Method and system for automatic connection processing of customer complaints and business work orders
By extracting emotional characteristics and complaint portraits in unstructured complaint information, and automatically generate and allocate work orders, the problem of low-efficiency complaint handling in the existing technology is solved, and more efficient and accurate customer complaint handling is achieved, and customers are improved. Satisfaction.
Patent Information
- Application Number
- CN202410851066.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-29
AI Technical Summary
The existing technology cannot effectively and automatically handle unstructured customer complaints, resulting in low-efficiency in handling complaints and inability to accurately grasp customer demands and prevent complaints.
A system was designed, including a complaint acquisition module, a feature extraction module and a work order generation module. By extracting emotional characteristics and complaint portraits in unstructured complaint information, the work orders are automatically generated and assigned to the most appropriate processing. terminal.
It improves the efficiency of complaint handling, accurately reflects customer feelings and needs, reduces misunderstandings and omissions, ensures that complaints are handled professionally, and improves customer satisfaction and service optimization.
Smart Images

Figure CN119048103A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing, and in particular to a method and system for automatically processing customer complaints and business work orders. Background Art
[0002] With the rapid development of social economy, people's material living standards are also constantly improving, which has led to an increasing demand for electricity and an increasing demand for power supply capacity and service quality. When the power supply service enjoyed by power customers does not meet their expectations, it will stimulate various demands from power customers and even lead to complaints from power customers. How to accurately grasp customer demands and effectively prevent customer complaints is the key to improving customer satisfaction with power supply services and maintaining a good image of power companies.
[0003] The customer service system is a channel for communicating with customers directly, and the database management system that stores call data has a huge amount of customer data, which is mainly divided into structured data and unstructured data. At present, the unstructured data of complaint content mainly relies on manual identification and cannot be automatically analyzed, resulting in low efficiency in handling customer complaints.
[0004] Therefore, it is necessary to provide a method and system for automatically linking customer complaints and business work orders to improve the efficiency of customer complaint handling. Summary of the invention
[0005] The present invention provides a system for automatically processing customer complaints and business work orders, including: a complaint acquisition module, used to obtain users' unstructured complaint information; a feature extraction module, used to extract emotional features and complaint portraits of the unstructured complaint information; a work order generation module, used to generate a complaint processing work order corresponding to the unstructured complaint information based on the emotional features and complaint portraits; and a work order allocation module, used to allocate the complaint processing work order to a complaint processing terminal.
[0006] Furthermore, the unstructured complaint information includes voice complaint information and / or text complaint information; the feature extraction module extracts emotional features of the unstructured complaint information, including: obtaining the user's historical complaint information, and determining the number of historical unstructured complaint information included in the historical complaint information; when the number of historical unstructured complaint information included in the historical complaint information is less than or equal to a preset number threshold, extracting the emotional features of the voice complaint information, and / or extracting the emotional features of the text complaint information; when the number of historical unstructured complaint information included in the historical complaint information is greater than a preset number threshold, determining the validity parameter of the unstructured complaint information based on the user's historical complaint information; when the validity parameter of the unstructured complaint information is greater than a preset validity parameter threshold, extracting the emotional features of the voice complaint information, and / or extracting the emotional features of the text complaint information.
[0007] Furthermore, the feature extraction module determines the validity parameters of the unstructured complaint information based on the historical complaint information of the user, including: for each piece of historical unstructured complaint information included in the historical complaint information, determining the authenticity parameter and information completeness parameter of the historical unstructured complaint information; determining the complaint authenticity ratio based on the authenticity parameter of each piece of the historical unstructured complaint information; determining the information completeness parameter mean and the information completeness parameter discrete value based on the information completeness parameter of each piece of the historical unstructured complaint information; determining the validity parameters of the unstructured complaint information based on the complaint authenticity ratio, the information completeness parameter mean and the information completeness parameter discrete value.
