Complaint event processing method, apparatus, device, and program product

Abstract

The application discloses a complaint event processing method, device, equipment and program product, and solves the problem that the existing method is difficult to identify, correlate and aggregate processing of mass complaints from different channels. The scheme includes: based on the complaint events of the target business from at least two different channels, an association graph is constructed; each node in the association graph represents a complaint event, and the edge connecting the nodes represents the association strength between the corresponding complaint events. The association graph is subjected to graph representation learning based on a business-aware attention mechanism, and the node features of each node in the association graph are obtained; the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the channel consistency and / or geographical area consistency between nodes are met. Based on the node features of each node in the association graph, the nodes in the association graph are clustered to obtain at least one complaint event cluster. For each complaint event cluster, the information of each complaint event contained is aggregated to generate a corresponding complaint report work order.

Description

Technical Field

[0001] This application relates to the field of complaint handling technology, and in particular to a complaint handling method, apparatus, equipment and program product. Background Technology

[0002] In customer service operations, businesses typically receive customer complaints through multiple channels, including hotlines, emails, online customer service, and mobile applications. When events such as regional network failures or business upgrades occur, the same issue is often repeatedly reported by different customers through different channels, resulting in fragmented and scattered complaint data. This poses a significant challenge to quickly identifying global events and responding efficiently.

[0003] Currently, existing technical solutions for integrating and processing such multi-channel complaints have significant shortcomings. They mainly rely on independent processing by each channel or aggregation methods based on simple rules and shallow text features, making it difficult to effectively identify and link complaint events originating from different channels, with varying expressions, but essentially reflecting the same core issue. This results in the system's inability to automatically detect cross-channel group complaints, leading to redundant investment of processing resources, slow event localization, and severely impacting the overall response timeliness and handling efficiency of enterprises for large-scale service issues. Summary of the Invention

[0004] This application proposes a method, apparatus, device, and program for handling complaint incidents, aiming to solve the problem in the prior art of difficulty in identifying, associating, and aggregating group complaints from different channels. Accordingly, the technical solution of this application is as follows: In a first aspect, embodiments of this application provide a method for handling complaint incidents, including: Based on complaint events from target businesses from at least two different channels, a correlation graph is constructed; wherein each node in the correlation graph represents a complaint event, and the edges connecting the nodes represent the correlation strength between the corresponding complaint events; The association graph is subjected to graph representation learning based on a business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weights between nodes based on whether the nodes satisfy channel consistency and / or geographical region consistency. Based on the node characteristics of each node in the association graph, the nodes in the association graph are clustered to obtain at least one cluster of complaint events; For each of the aforementioned complaint event clusters, information on each complaint event is aggregated to generate a corresponding complaint report work order.

[0005] Secondly, embodiments of this application provide a complaint handling device, comprising: The graph construction module constructs an association graph based on complaint events from target businesses from at least two different channels; wherein each node in the association graph represents a complaint event, and the edges connecting the nodes represent the association strength between the corresponding complaint events; The graph representation learning module performs graph representation learning on the association graph based on the business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the nodes satisfy channel consistency and / or geographical region consistency. The clustering execution module clusters the nodes in the association graph based on the node characteristics of each node in the association graph to obtain at least one cluster of complaint events. The report generation module aggregates information from each complaint event cluster and generates a corresponding complaint report work order.

[0006] Thirdly, embodiments of this application provide an electronic device, including: a processor; and a memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method described in the first aspect.

[0007] Fourthly, embodiments of this application provide a computer program product, the computer program product including a computer-readable storage medium storing a computer program operable to cause a computer to perform the method described in the first aspect.

[0008] This application's embodiments construct a correlation graph based on complaint events from at least two different channels, where nodes represent complaint events and edges represent the correlation strength between events. A business-aware attention mechanism is used to learn the representation of the correlation graph, obtaining feature representations for each node. This mechanism dynamically adjusts attention weights based on whether nodes satisfy channel consistency and / or geographical region consistency to enhance the correlation perception between complaint events with the same channel or adjacent regional attributes. Based on the learned node features, the nodes in the graph are clustered to form complaint event clusters reflecting the same group problem. Then, for each cluster, key information of each complaint event is aggregated to automatically generate a structured complaint report work order. The entire solution, through the aforementioned correlation graph modeling and representation learning, achieves intelligent identification of related complaints scattered across different channels; completes accurate aggregation of related complaint events through node clustering; and outputs standardized and unified work orders through information fusion and structured generation, thus systematically solving the problem of effectively identifying, aggregating, and standardizing cross-channel, multi-source complaints. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the first type of complaint handling method according to an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of a second flowchart of a complaint handling method according to an embodiment of this application.

