A full-channel data fusion customer portrait graph and intelligent attribution analysis method

Abstract

This invention relates to a method for creating a customer profile map and intelligent attribution analysis based on omnichannel data fusion. The method includes: obtaining a first feature set from multi-source heterogeneous raw data; extracting the time offset baseline and clock drift coefficient of nodes, compensating for discrepancies in conflicting timestamps, and obtaining a unified time-series benchmark through constraint propagation; extracting service commitment subjects and business process elements, using link association to obtain implicit associations between commitment elements and business nodes, and embedding business data under the unified time-series benchmark with the commitment link to obtain a second feature set; obtaining implicit associations and transmission paths between nodes to generate associated data for customer profile nodes; identifying abnormal nodes in complaint feedback and problem representation through cross-dimensional fusion; and tracing the source of the problem through reverse link tracing to obtain intelligent attribution results based on cause deconstruction. This method achieves a customer profile map and intelligent attribution analysis based on omnichannel data fusion.

Description

Technical Field

[0001] This invention belongs to the field of intelligent data analysis and customer service management technology, specifically involving a customer profile mapping and intelligent attribution analysis method based on omnichannel data fusion. Background Technology

[0002] In current multi-channel business operation scenarios, the following areas still need improvement in customer profile construction and attribution analysis: In today's era of deep integration between digital services and multi-channel business operations, customer interaction touchpoints in industries such as finance, e-commerce, and enterprise services are becoming increasingly diverse. The parallel service model of multiple channels, including telephone, online customer service, WeChat, SMS, and order platforms, has become the norm. This is accompanied by the massive generation of multi-source heterogeneous data, such as customer behavior information, service interaction logs, and transaction topology metadata. However, the industry currently faces numerous technical shortcomings and application pain points in processing channel data, constructing customer profiles, and conducting intelligent problem attribution analysis. These are specifically reflected in the following aspects: In customer-service agent interaction data, the analysis of the correlation between service commitment elements and business processes lacks depth and precision. Existing technologies can only achieve surface-level information extraction and cannot effectively identify the implicit relationships between core commitment elements such as the service commitment subject, commitment content, and fulfillment deadline, and business nodes such as telemarketing, payment collection, complaint handling, and identity verification. Furthermore, it is difficult to build a full-link traceability system for commitments from initiation and follow-up to fulfillment / non-fulfillment. This not only prevents the commitment agreements between agents and customers from being effectively linked to business processes and hinders accurate tracking of commitment fulfillment results, but also results in a lack of data support for evaluating business conversion effectiveness and agent service quality based on commitments.

[0003] The current customer profile analysis methods rely on a single dimension, failing to deeply integrate data from multiple dimensions such as customer attributes, historical interaction background, and agent service behavior. This hinders the uncovering of implicit connections and business transmission paths between nodes, resulting in incomplete and delayed identification of anomalous nodes representing issues like complaint feedback, service objections, and process bottlenecks. Furthermore, existing problem attribution analysis methods often focus on single problem symptoms, lacking the ability to trace back to the root cause from multiple dimensions, including time, business chain, and subject behavior. This makes it difficult to accurately pinpoint the triggering source, transmission chain, and core causes of problems, and also hinders the identification of the responsible party. Consequently, problem rectification lacks focus, leading to recurring similar issues and severely impacting customer service experience.

[0004] The lack of a comprehensive, quantitative monitoring and performance evaluation system for service agents means that existing technology cannot link agent service behaviors (such as adherence to key scripts, whether prohibited words were used, and processing time), commitment fulfillment, customer feedback, and business conversion results for analysis. This makes it difficult to accurately identify shortcomings and deficiencies in the agent service process and to form a scientific basis for agent performance evaluation, hindering the improvement of agent service capabilities and the optimization of service processes. Furthermore, insufficient exploration of individual customer needs leads to a lack of targeted service responses, making it difficult to accurately match customer needs and provide personalized services.

[0005] In summary, current technologies suffer from shortcomings in areas such as multi-channel data fusion, unified time-series benchmarks, correlation between commitments and business processes, anomaly identification, and multi-dimensional intelligent attribution. These shortcomings result in incomplete and inaccurate customer profile construction, a lack of scientific rigor and focus in problem attribution analysis, and insufficient data support for agent service monitoring and performance evaluation. Ultimately, this leads to problems such as low customer service quality, low business operational efficiency, and poor customer satisfaction and business conversion rates. Therefore, there is an urgent need to develop a method that can achieve deep multi-channel data fusion, accurately construct time-series customer profiles, and perform intelligent problem attribution analysis to address the pain points of existing technologies and improve the intelligence level of enterprise multi-channel business operations and customer service management. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention provides a method for omnichannel data fusion in customer profiling and intelligent attribution analysis. The objective of this invention can be achieved through the following technical solutions: S1: Obtain multi-source heterogeneous raw data by capturing customer behavior information, service interaction traces, and transaction topology metadata; S2: Extract customer ontology attributes and business data to obtain the first feature set; extract the time offset baseline and clock drift coefficient of the node through the generation node of the timestamp; compensate for the deviation of conflicting timestamps and obtain a unified time series benchmark through constraint propagation; obtain the implicit association between the commitment element and the business node through the link association by the service commitment subject and business link elements; and embed the business data of the unified time series benchmark with the commitment link through time-commitment two-dimensional binding to obtain the second feature set. S3: Integrate the temporal constraints and commitment association rules of the second feature set to generate associated data of customer profile nodes; perform cross-dimensional fusion of the associated data with the first feature set to identify abnormal nodes of complaint feedback and problem representation; combine the temporal conflict resolution results of the second feature set to trace the source of the problem through reverse link tracing and obtain intelligent attribution results of cause deconstruction.