[0008] Furthermore, the feature extraction module extracts emotional features of the voice complaint information, including: preprocessing the voice complaint information to obtain preprocessed voice complaint information; framing the preprocessed voice complaint information to obtain multiple frames of audio signals, performing Fourier transform on each frame of audio signals to obtain spectrum information, converting the spectrum information to Mel frequency, and calculating the logarithmic power spectrum after Mel filtering to obtain Mel spectrum features corresponding to the voice complaint information; performing empirical mode decomposition on the preprocessed voice complaint information to obtain speech connotation modal components and speech residuals, extracting sound features of the speech connotation modal components and sound features of the speech residuals; identifying emotional features of the voice complaint information based on the Mel spectrum features corresponding to the voice complaint information, the sound features of the speech connotation modal components and the sound features of the speech residuals through a speech emotion recognition model.
[0009] Furthermore, the feature extraction module extracts emotional features of the text complaint information, including: extracting emotional keywords and emotional symbols of the text complaint information; and identifying the emotional features of the text complaint information based on the extracted emotional keywords and emotional symbols of the text complaint information through a text emotion recognition model.
[0010] Furthermore, the feature extraction module extracts a complaint portrait of the unstructured complaint information, including: extracting key time information from the unstructured complaint information; extracting key location information from the unstructured complaint information; extracting complaint cause information from the unstructured complaint information; and generating a complaint portrait of the unstructured complaint information based on the key time information, key location information and complaint cause information.
[0011] Furthermore, the work order generation module is also used to: establish a knowledge graph, wherein the knowledge graph is used to record the association between sample complaint portraits and work order processing personnel; the work order generation module generates a complaint processing work order corresponding to the unstructured complaint information based on the emotional characteristics and complaint portraits, including: determining the processing priority of the unstructured complaint information based on the emotional characteristics; determining the optimal work order processing personnel based on the complaint portrait and the knowledge graph; and generating a complaint processing work order corresponding to the unstructured complaint information based on the processing priority and the optimal work order processing personnel.
[0012] Furthermore, the work order generation module is also used to: obtain historical complaint handling work orders within the current complaint cycle; based on the complaint portrait corresponding to the historical complaint handling work order and the complaint portrait corresponding to the complaint handling work order, determine whether the historical complaint handling work order is a similar historical complaint handling work order; when it is determined that a similar historical complaint handling work order exists, merge the similar historical complaint handling work order and the complaint handling work order, and adjust the processing priority of the merged complaint handling work order.
[0013] Furthermore, the feature extraction module includes a first scheduling virtual server and multiple feature extraction virtual servers, wherein the first scheduling virtual server is used to obtain the running status information of the multiple feature extraction virtual servers, and based on the running status information of the multiple feature extraction virtual servers, determine the optimal feature extraction virtual server from the multiple feature extraction virtual servers, and the optimal feature extraction virtual server is used to extract the emotional characteristics and complaint portraits of the unstructured complaint information; the work order generation module includes a second scheduling virtual server and multiple work order generation virtual servers, wherein the second scheduling virtual server is used to obtain the running status information of the multiple work order generation virtual servers, and based on the running status information of the multiple work order generation virtual servers, determine the optimal work order generation virtual server from the multiple work order generation virtual servers, and the optimal work order generation virtual server is used to generate a complaint processing work order corresponding to the unstructured complaint information.
[0014] The present invention provides a method for automatically processing customer complaints and business work orders, comprising: obtaining unstructured complaint information; extracting emotional characteristics and complaint portraits of the unstructured complaint information; generating a complaint processing work order corresponding to the unstructured complaint information based on the emotional characteristics and complaint portraits; and allocating the complaint processing work order to a complaint processing terminal.
[0015] Compared with the prior art, the method and system for automatically processing customer complaints and business work orders provided by the present invention have at least the following beneficial effects:
[0016] 1. By automatically extracting key emotional features and complaint portraits from unstructured information, we can quickly understand the core content and demands of the complaint, and quickly generate corresponding processing work orders. Compared with manually reading and understanding the content of the complaint one by one, this undoubtedly greatly improves the processing efficiency. Emotional features and complaint portraits can accurately reflect the dissatisfaction and expectations of the complainant, allowing the complaint handling team to quickly locate the problem and avoid misunderstanding or omission of the complaint content. Work orders based on emotional features and complaint portraits can more accurately reflect the customer's true feelings and needs, so as to better meet customer expectations during the processing process and improve customer satisfaction. By analyzing a large number of complaint emotional features and portraits, we can analyze the problems and deficiencies in services or products, so as to make targeted improvements and optimizations to reduce the recurrence of similar complaints.
