Bank customer complaint intelligent shunting and handling method and system based on NLP technology

Abstract

The application provides a bank customer complaint intelligent shunting treatment method and system based on NLP technology, comprising the following steps: step 1, establishing a bank customer complaint label system; step 2, establishing a complaint text intelligent classification rule library and a model library; step 3, through the classification rule library and the model library, the complaint text is subjected to label classification and auxiliary label classification; step 4, according to the label corresponding to the complaint handling department obtained after automatic or auxiliary label classification, intelligent shunting treatment is carried out; step 5, the complaint label system is subjected to label optimization and new iteration; and step 6, the complaint data is stored and displayed. The application utilizes new technical means such as artificial intelligence to more efficiently handle bank customer complaints, can effectively reduce the work load of customer service personnel, improve the work efficiency and quality of the customer service personnel, and realize comprehensive evaluation and assessment of the bank management department on the complaint situation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a bank customer complaint intelligent shunting method and system based on NLP technology. BACKGROUND

[0002] In recent years, with the complexity and diversification of banking business, the number of complaints brought by the increasing scale of business demand of bank customers is also increasing, and the efficiency of customer complaint management and processing plays an increasingly important role in consumer protection work and product service management. Under the background of strict supervision requirements, increasing customer demand and widespread social concern, using new technical means such as artificial intelligence to more efficiently handle bank customer complaints can effectively reduce the workload of customer service personnel, improve the work efficiency and quality of customer service personnel, and realize the comprehensive evaluation and assessment of the bank management department on the complaint situation.

[0003] Patent document CN113312924A (application number: CN202110695416.X) discloses a risk rule classification method and device based on NLP high-precision analysis label. Among them, the method comprises: obtaining risk scene information; generating a label analysis rule according to the risk scene information; classifying risk data through the label analysis rule to obtain a classification result; and outputting the classification result.

[0004] The existing bank customer complaint processing procedure is usually to classify customer complaint calls by customer service personnel, label or manually input complaint labels under the drop-down menu of the complaint management system, and then manually select the processing department corresponding to the complaint business type to process the complaint order. The main shortcomings are: there are many bank business channels and product categories, and new ones are continuously added, manual classification needs to compare different category labels, and then classify and transfer the complaint order, which is low in efficiency and easy to misclassify due to subjective judgment; a large amount of complaint data is expressed in text form, which hides a large number of classification features of complaint data in text expression, and it is easy to miss information when manually classifying and transferring, and it is difficult to fully express the information of complaint data; there is a lack of a large amount of data support for complaint data analysis, and the complaint disposal pressure is large, and it is difficult for the management department to comprehensively analyze the complaint situation. SUMMARY

[0005] In view of the defects in the prior art, the purpose of the present application is to provide a bank customer complaint intelligent shunting method and system based on NLP technology.

[0006] The bank customer complaint intelligent shunting method based on NLP technology provided by the present application comprises:

[0007] Step 1: Establish a bank customer complaint label system;

[0008] Step 2: Establish a complaint text intelligent classification rule base and model base;

[0009] Step 3: Through the classification rule base and model base, the complaint text is classified and auxiliary labeled;

[0010] Step 4: According to the complaint handling department corresponding label obtained after automatic or auxiliary label classification, intelligent shunt disposal is carried out;

[0011] Step 5: The complaint label system is optimized and iterated;

[0012] Step 6: Store and display the complaint data.

[0013] Preferably, the keywords involved in different categories form a rule base in the form of keyword-label;

[0014] Different complaint business types adopt different complaint business handling channels, and according to the complaint reasons and demands of bank customers, they are summarized to different complaint handling departments, forming a label and complaint handling department dispatching processing logic, which is solidified as a label dispatching logic rule base;

[0015] Using historical complaint data, sample labeling is carried out according to the established complaint label system, and NLP model training is carried out. The model input is Chinese string type complaint text data, and the model output is the probability of different complaint labels, forming an intelligent label classification model base.

[0016] Preferably, using the trained NLP model, the complaint text is labeled and classified, the customer service personnel answer the customer complaints and complete the complaint text summary, call the intelligent classification rule and NLP model for prediction, return the predicted label and probability, and directly automatically label the single classification label prediction probability far higher than other labels; When the prediction exists, the top 3 labels are returned, and the customer service personnel manually assist in label selection;

[0017] According to different types of labels, according to the label dispatching logic rule base, the program is used to automatically dispatch complaints to the corresponding complaint handling department for disposal and feedback.