[0007] As a preferred technical solution of the present invention, the acquisition of multi-source heterogeneous raw data is as follows: Acquire multi-source heterogeneous raw data, including fragmented customer behavior trajectories, service interaction conversation flows, and transaction link topology data; The multi-source heterogeneous raw data is subjected to format translation and semantic alignment. By using an edge-side data fidelity processing mechanism, the translated data is subjected to real-time noise reduction and integrity verification to obtain multi-source heterogeneous original data.

[0008] Specifically, the process of extracting customer ontology attributes and business data includes: The multi-source heterogeneous raw data is decomposed into feature dimensions based on the customer feature dimension spectrum. Cross-domain complementarity of corresponding features is achieved through a cross-data source feature association mechanism; Based on the feature semantic normalization algorithm, customer features are logically integrated and ambiguity is resolved; By using feature-structured modeling, integrated customer features are transformed into a first feature set that is interpretable.

[0009] Specifically, the process of extracting the time offset baseline and clock drift coefficient of the node includes: The generation node of the timestamp of the multi-source heterogeneous original data is traced based on data tracing technology; Based on time node clock calibration, the historical operating data of the generated node is analyzed to extract the time offset baseline of the node; By monitoring clock drift, the clock fluctuations of nodes are captured in real time, and the clock drift coefficient of the nodes is obtained.

[0010] Specifically, the process of compensating for the discrepancy in the conflict timestamps includes: Based on the extracted time offset baseline and clock drift coefficient, the deviation correction value of the conflicting timestamp is calculated; Dynamic time warping is used to compensate for the discrepancies in conflicting timestamps of the multi-source heterogeneous raw data. By using a time consistency verification mechanism, the compensated timestamps are verified to obtain preliminary calibrated time series data.

[0011] Specifically, the constraint propagation is as follows: Based on the logical sequence of business events, the time dependencies of business processes are obtained. The time series data is input into the constraint propagation model, and implicit time conflicts across channels and systems are resolved through rule matching and logical reasoning. Time series data are optimized based on the results of constraint propagation to obtain a time series baseline.

[0012] Specifically, the process of extracting the service commitment subject and business process elements includes: Extract commitment elements from heterogeneous raw data from multiple channels and sources; By using element analysis technology in business processes, business node information can be identified; By combining customer background information and basic agent service data, the extracted commitment elements and business node information are filtered to obtain an element set covering customer needs and agent services.

[0013] Specifically, the process of obtaining the implicit association between commitment elements and business nodes includes: The rule base is used to match and logically reason the set of elements to uncover the implicit associations between commitment elements and business nodes; Confidence assessments are performed on implicit associations to obtain reliable association pairs.

[0014] Specifically, the process of embedding business data under a unified time-series benchmark with the committed link includes: Based on the aforementioned association pairs, a commitment link topology is constructed to obtain the flow path of commitment information; By using a time-commitment dual-dimensional binding mechanism, the time-series business dataset is embedded with the commitment link topology to obtain the association mapping between time nodes and the corresponding commitment fulfillment status.

[0015] Specifically, the process of generating associated data for customer profile nodes includes: Based on the temporal constraint rules and commitment association rules of the second feature set, a node association model is constructed; The second feature set is input into the node association model to obtain the implicit associations and business transmission paths between nodes; By structurally encapsulating the associated data, associated data for customer profile nodes is generated.

[0016] Specifically, the process of identifying anomalous nodes in complaint feedback and problem representation includes: Integrate the associated data of the customer profile nodes with the first feature set; Based on the abnormal node identification model, feature thresholds and identification rules are set for complaint feedback and problem representation. The entire feature data network is input into the abnormal node identification model, and abnormal nodes are identified through feature matching and logical judgment. Based on the abnormal node verification mechanism, the identification results are confirmed to obtain the abnormal node set.

