Method for solving complaint work order governance by construction problem monitoring robot

Abstract

The invention discloses a method for solving complaint work order treatment by a construction problem monitoring robot, and the method comprises the steps: obtaining multi-channel customer complaint data, and classifying complaint work order problems; key information after complaint work order classification is extracted, and a structured work order task is generated; performing clustering analysis on the generated structured work order task to obtain a standard work order; constructing a proposal board and a common problem discovery model, and predicting a basic problem of the complaint work order; the predicted fundamental problem is diagnosed by using the affair map, and the root cause of the problem is output; and enterprise intelligence brain is built, customer service GPT is output, and solution suggestions are quickly provided for root causes of complaints. According to the invention, intelligent processing of customer complaints is realized, the problem solving efficiency is improved, the customer complaint rate is effectively reduced, and powerful support is provided for enterprises to improve customer satisfaction and service quality.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and total quality management, and in particular to a method for building a problem monitoring robot to solve the management of complaint work orders. Background Art

[0002] Complaint ticket problem management is an important part of corporate customer service. Accurately managing the number of complaint tickets is of great significance for companies to reasonably allocate resources and improve customer satisfaction.

[0003] At present, in most enterprises, all links in the complaint ticket processing process are completed manually, which is inefficient and prone to errors. At the same time, the connection between different links is not tight, and information transmission is not timely, which affects the rapid resolution of complaint issues. Therefore, in traditional problem management, there is a lack of full-process robot management, a lack of comprehensive quality management, and a lack of the combination of robot monitoring management and comprehensive quality management. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the background technology and to propose a method for building a problem monitoring robot to solve the complaint work order management.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for building a problem monitoring robot to solve the complaint work order management includes the following steps:

[0007] S1, obtains customer complaint data from multiple channels, uses natural language processing for semantic understanding and intent recognition, and classifies complaint ticket issues;

[0008] S2, extracts key information after complaint work order classification and generates structured work order tasks;

[0009] S3, cluster analysis is performed on the generated structured work order tasks to determine the resubmission and cross-level complaint prediction work orders and obtain standardized work orders;

[0010] S4, build a proposal dashboard and common problem discovery model, output the standardized work orders through the common problem discovery model, and then use Toyota's one-page minimalist thinking method to compare and display them, and predict the root problems of the complaint work orders;

[0011] S5, using the event map to diagnose the root problem of the prediction and output the root cause of the problem;

[0012] S6, build an enterprise intelligent brain, output customer service GPT, and quickly provide solutions to the root causes of complaints;

[0013] S7, after the problem is solved, the self-problem solving follow-up system is used to track new data from customer feedback in a timely manner, and continuous iterative optimization is carried out based on the new data from customer feedback to complete the closed loop.

[0014] Preferably, the S1 specifically includes:

[0015] Obtain customer complaint data from multiple channels, and record basic information such as relevant complaint channels, complaint time, complaint content, and customer information for subsequent processing;

[0016] Based on the complaint content obtained, natural language processing is used to perform semantic analysis, extract key information from the complaint content, and obtain data on the complaint object, problem type, and sentiment tendency;

[0017] Through intention identification, we can judge the customer's complaint intention and determine whether it is for consultation, suggestion or appeal, so as to take appropriate measures;

[0018] If the intent recognition result is a complaint, the complaint content is classified according to the result of semantic understanding to determine the specific complaint category, including product quality issues, service attitude issues or charging disputes.

[0019] Preferably, the S2 specifically includes:

[0020] According to the complaint time and complaint channel, the initial information of the work order is obtained to determine whether the complaint is made during working hours and obtain the priority of complaint handling;

[0021] By using the information of the complaining customer, we can judge the customer's historical complaint record and credit rating, and determine the importance of the customer;

[0022] According to the complaint content and the object of the complaint, obtain the keywords and topics of the problem and determine the problem classification;

[0023] Determine the assigned department based on the urgency and problem classification;

[0024] Obtain past processing records and judge processing progress through current processing status;

[0025] Based on the update of processing results, determine whether the problem has been resolved, obtain customer feedback, and determine customer satisfaction.