[0017] 2. Mel spectrum features are robust and can achieve good recognition results under low signal-to-noise ratio conditions, which helps to improve the accuracy of emotion recognition. The sound features of the speech connotation modal components and the sound features of the residual features can capture subtle changes in the speech signal and further enrich the information that the speech emotion recognition model relies on, thereby improving the accuracy of recognition.
[0018] 3. By analyzing the complaint profile, the nature of the complaint and customer needs can be understood more accurately. Combined with the knowledge graph, the sample that best matches the current complaint profile and the corresponding optimal handler can be found to ensure that the work order is assigned to the most appropriate handler. The knowledge graph can identify the expertise and experience of different work order handlers, so that the work order can be reasonably allocated according to the complaint profile. This helps to ensure that each work order is handled in the most professional way while avoiding waste of resources. By determining the processing priority of unstructured complaint information through emotional characteristics, it can ensure that urgent or important complaints are handled first. At the same time, the optimal work order handler determined based on the complaint profile and knowledge graph can quickly understand and handle the work order, thereby improving the overall processing efficiency. When the work order is assigned to a handler with relevant experience and expertise, the customer can get a more professional and efficient solution. This targeted service helps to improve customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:
[0020] Figure 1 It is a module diagram of a system for automatically processing customer complaints and business work orders according to some embodiments of this specification;
[0021] Figure 2 is a schematic diagram of a process for extracting emotional features of unstructured complaint information according to some embodiments of this specification;
[0022] Figure 3 is a schematic diagram of a knowledge graph according to some embodiments of this specification;
[0023] Figure 4 It is a flowchart of a method for automatically processing customer complaints and business work orders according to some embodiments of this specification. DETAILED DESCRIPTION
[0024] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.
[0025] Figure 1is a module diagram of a system for automatically processing customer complaints and business work orders according to some embodiments of this specification, such as Figure 1 As shown, a system for automatically processing customer complaints and business work orders may include a complaint acquisition module, a feature extraction module, a work order generation module and a work order allocation module.
[0026] The complaint acquisition module can be used to obtain users' unstructured complaint information.
[0027] In some embodiments, the unstructured complaint information includes voice complaint information and / or text complaint information. Specifically, the user's voice complaint information can be obtained through a complaint hotline, and the user's text complaint information can be obtained from a complaint website.
[0028] The feature extraction module can be used to extract emotional features and complaint portraits of unstructured complaint information.
[0029] In some embodiments, the feature extraction module includes a first scheduling virtual server and multiple feature extraction virtual servers, wherein the first scheduling virtual server is used to obtain the operating status information of multiple feature extraction virtual servers, and based on the operating status information of multiple feature extraction virtual servers, determine the optimal feature extraction virtual server from multiple feature extraction virtual servers, and the optimal feature extraction virtual server is used to extract emotional characteristics and complaint portraits of unstructured complaint information.
[0030] Specifically, the running status information of the feature extraction virtual server may include the running status information of the virtual server and the running status information of the physical server where the feature extraction virtual server is located.
[0031] Among them, the virtual server operation status information may include: CPU utilization: indicates the busyness of the CPU on the virtual server; memory usage: includes the amount of memory used and the amount of remaining memory, used to evaluate whether the server has enough memory to handle tasks; disk usage: displays the utilization rate and remaining space of the disk space on the virtual server to determine whether there is enough storage space; network bandwidth: the network bandwidth available to the virtual server; network traffic: real-time or historical network traffic data, used to detect whether there is network congestion or abnormal connection; the status of services running on the virtual server, including started, running, stopped and other states; log information of the virtual server to understand the operation status and error information of the service; response speed: the response speed of the virtual server to requests; stability: check the stability of the virtual server through monitoring tools, such as whether there are frequent downtime or performance degradation.
[0032] The operating status information of the physical server where the feature extraction virtual server is located may include the virtual server operating status information of all feature extraction virtual servers included in the physical server where the feature extraction virtual server is located and the health status of the hardware components (for example, CPU, memory, hard disk, network card, etc.) of the physical server where the feature extraction virtual server is located, including temperature, voltage, vibration, fan speed, disk errors, etc. The operating status information of the physical server where the feature extraction virtual server is located may also include the environmental information of the physical server where the feature extraction virtual server is located (for example, ambient temperature, ambient humidity, ambient smoke concentration, etc.).