[0012] Figure 3 This is a schematic diagram of the structure of the complaint handling device according to an embodiment of this application.

[0013] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0015] In customer service operations, enterprises receive complaints through multiple channels such as hotlines, emails, online customer service, and mobile applications. When regional failures or business problems occur, the same issue is often repeatedly reported through different channels, resulting in scattered and fragmented complaint data, making it extremely challenging to quickly identify global events. Existing technologies mostly rely on independent processing by channels or aggregation methods based on simple rules and shallow text features, making it difficult to effectively identify and associate group complaints originating from different channels, with varying expressions but essentially the same content. This leads to the system's inability to automatically detect cross-channel related events, resulting in redundant resource investment and response delays. To address this, this application proposes a complaint event processing method, apparatus, device, and program product, aiming to solve the problem of the difficulty in effectively identifying, associating, and aggregating cross-channel group complaints in existing technologies. The technical solutions provided by the various embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0016] One embodiment of this application provides a method for handling complaint incidents. Figure 1 This is a flowchart illustrating the procedures for handling this complaint, including: S101. Based on complaint events from target businesses from at least two different channels, construct an association graph; where each node in the association graph represents a complaint event, and the edges connecting the nodes represent the association strength between the corresponding complaint events.

[0017] The significance of this step is to transform the originally discrete and independent complaint data from multiple channels such as hotlines, emails, online customer service, and mobile applications into a structured graph model that can explicitly express the complex relationships between events. In real-world business scenarios, the target business could be network quality complaints or service processing failures in communication services. The same business issue might be reported by different users through different channels. For example, a regional base station failure might lead user A to report "signal interruption" via hotline, while user B might submit a complaint about "inability to access the internet" via an app. By constructing a relationship graph, these seemingly independent and differently described complaint events can be mapped to nodes in the graph, and their correlation strength across semantic, temporal, geographical, and channel dimensions can be calculated as edges, thus laying the data foundation for subsequent identification of cross-channel group complaints.

[0018] Specifically, this embodiment constructs the association graph as follows. For the set of complaint events after preprocessing and feature extraction, each complaint event corresponds to a feature vector containing multi-dimensional information such as semantics, channel, time sequence, and geography. The association strength between two complaint events is obtained by weighted calculation of multi-dimensional features. The association strength is determined based on at least one of semantic similarity weighting, time difference decay weighting, geographical similarity weighting, and channel consistency weighting. Among them, the semantic similarity weighting reflects the degree of correlation between the two events in text content; the time difference decay weighting reflects the relationship that the proximity of the events' occurrence times decreases as the time interval increases; the geographical similarity weighting measures the proximity of the two events in geographical location; and the channel consistency weighting adjusts the association weight of complaint events within the same channel and between different channels. For example, two complaint events belonging to the same hotline channel can be assigned a higher channel consistency weighting in the calculation. Based on this, this embodiment further adopts a dynamic dual-threshold determination strategy: an edge is established between the corresponding two nodes only when the calculated association strength exceeds the preset association threshold and the time interval between the two complaint events is less than the time window dynamically adjusted by the system according to the real-time complaint volume. Through the above method, this embodiment can adaptively construct and update a graph structure that reflects the actual correlation patterns between complaint events, thereby supporting the accurate identification and aggregation analysis of subsequent group complaint events.

[0019] Furthermore, if the calculation of association strength relies on semantic similarity weighting, this embodiment can also introduce a domain knowledge graph of the target business domain to enhance the depth and accuracy of semantic understanding before constructing the association graph based on multi-channel complaint events. This domain knowledge graph structures business entities, attributes, and their relationships. For each complaint event, the system first extracts key terms from its description and maps them to the corresponding business entity nodes in the knowledge graph. Then, based on the mapped business entity nodes and their contextual relationships in the graph, the semantic context corresponding to the complaint event is determined. On this basis, the system uses the obtained semantic context to perform semantic enhancement processing on the original complaint information to enrich its semantic representation. For example, for the non-standard expression "internet lag" in user complaints, the knowledge graph can associate it with the standard business entity "low mobile data rate" and its related fault modes, thereby fusing this structured semantic information into the original text vector to generate a semantically enhanced feature representation. This process significantly improves the accuracy of subsequent semantic similarity calculations, thereby enhancing the system's reliability in identifying cross-channel related complaints.

[0020] S102, perform graph representation learning on the association graph based on the business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the nodes satisfy channel consistency and / or geographical region consistency.

