A method and system for omnichannel customer relationship management

By constructing a weighted set of user interaction states and generating personalized verification tasks, combined with operation trajectory analysis, the problems of untimely updates to user profiles and separation of security verification were solved, achieving the unification of security verification and business profiles, and improving data timeliness and system efficiency.

CN122596981BActive Publication Date: 2026-11-17YIZHI TIMES (XIAMEN) DIGITAL TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610627365.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-05-08
Publication Date
2026-11-17
Estimated Expiration
2046-05-08

AI Technical Summary

Technical Problem

Existing technologies suffer from problems such as untimely updates to user profiles, separation of security verification processes from business data maintenance, and high resource consumption.

Method used

By receiving and parsing raw interaction logs from all channels, a weighted set of user interaction states is constructed. Based on the weights, business data fragments are selected to generate personalized verification tasks. The user's operation trajectory during the completion process is collected for double verification and scoring decisions. Finally, the verification score is directly fed back to the set of user interaction states to update the weights of each dimension in real time.

Benefits of technology

It unifies the processes of security verification and business profile maintenance, automatically updates data, reduces system operation and maintenance costs, improves verification relevance and efficiency, enhances security, and reduces the need for manual maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122596981B_ABST
    Figure CN122596981B_ABST
Patent Text Reader

Abstract

This invention discloses an omnichannel customer relationship management method and system, relating to the field of data management technology. The system includes: a log parsing module, a state construction module, a feature filtering module, a mask generation module, a feedback receiving module, a comparison and analysis module, a scoring decision module, and a weight update module. This invention constructs a weighted set of user interaction states by receiving and parsing raw interaction logs from all channels. Then, when a data access request is triggered, it filters business data fragments based on weights, generates personalized verification tasks, and collects the user's operation trajectory during the completion process for dual verification and scoring decisions. Finally, the verification score is directly fed back to the user interaction state set to update the weights of each dimension in real time. This achieves unified processes for security verification and business profile maintenance, automatic data updates, and reduced system operation and maintenance costs, solving the problems of untimely user profile updates and the separation of security verification processes from business data maintenance, which consumes significant resources.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to an omnichannel customer relationship management method and system. Background Technology

[0002] As enterprises evolve digitally, omnichannel customer relationship management systems (CRMs) need to integrate user interaction data from multiple touchpoints, such as websites and mobile applications, to build accurate user profiles and implement personalized customer management. User profiles are the core of understanding customer needs and making precise outreach; their accuracy and timeliness are paramount.

[0003] Currently, existing technologies for building user profiles typically rely on periodic offline analysis and model training of historical interaction logs. User profiles generated in this way are slowly updated and cannot reflect the latest changes in user interests and behaviors in a timely manner, remaining in a relatively static state. Furthermore, identity verification commonly employs generic CAPTCHA technologies unrelated to business logic, such as requiring users to identify distorted characters or select specific image types. This verification process is independent of customer relationship management processes, and the resulting user interaction data is discarded after verification, providing no value for business analysis.

[0004] Therefore, existing technologies have significant shortcomings: the updating and maintenance of user profiles is disconnected from the security verification process. On the one hand, dynamic updates to profiles require additional computing resources and complex scheduling, resulting in high costs and delays; on the other hand, the independent security verification step only consumes resources without providing any business benefits. This leads to problems in customer relationship management, such as untimely updates to user profiles and the separation of the security verification process from business data maintenance, which also consumes significant resources. Summary of the Invention

[0005] This application provides an omnichannel customer relationship management method and system, which solves the problems of untimely user profile updates, separation of security verification process and business data maintenance, and high resource consumption in the prior art. It realizes the unification of security verification and business profile maintenance processes, automatic data updates, and reduced system operation and maintenance costs.

[0006] This application provides an omnichannel customer relationship management method, which is applied to an omnichannel customer relationship management system and includes:

[0007] Receive raw interaction logs from different channels, parse the raw interaction logs to extract the behavior object identifier and timestamp information;

[0008] Based on the behavior object identifier and timestamp information, a set of user interaction states containing multiple data dimensions is constructed, and an initial activity weight is assigned to each dimension in the set;

[0009] When a data access request initiated by a client is detected, target feature data fragments are filtered out from the user interaction state set based on activity weight;

[0010] Perform a local masking operation on the target feature data fragment to generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering;

[0011] Receive feedback data from the client for filling missing value regions. The feedback data includes the filled content value and the operation trajectory coordinate sequence.