[0033] The health score of the feature-extracted virtual server may be determined based on the virtual server running status information through a virtual server status prediction model, wherein the virtual server status prediction model may be an LSTM (Long Short-Term Memory) model.
[0034] For any two feature extraction virtual servers running on the same physical server, the health scores of the two feature extraction virtual servers at multiple historical time points can be obtained, and the health association parameters of the two feature extraction virtual servers are calculated based on the health scores of the two feature extraction virtual servers at multiple historical time points according to the following formula:
[0035]
[0036] Among them, C (i,j) is the health association parameter between the i-th feature extraction virtual server and the j-th feature extraction virtual server, Extract the health score of the virtual server at the n1th historical time point for the i-th feature, The health score of the virtual server at the n1th historical time point is extracted for the jth feature, where N1 is the total number of sampled historical time points.
[0037] The health score of the physical server may be determined by a physical server prediction model based on the health status of the hardware components of the physical server and the environment information in which the physical server is located. The physical server prediction model may be an LSTM (Long Short-Term Memory) model.
[0038] For each feature extraction virtual server, the ranking score of the feature extraction virtual server can be calculated according to the following formula, and the feature extraction virtual server with the highest ranking score is selected as the optimal feature extraction virtual server:
[0039]
[0040] Among them, S iExtract the ranking score of the virtual server for the i-th feature, V (i,current) Extract the health score of the virtual server at the current time point for the i-th feature, V (j,current) is the health score of the jth feature extraction virtual server at the current time point, J is the total number of feature extraction virtual servers deployed on the physical server where the i-th feature extraction virtual server is located, V (i,physical) Extract the health score of the physical server where the virtual server is located at the current time point for the i-th feature, a1, a2 and a3 are preset weights, and a1, a2 and a3 are all greater than 0, a1+a2+a3=1, M 11 is the preset parameter, and M 11 Greater than 0.
[0041] Figure 2 is a flow chart of extracting emotional features of unstructured complaint information according to some embodiments of this specification, such as Figure 2 As shown, in some embodiments, the feature extraction module extracts emotional features of unstructured complaint information, including:
[0042] Obtaining historical complaint information of the user, and determining the amount of historical unstructured complaint information included in the historical complaint information;
[0043] When the amount of historical unstructured complaint information included in the historical complaint information is less than or equal to a preset amount threshold, extracting emotional features of the voice complaint information, and / or extracting emotional features of the text complaint information;
[0044] When the amount of historical unstructured complaint information included in the historical complaint information is greater than a preset amount threshold, determining the validity parameter of the unstructured complaint information based on the historical complaint information of the user;
[0045] When the validity parameter of the unstructured complaint information is greater than a preset validity parameter threshold, the emotional features of the voice complaint information are extracted, and / or the emotional features of the text complaint information are extracted.
[0046] In some embodiments, the feature extraction module determines the validity parameters of the unstructured complaint information based on the user's historical complaint information, including:
[0047] For each piece of historical unstructured complaint information included in the historical complaint information, determine the authenticity parameter and information completeness parameter of the historical unstructured complaint information;
[0048] Determine the authenticity ratio of complaints based on the authenticity parameters of each piece of historical unstructured complaint information;
[0049] Based on the information completeness parameter of each piece of historical unstructured complaint information, determine the mean value of the information completeness parameter and the discrete value of the information completeness parameter;
[0050] Based on the complaint authenticity ratio, the mean value of the information completeness parameter and the discrete value of the information completeness parameter, the validity parameters of unstructured complaint information are determined.
[0051] Specifically, the authenticity parameters of the historical unstructured complaint information may be determined based on the information fed back by the complaint processing terminal, and the complaint authenticity ratio may be the ratio of the historical unstructured complaint information determined to be authentic to the total number of historical unstructured complaint information.
[0052] The information completeness parameter can be used to reflect the completeness of the historical unstructured complaint information, including key time information, key location information and complaint reason information.
