, to form a final judgment threshold. The judgment threshold determines whether a candidate essential question qualifies for inclusion in the output set. The judgment threshold is not a fixed value, but rather a value obtained by context-adjusting preset output attributes. The formula for determining the judgment threshold corresponding to a candidate essential question can be: ; in, This refers to the judgment threshold corresponding to the essential question of the candidate; This refers to preset output attributes; It refers to the first A context adjustment function that dynamically adjusts weights based on the user's business environment.

[0066] It should be noted that the set of questions whose correlation strength attribute is greater than or equal to the judgment threshold are considered usable candidate essential questions. Multiple preset dimensions can include compliance and business priority dimensions. Weighting can be applied across multiple dimensions based on the target strategy. The target candidate essential question refers to the question ultimately determined for output after selecting from the usable candidate essential questions and the target strategy. The formula for determining the target candidate essential question can be: ; in, This refers to the essential issue of target candidates; This refers to the use of candidate essential problems; This is a set of compliance constraints used to eliminate problem types that are not allowed to be output. Examples include blacklisted problems, obsolete problems, and policy restrictions. This information serves as a business priority, used to weight and rank issues. For example, issues with greater impact, higher risk, or higher urgency are prioritized. This refers to the filtering and reordering function, which combines compliance constraints and priority rules to determine which issues are retained and in what order the retained issues are output.

[0067] Specifically, the final judgment threshold for each scenario is first calculated using preset output attributes and context adjustment functions. Then, the set of usable candidate questions is screened out using the association strength attribute. Subsequently, compliance constraints are introduced for filtering, and the questions are reordered in combination with business priorities. Finally, the set of target candidate essential questions to be output is selected by the filtering function and output as the attribution analysis results for the target users.

[0068] In this embodiment, the attribution path system is updated based on interactive behavior data, target candidate essential problems, and attribution analysis results. When the attribution path system meets preset conditions, the attribution analysis results of newly added manifestations are determined based on the attribution path system.

[0069] The attribution path system includes at least a graph structure and a set of representation clusters.

[0070] It's important to note that the attribution path system refers to an evolving mapping table or graph that traces from appearances to essence. It's a maintainable, scalable, and traceable knowledge structure that stores the mapping relationships between which appearances, through which contact chains, typically correspond to which essential questions. Furthermore, the attribution path system includes version and timestamp information for easy backtracking and auditing. Meeting preset conditions refers to a set of quality thresholds that must be met before a new mapping path can be formally incorporated into the attribution path system and used for subsequent automatic attribution. For example, meeting preset conditions might include: the mapping path's confidence reaching a threshold, sufficiently high similarity, a complete causal chain, a reasonable chronological order, and no violation of compliance rules or version management records.

[0071] Optionally, determine the vector distance, event difference information, and attribution link integrity between the newly added representation and the graph structure in the attribution path system; based on the vector distance, event difference information, and attribution link integrity, determine the attribution analysis results of the newly added representation.

[0072] It's important to note that vector distance refers to the distance between a newly added representation and existing representation nodes or path representations in the graph structure after encoding the new representation into a vector. For example, vector distance can be cosine distance or Euclidean distance. The closer the vector distance, the more likely the new representation is to follow an existing attribution path. Event difference information refers to whether the event sequence corresponding to the new representation is consistent with the typical triggering order of a fundamental problem, and whether the time interval is reasonable. If the time sequence does not conform to common-sense causality, the mapping is unreliable. Attribution link completeness refers to whether the "representation—touchpoint chain—fundamental problem" link is sufficiently complete and can form a multi-hop path that is business-explainable. The more complete and coherent the link, the more reliable the mapping; severe missing links or breakpoints tend to indicate low confidence.

[0073] It should also be noted that the system integrates vector distance, time difference information, and attribution path integrity to calculate the overall confidence score. When the overall confidence score reaches a preset condition, the corresponding candidate essential problem is used as the attribution analysis result of the new manifestation, and this mapping path is accumulated into the attribution path system for direct use by subsequent similar manifestations.