[0017] Specifically, the process of obtaining the intelligent attribution results of causal deconstruction includes: Based on the time conflict resolution results of the second feature set, the source tracing path and query rules are obtained; Tracing the source of the problem, its transmission path, and its scope of impact through reverse link tracing; By combining the corresponding data from the first feature set and the second feature set, the causes of the problem and the attribution of responsibility are analyzed; Intelligent attribution results are obtained through structured integration of attribution results.

[0018] The beneficial effects of this invention are as follows: it focuses on heterogeneous data processing, customer feature extraction, and timestamp calibration, and achieves heterogeneous data format translation and semantic alignment through general channel data collection and fidelity processing; it completes cross-domain feature complementarity and semantic normalization based on the customer feature dimension spectrum, making the first feature set highly interpretable; and it extracts the timestamp node offset baseline and drift coefficient through progressive calibration of data tracing, deviation compensation, and constraint propagation, constructing a unified time series benchmark and solving the pain point of traditional time series disorder.

[0019] By integrating customer background and agent service data for element filtering, the element set comprehensively covers customer needs, agent services, and business processes; and by establishing a precise mapping between time nodes and commitment fulfillment status, a second feature set with timeliness, relevance, and traceability is formed, enabling full-link tracking of commitments and providing a structured data foundation for agent service evaluation and performance analysis.

[0020] By integrating time-series and commitment association rules to construct a node association model, we can uncover implicit relationships and business transmission paths between nodes and fully reconstruct the customer's business flow logic. Relying on reverse link tracing technology, we can trace the source of problems, transmission links, and scope of impact by combining dual feature set data, and output structured intelligent attribution results, which greatly enhances the business guidance value of customer profile analysis. Attached Figure Description

[0021] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart illustrating a method for omnichannel data fusion in customer profiling and intelligent attribution analysis according to the present invention. Figure 2 This is a structural block diagram of the association analysis and intelligent attribution in this invention. Detailed Implementation

[0023] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0024] Please see Figure 1-2A method for omnichannel data fusion in customer profiling and intelligent attribution analysis includes: S1: Obtain multi-source heterogeneous raw data by capturing customer behavior information, service interaction traces, and transaction topology metadata; S2: Extract customer ontology attributes and business data to obtain the first feature set; extract the time offset baseline and clock drift coefficient of the node through the generation node of the timestamp; compensate for the deviation of conflicting timestamps and obtain a unified time series benchmark through constraint propagation; obtain the implicit association between the commitment element and the business node through the link association by the service commitment subject and business link elements; and embed the business data of the unified time series benchmark with the commitment link through time-commitment two-dimensional binding to obtain the second feature set. S3: Integrate the temporal constraints and commitment association rules of the second feature set to generate associated data of customer profile nodes; perform cross-dimensional fusion of the associated data with the first feature set to identify abnormal nodes of complaint feedback and problem representation; combine the temporal conflict resolution results of the second feature set to trace the source of the problem through reverse link tracing and obtain intelligent attribution results of cause deconstruction.

[0025] As a preferred technical solution of the present invention, the acquisition of multi-source heterogeneous raw data is as follows: Acquire multi-source heterogeneous raw data, including fragmented customer behavior trajectories, service interaction conversation flows, and transaction link topology data; The multi-source heterogeneous raw data is subjected to format translation and semantic alignment. By using an edge-side data fidelity processing mechanism, the translated data is subjected to real-time noise reduction and integrity verification to obtain multi-source heterogeneous original data.

[0026] In this embodiment, the multi-channel data acquisition gateway completes unified access to data from all channels. This gateway can simultaneously connect to multiple customer interaction ports, including telephone, online customer service, WeChat, SMS, order platforms, and enterprise-owned business systems. It synchronously collects multi-dimensional data such as fragmented behavioral patterns, service interaction conversations, and transaction link topology data generated by customers across various channels, forming a multi-source heterogeneous raw data pool to achieve full-domain capture of customer interaction data. Addressing the format differences between different channels, customized protocol parsing strategies are used to translate the format of structured and unstructured data, converting the raw data into a unified and parsable data format. Through semantic alignment processing of cross-channel text data, semantic normalization of text information is performed, eliminating semantic misunderstandings across channels. A dedicated data fidelity processing module is deployed at the edge to perform real-time noise reduction processing on the data that has undergone format translation and semantic alignment, filtering out invalid noise and interference information. Simultaneously, the integrity of the data is verified, and missing key fields are intelligently completed to obtain multi-source heterogeneous raw data.

[0027] Specifically, the process of extracting customer ontology attributes and business data includes: The multi-source heterogeneous raw data is decomposed into feature dimensions based on the customer feature dimension spectrum. Cross-domain complementarity of corresponding features is achieved through a cross-data source feature association mechanism; Based on the feature semantic normalization algorithm, customer features are logically integrated and ambiguity is resolved; By using feature-structured modeling, integrated customer features are transformed into a first feature set that is interpretable.