[0026] Preferably, the S3 specifically includes:

[0027] Conduct preliminary clustering analysis on structured work order tasks, use clustering algorithms to group similar work orders, determine the subject categories of the work orders, and obtain preliminary clustering results;

[0028] By analyzing the submission time and person-in-charge information, the efficiency of the work order processing is determined. If the processing cycle does not match the priority, it is identified as a potential cross-level complaint work order;

[0029] Combine the contact information and attachment information of the work order to obtain the complete communication record of the work order, analyze the communication content through information mining, determine whether there is a resubmission, and obtain the judgment result of the resubmission work order;

[0030] By analyzing the relationship between the submitter and the problem type, if a customer submits the same type of work order multiple times in a short period of time, he or she will be identified as a potential repeat customer and a list of repeat customers will be obtained.

[0031] Preferably, the S4 specifically includes:

[0032] S401, first construct a common problem discovery model, use historical data for training, and adjust hyperparameters to obtain an algorithm model for discovering common problems with an optimal loss function;

[0033] S402: For all the work orders received from the front end every day, a cluster analysis is performed on the work order acceptance content using a trained algorithm model for discovering common problems to identify potential common problems.

[0034] S403, using the Euclidean distance measurement method to compare the potential common problems found by the clustering model with the historical potential problems, if it is a new problem, it will trigger Toyota One Page to raise the problem, if it is an existing problem, it will not trigger Toyota One Page;

[0035] S404, based on Toyota's one-page management process, provides feedback on solutions to the issues raised and continuously tracks the status of problem solving.

[0036] Preferably, the S401 specifically includes:

[0037] The k-means clustering algorithm model is used to build a basic algorithm model for discovering common problems, and the training data set is selected as 6 months of work order data;

[0038] One month of historical work orders were extracted from the training data set, and the work order acceptance content was clustered using the k-means clustering algorithm model. The hyperparameter k was adjusted from 1 to 20, and the change in the loss value of the model was observed. The clustering results were compared with the quality inspection results to observe the clustering effect.

[0039] Repeat the above steps. When the hyperparameter k reaches the preset value, an algorithm model for discovering common problems is obtained with a stable clustering loss and good clustering effect.

[0040] Preferably, the S5 specifically includes:

[0041] S501, based on the historical work orders of the operator customer service for 6 months, analyze and extract the facts and causal relationship information from the acceptance and processing of the work orders and whether there are repeated complaints or complaints beyond the level, and build the facts and causal relationship map of the operator customer service;

[0042] S502, for the work orders received from the front end every day, the customer star rating, product type, service category, and acceptance content of the work orders are also extracted;

[0043] S503, search the causal graph constructed in S5011 according to the event obtained in S502, sort the causal relationship probability of the search results, take out the tail event with the highest probability and send it to the application system, namely M-Order Eagle Eye and M-Work Order Hotspot Eagle Eye.

[0044] Preferably, the S501 specifically includes:

[0045] Use rules to match causal mention text in causal events from the acceptance content and processing content of historical work orders;

[0046] Based on the BERT-based causal relationship extraction model, feature extraction is performed from the customer star rating, product type, service category, resubmission mark, and skip level mark attributes of historical work orders, and BiLSTM+CRF is used to output causal identification labels;

[0047] By calculating the strength of causal relationships and the probability of transition, the causal relationships of various events are adjusted, and finally the operator's cause-and-effect map is constructed.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] The present invention firstly realizes the visual management of work order governance by setting up development re-investment, cross-level risk monitoring dashboard, similar group complaint work order dashboard, proposal dashboard, M-order Eagle Eye and M-work order hot spot Eagle Eye; customer service GPT and other dashboards; in addition, the process of each link of complaint work order processing is connected in series from problem triggering, problem description, problem diagnosis, problem solution and problem solving self-follow-up, which greatly reduces manpower input, realizes intelligent processing of customer complaints, improves problem solving efficiency, effectively reduces customer complaint rate, and provides strong support for enterprises to improve customer satisfaction and service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for building a problem monitoring robot to solve complaint work order management proposed by the present invention;

[0051] Figure 2 This is a flow chart of S4 in a method for building a problem monitoring robot to solve complaint work order management proposed by the present invention;

[0052] Figure 3 This is a flowchart of S5 in a method for building a problem monitoring robot to solve complaint work order management proposed by the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] In the description of the present invention, it is necessary to understand that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0055] Reference Figure 1-3 ,A method for building a problem monitoring robot to solve the complaint work order management includes the following steps:

[0056] S1, obtains customer complaint data from multiple channels, uses natural language processing for semantic understanding and intent recognition, and classifies complaint ticket issues;

[0057] Specifically include:

[0058] Obtain customer complaint data from multiple channels, and record basic information such as relevant complaint channels, complaint time, complaint content, and customer information for subsequent processing; use natural language processing to perform semantic analysis based on the acquired complaint content, extract key information from the complaint content, and obtain the complaint object, problem type, and sentiment tendency data; use intent recognition to determine the customer's complaint intention and determine whether it is for consultation, suggestion, or appeal, so as to take appropriate measures; if the intent recognition result is an appeal, classify the complaint content according to the result of semantic understanding, and determine the specific complaint classification, including product quality issues, service attitude issues, or fee disputes.

[0059] When obtaining multi-channel customer complaint data, you can collect data from different platforms (such as social media, mail systems, online chat tools, etc.) through integrated APIs and store them in a centralized database.

[0060] S2, extracts key information after complaint work order classification and generates structured work order tasks;

[0061] Specifically include:

[0062] According to the complaint time and complaint channel, the initial information of the work order is obtained to determine whether the complaint is made during working hours and obtain the priority of complaint handling; through the information of the complaining customer, the customer's historical complaint record and credit rating are determined to determine the importance of the customer; according to the complaint content and the object of the complaint, the keywords and topics of the problem are obtained to determine the problem classification; according to the urgency and problem classification, the assigned department is determined; through the current processing status, past processing records are obtained to determine the processing progress; according to the update of the processing results, it is determined whether the problem has been resolved, the customer's feedback is obtained, and the customer satisfaction is determined, thereby realizing the establishment of structured work order tasks.

[0063] S3, cluster analysis is performed on the generated structured work order tasks to determine the resubmission and cross-level complaint prediction work orders and obtain standardized work orders;

[0064] Specifically include:

[0065] Conduct a preliminary clustering analysis on structured work order tasks, use a clustering algorithm to group similar work orders, determine the subject category of the work orders, and obtain preliminary clustering results; determine the processing efficiency of the work order by analyzing the submission time and the person in charge information. If the processing cycle does not match the priority, it is identified as a potential cross-level complaint work order; obtain the complete communication record of the work order by combining the contact information and attachment information of the work order, and analyze the communication content through information mining to determine whether there is a resubmission and obtain the judgment result of the resubmitted work order; analyze the relationship between the submitter and the problem type, if a customer submits the same type of work order multiple times in a short period of time, it is identified as a potential resubmitted customer and a list of resubmitted customers is obtained.

[0066] S4, build a proposal dashboard and common problem discovery model, output the standardized work orders through the common problem discovery model, and then use Toyota's one-page minimalist thinking method to compare and display them, and predict the root problems of the complaint work orders;

[0067] Specifically include:

[0068] S401, first construct a common problem discovery model, use historical data for training, and adjust hyperparameters to obtain an algorithm model for discovering common problems with an optimal loss function;

[0069] S402: For all the work orders received from the front end every day, a cluster analysis is performed on the work order acceptance content using a trained algorithm model for discovering common problems to identify potential common problems.

[0070] S403, using the Euclidean distance measurement method to compare the potential common problems found by the clustering model with the historical potential problems, if it is a new problem, it will trigger Toyota One Page to raise the problem, if it is an existing problem, it will not trigger Toyota One Page;

[0071] S404, based on Toyota's one-page management process, provides feedback on solutions to the issues raised and continuously tracks the status of problem solving.

[0072] S401 specifically includes:

[0073] The k-means clustering algorithm model is used to construct a basic algorithm model for discovering common problems. The training data set is selected as 6 months of work order data. The work order data here is the data after preprocessing measures such as cleaning, standardization and feature selection; 1 month of historical work orders are extracted from the training data set, and the work order acceptance content is clustered using the k-means clustering algorithm model. The hyperparameter k is adjusted from 1 to 20 from small to large, the change in the model's loss value is observed, and the clustering results are compared with the quality inspection results to observe the clustering effect; repeat the above steps. When the hyperparameter k reaches the preset value, an algorithm model for discovering common problems with stable clustering loss and good clustering effect is obtained.