[0074] Specifically, it not only attributes the currently identified phenomena to their underlying causes, but also distills this attribution process into reusable knowledge. That is, when a new phenomenon appears, it attempts to establish a mapping between it and known candidate underlying problems, and this mapping is added to the system's "phenomenon-touchpoint chain-underlying problem" relationship network, thus forming an attribution path system that continuously evolves with the data. Once this attribution path system accumulates to meet preset quality conditions—for example, sufficiently high confidence, sufficiently complete links, or version review—it can be directly used to provide attribution analysis results for subsequent similar new phenomena.

[0075] The technical solution of this disclosure first acquires interactive behavior data of a target user under multiple reference indicators to determine a behavior representation vector based on the interactive behavior data. Then, based on the behavior representation vector, it determines the isolated anomaly class representation and the structural representation corresponding to the target user. The isolated anomaly class representation reflects the anomaly data determined by considering interactive behavior data under different reference indicators individually, while the structural representation is the anomaly data determined from the interactive behavior data after combining multiple reference indicators. Next, based on pre-configured custom rules and rule matching functions, it determines the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structural representation. Further, based on the event chain attributes, sentiment evaluation attributes, and a pre-created graph structure, it determines the association strength attribute with candidate essential questions. The graph structure includes at least one node, and each node corresponds to a candidate essential question. Finally, based on the association strength attribute, it determines the attribution analysis result associated with the target user. This solves the problem in the prior art where the identification of user behavior anomalies and the attribution of problems rely on a preset rule base for matching and judgment, resulting in insufficient adaptability to emerging or evolving behavioral patterns, leading to problems such as recognition lag, misjudgment, or missed judgment. This disclosure implements a method to determine the attribution analysis results associated with the target user based on the behavior representation vector corresponding to the interaction behavior data, thereby improving the comprehensiveness and accuracy of anomaly detection and achieving a more accurate attribution analysis result for the target user.

[0076] Example 2 Figure 2This is a flowchart illustrating the data processing method provided in this embodiment of the invention. Based on the foregoing embodiments, it provides a more detailed explanation of how to determine the event chain attributes and sentiment evaluation attributes corresponding to isolated abnormal class representations and structured class representations based on pre-configured custom rules and rule matching functions. For specific implementation details, please refer to the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0077] like Figure 2 As shown, the method specifically includes the following steps: S210. Obtain interactive behavior data of the target user under multiple reference indicators, and determine the behavior representation vector based on the interactive behavior data.

[0078] S220. Based on the behavior representation vector, determine the isolated anomaly class representation and structural representation corresponding to the target user.

[0079] S230. Based on the interactive behavior data corresponding to isolated anomaly-type representations and structural representations, use rule matching functions combined with custom rules to analyze the interactive behavior data and determine the event chain attributes corresponding to the interactive behavior data.

[0080] Interactive behavior data corresponding to isolated anomalies refers to the user event sequences and related text identified as isolated anomalies. For example, this data could include a list of events, channels, actions taken, and feedback content during a time slice when a state transition or behavioral shift occurs. Interactive behavior data corresponding to structural anomalies refers to the combined behavioral data identified as structural anomalies. This data typically includes multiple event sequences covered by composite behavioral vectors spliced ​​across time, high-frequency co-occurring behavioral sequences at the group level, or original event and text data corresponding to high-frequency event sequences.

[0081] It should be noted that a custom rule set can be represented as: ; in, This refers to each custom rule. When analyzing interactive behavior data, in addition to custom rules, built-in rule sets can also be used. A built-in rule set can be represented as: ; in, This refers to each rule in the built-in rule set. The rule matching function can be expressed as: ; It should be noted that, This refers to the event chain attribute corresponding to the interactive behavior data; It refers to the candidate event sequence; the candidate event sequence is a serialized representation extracted from interaction behavior data and organized in chronological order, used for rule matching and event chain identification. Refers to rules or The corresponding weight parameters; This refers to a Boolean function indicating whether a rule is matched; it takes a value of 1 if a rule is matched and a value of 0 if a rule is not matched. By combining the rule matching function with custom rules, the determination of whether a certain event chain pattern appears in a candidate event sequence is transformed into a computable result.