[0053] The mean value and discrete value of the information completeness parameter can be calculated according to the following formula:
[0054]
[0055] Among them, M k is the mean value of the information completeness parameter of the kth user, is the information completeness parameter of the n2th historical unstructured complaint information of the kth user, and N2 is the total number of historical unstructured complaint information of the kth user.
[0056]
[0057] Among them, W k is the discrete value of the information integrity parameter of the kth user.
[0058] The validity parameter can be calculated according to the following formula:
[0059]
[0060] Among them, P k is the validity parameter of the kth user, R k is the authenticity ratio of the complaint of the kth user, M0 is the preset mean value of the information completeness parameter, W0 is the preset discrete value of the information completeness parameter, a4 and a5 are preset weights, and a4 and a5 are both greater than 0, a4+a5=1.
[0061] The validity parameter of the unstructured complaint information is the validity parameter of the user who initiated the unstructured complaint information.
[0062] In some embodiments, the feature extraction module extracts emotional features of voice complaint information, including:
[0063] Preprocessing the voice complaint information to obtain preprocessed voice complaint information;
[0064] The pre-processed voice complaint information is divided into frames to obtain multiple frames of audio signals, and each frame of audio signals is subjected to Fourier transform to obtain spectrum information, which is then converted to Mel frequency. The logarithmic power spectrum after Mel filtering is calculated to obtain Mel spectrum features corresponding to the voice complaint information.
[0065] Performing empirical mode decomposition on the preprocessed voice complaint information to obtain voice connotation modal components and voice residuals, and extracting the sound features of the voice connotation modal components and the sound features of the voice residuals;
[0066] The emotional characteristics of the speech complaint information are identified through a speech emotion recognition model based on the Mel spectrum features corresponding to the speech complaint information, the sound features of the speech connotation modal components and the sound features of the speech residual. The speech emotion recognition model can be a Convolutional Neural Network (CNN) model.
[0067] Specifically, for each speech connotation modal component, the speech rate characteristics, fundamental frequency, spectrum envelope and harmonic structure of the speech connotation modal component are determined. The speech connotation modal component can be divided into multiple speech connotation modal component segments according to a preset time window. For each speech connotation modal component segment, the sound intensity mean and sound intensity discrete value of the speech connotation modal component segment are determined. The calculation method of the sound intensity discrete value is similar to the calculation direction of the information completeness parameter discrete value, which will not be repeated here. The sound characteristics of the speech connotation modal component may include the speech rate characteristics, fundamental frequency, spectrum envelope and harmonic structure of the speech connotation modal component, the sound intensity mean and sound intensity discrete value of each speech connotation modal component segment;
[0068] The speech rate characteristics, fundamental frequency, spectral envelope and harmonic structure of the speech residual are determined, and the speech residual can be divided into multiple speech residual segments according to a preset time window. For each speech residual segment, the sound intensity mean and sound intensity discrete value of the speech residual segment can be determined. The sound characteristics of the speech residual can include the speech rate characteristics, fundamental frequency, spectral envelope and harmonic structure of the speech residual, and the sound intensity mean and sound intensity discrete value of the speech residual segment.
[0069] In some embodiments, the feature extraction module extracts emotional features of text complaint information, including:
[0070] Extract emotional keywords and emotional symbols from text complaint information. For example, emotional keywords may include "angry", "angry", etc., and emotional symbols may include "?", "!", etc.;
[0071] The emotional features of the text complaint information are identified through a text emotion recognition model based on the emotional keywords and emotional symbols extracted from the text complaint information, wherein the text emotion recognition model can be a recurrent neural network (RNN) model.
[0072] Specifically, training the text emotion recognition model includes:
[0073] The text emotion recognition model encodes the emotion keywords and emotion symbols of the sample text complaint information to obtain a fusion vector;
[0074] The text emotion recognition model performs classification processing based on the fusion vector to obtain the emotion category prediction result;
[0075] Determine the emotion recognition loss between the emotion category prediction result and the emotion category of the sample text complaint information, and update the parameters of the text emotion recognition model according to the emotion recognition loss;
[0076] In response to the emotion recognition loss converging, the training of the text emotion recognition model is completed.