[0074] In cluster analysis, the K-means clustering algorithm is used to cluster the data. The coefficient k is calculated. According to the inflection point method (elbow method), k is taken from 1 to 20 from small to large, and the clustering loss corresponding to each value of k from 1 to 20 is calculated (the clustering loss can be obtained by calculating the distance between each data sample and the centroid of the cluster to which it belongs, and adding all distances). As the k value increases, the clustering loss will decrease. Pay attention to the slope of this change. When K>=10, the slope suddenly changes from large to small, and then changes slowly. It is considered that the k=10 value is a suitable k value. The k value is (10). In terms of data selection, for example, the data from (2023-01-01) to (2023-06-30) (a total of 18,780 work orders in 6 months are selected as the data set)

[0075] S5, using the event map to diagnose the root problem of the prediction and output the root cause of the problem;

[0076] Specifically include:

[0077] S501, based on the historical work orders of the operator customer service for 6 months, analyze and extract the facts and causal relationship information from the acceptance and processing of the work orders and whether there are repeated complaints or complaints beyond the level, and build the facts and causal relationship map of the operator customer service;

[0078] S502, for the work orders received from the front end every day, the customer star rating, product type, service category, and acceptance content of the work orders are also extracted;

[0079] S503, search the causal graph constructed in S5011 according to the event obtained in S502, sort the causal relationship probability of the search results, take out the tail event with the highest probability and send it to the application system, namely M-Order Eagle Eye and M-Work Order Hotspot Eagle Eye.

[0080] S501 specifically includes:

[0081] Rules of Use<Pattern,Constraint,Priority> Match the causal mention text in the causal events from the acceptance content and processing content of historical work orders, where Pattern is a regular expression for extracting causal relationships; Constraint is a syntactic structure constraint; Priority indicates the decision if multiple Patterns match successfully; based on the BERT-based causal relationship extraction model, feature extraction is performed from the customer star rating, product type, service category, re-submission mark and skip-level mark attributes of historical work orders, and BiLSTM+CRF is used to output causal identification labels; by calculating the causal relationship strength and transition probability, the causal relationship of each event is adjusted, and finally a cause-and-effect graph of the operator is constructed.

[0082] S6, build an enterprise intelligent brain, output customer service GPT, and quickly provide solutions to the root causes of complaints;

[0083] S7, after the problem is solved, the self-problem solving follow-up system is used to track new data from customer feedback in a timely manner, and continuous iterative optimization is carried out based on the new data from customer feedback to complete the closed loop.

[0084] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for building a problem monitoring robot to solve the complaint work order management, characterized in that: The following steps are involved: S1, obtains customer complaint data from multiple channels, uses natural language processing for semantic understanding and intent recognition, and classifies complaint ticket issues; S2, extracts key information after complaint work order classification and generates structured work order tasks; S3, cluster analysis is performed on the generated structured work order tasks to determine the resubmission and cross-level complaint prediction work orders and obtain standardized work orders; S4, build a proposal dashboard and common problem discovery model, output the standardized work orders through the common problem discovery model, and then use Toyota's one-page minimalist thinking method to compare and display them, and predict the root problems of the complaint work orders; S5, using the event map to diagnose the root problem of the prediction and output the root cause of the problem; S6, build an enterprise intelligent brain, output customer service GPT, and quickly provide solutions to the root causes of complaints; S7, after the problem is solved, the self-problem solving follow-up system is used to track new data from customer feedback in a timely manner, and continuous iterative optimization is carried out based on the new data from customer feedback.

2. According to claim 1, a method for solving complaint work order management by building a problem monitoring robot is characterized in that: The S1 specifically includes: Obtain customer complaint data from multiple channels, and record basic information such as relevant complaint channels, complaint time, complaint content, and customer information for subsequent processing; Based on the complaint content obtained, natural language processing is used to perform semantic analysis, extract key information from the complaint content, and obtain data on the complaint object, problem type, and sentiment tendency; Through intention identification, we can judge the customer's complaint intention and determine whether it is for consultation, suggestion or appeal, so as to take appropriate measures; If the intent recognition result is a complaint, the complaint content is classified according to the result of semantic understanding to determine the specific complaint category, including product quality issues, service attitude issues or charging disputes.