[0012] The filled content value is compared with the original value of the target feature data segment to generate the content matching result, and the vector offset of the operation trajectory coordinate sequence is calculated at the same time to generate the operation confidence coefficient.

[0013] Based on the content matching results and the operation trajectory coordinate sequence, calculate the comprehensive verification score for this interaction. If the comprehensive verification score meets the preset threshold, respond to the data access request.

[0014] Using the comprehensive verification score as a feedback factor, the activity weight of the corresponding dimension in the user interaction state set is directly updated, thereby changing the selection priority for generating the next verification structure to be completed.

[0015] Furthermore, the user interaction state set, constructed based on the behavior object identifier and timestamp information, includes multiple data dimensions, including:

[0016] The original interaction logs are arranged in reverse order of timestamps, and the behavior object identifiers within the most recent preset period are extracted as candidate dimensions.

[0017] Calculate the frequency of each candidate dimension in the original interaction log, and store the candidate dimensions whose frequency exceeds the frequency threshold into the user interaction state set;

[0018] Based on the time interval between the timestamp and the current time, the initial active weight of each stored dimension is calculated using the time decay function, so that the dimension with the smaller time interval has a larger initial active weight.

[0019] Furthermore, the step of performing a local masking operation on the target feature data fragment to generate a verification structure to be completed, containing regions of missing values, includes:

[0020] Identify the data type of the target feature data fragment. If the data type is text, retain the first and last characters and replace the remaining characters with placeholders to form a missing value region.

[0021] If the data type is image index, the image processing interface is called to divide the image into blocks, and at least one image block is randomly removed to form a missing value region.

[0022] The processed data is encapsulated into a JSON-formatted rendering instruction package, which is the structure to be completed and verified.

[0023] Furthermore, the feedback data returned by the receiving client for completing the missing value region includes:

[0024] During the rendering of the rendering instruction package on the client side, an event listener is started to capture the cursor's movement coordinates on the screen at a frequency of milliseconds;

[0025] Record the duration a user spends from confirming a missing value area to submitting the populated content value;

[0026] The movement coordinates are combined to form an operation trajectory coordinate sequence, and the user-input fill values ​​and dwell time are packaged into completion feedback data.

[0027] Furthermore, the step of comparing the filled content value with the original value of the target feature data segment to generate a content matching result, and simultaneously calculating the vector offset of the operation trajectory coordinate sequence to generate an operation confidence coefficient, includes:

[0028] Determine whether the filled content value is completely consistent with the original value of the target feature data segment. If they are consistent, mark the content matching result as true; otherwise, mark it as false.

[0029] Calculate the rate of change of distance between adjacent coordinate points in the operation trajectory coordinate sequence to obtain the acceleration sequence;

[0030] Calculate the standard deviation of the acceleration sequence, and use the reciprocal of the standard deviation as the operational confidence coefficient after normalization.

[0031] Furthermore, based on the content matching results and the operation trajectory coordinate sequence, the comprehensive verification score for this interaction is calculated, including:

[0032] A base score is preset;

[0033] If the content matching result is true, the product of the operation confidence coefficient and the preset weight is added to the base score to obtain the comprehensive verification score;

[0034] If the content matching result is false, the comprehensive verification score will be set to zero and the exception recording process will be triggered.

[0035] Furthermore, the step of using the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set includes:

[0036] Obtain the current active weight of the selected target feature data fragment in the user interaction state set;

[0037] Calculate the difference between the comprehensive verification score and the preset benchmark score to obtain the weight adjustment step size;

[0038] Add the current activity weight to the weight adjustment step size to obtain the updated activity weight, and write the updated activity weight into the user interaction state set.

[0039] Furthermore, writing the updated activity weight into the user interaction state set includes:

[0040] Check if the updated activity weight is lower than the elimination threshold;

[0041] If the value exceeds the elimination threshold, the data for that dimension will be retained in the database.

[0042] If the value is below the elimination threshold, the feature data fragment of the corresponding dimension will be removed from the user interaction state set, and the dimension will be excluded in the next selection of target feature data fragments.

[0043] Furthermore, if the threshold is lower than the elimination threshold, the corresponding feature data fragment will be removed from the user interaction state set, including:

[0044] Check if there are any sub-dimensions in the original interaction log that are related to the removed dimension;

[0045] If it exists, extract the secondary dimension and add it to the user interaction state set according to the initial activity weight to fill the gap after removal.