[0028] In this embodiment, a comprehensive customer feature dimension spectrum is pre-constructed based on business scenarios and analytical needs. This spectrum fully covers core dimensions such as basic customer attributes, product preferences, business participation behavior, historical interaction records, and complaint / inquiry characteristics. Each core dimension is further subdivided into corresponding sub-feature dimensions, forming a hierarchical and complete feature decomposition system. Based on this customer feature dimension spectrum, high-fidelity multi-source heterogeneous raw data is subjected to targeted feature dimension decomposition and extraction, achieving accurate customer feature extraction. Through a cross-data source feature association mechanism, a mapping relationship for the same customer feature under different scenarios is established, linking and fusing offline transaction features with online interaction features and behavioral features from different platforms, achieving cross-domain complementarity of customer features. A feature semantic normalization algorithm is used to standardize all extracted customer features, logically integrating and resolving semantic ambiguities and redundant associations in feature descriptions, unifying the semantic definition and representation form of each feature dimension. Through feature-based structured modeling technology, discrete customer feature data is systematically integrated and modeled to construct a structured feature set with clear hierarchical relationships and accurate feature representation. This is transformed into a first feature set with strong interpretability. The modeling object is the integrated customer feature set F={f1,f2,...,f...} after customer feature dimension hierarchy decomposition, cross-data source feature association, and feature semantic normalization. m}, where m is the total number of dimensions of customer features, and all features have undergone ambiguity resolution and semantic unification.

[0029] Specifically, the process of extracting the time offset baseline and clock drift coefficient of the node includes: The generation node of the timestamp of the multi-source heterogeneous original data is traced based on data tracing technology; Based on time node clock calibration, the historical operating data of the generated node is analyzed to extract the time offset baseline of the node; By monitoring clock drift, the clock fluctuations of nodes are captured in real time, and the clock drift coefficient of the nodes is obtained.

[0030] In this embodiment, the entire process of generating timestamps in multi-source heterogeneous raw data is traced based on data source tracing to locate the corresponding generation nodes, including various business systems, data acquisition terminals, transmission gateways, and channel service ports. For each located timestamp generation node, timestamp data generated during its long-term historical operation is retrieved, and clock calibration analysis is performed in conjunction with standard time. Through multi-dimensional statistics and pattern mining of historical operation data, the inherent time offset baseline of the generation node is extracted. During the actual operation of each generation node, the time operation status of the node is monitored in real time through a clock drift monitoring mechanism, continuously capturing the clock fluctuation of the node, dynamically recording the drift changes of the node time caused by factors such as runtime, network environment, and device status, and obtaining the clock drift coefficient of each generation node based on the monitoring data mining of the node clock drift patterns.

[0031] Specifically, the process of compensating for the discrepancy in the conflict timestamps includes: Based on the extracted time offset baseline and clock drift coefficient, the deviation correction value of the conflicting timestamp is calculated; Dynamic time warping is used to compensate for the discrepancies in conflicting timestamps of the multi-source heterogeneous raw data. By using a time consistency verification mechanism, the compensated timestamps are verified to obtain preliminary calibrated time series data.

[0032] In this embodiment, the time offset baseline and clock drift coefficient of each timestamp generation node are combined, along with relevant information such as the actual generation time of the timestamp and the operating status of the generation node. A customized deviation calculation model is used to analyze all timestamps with time conflicts in the multi-source heterogeneous raw data, calculating the deviation correction value for each conflicting timestamp. Dynamic time warping technology is employed, using a standard time axis as a unified benchmark. Based on the calculated deviation correction values, point-by-point deviation compensation and time alignment are performed on the conflicting timestamps in the multi-source heterogeneous raw data, accurately correcting the deviation between node time and standard time, and initially resolving explicit time conflicts across channels and systems. A strict time consistency verification mechanism is established, and time verification rules are set. Based on these rules, comprehensive rule matching and logical verification are performed on the timestamps that have completed deviation compensation, resulting in preliminarily calibrated time-series data.

[0033] Specifically, the constraint propagation is as follows: Based on the logical sequence of business events, the time dependencies of business processes are obtained. The time series data is input into the constraint propagation model, and implicit time conflicts across channels and systems are resolved through rule matching and logical reasoning. Time series data are optimized based on the results of constraint propagation to obtain a time series baseline.