3. According to claim 1, a method for solving complaint work order management by building a problem monitoring robot is characterized in that: The S2 specifically includes: According to the complaint time and complaint channel, obtain the initial information of the work order, determine whether the complaint is made during working hours, and obtain the priority of complaint handling; By using the information of the complaining customer, we can judge the customer's historical complaint record and credit rating, and determine the importance of the customer; According to the complaint content and the object of the complaint, obtain the keywords and topics of the problem and determine the problem classification; Determine the assigned department based on the urgency and problem classification; Obtain past processing records and judge processing progress through current processing status; Based on the update of processing results, determine whether the problem has been resolved, obtain customer feedback, and determine customer satisfaction.

4. According to claim 1, a method for solving complaint work order management by building a problem monitoring robot is characterized in that: The S3 specifically includes: Conduct preliminary clustering analysis on structured work order tasks, use clustering algorithms to group similar work orders, determine the subject categories of the work orders, and obtain preliminary clustering results; By analyzing the submission time and person-in-charge information, the efficiency of the work order processing is determined. If the processing cycle does not match the priority, it is identified as a potential cross-level complaint work order; Combine the contact information and attachment information of the work order to obtain the complete communication record of the work order, analyze the communication content through information mining, determine whether there is a resubmission, and obtain the judgment result of the resubmission work order; By analyzing the relationship between the submitter and the problem type, if a customer submits the same type of work order multiple times in a short period of time, he or she will be identified as a potential repeat customer and a list of repeat customers will be obtained.

5. According to claim 1, a method for solving complaint work order management by building a problem monitoring robot is characterized in that: The S4 specifically includes: S401, first construct a common problem discovery model, use historical data for training, and adjust hyperparameters to obtain an algorithm model for discovering common problems with an optimal loss function; S402: For all the work orders received from the front end every day, a cluster analysis is performed on the work order acceptance content using a trained algorithm model for discovering common problems to identify potential common problems. S403, using the Euclidean distance measurement method to compare the potential common problems found by the clustering model with the historical potential problems, if it is a new problem, it will trigger Toyota One Page to raise the problem, if it is an existing problem, it will not trigger Toyota One Page; S404, based on Toyota's one-page management process, provides feedback on solutions to the issues raised and continuously tracks the status of problem solving.

6. According to claim 4, a method for solving complaint work order management by using a construction problem monitoring robot is characterized in that: The S401 specifically includes: The k-means clustering algorithm model is used to build a basic algorithm model for discovering common problems, and the training data set is selected as 6 months of work order data; One month of historical work orders were extracted from the training data set, and the work order acceptance content was clustered using the k-means clustering algorithm model. The hyperparameter k was adjusted from 1 to 20, and the change in the loss value of the model was observed. The clustering results were compared with the quality inspection results to observe the clustering effect. Repeat the above steps. When the hyperparameter k reaches the preset value, an algorithm model for discovering common problems is obtained with a stable clustering loss and good clustering effect.

7. According to claim 1, a method for solving complaint work order management by building a problem monitoring robot is characterized in that: The S5 specifically includes: S501, based on the historical work orders of the operator customer service for 6 months, analyze and extract the facts and causal relationship information from the acceptance and processing of the work orders and whether there are repeated complaints or complaints beyond the level, and build the facts and causal relationship map of the operator customer service; S502, for the work orders received from the front end every day, the customer star rating, product type, service category, and acceptance content of the work orders are also extracted; S503, search the causal graph constructed in S5011 according to the event obtained in S502, sort the causal relationship probability of the search results, take out the tail event with the highest probability and send it to the application system, namely M-Order Eagle Eye and M-Work Order Hotspot Eagle Eye.

8. The method for solving the complaint work order management by building a problem monitoring robot according to claim 7 is characterized in that: The S501 specifically includes: Use rules to match causal mention text in causal events from the acceptance content and processing content of historical work orders; Based on the BERT-based causal relationship extraction model, feature extraction is performed from the customer star rating, product type, service category, resubmission mark, and skip level mark attributes of historical work orders, and BiLSTM+CRF is used to output causal identification labels; By calculating the strength of causal relationships and the probability of transition, the causal relationships of various events are adjusted, and ultimately the operator's cause-and-effect map is constructed.

Images