[0046] This application provides an omnichannel customer relationship management system for implementing an omnichannel customer relationship management method, including: a log parsing module, a status construction module, a feature filtering module, a mask generation module, a feedback receiving module, a comparison analysis module, a scoring decision module, and a weight update module;

[0047] The log parsing module is used to receive raw interaction logs from different channels and parse the raw interaction logs to extract the behavior object identifier and timestamp information.

[0048] The state construction module is used to construct a set of user interaction states containing multiple data dimensions based on the behavior object identifier and timestamp information, and to assign an initial activity weight to each dimension in the set;

[0049] The feature filtering module is used to filter out target feature data fragments from the user interaction state set based on the activity weight when a data access request initiated by the client is detected.

[0050] The mask generation module is used to perform a local masking operation on the target feature data fragment, generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering;

[0051] The feedback receiving module is used to receive completion feedback data for missing value regions returned by the client. The completion feedback data includes the filled content value and the operation trajectory coordinate sequence.

[0052] The comparison and analysis module is used to compare the filled content value with the original value of the target feature data segment to generate a content matching result, and at the same time calculate the vector offset of the operation trajectory coordinate sequence to generate an operation confidence coefficient.

[0053] The scoring decision module is used to calculate the comprehensive verification score of this interaction based on the content matching result and the operation trajectory coordinate sequence. If the comprehensive verification score meets the preset threshold, it will respond to the data access request.

[0054] The weight update module is used to use the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set, so as to change the screening priority for generating the verification structure to be completed next time.

[0055] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0056] This application provides a method and system for omnichannel customer relationship management, which involves parsing omnichannel interaction logs to construct a set of user interaction states and assigning active weights to dimensions; filtering target feature data fragments based on weights, generating a verification structure to be completed through partial masking, and distributing it; receiving the fill content and operation trajectory feedback from the client, performing content matching and operation confidence coefficient analysis, calculating a comprehensive verification score for decision-making; and using the score to update the active weights of the corresponding dimensions in the set.

[0057] In this process, verification tasks are generated using the user's own business history data, so that the verification process is naturally associated with the user profile. While completing identity verification, the user's status is lightly detected, improving the relevance and efficiency of verification.

[0058] Furthermore, by analyzing the trajectory coordinate sequence during user completion operations, the operation confidence coefficient is calculated and combined with the content correctness for evaluation, effectively distinguishing between human and machine operations, enhancing security without the need for additional biometric data collection.

[0059] Ultimately, the verification score is directly used to adjust the weights in the user interaction state set, so that each security interaction automatically drives the cleaning and optimization of business profile data, continuously maintains data timeliness, reduces the need for manual maintenance, and achieves the integration of security and data management. Attached Figure Description

[0060] Figure 1 A flowchart of an omnichannel customer relationship management method provided in this application embodiment;

[0061] Figure 2 This is a schematic diagram of the structure of an omnichannel customer relationship management system provided in an embodiment of this application. Detailed Implementation

[0062] This application provides an omnichannel customer relationship management method and system, which solves the problems of untimely user profile updates, separation of security verification process and business data maintenance, and high resource consumption in the prior art. By receiving and parsing the original interaction logs of all channels, a weighted set of user interaction states is constructed. Then, when a data access request is triggered, business data fragments are selected according to the weights, personalized verification tasks are generated, and the operation trajectory of the user in the completion process is collected for dual verification and scoring decision. Finally, the verification score is directly fed back to the set of user interaction states to update the weights of each dimension in real time. This achieves the unification of the security verification and business profile maintenance process, automatic data updates, and reduced system operation and maintenance costs.

[0063] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0064] like Figure 1 As shown in the figure, this application provides an omnichannel customer relationship management method, which is applied in an omnichannel customer relationship management system, including:

[0065] Receive raw interaction logs from different channels, parse the raw interaction logs to extract the behavior object identifier and timestamp information;

[0066] Raw interaction logs from different channels such as web pages and mobile applications must follow a unified logging standard and include at least a user identifier, an action object identifier, and a timestamp.

[0067] The required structured fields are extracted from unstructured log text. User identifiers are used to associate all data of the same user; behavior object identifiers are the basis for building user interest dimensions, including product IDs, article IDs, page URLs, etc.; timestamps are used to calculate the recency of behavior.

[0068] The parsed data is temporarily stored in memory or a cached database in the form of a structured list or data table.

[0069] Based on the behavior object identifier and timestamp information, a set of user interaction states containing multiple data dimensions is constructed, and an initial activity weight is assigned to each dimension in the set;

[0070] The user interaction state set is a dynamic collection of key-value pairs or a small database table, where the key is the identifier of the behavior object, representing an interest dimension, and the value is the current active weight of the interest dimension.