[0018] Preferably, the newly added complaint business labels and product labels with a number lower than the preset threshold are automatically added to the rule base, and if the complaint summary contains related keywords, they are automatically matched and labeled. When new labels and new keywords are added, based on the corresponding relationship of existing keywords and categories, if there is a conflict, the conflicting keywords are modified accordingly, and a semantic similarity model is added for similarity prediction, and then auxiliary intelligent label classification is carried out according to semantic indexing.

[0019] Preferably, the complaint number of the complaint form that has been marked and the corresponding label are added to the database, visualized display is carried out through query, screening and statistical operation, and meanwhile, complaint summary and label data are stored through a distributed search and analysis engine ElasticSearch, so as to realize overall analysis of complaints in the whole bank, analysis of complaints in each management department, analysis of complaints in key areas, analysis of complaints in each branch, statistical analysis of different dimension labels, and quality and efficiency analysis of complaint processing.

[0020] The bank customer complaint intelligent shunting and handling system based on the NLP technology provided by the application comprises:

[0021] Module M1: establishing a bank customer complaint label system;

[0022] Module M2: establishing a complaint text intelligent classification rule library and a model library;

[0023] Module M3: through the classification rule library and the model library, complaint text is classified and auxiliary classified;

[0024] Module M4: according to the label corresponding to the complaint handling department obtained after automatic or auxiliary label classification, intelligent shunting and handling are carried out;

[0025] Module M5: label optimization and new iteration are carried out on the complaint label system;

[0026] Module M6: complaint data are stored and displayed.

[0027] Preferably, a rule library in the form of keyword-label is formed for keywords involved in different categories;

[0028] Different complaint business types adopt different complaint business handling channels, and according to the complaint reasons and complaint demands of bank customers, different complaint handling departments are concluded, a label and complaint handling department dispatching and processing logic is formed, and is solidified as a label dispatching logic rule library;

[0029] Using historical complaint data, sample labeling is carried out according to the established complaint label system, and NLP model training is carried out, the model input is Chinese string type complaint text data, the model output is the probability of different complaint labels, and an intelligent label classification model library is formed.

[0030] Preferably, using the trained NLP model, complaint text is classified and marked, customer service personnel answer customer complaints and complete complaint text summary, intelligent classification rules and NLP models are called for prediction, the predicted labels and probabilities are returned, and the result that the prediction probability of a single classification label is much higher than that of other labels is directly and automatically marked; when the prediction exists and the first few probabilities are close, the top 3 labels are returned, and customer service personnel manually assist in label selection;

[0031] According to different types of labels, according to the label order allocation logic rule base, the program is used to automatically allocate complaints to the corresponding complaint handling department for disposal and feedback.

[0032] Preferably, the newly added complaint business label and product label with a quantity lower than a preset threshold are automatically added to the rule base, and if the complaint summary contains relevant keywords, the keywords are automatically matched and labeled, and when the new label and new keyword are added, based on the corresponding relationship of the existing keywords and categories, if there is a conflict, the conflicting keywords are modified accordingly, and a semantic similarity model is added for similarity prediction, and then semantic indexing is used for auxiliary intelligent label classification.

[0033] Preferably, the complaint number and corresponding label of the complaint order that has been labeled are added to the database, and visual display is performed through query, screening and statistical operation, and complaint summary and label data are stored through a distributed search and analysis engine ElasticSearch, to realize overall analysis of complaints in the whole bank, complaint situation analysis of each management department, complaint situation analysis of key areas, complaint situation analysis of each branch, statistical analysis of different dimension labels, and quality and efficiency analysis of complaint handling.

[0034] Compared with the prior art, the present application has the following beneficial effects:

[0035] (1) The present application uses new technical means such as artificial intelligence to more efficiently handle bank customer complaints, which can effectively reduce the workload of customer service personnel, improve the work efficiency and quality of customer service personnel, and realize comprehensive evaluation and assessment of the complaint situation by the bank management department.