[0082] It should also be noted that, in order to adapt to business scenarios with changing behavioral characteristics, custom rules support on-demand expansion. That is, when a new custom rule is added, a rule matching function corresponding to the new custom rule is automatically generated based on the event sequence matching conditions and corresponding weight parameters of the new custom rule. The rule matching function is then configured into the rule matching function set, so that the new custom rule is added to the custom rule set and the rule set formed together with the built-in rule set, for use in rule matching of candidate event sequences and determination of event chain attributes.

[0083] Specifically, event chain identification is performed based on the interactive behavior data corresponding to isolated anomaly-type or structural-type representations. First, candidate event sequences are extracted from the interactive behavior data and organized in chronological order. Then, the candidate event sequences are input into a rule matching function. Combined with custom rules, the rule matching function is executed one by one. A boolean result of 1 for a match and 0 for a miss indicates whether the corresponding event chain pattern has appeared. The event chain attributes are obtained by weighting and summarizing according to the rule weights.

[0084] S240. Determine the emotional assessment attribute based on the first emotional attribute corresponding to the interaction behavior data output by the first model and the emotional attribute adjustment factor determined by a preset number of plug-in functions.

[0085] The plugin function is determined based on the frequency of negative keywords in the interaction behavior data, industry factors, keyword set, industry sentiment weight, and contextual brightness attributes.

[0086] The first model can be a basic emotion model. The first emotion attribute refers to the initial emotion result, which can be an emotion polarity score, with a value between 0 and 1. The preset number of plug-in functions refers to a number of programmable emotion enhancement modules that can be stacked. Correcting the first emotion attribute based on the preset number of plug-in functions can improve the adaptability to different industries and scenarios. The emotion attribute adjustment factor is a correction coefficient calculated by the plug-in functions and used to amplify or weaken the first emotion attribute.

[0087] It should be noted that the calculation of the plug-in function will take into account three types of information: first, the frequency of occurrence of negative keywords; second, industry factors, to reflect the sentiment weight contribution of keywords in a specific industry; and third, the contextual attributes, namely contextual sensitivity, to reflect whether the sentiment is more intense or more sensitive in business contexts such as package type, channel type, and holidays.

[0088] It should also be noted that the formula for determining the attributes of emotional assessment can be: ; in, This refers to the emotional assessment attribute; This refers to the first sentiment attribute output by the first model, which corresponds to the interaction behavior data. Indicates the first Each plugin function can determine an emotional attribute adjustment factor, which is used to enhance or modify the primary emotional attribute. Furthermore, the formula for determining the emotional attribute adjustment factor based on the plugin function can be: ; in, This refers to the frequency of negative keywords, which can be determined by dividing the number of negative keywords by the total number of occurrences. This represents the sentiment-weighted score of keywords within a specific industry. The set of keywords appearing in the text. Industry sentiment weight for keywords. It can be used to measure the emotional intensity of the text in the current business context. , as well as These are parameter weights, which can be adjusted manually.

[0089] Specifically, when performing contextualized correction on the text sentiment corresponding to interactive behavior data and outputting sentiment assessment attributes, the first sentiment attribute output by the first model for the interactive behavior data is first obtained. Then, sentiment attribute adjustment factors are calculated based on a preset number of plug-in functions, and the first sentiment attribute is enhanced or corrected to improve the accuracy and adaptability of sentiment recognition in different industries and business scenarios, ultimately obtaining the sentiment assessment attribute.