[0071] By transforming raw, discrete interaction logs into a set of user interaction states, personalized verification challenges can be dynamically generated based on the user's latest interest intensity. Activity weight is a relative value, and its initial value is calculated based on two principles: frequency of recent behavior and the recentity of the behavior. The set of user interaction states is used to create real-time user profiles.

[0072] When a data access request initiated by a client is detected, target feature data fragments are filtered out from the user interaction state set based on activity weight;

[0073] When security verification is required, the system does not randomly select verification content, but intelligently filters based on the current set of user interaction states.

[0074] The probability of each dimension being selected Its active weight Proportional, that is .

[0075] Where i is the specified dimension number, and j is the dimension summation index. It is the total number of dimensions in the set. This indicates a summation operation.

[0076] Since recently active and high-weighted interest dimensions are more likely to be selected, the generated verification challenges are highly relevant to the user's current cognition, which improves the user-friendliness of the challenges and their effectiveness as a behavior detector.

[0077] Perform a local masking operation on the target feature data fragment to generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering;

[0078] Partial masking refers to hiding key information while preserving some contextual information, creating a fill-in-the-blank or completion task. The generated structure is a structured data object, such as JSON, containing task description, masked data, answer index, and other information. Its purpose is to transform selected information, i.e., target feature data fragments, into an interactive human-computer verification task, presenting a challenge on the client side that requires user cognition and operation to complete, rather than simple image recognition or slider dragging.

[0079] Receive feedback data from the client for filling missing value regions. The feedback data includes the filled content value and the operation trajectory coordinate sequence.

[0080] The completion feedback data returned by the client includes explicit answers (i.e., the values ​​to be filled) and implicit behavioral data (i.e., the sequence of operation trajectory coordinates). The sequence of operation trajectory coordinates is a series of triples arranged in chronological order. ,in , These are the horizontal and vertical pixel coordinates in the screen coordinate system. This is a timestamp in milliseconds relative to the start of the operation. This data is the primary basis for analyzing the naturalness of user actions.

[0081] The filled content value is compared with the original value of the target feature data segment to generate the content matching result, and the vector offset of the operation trajectory coordinate sequence is calculated at the same time to generate the operation confidence coefficient.

[0082] Content matching verifies whether the user knows the answer and generates a boolean value.

[0083] Operation confidence coefficient analysis verifies whether the user's response conforms to natural human interaction patterns, generating a scalar value between 0 and 1 for the operation confidence coefficient. This dual verification increases the difficulty for web crawlers or automated scripts to simulate authentication.

[0084] Based on the content matching results and the operation trajectory coordinate sequence, calculate the comprehensive verification score for this interaction. If the comprehensive verification score meets the preset threshold, respond to the data access request.

[0085] Comprehensive verification score The calculation formula is .

[0086] in, It is a Boolean value. It is the base score. These are the weighting coefficients for the confidence coefficient. This is the operational confidence coefficient.

[0087] When Boolean value When the answer is false, since a correct answer is a fundamental prerequisite, no matter how natural the operation is, the overall verification score will be affected. All are zero; verification threshold It is a decision boundary determined in advance through experiments; and then the continuous analysis results are transformed into explicit pass or reject instructions.

[0088] Using the comprehensive verification score as a feedback factor, the activity weight of the corresponding dimension in the user interaction state set is directly updated, thereby changing the selection priority for generating the next verification structure to be completed.

[0089] The verification results are combined with the verification score as a feedback signal to adjust the weights of the corresponding dimensions in the user interaction state set.

[0090] Set the weight update formula as follows: .

[0091] in, The updated weights; To update the previous weights; It is the learning rate, used to control the step size of updates; It is the expected comprehensive verification score benchmark value. This is an error or deviation signal.

[0092] If the verification results are combined with the verification score Higher than the comprehensive verification score benchmark If the interest is strong, the interest dimension is enhanced; otherwise, it is weakened. The system then dynamically adjusts the assessment of the intensity of the user's interest based on the user's actual interaction.

[0093] Furthermore, the user interaction state set, constructed based on the behavior object identifier and timestamp information, includes multiple data dimensions, including:

[0094] The original interaction logs are arranged in reverse order of timestamps, and the behavior object identifiers within the most recent preset period are extracted as candidate dimensions.