[0077] In some embodiments, the feature extraction module extracts a complaint profile of the unstructured complaint information, including:
[0078] Extracting key time information from unstructured complaint information, where the key time information can characterize the event at which the complaint event occurred;
[0079] Extract key location information from unstructured complaint information, where the key location information can represent the location where the complaint event occurred;
[0080] Extract complaint reason information from unstructured complaint information, where key location information can characterize the cause of the complaint event (e.g., bad attitude of staff, equipment damage, etc.);
[0081] Generate a complaint profile of unstructured complaint information based on key time information, key location information and complaint reason information.
[0082] Specifically, the feature extraction module can extract key time information, key location information and complaint cause information from unstructured complaint information through an information extraction model, wherein the information extraction model can be a recurrent neural network (RNN) model.
[0083] The work order generation module can be used to generate complaint handling work orders corresponding to unstructured complaint information based on emotional characteristics and complaint portraits.
[0084] In some embodiments, the work order generation module includes a second scheduling virtual server and multiple work order generation virtual servers, wherein the second scheduling virtual server is used to obtain the operating status information of the multiple work order generation virtual servers, and based on the operating status information of the multiple work order generation virtual servers, determine the optimal work order generation virtual server from the multiple work order generation virtual servers, and the optimal work order generation virtual server is used to generate a complaint handling work order corresponding to the unstructured complaint information.
[0085] Specifically, the second scheduling virtual server generates running status information of virtual servers based on multiple work orders, and the method of determining the optimal work order generation virtual server from multiple work order generation virtual servers is similar to the method of the first scheduling virtual server extracting running status information of virtual servers based on multiple features, and determining the optimal feature extraction virtual server from multiple feature extraction virtual servers. For more descriptions, please refer to the above and will not be repeated here.
[0086] In some embodiments, the work order generation module is also used to: establish a knowledge graph, wherein the knowledge graph is used to record the association relationship between the sample complaint portrait and the work order processing personnel.
[0087] Specifically, Figure 3 is a schematic diagram of a knowledge graph according to some embodiments of this specification, such as Figure 3 As shown, sample complaint portrait 1 is associated with work order handler 1, work order handler 2, work order handler 3 and work order handler 4, and sample complaint portrait 2 is associated with work order handler 5, work order handler 6 and work order handler 7.
[0088] In some embodiments, the work order generation module generates a complaint handling work order corresponding to the unstructured complaint information based on the emotional features and the complaint profile, including:
[0089] Based on the emotional characteristics, determine the processing priority of unstructured complaint information, where the higher the degree of anger represented by the emotional characteristics, the higher the processing priority;
[0090] Determine the best work order handler based on complaint profiles and knowledge graphs;
[0091] Generate complaint handling work orders corresponding to unstructured complaint information based on processing priority and optimal work order handlers.
[0092] Specifically, for each sample complaint portrait, the portrait similarity between the sample complaint portrait and the complaint portrait can be calculated, and the sample complaint portrait with a portrait similarity greater than a preset portrait similarity threshold and the largest portrait similarity is taken as the target sample complaint portrait. The portrait similarity between the sample complaint portrait and the complaint portrait can be calculated through a similarity determination model, wherein the similarity determination model can be a convolutional neural network (CNN) model.
[0093] According to the knowledge graph, the work order handlers associated with the target sample complaint profile are taken as candidate work order handlers, and the work order saturation of each candidate work order handler is determined. According to the historical user evaluation information of the candidate work order handlers, the service quality parameters of the candidate work order handlers are determined. Based on the work order saturation and service quality parameters of the candidate work order handlers, the work order matching score of the candidate work order handlers is calculated according to the following formula:
[0094]
[0095] Among them, S e is the work order matching score of the e-th candidate work order handler, M 12 is the preset parameter, and M 12 greater than 0, a6 and a7 are preset weights, and a6+a7=1, D e is the work order saturation of the e-th candidate work order processor, P e is the service quality parameter of the e-th candidate work order handler, P current The service quality parameter of the current demand is determined based on the emotional characteristics. It can be understood that the higher the degree of anger represented by the emotional characteristics, the greater the value of the service quality parameter of the current demand.
[0096] The candidate work order handler with the largest work order matching score is selected as the optimal work order handler.