[0090] For example, if multiple typical negative sentiment words are detected in the interaction behavior data, and the feedback time is just before a holiday, the primary sentiment attribute will be enhanced using multiple factors based on high keyword density, high keyword sentiment weight, and strong scene sensitivity. Ultimately, the interaction behavior data will be judged to have a strong negative sentiment tendency, thereby increasing its correlation strength with essential issues such as abnormal service touchpoints and deficiencies in the reminder mechanism.

[0091] S250. Based on the event chain attributes, sentiment assessment attributes, and the pre-created graph structure, determine the correlation strength attributes between the candidate essential questions and the relevant issues.

[0092] S260. Based on the association strength attribute, determine the attribution analysis results associated with the target user.

[0093] The technical solution of this disclosure first acquires interactive behavior data of a target user under multiple reference indicators, determines a behavior representation vector based on the interactive behavior data, and determines the isolated anomaly class representation and structural representation corresponding to the target user based on the behavior representation vector. Then, based on the interactive behavior data corresponding to the isolated anomaly class representation and structural representation, an interactive behavior data analysis is performed using a rule matching function combined with custom rules to determine the event chain attributes corresponding to the interactive behavior data. Furthermore, based on the first sentiment attribute output by the first model corresponding to the interactive behavior data and the sentiment attribute adjustment factor determined by a preset number of plugin functions, a sentiment evaluation attribute is determined. Further, based on the event chain attribute, sentiment evaluation attribute, and a pre-created graph structure, the association strength attribute with the candidate essential question is determined. Finally, based on the association strength attribute, the attribution analysis result associated with the target user is determined. By fusing multidimensional interactive behavior data with textual sentiment information for modeling, both isolated anomalies and structural anomalies can be discovered simultaneously, improving the coverage and accuracy of anomaly detection. Furthermore, interpretable event chain attributes can be extracted using scalable custom rules and rule matching functions, and the sentiment results can be corrected by combining industry and context plugins, enhancing cross-scene adaptability.

[0094] Example 3 As an optional embodiment of the present invention, an example is provided to further illustrate the invention.

[0095] It should be noted that, see Figure 3 To monitor user behavior, user interaction data at various business touchpoints is collected, forming a time series and encoding features to prepare for subsequent anomaly identification and similarity calculation. When identifying potential new phenomena, phenomena can be determined in two ways: isolated anomaly identification through single-point or local mutations, and composite structural identification through multi-event combinations or path patterns. Simultaneously, novelty assessment is performed to determine whether a potential new phenomenon truly is a new phenomenon. Further, abnormal event chain identification and programmable sentiment analysis are conducted, determining event chain attributes and sentiment evaluation attributes. After determining these attributes, a dynamic causal graph is constructed, calculating the correlation strength of each candidate essential question and controlling the output to determine the final attribution result. Similarity calculation and confidence filtering are performed on the mapping from new phenomena to essential questions, and the data is written back to the graph and maintained, forming a continuously evolving attribution path system.

[0096] The technical solution of this disclosure determines event chain attributes, new manifestations, sentiment assessment attributes, causal graph mapping, and weights based on the input interactive behavior data. Ultimately, it outputs attribution analysis results, which can more comprehensively cover single-point mutations and multi-event combination patterns, reduce the false negative rate, and improve the accuracy and stability of locating the root cause of problems.

[0097] Example 4 Figure 4 This is a schematic diagram of the structure of the data processing apparatus provided in the embodiments of this disclosure, as shown below. Figure 4 As shown, the device includes: a behavior representation vector determination module 310, an appearance determination module 320, an event chain attribute determination module 330, an association strength attribute determination module 340, and an attribution analysis result determination module 350.