[0095] Calculate the frequency of each candidate dimension in the original interaction log, and store the candidate dimensions whose frequency exceeds the frequency threshold into the user interaction state set;

[0096] Based on the time interval between the timestamp and the current time, the initial active weight of each stored dimension is calculated using the time decay function, so that the dimension with the smaller time interval has a larger initial active weight.

[0097] Calculate the frequency of candidate dimensions within the most recent L days and compare it with a frequency threshold. By filtering out accidental and meaningless interaction noise, ensure that the dimensions in the set represent the user's stable or recent interests.

[0098] Initial active weight The time decay function is derived from the time decay function. .

[0099] in, It is a natural constant; It is the attenuation rate constant; From the moment the action last occurred up to the current moment The time interval, .

[0100] It is the attenuation rate constant. This is used to determine how quickly the weight decays over time. For example, if the half-life is set to 7 days, then... .

[0101] when Smaller values ​​indicate that more recent actions receive a higher weight close to 1, while... Larger values ​​indicate that older behaviors have weights that are close to 0.

[0102] Furthermore, the step of performing a local masking operation on the target feature data fragment to generate a verification structure to be completed, containing regions of missing values, includes:

[0103] Identify the data type of the target feature data fragment. If the data type is text, retain the first and last characters and replace the remaining characters with placeholders to form a missing value region.

[0104] If the data type is image index, the image processing interface is called to divide the image into blocks, and at least one image block is randomly removed to form a missing value region.

[0105] The processed data is encapsulated into a JSON-formatted rendering instruction package, which is the structure to be completed and verified.

[0106] By retaining the first and last characters and masking the middle part, text anonymization and CAPTCHA generation are achieved, which can provide some contextual clues while protecting core information.

[0107] Image restoration and verification are performed by dividing the image into blocks and randomly removing some blocks. The grid size and the number of mask blocks need to be preset based on the image complexity and user experience.

[0108] The masked data, the location information of the missing regions, the unique identifier of the original data fragment, and the ID of this verification session are encapsulated together into a structured JSON object. This JSON object is the verification structure to be completed, which can be sent to the client via WebSocket or HTTP response body.

[0109] Furthermore, the feedback data returned by the receiving client for completing the missing value region includes:

[0110] During the rendering of the rendering instruction package on the client side, an event listener is started to capture the cursor's movement coordinates on the screen at a frequency of milliseconds;

[0111] Record the duration a user spends from confirming a missing value area to submitting the populated content value;

[0112] The movement coordinates are combined to form an operation trajectory coordinate sequence, and the user-input fill values ​​and dwell time are packaged into completion feedback data.

[0113] The operation trajectory is collected at a millisecond-level frequency of no less than 50Hz (i.e., 20 millisecond intervals) to obtain sufficiently fine coordinate and time data in order to calculate differential features such as speed and acceleration, thereby accurately distinguishing between human and machine operation modes.

[0114] Dwell time is used to reflect the user's thinking or reaction time, serving as an auxiliary feature for judging the naturalness of the operation, and can be integrated into operation confidence coefficient analysis or anomaly detection.

[0115] Furthermore, the step of comparing the filled content value with the original value of the target feature data segment to generate a content matching result, and simultaneously calculating the vector offset of the operation trajectory coordinate sequence to generate an operation confidence coefficient, includes:

[0116] Determine whether the filled content value is completely consistent with the original value of the target feature data segment. If they are consistent, mark the content matching result as true; otherwise, mark it as false.

[0117] Calculate the rate of change of distance between adjacent coordinate points in the operation trajectory coordinate sequence to obtain the acceleration sequence;

[0118] Calculate the standard deviation of the acceleration sequence, and use the reciprocal of the standard deviation as the operational confidence coefficient after normalization.

[0119] Let the sequence of coordinates of the operation trajectory be m is the index of the coordinate point.

[0120] First, calculate the pixel distance and instantaneous velocity between adjacent coordinate points.

[0121] For the k-th operation trajectory coordinate point, k ranges from 1 to m-1:

[0122] The instantaneous velocity of the kth operation trajectory coordinate point and instantaneous acceleration The specific method of obtaining it is as follows:

[0123] Instantaneous velocity: .in, This is the time index corresponding to the kth operation trajectory coordinate point.

[0124] Instantaneous acceleration: This leads to the acceleration sequence A.

[0125] Operational confidence coefficient The calculation formula is as follows: .