[0097] In some embodiments, the work order generation module is further configured to:
[0098] Obtain historical complaint handling tickets within the current complaint cycle (e.g., 1 day, 3 days, etc.);
[0099] Based on the complaint profile corresponding to the historical complaint handling work order and the complaint profile corresponding to the complaint handling work order, determine whether the historical complaint handling work order is a similar historical complaint handling work order;
[0100] When it is determined that there are similar historical complaint handling work orders, the similar historical complaint handling work orders and the complaint handling work orders are merged, and the processing priority of the merged complaint handling work orders is adjusted.
[0101] Specifically, the similarity determination model can be used to calculate the image similarity between the complaint image corresponding to the historical complaint handling work order and the complaint image corresponding to the complaint handling work order. If the image similarity is greater than a preset image similarity threshold, the historical complaint handling work order is determined to be a similar historical complaint handling work order. The larger value of the processing priority of the similar historical complaint handling work order and the processing priority of the complaint handling work order is used as the processing priority of the merged complaint handling work order.
[0102] The work order allocation module can be used to allocate complaint handling work orders to complaint handling terminals.
[0103] Specifically, the work order allocation module can be used to allocate the complaint handling work order to the complaint handling terminal of the optimal work order processing personnel.
[0104] Figure 4 is a flowchart of a method for automatically processing customer complaints and business work orders according to some embodiments of this specification, such as Figure 4 As shown, a method for automatically processing customer complaints and business work orders may include the following steps.
[0105] Obtain unstructured complaint information;
[0106] Extract emotional features and complaint profiles from unstructured complaint information;
[0107] Generate complaint handling work orders corresponding to unstructured complaint information based on emotional characteristics and complaint portraits;
[0108] Assign the complaint handling work order to the complaint handling terminal.
[0109] A method for automatically connecting and processing customer complaints and business work orders can be executed by a system for automatically connecting and processing customer complaints and business work orders. For more descriptions of a method for automatically connecting and processing customer complaints and business work orders, please refer to the relevant description of a system for automatically connecting and processing customer complaints and business work orders, which will not be repeated here.
[0110] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.
Claims
1. A system for automatically processing customer complaints and business work orders, characterized in that: include: Complaint acquisition module, used to obtain users' unstructured complaint information; A feature extraction module, used to extract emotional features and complaint portraits of the unstructured complaint information; A work order generation module, used to generate a complaint handling work order corresponding to the unstructured complaint information based on the emotional characteristics and complaint portrait; The work order allocation module is used to allocate the complaint handling work order to the complaint handling terminal.
2. A system for automatically processing customer complaints and business work orders according to claim 1, characterized in that: The unstructured complaint information includes voice complaint information and / or text complaint information; The feature extraction module extracts the emotional features of the unstructured complaint information, including: Acquire historical complaint information of the user, and determine the amount of historical unstructured complaint information included in the historical complaint information; When the amount of historical unstructured complaint information included in the historical complaint information is less than or equal to a preset amount threshold, extracting emotional features of the voice complaint information, and / or extracting emotional features of the text complaint information; When the amount of historical unstructured complaint information included in the historical complaint information is greater than a preset amount threshold, determining a validity parameter of the unstructured complaint information based on the historical complaint information of the user; When the validity parameter of the unstructured complaint information is greater than a preset validity parameter threshold, the emotional features of the voice complaint information are extracted, and / or the emotional features of the text complaint information are extracted.
3. A system for automatically processing customer complaints and business work orders according to claim 2, characterized in that: The feature extraction module determines the validity parameter of the unstructured complaint information based on the historical complaint information of the user, including: For each piece of historical unstructured complaint information included in the historical complaint information, determining an authenticity parameter and an information completeness parameter of the historical unstructured complaint information; Determining a complaint authenticity ratio based on the authenticity parameters of each piece of the historical unstructured complaint information; Based on the information completeness parameter of each piece of the historical unstructured complaint information, determine the mean value of the information completeness parameter and the discrete value of the information completeness parameter; Based on the complaint authenticity ratio, the information completeness parameter mean and the information completeness parameter discrete value, the validity parameter of the unstructured complaint information is determined.