[0098] The system includes a behavior representation vector determination module, used to acquire interactive behavior data of a target user under multiple reference indicators, and to determine a behavior representation vector based on the interactive behavior data; a representation determination module, used to determine the isolated anomaly representation and structural representation corresponding to the target user based on the behavior representation vector, wherein the isolated anomaly representation reflects the anomaly data determined by considering the interactive behavior data under different reference indicators individually, and the structural representation is the anomaly data determined from the interactive behavior data after combining multiple reference indicators; an event chain attribute determination module, used to determine the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly representation and the structural representation based on pre-configured custom rules and rule matching functions; a correlation strength attribute determination module, used to determine the correlation strength attribute between the event chain attribute, the sentiment evaluation attribute, and a pre-created graph structure; wherein the graph structure includes at least one node, the node corresponding to the candidate essential question; and an attribution analysis result determination module, used to determine the attribution analysis result associated with the target user based on the correlation strength attribute.

[0099] The technical solution of this disclosure first acquires interactive behavior data of a target user under multiple reference indicators to determine a behavior representation vector based on the interactive behavior data. Then, based on the behavior representation vector, it determines the isolated anomaly class representation and the structural representation corresponding to the target user. The isolated anomaly class representation reflects the anomaly data determined by considering interactive behavior data under different reference indicators individually, while the structural representation is the anomaly data determined from the interactive behavior data after combining multiple reference indicators. Next, based on pre-configured custom rules and rule matching functions, it determines the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structural representation. Further, based on the event chain attributes, sentiment evaluation attributes, and a pre-created graph structure, it determines the association strength attribute with candidate essential questions. The graph structure includes at least one node, and each node corresponds to a candidate essential question. Finally, based on the association strength attribute, it determines the attribution analysis result associated with the target user. This solves the problem in the prior art where the identification of user behavior anomalies and the attribution of problems rely on a preset rule base for matching and judgment, resulting in insufficient adaptability to emerging or evolving behavioral patterns, leading to problems such as recognition lag, misjudgment, or missed judgment. This disclosure implements a method to determine the attribution analysis results associated with the target user based on the behavior representation vector corresponding to the interaction behavior data, thereby improving the comprehensiveness and accuracy of anomaly detection and achieving a more accurate attribution analysis result for the target user.

[0100] Based on the above technical solutions, the behavior representation vector determination module 310 is further configured to acquire the target user's interaction behavior data under multiple reference indicators, wherein the interaction behavior data includes at least the tariff operation behavior data, service access data, business processing data, and user subjective feedback data associated with the current terminal; the interaction behavior data is encoded according to the event type corresponding to the interaction behavior data, and features are extracted from the interaction behavior data according to the first keyword information to obtain structured topic features; the interaction behavior data is extracted according to the behavior occurrence time information in the same event type, and the extracted features are encoded by sequence hashing to obtain encoded features; the behavior representation vector of the target user is obtained by concatenating the event type, the topic features, and the encoded features.

[0101] Based on the above technical solutions, the appearance determination module 320 includes an isolated anomaly appearance determination submodule and a structured appearance determination submodule.

[0102] The isolated anomaly identification submodule is used to aggregate the behavior representation vector based on a preset aggregation time length and reference indicators to obtain a first behavior vector, and then normalize the first behavior vector to obtain a second behavior vector. The normalization process is performed by combining the first behavior vectors of all users. Based on the second behavior vector, the module determines the behavior state region where the target user's interaction behavior data resides. Different behavior state regions correspond to different indicator thresholds, and these thresholds are dynamically changed based on a smoothing factor. Based on the comprehensive state vector corresponding to a preset time selection slice, the module determines the behavior path evolution direction vector of the target user under that preset time selection slice. Based on the behavior path evolution direction vector corresponding to the same preset time selection slice, the associated behavior state region, and the comprehensive state vector, the module determines whether the target user's interaction behavior data constitutes an isolated anomaly.

[0103] Based on the above technical solutions, the isolated anomaly representation determination submodule is further configured to determine whether a transition has occurred in the behavior state partition corresponding to the preset time selection slice based on the comprehensive state vector and the behavior state region to which the comprehensive state vector belongs; if a transition occurs, it is determined that the target user's interaction behavior data contains isolated anomaly representations; and / or, based on the behavior path evolution direction vector, determine behavior direction mutation information; when the behavior direction mutation information meets preset conditions, it is determined that the target user's interaction behavior data contains isolated anomaly representations.