[0034] This embodiment comprehensively analyzes the actual workflow logic of all business processes, including telemarketing, payment collection, complaint handling, identity verification, transaction processing, and service follow-up. Combining business operation standards and actual operational scenarios, it clarifies the sequence, connection, and temporal correlation between various business events, forming a clear time dependency relationship between each business process and constructing a comprehensive business time constraint rule base. The initially calibrated time-series data is input into a pre-built constraint propagation model. This model, equipped with a rule matching and logical reasoning engine, performs in-depth logical analysis and verification of the time-series data based on the business time constraint rule base. It uncovers and identifies hidden time conflicts in cross-channel and cross-system data, such as logical contradictions in the recorded times of the same business event in different systems, or time connections between business processes that do not conform to actual operational patterns. Logical reasoning is used to accurately resolve these hidden time conflicts. The time-series data after constraint propagation processing undergoes further optimization. Timeline smoothing technology eliminates minor time fluctuations in the data, and the time-series data is comprehensively regulated and calibrated to form a time-series benchmark for business process data.

[0035] Specifically, the process of extracting the service commitment subject and business process elements includes: Extract commitment elements from heterogeneous raw data from multiple channels and sources; By using element analysis technology in business processes, business node information can be identified; By combining customer background information and basic agent service data, the extracted commitment elements and business node information are filtered to obtain an element set covering customer needs and agent services.

[0036] In this embodiment, service commitment-related elements are extracted from multi-source heterogeneous raw data collected from multiple channels. Using technologies such as named entity recognition and keyword extraction, core commitment elements, including the commitment subject, content, fulfillment timeframe, and achievement conditions, are accurately extracted from customer-agent interaction conversations and service records. Through business process element analysis technology, based on a pre-set business node feature library, feature matching and logical parsing are performed on the raw data to identify business node information corresponding to each business process, including: node type, core operation process, key verification items, associated participating entities, and business connection requirements, achieving a systematic decomposition of business node information. Customer background information, such as historical interaction records, product preferences, complaint and consultation backgrounds, and business processing records, is retrieved. Simultaneously, basic agent service data, such as service type, business permissions, service skills, and historical service records, is retrieved and used as filtering criteria to screen the extracted commitment elements and business node information for relevance and effectiveness. Redundant information irrelevant to customer needs, agent services, and business operations is eliminated, forming a high-quality element set comprehensively covering customer needs and agent services.

[0037] Specifically, the process of obtaining the implicit association between commitment elements and business nodes includes: The rule base is used to match and logically reason the set of elements to uncover the implicit associations between commitment elements and business nodes; Confidence assessments are performed on implicit associations to obtain reliable association pairs.

[0038] In this embodiment, a comprehensive association rule base is pre-constructed by integrating core elements such as enterprise business rules, service operation specifications, and commitment fulfillment logic. This rule base includes the association logic between commitment elements and business nodes, commitment management requirements for each business link, and the corresponding relationships for the fulfillment of different types of commitments. A high-quality set of elements is input into this association rule base. Through semantic matching and logical reasoning algorithms, a deep association analysis is performed on the information of commitment elements and business nodes to uncover hidden implicit relationships between them. This clarifies the core association relationships, such as the business nodes corresponding to different types of commitment elements, the commitment agreements contained in each business node, and the association logic between commitment fulfillment and the advancement of business nodes. Based on an association degree calculation model, a comprehensive quantitative evaluation and confidence calculation are performed on all uncovered implicit association relationships from multiple dimensions, including semantic similarity, business logic matching degree, and data correlation. A reasonable confidence threshold is set, and association relationships with confidence levels higher than the threshold are selected to form highly reliable association pairs.

[0039] Specifically, the process of embedding business data under a unified time-series benchmark with the commitment link includes: constructing a commitment link topology based on the association pair and obtaining the flow path of commitment information; By using a time-commitment dual-dimensional binding mechanism, the time-series business dataset is embedded with the commitment link topology to obtain the association mapping between time nodes and the corresponding commitment fulfillment status.

[0040] In this embodiment, based on highly reliable association pairs, and strictly following the enterprise's business flow logic and the entire commitment fulfillment process, a complete commitment link topology is constructed. This clearly outlines the entire flow path of a commitment from initiation, follow-up, fulfillment to result feedback, obtaining the transmission logic, status change nodes, and connection requirements of commitment information between various business nodes. This achieves visualization and structured presentation of the entire commitment lifecycle flow path. All business data under a unified time-series benchmark are systematically sorted and integrated along a timeline to form a time-series business dataset. Through a time-commitment dual-dimensional binding mechanism, the time-series business dataset is deeply embedded with the commitment link topology. Each time node in the time-series business dataset is precisely bound to the corresponding commitment fulfillment status in the commitment link topology, establishing a three-dimensional association mapping of time node-business link-commitment fulfillment status. This achieves deep integration of business data and the commitment link, allowing the commitment fulfillment process to be accurately traced along the timeline, ultimately forming a second feature set that combines time-series continuity, business relevance, and commitment traceability.

[0041] Specifically, the process of generating associated data for customer profile nodes includes: Based on the temporal constraint rules and commitment association rules of the second feature set, a node association model is constructed; The second feature set is input into the node association model to obtain the implicit associations and business transmission paths between nodes; By structurally encapsulating the associated data, associated data for customer profile nodes is generated.