4. A system for automatically processing customer complaints and business work orders according to claim 2, characterized in that: The feature extraction module extracts the emotional features of the voice complaint information, including: Preprocessing the voice complaint information to obtain preprocessed voice complaint information; The preprocessed voice complaint information is divided into frames to obtain multiple frames of audio signals, each frame of audio signals is subjected to Fourier transform to obtain spectrum information, the spectrum information is converted to Mel frequency, and the logarithmic power spectrum after Mel filtering is calculated to obtain Mel spectrum features corresponding to the voice complaint information; Performing empirical mode decomposition on the preprocessed voice complaint information to obtain voice connotation modal components and voice residuals, and extracting the sound features of the voice connotation modal components and the sound features of the voice residuals; The emotional characteristics of the voice complaint information are identified through a voice emotion recognition model based on the Mel spectrum characteristics corresponding to the voice complaint information, the sound characteristics of the voice connotation modal component and the sound characteristics of the voice residual.
5. A system for automatically processing customer complaints and business work orders according to claim 2, characterized in that: The feature extraction module extracts the emotional features of the text complaint information, including: Extract emotional keywords and emotional symbols from the text complaint information; The emotional features of the text complaint information are identified through a text emotion recognition model based on the emotional keywords and emotional symbols extracted from the text complaint information.
6. A system for automatically processing customer complaints and business work orders according to any one of claims 1 to 5, characterized in that: The feature extraction module extracts the complaint profile of the unstructured complaint information, including: extracting key time information from the unstructured complaint information; extracting key location information from the unstructured complaint information; extracting complaint reason information from the unstructured complaint information; Based on the key time information, key location information and complaint reason information, a complaint profile of the unstructured complaint information is generated.
7. A system for automatically processing customer complaints and business work orders according to any one of claims 1 to 5, characterized in that: The work order generation module is also used for: Establishing a knowledge graph, wherein the knowledge graph is used to record the association between sample complaint profiles and work order processing personnel; The work order generation module generates a complaint handling work order corresponding to the unstructured complaint information based on the emotional characteristics and the complaint portrait, including: Based on the emotional characteristics, determining a processing priority of the unstructured complaint information; Determine the optimal work order processing personnel based on the complaint profile and the knowledge graph; Based on the processing priority and the optimal work order processing personnel, a complaint processing work order corresponding to the unstructured complaint information is generated.
8. A system for automatically processing customer complaints and business work orders according to claim 7, characterized in that: The work order generation module is also used for: Obtain historical complaint handling work orders within the current complaint cycle; Based on the complaint portrait corresponding to the historical complaint handling work order and the complaint portrait corresponding to the complaint handling work order, determining whether the historical complaint handling work order is a similar historical complaint handling work order; When it is determined that there is a similar historical complaint handling work order, the similar historical complaint handling work order and the complaint handling work order are merged, and the processing priority of the merged complaint handling work order is adjusted.
9. A system for automatically processing customer complaints and business work orders according to any one of claims 1 to 5, characterized in that: The feature extraction module includes a first scheduling virtual server and multiple feature extraction virtual servers, wherein the first scheduling virtual server is used to obtain the running status information of the multiple feature extraction virtual servers, and based on the running status information of the multiple feature extraction virtual servers, determine the optimal feature extraction virtual server from the multiple feature extraction virtual servers, and the optimal feature extraction virtual server is used to extract the emotional features and complaint portraits of the unstructured complaint information; The work order generation module includes a second scheduling virtual server and multiple work order generation virtual servers, wherein the second scheduling virtual server is used to obtain the operating status information of the multiple work order generation virtual servers, and based on the operating status information of the multiple work order generation virtual servers, determine the optimal work order generation virtual server from the multiple work order generation virtual servers, and the optimal work order generation virtual server is used to generate a complaint handling work order corresponding to the unstructured complaint information.
10. A method for automatically processing customer complaints and business work orders, characterized in that: include: Obtain unstructured complaint information; Extracting emotional features and complaint portraits of the unstructured complaint information; Based on the emotional characteristics and complaint portrait, a complaint handling work order corresponding to the unstructured complaint information is generated; The complaint handling work order is distributed to the complaint handling terminal.
Citation Information
Cited By
Work order generation method and device based on natural language processing, equipment and medium
CN119761334A
Alarm-receiving complaint handling platform and method
CN120338274A
Reason analysis method, system and related device
CN121561138A
A cause analysis method, system and related devices
CN121561138B