[0104] The structural representation determination submodule is used to select a central trajectory from multiple representation cluster sets associated with existing representation recognition results, and determine a representative path set. The central trajectory characterizes the representative path of the corresponding representation cluster set, and the representative path set includes multiple representative paths that reflect the occurrence chain of behavioral interaction data. Based on the representation paths of the target user occurring in multiple time slices, time concatenation is performed to determine a composite behavior vector. Furthermore, multiple users' frequently occurring behavior sequences under the same reference indicators are determined, and a high-frequency event sequence is determined based on a sliding time window. Based on the composite behavior vector, the behavior sequence, the high-frequency event sequence, and the representative path set, a coverage evaluation result, the business tag offset information involved in the target user, and the business indicator offset information are determined. When at least two of the coverage evaluation result, the business tag offset information, and the business indicator offset information meet preset conditions, the target user's interaction behavior data, the composite behavior vector, the behavior sequence, and the high-frequency event sequence are determined as structural representations.

[0105] Based on the above technical solutions, the event chain attribute determination module 330 is further configured to analyze the interaction behavior data corresponding to the isolated abnormal class representation and the structural class representation using a rule matching function combined with custom rules to determine the event chain attribute corresponding to the interaction behavior data; and to determine the sentiment evaluation attribute based on the first sentiment attribute output by the first model corresponding to the interaction behavior data and the sentiment attribute adjustment factor determined by a preset number of plug-in functions; wherein, the plug-in functions are determined based on the frequency of occurrence of negative keywords, industry factors, keyword set, industry sentiment weight, and contextual brightness attributes in the interaction behavior data.

[0106] Based on the above technical solutions, the association strength attribute determination module 340 is further used to process the event chain attribute and sentiment evaluation attribute based on the objective function and auxiliary function, combined with the graph structure, to obtain the association strength attribute of the candidate essential question relative to the interaction behavior data.

[0107] Based on the above technical solutions, the attribution analysis result determination module 350 is further configured to determine the judgment threshold corresponding to the candidate essential problem according to the preset output attributes and context adjustment function; determine the usable candidate essential problem according to the judgment threshold corresponding to each candidate essential problem and the association strength attribute; determine the target candidate essential problem according to the usable candidate essential problem and the target strategy; wherein, the target strategy is used to reflect weighted information in multiple preset dimensions; and determine the attribution analysis result according to the target candidate essential problem.

[0108] Based on the above technical solutions, the device further includes: an attribution path system update module, used to update the attribution path system based on the interaction behavior data, the target candidate essential problem and the attribution analysis results, so as to determine the attribution analysis results of the newly added representations based on the attribution path system when the attribution path system meets preset conditions; wherein, the attribution path system includes at least a graph structure and a representation cluster set.

[0109] Based on the above technical solutions, the attribution path system update module is further used to determine the vector distance, event difference information, and attribution link integrity between the newly added representation and the graph structure in the attribution path system; and to determine the attribution analysis result of the newly added representation based on the vector distance, the event difference information, and the attribution link integrity.

[0110] The data processing apparatus provided in this disclosure can execute the data processing method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects for executing the method.

[0111] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this disclosure.

[0112] Example 5 Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Refer to the following... Figure 5 It illustrates an electronic device suitable for implementing embodiments of the present disclosure (e.g., Figure 5 The diagram below shows the structure of the terminal device or server 500. The terminal device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and vehicle terminals (e.g., vehicle navigation terminals). Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0113] like Figure 5 As shown, electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from storage device 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. An edit / output (I / O) interface 505 is also connected to bus 504.

[0114] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0115] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0116] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0117] The electronic device provided in this embodiment and the data processing method provided in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0118] Example 6 This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the data processing method provided in the above embodiments.