[0021] The significance of this step lies in the fact that traditional graph attention mechanisms rely solely on the similarity between node features to calculate attention weights, failing to incorporate actual business rules such as channel relevance and regional proximity. This results in limited accuracy in identifying related events in complex and ever-changing complaint scenarios. Through a business-aware attention mechanism, the system can dynamically adjust the attention weights between nodes using business rules as bias terms, thereby enhancing the model's adaptability to business logic. Channel consistency refers to whether two complaint nodes originate from the same information channel; for example, complaints from the same hotline often show higher consistency in problem description and processing flow. Geographical consistency refers to whether two complaint nodes belong to the same business region; for example, user complaints within the coverage area of ​​the same base station may reflect the same network failure. These business rules help to more accurately capture potential correlations between cross-channel complaints.

[0022] In this implementation, a multi-head attention mechanism is used for graph representation learning. For each node in the association graph, its initial features are a comprehensive feature vector after semantic enhancement and multi-dimensional fusion. In each layer of attention calculation, the attention coefficients of a node to itself and its neighbors are determined by feature similarity and business rules. Business rules are implemented through a business rule function that integrates multiple business dimensions such as channel consistency, regional consistency, and channel complementarity. For example, the channel consistency function outputs a higher value when both endpoints originate from the same channel, and the regional consistency function outputs a higher value when both endpoints belong to the same administrative region. The outputs of these business rules are injected as bias terms into the calculation of the attention coefficients, enabling the model to enhance the association weights when facing complaint events from the same channel, the same region, or with complementary information. Through multi-layered attention propagation, each node ultimately learns a comprehensive feature representation containing local graph structure, node features, and business rules, thus providing more discriminative node features for subsequent clustering analysis.

[0023] S103. Based on the node characteristics of each node in the association graph, cluster the nodes in the association graph to obtain at least one cluster of complaint events.

[0024] The significance of this step is to group the node features, which contain rich correlation information obtained after graph representation learning, thereby automatically aggregating multiple complaint events scattered across the graph but essentially reflecting the same core issue into a single set. Here, node features are low-dimensional vector representations refined through a business-aware attention mechanism, comprehensively reflecting the semantics, channel, temporal sequence, and geographical attributes of complaint events. A complaint event cluster is a set composed of these feature-similar, closely related nodes, with each cluster representing a potential group complaint event. For example, all complaints about "weak 5G signal coverage in a certain area," regardless of whether they originate from a hotline, app, or online customer service, and regardless of whether they are specifically described as "slow internet" or "poor signal," are clustered into the same cluster for easier unified processing and work order generation.

[0025] In its specific implementation, this embodiment employs an adaptive hierarchical clustering method based on density peaks. First, cosine distances between nodes are calculated based on their characteristics in the association graph. Then, based on the statistical distribution of distances between all nodes within the current time window, a cutoff distance value is dynamically determined; this value serves as a distance threshold for determining whether any two nodes are neighbors. For each node, the number of neighboring nodes whose distance to itself is less than the cutoff distance value is counted to determine the node's local density. Before selecting core nodes as cluster centers from all nodes, this embodiment corrects the node's local density based on the credibility of the source channel of the complaint event corresponding to each node. This positively enhances the local density of complaint nodes from channels with high historical accuracy and timely responses, thereby increasing their importance in the clustering process. Next, based on each node's local density and its distance to the nearest node with higher local density, this embodiment selects nodes with high local density and farthest distances from higher-density nodes as core nodes, i.e., the cluster centers of the complaint event clusters. Finally, for each non-core node, it is assigned to the cluster of complaint events corresponding to the nearest core node with the highest local density.

[0026] Furthermore, the clustering in this embodiment is performed dynamically over consecutive time windows. The system calculates the overlap (such as the Jaccard similarity coefficient) between the clustering results of the current time window and the clustering results of the previous adjacent time window. If this overlap is lower than a threshold dynamically determined based on historical fluctuations, the clustering process for the current time window is adjusted, for example by fine-tuning the distance metric or recalibrating the local density calculation, to ensure that the clustering results remain stable over time and to avoid drastic fluctuations in the clustering results for the same persistent event between adjacent windows. Through the above-described clustering process, which includes adaptive parameter determination, channel weight correction, and temporal consistency constraints, this embodiment can achieve accurate, robust, and business-interpretable aggregation of related complaint events.

[0027] S104: For each cluster of complaint events, aggregate the information of each complaint event contained therein and generate the corresponding complaint report work order.

[0028] The significance of this step is to integrate multiple complaint events reflecting the same group problem, formed through intelligent identification and aggregation, into a unified, standardized, and complete set of handling guidelines. This addresses the problems of fragmented information and scattered key information across various channels in the existing point-to-point processing model, resulting in low-quality work orders and difficulties in decision-making. For example, a cluster of complaint events regarding "network latency in XX park" may include dozens of specific complaints from channels such as hotlines, WeChat, and emails, each with different perspectives, levels of detail, and wording. Through this step, the system can automatically integrate this multi-source information to generate a standardized work order that clearly describes the core problem, clearly indicates the scope of impact, and severity level. This directly serves the operations and maintenance team in fault location and repair, greatly improving the efficiency and accuracy of work order generation from complaint reception to work order generation.