[0042] In this embodiment, based on the temporal constraint rules and commitment association rules in the second feature set, and integrating the flow logic of the entire business process, customer behavior patterns, and service operation standards, a multi-dimensional node association model is constructed, comprising a node feature layer, an association rule layer, and a transmission path layer. This multi-dimensional node association model is a three-layer coupled architecture. The node feature layer is the basic layer, completing the quantitative representation and dimensional mapping of node features; the association rule layer is the intermediate layer, constructing a node association judgment system based on the feature layer output and business rules; and the transmission path layer is the output layer, mining the business flow logic and path weights between nodes, ultimately achieving a dual output of implicit node associations and transmission paths. The three-layer architecture achieves data interoperability and logical linkage through feature vectors and rule matrices. The attributes, behaviors, temporal sequences, commitments, and other multi-dimensional information of each node in the customer profile are transformed into standardized feature vectors. The formula is: Let any node in the customer profile be vi, and its feature vector is: , Node basic attribute feature sub-vectors (such as node type, business process, channel type, etc., quantized by one-hot encoding / tag encoding) Node interaction behavior feature sub-vectors (such as agent service behavior, customer feedback behavior, operation frequency, etc., after normalization and quantization) Node time-series feature sub-vectors (such as node occurrence time, duration, and time interval with upstream and downstream nodes, normalized based on a unified time-series benchmark) Node commitment associated feature sub-vectors (such as the commitment status, commitment type, and performance progress of the node, which are numerically encoded) Feature vector normalization: , Ensure that the dimensions of each feature are consistent and the value range is [0,1].

[0043] Final node feature layer output: Standardized node feature matrix , where n is the total number of nodes in the customer profile.

[0044] Association Rule Layer: Node Association Logic Determination and Weight Calculation: Based on the node feature matrix, combined with business preset rules and association degree algorithm, an association rule matrix between nodes is constructed to determine whether there is an association between nodes and the strength of the association.

[0045] Sim(v i ,v j )∈[0,1], the larger the value, the more likely the node v is to be in the range of 0,1. i With v j The higher the feature similarity, the stronger the basic association.

[0046] Business rule weight correction: Introduce a business rule weight matrix W=(w ij )n×n,w ij For node v pre-defined based on business logic i With v j The rule association weights (e.g., the rule weight between the telemarketing node and the order placement node is set to 0.9, and the rule weight between the complaint node and the payment collection node is set to 0.1), w ij ∈[0,1].

[0047] Corrected node correlation: , Where α is the feature similarity weight coefficient, α∈[0,1], which can be optimized according to the business scenario (e.g., α=0.7 for data-driven and α=0.3 for rule-driven), Rel(v i ,v j The final correlation degree between nodes is ∈[0,1].

[0048] Association rule matrix construction: The association rule layer outputs the association rule matrix Rel=(Rel(v i ,v j If Rel(v) = n×n, then Rel(v) i ,v j If v ≥ θ (θ is the correlation threshold, preset to 0.5), then v is determined to be... i With v j There is an implicit association; otherwise, there is no association.

[0049] Transmission Path Layer: Node Transmission Path Mining and Path Weight Calculation Core objective: Based on the association rule matrix, mine the business transmission direction and path between nodes, calculate the weight of each path, and reconstruct the node flow logic of the customer profile.

[0050] Node propagation direction determination (based on timing constraints): Let node v i The occurrence time is t i v j The occurrence time is t j Based on a unified time series benchmark, if t j -t i >0 and Rel(vi,v) j If )≥θ, then it is determined that there exists a value from v. i to v j The direction of conduction is denoted as v. i →v j .

[0051] Calculation of conduction path weights: For a single conduction path, P = [v1→v2→⋯→v] m The weight of ] is the product of the correlation between the nodes in the path: , Filter all paths with W(P)≥θP (θP is the path weight threshold, preset to 0.3) as core transmission paths. The transmission path layer finally outputs a directed graph G=(V,E,W) (V is the set of nodes, E is the set of transmission edges, and W is the edge weight / path weight) and a set of core transmission paths.

[0052] Reasoning process: Input the temporal constraint rules and commitment association rules of the second feature set, and extract all nodes in the customer profile; Each node is decomposed into multi-dimensional features, quantized and normalized to obtain a standardized feature vector, and a node feature matrix is ​​constructed (output of the node feature layer). Calculate the feature similarity between nodes, combine it with the business rule weight correction to obtain the final correlation degree, construct the correlation rule matrix, and determine the implicit correlation between nodes (output of the correlation rule layer). Based on a unified temporal reference, the direction of node propagation is determined, the weight of each propagation path is calculated, the core propagation paths are selected, and the node-associated directed graph and the core propagation path set (output of the propagation path layer) are output. Based on the output results, customer profile node association data is generated, realizing the dual mining of implicit node associations and business transmission paths.