[0119] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0120] In some implementations, the server may communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and may interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0121] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0122] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: Acquire interactive behavior data of target users under multiple reference indicators, and determine a behavior representation vector based on the interactive behavior data; Based on the behavior representation vector, the isolated anomaly class representation and the structural representation corresponding to the target user are determined. The isolated anomaly class representation is used to reflect the abnormal data determined by considering the interaction behavior data under different reference indicators individually. The structural representation is the abnormal data determined from the interaction behavior data by combining multiple reference indicators. Based on pre-configured custom rules and rule matching functions, the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structure class representation are determined. Based on the event chain attributes, sentiment assessment attributes, and a pre-created graph structure, the correlation strength attribute between the event chain and the candidate essential question is determined; wherein, the graph structure includes at least one node, and the node corresponds to the candidate essential question. Based on the association strength attribute, the attribution analysis results associated with the target user are determined.

[0123] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0124] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0125] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0126] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0127] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0128] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0129] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0130] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A data processing method, characterized in that, include: Acquire interactive behavior data of target users under multiple reference indicators, and determine a behavior representation vector based on the interactive behavior data; Based on the behavior representation vector, the isolated anomaly class representation and the structural representation corresponding to the target user are determined. The isolated anomaly class representation is used to reflect the abnormal data determined by considering the interaction behavior data under different reference indicators individually. The structural representation is the abnormal data determined from the interaction behavior data by combining multiple reference indicators. Based on pre-configured custom rules and rule matching functions, the event chain attributes and sentiment evaluation attributes corresponding to the isolated anomaly class representation and the structure class representation are determined. Based on the event chain attributes, sentiment assessment attributes, and a pre-created graph structure, the correlation strength attribute between the event chain and the candidate essential question is determined; wherein, the graph structure includes at least one node, and the node corresponds to the candidate essential question. Based on the association strength attribute, the attribution analysis results associated with the target user are determined.

2. The method according to claim 1, characterized in that, The step of acquiring target user interaction behavior data under multiple reference metrics, and determining behavior representation vectors based on the interaction behavior data, includes: The interaction behavior data of the target user under multiple reference indicators is obtained, wherein the interaction behavior data includes at least the tariff operation behavior data, service access data, business processing data and user subjective feedback data associated with the current terminal. Based on the event type corresponding to the interactive behavior data, the interactive behavior data is encoded, and features are extracted from the interactive behavior data based on the first keyword information to obtain structured topic features; The interactive behavior data is extracted based on the behavior occurrence time information of the same event type, and the extracted features are encoded by sequence hashing to obtain encoded features; The behavior representation vector of the target user is obtained by concatenating the event type, the topic features, and the encoding features.

3. The method according to claim 1, characterized in that, Based on the behavior representation vector, determine the isolated anomaly class representation corresponding to the target user, including: Based on the preset aggregation time length and reference indicators, the behavior representation vector is aggregated to obtain a first behavior vector, and the first behavior vector is normalized to obtain a second behavior vector. The normalization process is performed by combining the first behavior vectors of all users. Based on the second behavior vector, the behavior state zone where the target user's interaction behavior data is located is determined, wherein different behavior state zones correspond to different indicator thresholds, and the indicator thresholds are dynamically changed based on a smoothing factor; Based on the comprehensive state vector corresponding to the selected slice at the preset time, the behavioral path evolution direction vector of the target user under the selected slice at the preset time is determined. Based on the behavioral path evolution direction vector, the behavioral state region, and the comprehensive state vector corresponding to the selected slice at the same preset time, it is determined whether the interactive behavior data of the target user is an isolated abnormal phenomenon.