[0029] In specific implementation, this embodiment performs the following operations for each cluster of complaint events. First, based on the credibility of the channels to which each complaint event belongs within the cluster, the text information of the corresponding complaint events is weighted and fused to generate a concise work order summary. The credibility of a channel is comprehensively evaluated based on multiple dimensions such as the channel's historical complaint accuracy, response timeliness, and information completeness. For example, channels with high historical work order verification accuracy and well-structured complaint content are assigned higher credibility weights. During the weighted fusion process, complaint content from high-credibility channels is given higher weights, thus ensuring that the core description of the work order summary is more reliable. Subsequently, based on the generated work order summary, the system uses information extraction and classification technology to automatically extract and structure key information, ultimately generating a complaint report work order containing standardized fields. The key fields of this work order include at least the automatically identified problem type (e.g., "network access failure"), a standardized description of the affected area (e.g., ZZ Science and Technology Park, YY District, XX City), and a severity level assessed based on factors such as the number of complaints and the scope of impact. Through this intelligent work order generation process based on channel characteristic awareness, this embodiment realizes the automated conversion from multi-source, heterogeneous original complaints to high-quality, operable standard work orders.

[0030] In summary, the device in this embodiment constructs a correlation graph based on complaint events from target businesses originating from at least two different channels. Nodes in the graph represent complaint events, and edges represent the correlation strength between events. A business-aware attention mechanism is used to learn the representation of the correlation graph, obtaining feature representations for each node. This mechanism dynamically adjusts attention weights based on whether nodes satisfy channel consistency and / or geographical region consistency to enhance the perception of correlations between complaint events with the same channel or adjacent regional attributes. Based on the learned node features, the nodes in the graph are clustered to form clusters of complaint events reflecting the same group problem. Then, for each cluster, key information from each complaint event is aggregated to automatically generate a structured complaint report work order. The entire solution, through the aforementioned correlation graph modeling and representation learning, achieves intelligent identification of related complaints scattered across different channels; accurately aggregates related complaint events through node clustering; and outputs standardized and unified work orders through information fusion and structured generation, thus systematically solving the problem of effectively identifying, aggregating, and standardizing cross-channel, multi-source complaints.

[0031] Figure 2 This embodiment demonstrates the complete processing flow of the method described. First, the system collects and uniformly transforms raw complaint data from channels such as hotlines, emails, online customer service, and mobile applications through a multi-channel complaint data access module, and adaptively adjusts the data aggregation range using a dynamic time window mechanism. Next, the data preprocessing and feature extraction module cleans, standardizes, and performs deep semantic parsing on the data, constructing a comprehensive feature vector containing multi-dimensional information including semantics, channel, time series, and geography. Based on this, the dynamic complaint association graph construction module constructs a dynamic graph structure with complaint events as nodes and association relationships as edges, based on an association strength calculation method that integrates semantic similarity, time decay, geographical similarity, and channel consistency factors. To deepen semantic understanding, a knowledge graph-based semantic enhancement module can be used to enhance and complete the semantic representation of the complaint text using a domain knowledge graph. Subsequently, the business-aware attention module uses a graph attention mechanism that introduces business rules such as channel consistency and regional consistency as bias terms to perform representation learning on the dynamic graph, obtaining refined node features. Based on the aforementioned node characteristics, the complaint event clustering analysis module employs an adaptive hierarchical clustering algorithm that combines dynamic density thresholds, channel credibility weighting, and temporal consistency verification to aggregate related complaints into event clusters. For each event cluster, the standard work order generation module, based on channel characteristics, automatically generates a structured standard complaint report work order by quantifying channel credibility and content complementarity and integrating multi-source information. Finally, the system feedback and model optimization module continuously optimizes algorithm parameters and updates the knowledge base based on work order processing effect data, forming a complete intelligent processing and self-evolution closed loop.

[0032] The overall process of this embodiment will be described in detail below with reference to a specific implementation method.

[0033] This embodiment provides an intelligent processing solution for multi-channel complaint events based on dynamic association graphs. The solution first collects complaint data from multiple channels such as hotlines, emails, online customer service, and mobile applications through a unified data interface, and then uses a standardized data model to uniformly transform the original heterogeneous information, extracting core fields such as timestamps, channel sources, geographical locations, and complaint content text, providing a structured foundation for subsequent analysis.