[0126] in, It is the standard deviation of the acceleration sequence, characterizing the regularity of the operation. Natural, smooth human operations typically involve gradual changes in acceleration, therefore the standard deviation is... The standard deviation is relatively small; while machine scripts or rigid operations can produce drastic and irregular accelerations, thus leading to a larger standard deviation. Larger.

[0127] It is a very small positive number, such as Used to prevent when standard deviation The calculation error is that the denominator is zero when the denominator is 0.

[0128] Operational confidence coefficient It belongs to the interval (0,1). The closer the value is to 1, the smoother and more natural the operation.

[0129] Furthermore, based on the content matching results and the operation trajectory coordinate sequence, the comprehensive verification score for this interaction is calculated, including:

[0130] A base score is preset;

[0131] If the content matching result is true, the product of the operation confidence coefficient and the preset weight is added to the base score to obtain the comprehensive verification score;

[0132] If the content matching result is false, the comprehensive verification score will be set to zero and the exception recording process will be triggered.

[0133] Set base score =60, Operational weighting coefficient =40, verification threshold =70.

[0134] The correctness of the content matching result, i.e., the Boolean value, is used as a veto factor. Only when the content is correct will the operation confidence coefficient, which represents the naturalness of the operation, be included in the score.

[0135] Furthermore, the step of using the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set includes:

[0136] Obtain the current active weight of the selected target feature data fragment in the user interaction state set;

[0137] Calculate the difference between the comprehensive verification score and the preset benchmark score to obtain the weight adjustment step size;

[0138] Add the current activity weight to the weight adjustment step size to obtain the updated activity weight, and write the updated activity weight into the user interaction state set.

[0139] Set learning rate =0.1, baseline score =75, the calculated comprehensive verification score, and then the new weights are derived. (Base score...) Set at the verification threshold Therefore:

[0140] If the overall verification score is higher than the benchmark score, the verification is considered successful and its weight will be increased.

[0141] If the overall verification score is lower than the verification threshold, the verification is considered a failure and its weight will be reduced.

[0142] If the overall verification score is not lower than the verification threshold and not higher than the benchmark score, it is considered a passable verification. However, the weight will still be slightly reduced to reflect negative feedback on suspicious behavior.

[0143] Furthermore, writing the updated activity weight into the user interaction state set includes:

[0144] Check if the updated activity weight is lower than the elimination threshold;

[0145] If the value exceeds the elimination threshold, the data for that dimension will be retained in the database.

[0146] If the value is below the elimination threshold, the feature data fragment of the corresponding dimension will be removed from the user interaction state set, and the dimension will be excluded in the next selection of target feature data fragments.

[0147] Set a low elimination threshold ,For example =0.05.

[0148] Set elimination threshold It also removes dimensions with excessively low weights to prevent the infinite expansion of the user interaction state set, facilitates the automatic cleanup of long-term irrelevant or negative feedback points of interest, and maintains the timeliness and simplicity of the set.

[0149] Furthermore, if the threshold is lower than the elimination threshold, the corresponding feature data fragment will be removed from the user interaction state set, including:

[0150] Check if there are any sub-dimensions in the original interaction log that are related to the removed dimension;

[0151] If it exists, extract the secondary dimension and add it to the user interaction state set according to the initial activity weight to fill the gap after removal.

[0152] When an interest dimension is eliminated, the system queries the original interaction log database for other related dimensions based on a pre-defined association table. This association table is constructed based on product category relationships, item similarity in collaborative filtering of user behavior co-occurrence analysis, or manually defined business rules. For example, the association table is predefined based on business rules, such as: "Purchased a mobile phone" is associated with "Browsed mobile phone cases"; "Inquired about product A" is associated with "Inquired about competitor B".

[0153] If a related secondary object identifier is found, the system uses this secondary object identifier as a new candidate dimension, performs frequency filtering over the past L days and calculates time decay weights to obtain an initial activity weight. If the initial activity weight is greater than or equal to a threshold for introducing a new dimension, it is added to the user interaction state set as a new dimension, thereby achieving dynamic updates and supplementation of the set's content. This enables dynamic metabolism of the set's content and intelligent expansion based on business knowledge, allowing the system to proactively explore users' potential related interests, rather than relying solely on historical behavior.

[0154] like Figure 2 As shown in the figure, this application provides an omnichannel customer relationship management system for implementing the omnichannel customer relationship management method, including: a log parsing module, a status construction module, a feature filtering module, a mask generation module, a feedback receiving module, a comparison analysis module, a scoring decision module, and a weight update module;

[0155] The log parsing module is used to receive raw interaction logs from different channels and parse the raw interaction logs to extract the behavior object identifier and timestamp information.