[0053] The second feature set is fully input into the constructed node association model. Through feature extraction, rule matching, and logical reasoning, the model deeply mines the implicit relationships between nodes in the customer business profile, and simultaneously outlines the complete business transmission path of the customer between each business node, clearly reconstructing the true flow logic of the customer profile from contacting the business, participating in the business, to completing or terminating the business. The implicit relationships between nodes, the complete business transmission path, and the core data such as the attribute information of each node, the interaction behavior between the customer and the agent, the time sequence information, and the commitment fulfillment status are systematically and structurally encapsulated. They are integrated and processed according to a unified data format and standard to generate standardized and parsable customer profile node association data, which can fully present the full-link characteristics of the customer business profile.

[0054] Specifically, the process of identifying anomalous nodes in complaint feedback and problem representation includes: Integrate the associated data of the customer profile nodes with the first feature set; Based on the abnormal node identification model, feature thresholds and identification rules are set for complaint feedback and problem representation. The entire feature data network is input into the abnormal node identification model, and abnormal nodes are identified through feature matching and logical judgment. Based on the abnormal node verification mechanism, the identification results are confirmed to obtain the abnormal node set.

[0055] In this embodiment, the generated customer profile node association data is deeply integrated with the first feature set across dimensions. This integrates full-dimensional information such as customer ontology attributes, business participation characteristics, node relationships, agent service behavior, and commitment fulfillment status, constructing a comprehensive feature data network covering customers, agents, business, and commitments. Using an abnormal node identification model, combined with the enterprise's customer service scenarios, business operation needs, and quality control standards, corresponding feature thresholds and identification rules are set for various problem representations such as complaint feedback, service objections, process bottlenecks, unfulfilled commitments, and substandard agent service, forming a complete abnormal node identification system. The comprehensive feature data network is input into the abnormal node identification model. Through feature matching and logical judgment, the model performs comprehensive anomaly identification and analysis on all nodes in the customer business profile, accurately locating abnormal nodes corresponding to various problem representations. A strict abnormal node verification mechanism is established. Combining subsequent customer feedback, agent service quality inspection results, business process review status, and other information, the abnormal nodes identified by the model are further verified and confirmed to obtain an abnormal node set.

[0056] Specifically, the process of obtaining the intelligent attribution results of cause deconstruction includes: obtaining the source tracing path and query rules based on the time conflict resolution results of the second feature set; Tracing the source of the problem, its transmission path, and its scope of impact through reverse link tracing; By combining the corresponding data from the first feature set and the second feature set, the causes of the problem and the attribution of responsibility are analyzed; Intelligent attribution results are obtained through structured integration of attribution results.

[0057] In this embodiment, based on the time conflict resolution results of the second feature set and a unified time series benchmark, combined with the complete transmission path of customer business, a standardized reverse tracing path and data query rules are formed, providing clear and standardized logical guidance and operational basis for tracing the source of the problem. Using the confirmed set of abnormal nodes as the starting point for tracing, reverse tracing and in-depth analysis are conducted along the transmission path of customer business through reverse link tracing technology to locate the triggering source of the problem. Simultaneously, the intermediate transmission links of the problem, the impact of each link, and the actual customer groups, related business nodes, and agent service scope affected by the problem are clearly identified, clarifying the entire process of the problem's propagation from its inception to its manifestation. By integrating customer feature data from the first feature set with full-dimensional data such as business, commitment, and time series from the second feature set, in-depth analysis and mining of the causes of the problem are conducted from multiple dimensions, including the customer side, agent side, business process side, and commitment fulfillment side, clarifying the core triggers of the problem. At the same time, combined with the enterprise's organizational structure, business responsibility division, and service control requirements, the responsible parties corresponding to the problem are accurately defined. All the results of the source analysis, including problem characterization, triggering source, core cause, responsibility attribution, scope of impact, and transmission path, are structured and integrated, and presented in a unified format and standard. The output is a smart attribution result with clear cause deconstruction, clear responsibility division, and strong implementability. This result can provide targeted direction and basis for enterprises to rectify problems, optimize customer service training, adjust business processes, and improve service quality.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for customer profiling and intelligent attribution analysis based on omnichannel data fusion, characterized in that, include: S1: Obtain multi-source heterogeneous raw data by capturing customer behavior information, service interaction traces, and transaction topology metadata; S2: Extract customer ontology attributes and business data to obtain the first feature set; extract the time offset baseline and clock drift coefficient of the node through the timestamp generation node; Deviation compensation is performed on conflicting timestamps, and a unified time series benchmark is obtained through constraint propagation; by using the service commitment subject and business link elements, the implicit association between commitment elements and business nodes is obtained through link association; by time-commitment dual-dimensional binding, the business data of the unified time series benchmark is embedded with the commitment link to obtain the second feature set. S3: Integrate the temporal constraints and commitment association rules of the second feature set to generate associated data of customer profile nodes; perform cross-dimensional fusion of the associated data with the first feature set to identify abnormal nodes of complaint feedback and problem representation; combine the temporal conflict resolution results of the second feature set to trace the source of the problem through reverse link tracing and obtain intelligent attribution results of cause deconstruction.