[0034] To address the issue of sparse and dispersed early signals in the context of widespread complaints, this embodiment introduces a real-time data stream processing mechanism based on dynamic time windows. Dynamic Time Window The size is adaptively adjusted according to the real-time status of the system using the following formula: .in, As the base window size, To adjust the attenuation coefficient of the response sensitivity, This represents the highest number of complaints in history. Based on the current number of complaints, This is the minimum window offset. This mechanism allows the system to automatically expand the window to gather more relevant data when complaint volume increases, and shrink the window to reduce overhead during stable periods.

[0035] In the data preprocessing and feature extraction stage, this embodiment cleans, de-identifies, and spatiotemporally standardizes the raw data, and employs deep semantic parsing techniques (including domain segmentation and entity recognition) to understand the complaint content. Ultimately, each complaint is constructed as a comprehensive feature vector. :

[0036] in, For deep semantic feature vectors, For channel feature vectors, For time periodic feature vectors, This represents the customer and geographic feature vectors, with ⊕ indicating a vector concatenation operation. All numerical features have been normalized.

[0037] To enhance semantic understanding, this embodiment can also introduce a knowledge graph of the target business domain, learn entity embeddings through the TransR algorithm, and enhance the original semantic vector using a knowledge injection mechanism to generate an enhanced semantic vector. :

[0038] in, Indicates entities identified from the complaint text. The set of adjacent entities in a knowledge graph; The embedding vector of the adjacent entity; The relational path weights are dynamically calculated based on graph topology and semantic relevance. An enhancement coefficient is used to control the intensity of knowledge injection. This mechanism improves the ability to resolve technical terms and vague expressions.

[0039] The core element lies in constructing a dynamic complaint association graph and performing representation learning. In this embodiment, each complaint event is modeled as a graph node, with its initial features being a comprehensive feature vector enhanced by knowledge. The strength of the association between nodes The calculation is performed using a weighted formula that integrates semantic similarity, time difference decay, geographical similarity, and channel consistency factors.

[0040] in, This is a channel consistency factor used to adjust the correlation weight between complaints within the same channel and those across channels; Cosine similarity of semantic vectors; Indicates the time difference between two events; Indicates the similarity of geographical features; , , , It is an adjustable hyperparameter. The channel consistency factor is defined as follows: :

[0041] in, This is the cross-channel correlation attenuation coefficient. Only when... The event exceeds a preset threshold and the time interval between the two events is less than that of the dynamic window. At that time, edges are established between nodes to form a dynamic graph structure that reflects the relationship patterns of real-world events.

[0042] To further quantify the business correlations between events, this embodiment employs a business-aware graph attention mechanism for deep representation learning of the correlation graph. This mechanism injects business rules such as channel consistency and region consistency as bias terms into traditional feature-similarity-based attention calculations. For nodes... In its first Layer characteristics Composed of multiple attention outputs spliced ​​together:

[0043] in, This represents a vector concatenation operation; For the total number of attention heads; It is a non-linear activation function; It is the first Each attention head has a unique learnable weight matrix; Represents a node The set of neighboring nodes; and k represents the node For itself and its neighboring nodes In the Normalized attention coefficients on each attention head.

[0044] Attention coefficient k is determined by both feature similarity and business rules:

[0045] in, It is a learnable weight vector; It is a nonlinear activation function; For business rule functions:

[0046] , , These are adjustable business weight parameters. The definitions of each business rule function are as follows: Channel consistency function: ; Regional consistency function: ;in This represents the location code extracted from geographic features.

[0047] Channel complementarity function: ;in and It is a channel and Learnable embedding vectors, It is a learnable weight vector. For activation function, This indicates element-wise multiplication.

[0048] Through multiple layers of attention propagation, each node ultimately learns a comprehensive feature representation that includes graph structure, node features, and business logic. .

[0049] Based on the learned node features, this embodiment employs an adaptive hierarchical clustering algorithm based on density peaks to cluster nodes in the graph, forming clusters of complaint events reflecting the same group-related issues. The algorithm first calculates the cosine similarity matrix between nodes. , of which elements And define the normalized distance .

[0050] The cutoff distance is dynamically determined by analyzing the statistical distribution of real-time distances between nodes. :

[0051] in, The mean distance between all node pairs; Standard deviation; This indicates that the median is taken. This mechanism replaces fixed parameter settings, thus adapting to fluctuations in the complaint data stream.

[0052] Calculate the local density of nodes At the same time, the credibility of the corresponding channels was also taken into account for correction:

[0053] in This is an indicator function that takes the value 1 when the condition is true and 0 otherwise. Channel credibility is evaluated through dimensions such as historical accuracy and timeliness, giving higher weight to complaints from highly credible channels in clustering. Furthermore, by performing clustering over continuous time windows and optimizing the overlap of results from adjacent windows, the stability of the clustering results over time is ensured.