[0156] The state construction module is used to construct a set of user interaction states containing multiple data dimensions based on the behavior object identifier and timestamp information, and to assign an initial activity weight to each dimension in the set;

[0157] The feature filtering module is used to filter out target feature data fragments from the user interaction state set based on the activity weight when a data access request initiated by the client is detected.

[0158] The mask generation module is used to perform a local masking operation on the target feature data fragment, generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering;

[0159] The feedback receiving module is used to receive completion feedback data for missing value regions returned by the client. The completion feedback data includes the filled content value and the operation trajectory coordinate sequence.

[0160] The comparison and analysis module is used to compare the filled content value with the original value of the target feature data segment to generate a content matching result, and at the same time calculate the vector offset of the operation trajectory coordinate sequence to generate an operation confidence coefficient.

[0161] The scoring decision module is used to calculate the comprehensive verification score of this interaction based on the content matching result and the operation trajectory coordinate sequence. If the comprehensive verification score meets the preset threshold, it will respond to the data access request.

[0162] The weight update module is used to use the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set, so as to change the screening priority for generating the verification structure to be completed next time.

[0163] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0164] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0165] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0167] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0168] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for omnichannel customer relationship management, characterized in that, Includes the following steps: Receive raw interaction logs from different channels, parse the raw interaction logs to extract the behavior object identifier and timestamp information; Based on the behavior object identifier and timestamp information, a set of user interaction states containing multiple data dimensions is constructed, and an initial activity weight is assigned to each dimension in the set; Based on the behavior object identifier and timestamp information, a user interaction state set containing multiple data dimensions is constructed, including: The original interaction logs are arranged in reverse order of timestamps, and the behavior object identifiers within the most recent preset period are extracted as candidate dimensions. Calculate the frequency of each candidate dimension in the original interaction log, and store the candidate dimensions whose frequency exceeds the frequency threshold into the user interaction state set; Based on the time interval between the timestamp and the current time, the initial active weight of each stored dimension is calculated using the time decay function, so that the dimension with the smaller time interval has a larger initial active weight; When a data access request initiated by a client is detected, target feature data fragments are filtered out from the user interaction state set based on activity weight; Perform a local masking operation on the target feature data fragment to generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering; Receive feedback data from the client for filling missing value regions. The feedback data includes the filled content value and the operation trajectory coordinate sequence. The filled content value is compared with the original value of the target feature data segment to generate the content matching result, and the vector offset of the operation trajectory coordinate sequence is calculated at the same time to generate the operation confidence coefficient. Simultaneously, the vector offset of the operation trajectory coordinate sequence is calculated to generate the operation confidence coefficient, including: Determine whether the filled content value is completely consistent with the original value of the target feature data segment. If they are consistent, mark the content matching result as true; otherwise, mark it as false. Calculate the rate of change of distance between adjacent coordinate points in the operation trajectory coordinate sequence to obtain the acceleration sequence; Calculate the standard deviation of the acceleration sequence, and normalize the reciprocal of the standard deviation as the operation confidence coefficient; Based on the content matching results and the operation trajectory coordinate sequence, calculate the comprehensive verification score for this interaction. If the comprehensive verification score meets the preset threshold, respond to the data access request. Using the comprehensive verification score as a feedback factor, the activity weight of the corresponding dimension in the user interaction state set is directly updated, thereby changing the selection priority for generating the next verification structure to be completed.

2. The omnichannel customer relationship management method as described in claim 1, characterized in that, The step of performing a local masking operation on the target feature data fragment to generate a verification structure to be completed, including the region containing missing values, includes: Identify the data type of the target feature data fragment. If the data type is text, retain the first and last characters and replace the remaining characters with placeholders to form a missing value region. If the data type is image index, the image processing interface is called to divide the image into blocks, and at least one image block is randomly removed to form a missing value region. The processed data is encapsulated into a JSON-formatted rendering instruction package, which is the structure to be completed and verified.

3. The omnichannel customer relationship management method as described in claim 2, characterized in that, The feedback data returned by the receiving client for completing the missing value region includes: During the rendering of the rendering instruction package on the client side, an event listener is started to capture the cursor's movement coordinates on the screen at a frequency of milliseconds; Record the duration a user spends from confirming a missing value area to submitting the populated content value; The movement coordinates are combined to form an operation trajectory coordinate sequence, and the user-input fill values ​​and dwell time are packaged into completion feedback data.