4. The method according to claim 3, characterized in that, The step of determining whether the target user's interaction behavior data is an isolated anomaly based on the behavioral path evolution direction vector, the behavioral state region, and the comprehensive state vector corresponding to the selected slice at the same preset time includes: Based on the comprehensive state vector and the behavior state region to which the comprehensive state vector belongs, determine whether a transition has occurred in the behavior state partition corresponding to the preset time selection slice. If a transition occurs, it is determined that the target user's interaction behavior data exhibits isolated anomaly-like phenomena; and / or, Based on the behavioral path evolution direction vector, determine the behavioral direction mutation information; When the behavioral direction mutation information meets the preset conditions, it is determined that the target user's interactive behavior data has an isolated abnormal class manifestation.

5. The method according to claim 1, characterized in that, Based on the behavioral representation vector, structural representations are determined, including: Select a central trajectory from multiple sets of appearance clusters associated with existing appearance recognition results to determine a set of representative paths. The central trajectory is used to characterize the representative path of the corresponding appearance cluster set. The set of representative paths includes multiple representative paths, which are used to reflect the occurrence link of behavioral interaction data. Based on the apparent paths of the target user in multiple time slices, time splicing is performed to determine the composite behavior vector, and the behavior sequences of multiple users that occur frequently under the same reference indicators are determined, as well as the high-frequency event sequence is determined based on the sliding time window. Based on the composite behavior vector, the behavior sequence, the high-frequency event sequence, and the representative path set, determine the coverage assessment result, the business tag offset information and business indicator offset information involved in the target user; When at least two of the coverage assessment results, the business tag offset information, and the business indicator offset information meet preset conditions, the target user's interaction behavior data, the composite behavior vector, the behavior sequence, and the high-frequency event sequence are determined as structural representations.

6. The method according to claim 1, characterized in that, The process of determining the event chain attributes and sentiment assessment attributes corresponding to the isolated anomaly class representations and the structure class representations based on pre-configured custom rules and rule matching functions includes: Based on the interactive behavior data corresponding to the isolated anomaly class representation and the structural class representation, the interactive behavior data is analyzed using a rule matching function combined with custom rules to determine the event chain attributes corresponding to the interactive behavior data. The emotional assessment attribute is determined based on the first emotional attribute corresponding to the interactive behavior data output by the first model and the emotional attribute adjustment factor determined by a preset number of plug-in functions. The plugin function is determined based on the frequency of negative keywords, industry factors, keyword set, industry sentiment weight, and contextual brightness attributes in the interaction behavior data.

7. The method according to claim 1, characterized in that, The step of determining the correlation strength attribute between the event chain attribute, sentiment assessment attribute, and pre-created graph structure and the candidate essential question includes: Based on the objective function and auxiliary function, and combined with the graph structure, the event chain attributes and sentiment evaluation attributes are processed to obtain the correlation strength attribute of the candidate essential question relative to the interaction behavior data.

8. The method according to claim 1, characterized in that, The step of determining the attribution analysis result associated with the target user based on the association strength attribute includes: Based on the preset output attributes and context adjustment function, determine the judgment threshold corresponding to the candidate essential question; Based on the judgment threshold corresponding to each candidate essential question and the correlation strength attribute, the usable candidate essential questions are determined; Based on the available candidate essential problems and target strategies, a target candidate essential problem is determined; wherein, the target strategy is used to reflect weighted information across multiple preset dimensions; The attribution analysis results are determined based on the target candidate essential problem.

9. The method according to claim 8, characterized in that, The method further includes: Based on the interactive behavior data, the target candidate essential problem, and the attribution analysis results, the attribution path system is updated so that when the attribution path system meets preset conditions, the attribution analysis results of the newly added manifestations are determined based on the attribution path system. The attribution path system includes at least a graph structure and a set of representation clusters.

10. The method according to claim 9, characterized in that, Based on the aforementioned attribution path system, the attribution analysis results for newly added representations are determined, including: Determine the vector distance, event difference information, and attribution link integrity between the newly added representation and the graph structure in the attribution path system; The attribution analysis results for the newly added representations are determined based on vector distance, the event difference information, and the attribution link integrity.

Images