[0054] Finally, for each cluster, a cluster of complaint events is generated. This embodiment implements intelligent generation of standard work orders based on channel characteristic awareness. First, it assesses the reliability of each channel based on its historical accuracy, timeliness, and other information. :

[0055] in, Indicates channel Historical accuracy; Characterizes its response timeliness; Measuring the consistency of its content; , , Adjustable weight parameters are assigned to each dimension. The information complementarity between different channels is also analyzed.

[0056] Then, a condensed work order summary is automatically generated using a weighted fusion algorithm. :

[0057] in, It is a complementary enhancing factor; It is a complaint The final weight; It is the basic weight calculated based on semantic similarity; (.) indicates a summary extraction function (such as TextRank). Finally, key information such as the problem type, affected area, and severity level are automatically extracted from the summary to form a structured, standardized complaint report work order, completing the intelligent conversion from multi-source, heterogeneous complaints to standardized handling instructions.

[0058] To ensure continuous system optimization, this embodiment also establishes a feedback closed-loop mechanism, automatically adjusting key algorithm parameters based on actual performance data from work order processing. Optimization objective function Defined as:

[0059] in, Efficiency score based on work order processing time; A service quality score based on customer satisfaction assessment; This is a penalty term for incorrect clustering in the system. , , These are the weighting coefficients used to balance the various objectives; This represents the set of system parameters. The parameter update method is... ,in This is the learning rate. Simultaneously, the business knowledge base is dynamically updated based on newly confirmed high-confidence knowledge triples: ,in Indicates the confidence level. This is the confidence threshold. This continuously improves the system's accuracy and scenario adaptability.

[0060] It should be noted that, in describing the technical solution of this embodiment, the mathematical symbols and variables involved are clearly defined in the formula or context in which they first appear. It is particularly important to note that some symbols may appear repeatedly in different steps or formulas of the solution, but their specific meaning strictly depends on the current technical context and expression in which they appear. For example, the symbol... In dynamic time window formula The value in the formula represents the sensitivity adjustment coefficient that controls the decay rate of the control window; while in the knowledge enhancement formula... In the middle, this represents the enhancement coefficient that controls the intensity of knowledge injection. Similarly, , , , The symbols for parameters or weights may also have different roles and ranges of values ​​in different formulas. Furthermore, eigenvectors (such as...) , The specific meaning of a symbol also evolves with the processing stage (such as original features, features after knowledge enhancement, and features after graph representation learning). Therefore, when understanding any symbol, the reader should refer to its definition in the directly related current formula or paragraph description.

[0061] In addition, corresponding to Figure 1 The method shown in this embodiment, in another embodiment, also provides a complaint handling device. Wherein, Figure 3 This is a schematic diagram of the complaint handling device 300, including: The graph construction module 310 constructs an association graph based on complaint events from target businesses from at least two different channels; wherein each node in the association graph represents a complaint event, and the edges connecting the nodes represent the association strength between the corresponding complaint events.

[0062] The graph representation learning module 320 performs graph representation learning on the association graph based on a business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weights between nodes based on whether the nodes satisfy channel consistency and / or geographical region consistency.

[0063] The clustering execution module 330 clusters the nodes in the association graph based on the node characteristics of each node in the association graph to obtain at least one cluster of complaint events.

[0064] The report generation module 340 aggregates the information of each complaint event contained in each complaint event cluster and generates a corresponding complaint report work order.

[0065] This embodiment's device constructs a correlation graph based on complaint events from at least two different channels, where nodes represent complaint events and edges represent the correlation strength between events. A business-aware attention mechanism is used to learn the representation of the correlation graph, obtaining feature representations for each node. This mechanism dynamically adjusts attention weights based on whether nodes satisfy channel consistency and / or geographical region consistency to enhance the correlation perception between complaint events with the same channel or adjacent regional attributes. Based on the learned node features, the nodes in the graph are clustered to form complaint event clusters reflecting the same group problem. Then, for each cluster, key information from each complaint event is aggregated to automatically generate a structured complaint report work order. The entire solution, through the aforementioned correlation graph modeling and representation learning, achieves intelligent identification of related complaints scattered across different channels; accurately aggregates related complaint events through node clustering; and outputs standardized and unified work orders through information fusion and structured generation, thus systematically solving the problem of effectively identifying, aggregating, and standardizing cross-channel, multi-source complaints.

[0066] Optionally, the association strength is determined based on at least one of semantic similarity weighted value, time difference attenuation weighted value, geographical similarity weighted value, and channel consistency weighted value.