4. The omnichannel customer relationship management method as described in claim 1, characterized in that, The comprehensive verification score for this interaction is calculated based on the content matching results and the operation trajectory coordinate sequence, including: A base score is preset; If the content matching result is true, the product of the operation confidence coefficient and the preset weight is added to the base score to obtain the comprehensive verification score; If the content matching result is false, the comprehensive verification score will be set to zero and the exception recording process will be triggered.

5. The omnichannel customer relationship management method as described in claim 1, characterized in that, The step of using the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set includes: Obtain the current active weight of the selected target feature data fragment in the user interaction state set; Calculate the difference between the comprehensive verification score and the preset benchmark score to obtain the weight adjustment step size; Add the current activity weight to the weight adjustment step size to obtain the updated activity weight, and write the updated activity weight into the user interaction state set.

6. The omnichannel customer relationship management method as described in claim 5, characterized in that, The step of writing the updated activity weight into the user interaction state set includes: Check if the updated activity weight is lower than the elimination threshold; If the value exceeds the elimination threshold, the data for that dimension will be retained in the database. If the value is below the elimination threshold, the feature data fragment of the corresponding dimension will be removed from the user interaction state set, and the dimension will be excluded in the next selection of target feature data fragments.

7. The omnichannel customer relationship management method as described in claim 6, characterized in that, If the threshold is lower than the elimination threshold, the corresponding feature data fragment will be removed from the user interaction state set, including: Check if there are any sub-dimensions in the original interaction log that are related to the removed dimension; If it exists, extract the secondary dimension and add it to the user interaction state set according to the initial activity weight to fill the gap after removal.

8. An omnichannel customer relationship management system, used to implement the omnichannel customer relationship management method according to any one of claims 1-7, characterized in that, It includes: log parsing module, state construction module, feature filtering module, mask generation module, feedback receiving module, comparison and analysis module, scoring decision module, and weight update module; The log parsing module is used to receive raw interaction logs from different channels and parse the raw interaction logs to extract the behavior object identifier and timestamp information. The state construction module is used to construct a set of user interaction states containing multiple data dimensions based on the behavior object identifier and timestamp information, and to assign an initial activity weight to each dimension in the set; Based on the behavior object identifier and timestamp information, a user interaction state set containing multiple data dimensions is constructed, including: The original interaction logs are arranged in reverse order of timestamps, and the behavior object identifiers within the most recent preset period are extracted as candidate dimensions. Calculate the frequency of each candidate dimension in the original interaction log, and store the candidate dimensions whose frequency exceeds the frequency threshold into the user interaction state set; Based on the time interval between the timestamp and the current time, the initial active weight of each stored dimension is calculated using the time decay function, so that the dimension with the smaller time interval has a larger initial active weight; The feature filtering module is used to filter out target feature data fragments from the user interaction state set based on the activity weight when a data access request initiated by the client is detected. The mask generation module is used to perform a local masking operation on the target feature data fragment, generate a verification structure to be completed containing the missing value region, and send the verification structure to be completed to the client for rendering; The feedback receiving module is used to receive completion feedback data for missing value regions returned by the client. The completion feedback data includes the filled content value and the operation trajectory coordinate sequence. The comparison and analysis module is used to compare the filled content value with the original value of the target feature data segment to generate a content matching result, and at the same time calculate the vector offset of the operation trajectory coordinate sequence to generate an operation confidence coefficient. Simultaneously, the vector offset of the operation trajectory coordinate sequence is calculated to generate the operation confidence coefficient, including: Determine whether the filled content value is completely consistent with the original value of the target feature data segment. If they are consistent, mark the content matching result as true; otherwise, mark it as false. Calculate the rate of change of distance between adjacent coordinate points in the operation trajectory coordinate sequence to obtain the acceleration sequence; Calculate the standard deviation of the acceleration sequence, and normalize the reciprocal of the standard deviation as the operation confidence coefficient; The scoring decision module is used to calculate the comprehensive verification score of this interaction based on the content matching result and the operation trajectory coordinate sequence. If the comprehensive verification score meets the preset threshold, it will respond to the data access request. The weight update module is used to use the comprehensive verification score as a feedback factor to directly update the activity weight of the corresponding dimension in the user interaction state set, so as to change the screening priority for generating the verification structure to be completed next time.

Citation Information

Patent Citations

  • User portrait construction method and system based on multi-source heterogeneous data fusion

    CN121980062A

  • Conversational ai response generation system and security control method based on privacy sensitivity

    KR102911905B1