2. The method according to claim 1, characterized in that, The acquisition of multi-source heterogeneous raw data is as follows: Acquire multi-source heterogeneous raw data, including fragmented customer behavior trajectories, service interaction conversation flows, and transaction link topology data; The multi-source heterogeneous raw data is subjected to format translation and semantic alignment. By using an edge-side data fidelity processing mechanism, the translated data is subjected to real-time noise reduction and integrity verification to obtain multi-source heterogeneous original data.

3. The method according to claim 1, characterized in that, The specific process for extracting customer ontology attributes and business data includes: The multi-source heterogeneous raw data is decomposed into feature dimensions based on the customer feature dimension spectrum. Cross-domain complementarity of corresponding features is achieved through a cross-data source feature association mechanism; Based on the feature semantic normalization algorithm, customer features are logically integrated and ambiguity is resolved; By using feature-structured modeling, integrated customer features are transformed into a first feature set that is interpretable.

4. The method according to claim 1, characterized in that, The specific process of extracting the time offset baseline and clock drift coefficient of the node includes: The generation node of the timestamp of the multi-source heterogeneous original data is traced based on data tracing technology; Based on time node clock calibration, the historical operating data of the generated node is analyzed to extract the time offset baseline of the node; By monitoring clock drift, the clock fluctuations of nodes are captured in real time, and the clock drift coefficient of the nodes is obtained.

5. The method according to claim 1, characterized in that, The specific process for compensating for discrepancies in conflict timestamps includes: Based on the extracted time offset baseline and clock drift coefficient, the deviation correction value of the conflicting timestamp is calculated; Dynamic time warping is used to compensate for the discrepancies in conflicting timestamps of the multi-source heterogeneous raw data. By using a time consistency verification mechanism, the compensated timestamps are verified to obtain preliminary calibrated time series data.

6. The method according to claim 1, characterized in that, The constraint propagation is as follows: Based on the logical sequence of business events, the time dependencies of business processes are obtained. The time series data is input into the constraint propagation model, and implicit time conflicts across channels and systems are resolved through rule matching and logical reasoning. Time series data are optimized based on the results of constraint propagation to obtain a time series baseline.

7. The method according to claim 1, characterized in that, The specific process for extracting the service commitment subject and business process elements includes: Extract commitment elements from heterogeneous raw data from multiple channels and sources; By using element analysis technology in business processes, business node information can be identified; By combining customer background information and basic agent service data, the extracted commitment elements and business node information are filtered to obtain an element set covering customer needs and agent services.

8. The method according to claim 1, characterized in that, The specific process for obtaining the implicit association between the commitment elements and business nodes includes: The rule base is used to match and logically reason the set of elements to uncover the implicit associations between commitment elements and business nodes; Confidence assessments are performed on implicit associations to obtain reliable association pairs.

9. The method according to claim 1, characterized in that, The specific process of embedding business data under a unified time-series benchmark with the committed link includes: Based on the aforementioned association pairs, a commitment link topology is constructed to obtain the flow path of commitment information; By using a time-commitment dual-dimensional binding mechanism, the time-series business dataset is embedded with the commitment link topology to obtain the association mapping between time nodes and the corresponding commitment fulfillment status.

10. The method according to claim 1, characterized in that, The specific process of generating associated data for customer profile nodes includes: Based on the temporal constraint rules and commitment association rules of the second feature set, a node association model is constructed; The second feature set is input into the node association model to obtain the implicit associations and business transmission paths between nodes; By structurally encapsulating the associated data, associated data for customer profile nodes is generated.

11. The method according to claim 1, characterized in that, The specific process for identifying anomalous nodes in complaint feedback and problem representation includes: Integrate the associated data of the portrait nodes with the first feature set; Based on the abnormal node identification model, feature thresholds and identification rules are set for complaint feedback and problem representation. The entire feature data network is input into the abnormal node identification model, and abnormal nodes are identified through feature matching and logical judgment. Based on the abnormal node verification mechanism, the identification results are confirmed to obtain the abnormal node set.

12. The method according to claim 1, characterized in that, The specific process for obtaining the intelligent attribution results of causal deconstruction includes: Based on the time conflict resolution results of the second feature set, the source tracing path and query rules are obtained; Tracing the source of the problem, its transmission path, and its scope of impact through reverse link tracing; By combining the corresponding data from the first feature set and the second feature set, the causes of the problem and the attribution of responsibility are analyzed; Intelligent attribution results are obtained through structured integration of attribution results.

Images