[0067] Optionally, if the association strength is calculated based on a semantic similarity weighted value, then before constructing the association graph based on complaint events from at least two different channels of the target business, the graph construction module 310 is further configured to: obtain a domain knowledge graph of the target business domain, the domain knowledge graph containing business entities, attributes, and relationships between entities; extract key terms from the information of each complaint event to map them to the corresponding business entity nodes in the domain knowledge graph; determine the semantic context corresponding to each complaint event based on the domain knowledge graph and its mapped business entity nodes; and perform semantic enhancement processing on the information of the corresponding complaint event based on the determined semantic context.

[0068] Optionally, the clustering execution module 330 clusters the nodes in the association graph based on the node characteristics of each node in the association graph, including: calculating the distance between nodes based on the node characteristics of each node in the association graph; determining a cutoff distance value based on the statistical distribution of the distances between all nodes, wherein the cutoff distance value is used as a distance threshold for determining whether any two nodes are neighboring nodes; for each node, counting the number of neighboring nodes whose distance to itself is less than the cutoff distance value to determine the corresponding local density; selecting at least one core node from all nodes as the cluster center of the complaint event cluster based on the local density of each node and its distance to the nearest node with higher local density; and assigning each non-core node to the complaint event cluster corresponding to the nearest core node with higher local density.

[0069] Optionally, before selecting at least one core node from all nodes as the cluster center of the complaint event cluster, the clustering execution module 330 is further configured to: modify the local density of each node based on the credibility of the source channel of the complaint event corresponding to each node, so that the modified local density is positively correlated with the credibility.

[0070] Optionally, the clustering is performed dynamically over consecutive time windows, and the clustering for each time window is completed independently based on the complaint event nodes that arrive within that time window; the clustering execution module 330 is further configured to: calculate the overlap between the clustering result of the current time window and the clustering result of the previous adjacent time window; when the overlap is lower than a preset threshold, adjust the clustering assignment of the current time window, the adjustment including adjusting at least one of the local density calculation of nodes or the distance measurement between nodes, so that the clustering result of the current time window approaches the clustering result of the previous window.

[0071] Optionally, the report generation module 340 aggregates information from each complaint event cluster to generate a corresponding complaint report work order for each complaint event cluster, including: performing the following for each complaint event cluster: weighting and fusing the information of each complaint event based on the credibility of the channel to which each complaint event belongs, and generating a work order summary; extracting and structuring key information based on the work order summary, and generating a complaint report work order containing the problem type, affected area, and severity level; wherein, the credibility of the channel is determined based on at least one of the channel's historical complaint accuracy rate, response timeliness, and information completeness.

[0072] It should be noted that the complaint handling device in this embodiment can be used as... Figure 1 The execution body of the method shown is therefore able to achieve... Figure 1 The steps and functions of the method shown are illustrated.

[0073] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Please refer to it. Figure 4 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0074] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0075] Memory is used to store computer programs. Specifically, a computer program may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides the computer program to the processor.

[0076] Specifically, the processor reads the corresponding computer program from non-volatile memory into memory and then runs it, forming the above-mentioned logical structure. Figure 3 The complaint handling device shown. Correspondingly, the processor executes the program stored in the memory, and specifically performs the following operations: Based on complaint events from target businesses from at least two different channels, a correlation graph is constructed; wherein each node in the correlation graph represents a complaint event, and the edges connecting the nodes represent the correlation strength between the corresponding complaint events.

[0077] The association graph is subjected to graph representation learning based on the business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the nodes satisfy channel consistency and / or geographical region consistency.

[0078] Based on the node characteristics of each node in the association graph, the nodes in the association graph are clustered to obtain at least one cluster of complaint events.

[0079] For each of the aforementioned complaint event clusters, information on each complaint event is aggregated to generate a corresponding complaint report work order.

[0080] The above is as described in this instruction manual. Figure 1The complaint handling method disclosed in the illustrated embodiments can be applied to a processor and implemented by the processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the processor's hardware or by instructions in software form. The processor mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0081] Of course, in addition to software implementation, the electronic device described in this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0082] Furthermore, embodiments of this application also propose a computer program product, including a computer-readable storage medium storing one or more computer programs, the one or more computer programs including instructions.

[0083] When the aforementioned instructions are executed by a portable electronic device that includes multiple applications, they enable the portable electronic device to perform... Figure 1 The steps in the method shown include: Based on complaint events from target businesses from at least two different channels, a correlation graph is constructed; wherein each node in the correlation graph represents a complaint event, and the edges connecting the nodes represent the correlation strength between the corresponding complaint events.

[0084] The association graph is subjected to graph representation learning based on the business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the nodes satisfy channel consistency and / or geographical region consistency.

[0085] Based on the node characteristics of each node in the association graph, the nodes in the association graph are clustered to obtain at least one cluster of complaint events.

[0086] For each of the aforementioned complaint event clusters, information on each complaint event is aggregated to generate a corresponding complaint report work order.

[0087] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0089] The above are merely embodiments of this specification and are not intended to limit the scope of this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification. Furthermore, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of this document.

Claims

1. A method for handling complaint incidents, characterized in that, include: Based on complaint events from target businesses from at least two different channels, a correlation graph is constructed; wherein each node in the correlation graph represents a complaint event, and the edges connecting the nodes represent the correlation strength between the corresponding complaint events; The association graph is subjected to graph representation learning based on a business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weights between nodes based on whether the nodes satisfy channel consistency and / or geographical region consistency. Based on the node characteristics of each node in the association graph, the nodes in the association graph are clustered to obtain at least one cluster of complaint events; For each of the aforementioned complaint event clusters, information on each complaint event is aggregated to generate a corresponding complaint report work order.

2. The method according to claim 1, characterized in that, The association strength is determined based on at least one of semantic similarity weighting, time difference attenuation weighting, geographical similarity weighting, and channel consistency weighting.

3. The method according to claim 2, characterized in that, If the association strength is calculated based on a semantic similarity weighted value, then before constructing the association graph based on complaint events from at least two different channels, the method further includes: Obtain the domain knowledge graph of the target business domain, which includes business entities, attributes, and relationships between entities; For each complaint incident, key terms are extracted and mapped to the corresponding business entity nodes in the domain knowledge graph. Based on the domain knowledge graph, for each complaint event, the semantic context corresponding to the complaint event is determined according to the business entity node it is mapped to; Based on the determined semantic context, semantic enhancement processing is performed on the information of the corresponding complaint event.

4. The method according to claim 2, characterized in that, Based on the node features of each node in the association graph, the nodes in the association graph are clustered, including: Based on the node characteristics of each node in the association graph, the distance between nodes is calculated; Based on the statistical distribution of distances between all nodes, a cutoff distance value is determined, which serves as a distance threshold for determining whether any two nodes are neighboring nodes. For each node, count the number of neighboring nodes whose distance to itself is less than the cutoff distance value to determine the corresponding local density; Based on the local density of each node and its distance to the nearest node with a higher local density, at least one core node is selected from all nodes as the cluster center of the complaint event cluster. For each non-core node, it is assigned to the cluster of complaint events corresponding to the nearest core node with higher local density.

5. The method according to claim 4, characterized in that, Before selecting at least one core node from all nodes as the cluster center of the complaint event cluster, the method further includes: Based on the credibility of the source channel of the complaint event corresponding to each node, the local density of each node is modified so that the modified local density is positively correlated with the credibility.

6. The method according to claim 4, characterized in that, The clustering is performed dynamically over consecutive time windows, and the clustering for each time window is completed independently based on the complaint event nodes that arrive within that time window; the method further includes: Calculate the overlap between the clustering results of the current time window and the clustering results of the previous adjacent time window; When the overlap is lower than a preset threshold, the clustering assignment of the current time window is adjusted. The adjustment includes adjusting at least one of the local density calculation of nodes or the distance metric between nodes, so that the clustering result of the current time window is close to the clustering result of the previous window.

7. The method according to claim 1, characterized in that, For each of the aforementioned complaint event clusters, information on each included complaint event is aggregated to generate a corresponding complaint report work order, including: For each of the aforementioned complaint event clusters, the following steps are performed: based on the credibility of the channel to which each complaint event belongs, the information of each complaint event is weighted and integrated to generate a work order summary; based on the work order summary, key information is extracted and structured to generate a complaint report work order containing the problem type, affected area, and severity level; The credibility of the channel is determined based on at least one of the following: historical complaint accuracy, response timeliness, and information completeness.

8. A complaint handling device, characterized in that, include: The graph construction module constructs an association graph based on complaint events from target businesses from at least two different channels; wherein each node in the association graph represents a complaint event, and the edges connecting the nodes represent the association strength between the corresponding complaint events; The graph representation learning module performs graph representation learning on the association graph based on the business-aware attention mechanism to obtain the node features of each node in the association graph; wherein, the business-aware attention mechanism is configured to adjust the attention weight between nodes according to whether the nodes satisfy channel consistency and / or geographical region consistency. The clustering execution module clusters the nodes in the association graph based on the node characteristics of each node in the association graph to obtain at least one cluster of complaint events. The report generation module aggregates information from each complaint event cluster and generates a corresponding complaint report work order.

9. An electronic device, comprising: processor; And a memory arranged to store computer-executable instructions, characterized in that, when executed, the executable instructions cause the processor to perform the method as described in any one of claims 1 to 7.

10. A computer program product, the computer program product comprising a computer-readable storage medium storing a computer program, characterized in that, The computer program is operable to cause the computer to perform the method as described in any one of claims 1 to 7.

Images