[0036] (2) The present application provides a solution for dynamically adding and optimizing labels in actual application scenarios, without the need to retrain the label classification model, effectively reducing the iteration cost after the model goes online. BRIEF DESCRIPTION OF DRAWINGS

[0037] Other features, objects and advantages of the present application will become more apparent through reading the following detailed description of the non-limiting embodiments with reference to the accompanying drawings:

[0038] Figure 1 Complaint intelligent shunt disposal flow chart;

[0039] Figure 2 System framework diagram. DETAILED DESCRIPTION

[0040] The application will be described in detail below with specific examples. The following examples will help those skilled in the art to further understand the application, but do not limit the application in any form. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the application. These are within the scope of the present application.

[0041] Example 1

[0042] The application provides a bank customer complaint intelligent shunting method based on NLP technology, which includes a complaint label category system design, a rule library and an NLP model library for automatically classifying complaint summaries and giving category probabilities for automatic tagging and auxiliary tagging, automatically assigning to complaint handling departments, supporting flexible addition of labels, and storing and statistical analysis of complaint data.

[0043] As Figure 1 , specifically comprising the following steps:

[0044] Step 1: Establish a bank customer complaint label system.

[0045] The label categories include but are not limited to complaint customer labels, complaint business labels, complaint business handling channels, complaint reasons, complaint products, complaint demands, and complaint handling departments. Each label category can have multiple levels of labels, and the bank business, channel, product, and reason are designed in detail. Through the description of multi-dimensional labels, the problems of inconsistent statistical standards and unclear complaint flow and disposal due to different classification standards are solved.

[0046] Step 2: Establish a complaint text intelligent classification rule library and model library.

[0047] For different categories involving keywords, form a "keyword-label" rule library.

[0048] Different complaint business types and different complaint business handling channels will be summarized into different complaint handling departments according to the complaint reasons and complaint demands of bank customers, thus a label and complaint handling department assignment processing logic is needed, which is solidified as a "label assignment logic" rule library.

[0049] Use historical complaint data to form sample labeling according to the complaint label system formed in step 1, and perform NLP model training. The model input is the text of the complaint summary, and the data type is Chinese string. The model output is the probability of different complaint labels, forming an intelligent label classification model library.

[0050] Also include text classification model, information extraction model, semantic similarity model, input is string text data; The output of the text classification model is the probability of different labels, the output of the information extraction model is the entity information in different texts, the purpose is to enhance the effect of the text classification model, the output of the semantic similarity model is the label type of the training data closest to the sentence meaning vector obtained by the approximate nearest neighbor algorithm. The text classification model can convert the task of complaint label classification into a text classification problem; The information extraction model can extract entity information in the complaint summary to provide higher-order features for text classification; The semantic similarity model can quickly find the complaint type closest to the semantic vector through the approximate nearest neighbor algorithm to assist in identifying new samples, to some extent, to solve the problem of insufficient training data for some categories.

[0051] Step 3: automatic label classification and auxiliary label classification.

[0052] Use the trained NLP model to classify and label the complaint text. The customer service personnel listen to the customer complaints and complete the complaint text summary, and can call the intelligent classification rules and NLP models implemented in step 2 to make predictions, and return the predicted labels and probabilities. The result of the single classification label prediction probability is much higher than that of other labels, which can be directly automatically labeled; When the prediction exists, the variance of the first few items is small, that is, the probability is close, return the top 3 labels, and the customer service personnel manually assist in label selection.

[0053] Step 4: automatic allocation.

[0054] According to the complaint handling department corresponding to the label obtained after automatic or auxiliary label classification, intelligent shunt disposal is carried out. According to different types of labels, according to the "label allocation logic" rule library in step two, the program is used to automatically allocate complaints to the corresponding complaint handling department to dispose and feedback the complaints.

[0055] Step 5: complaint label system label optimization and new iteration.

[0056] For a small number of new complaint business labels, product labels, etc., they can be automatically added to the rule library, and if the complaint summary contains related keywords, they can be automatically matched and labeled. When new labels and new keywords are added, the system will list the corresponding relationship between the existing keywords and categories, and no conflict can be generated. If there is a conflict, the conflicting keywords need to be modified accordingly. At the same time, join the semantic similarity model for similarity prediction, and then assist in intelligent label classification according to the semantic index.

[0057] Step 6: complaint data storage and display.

[0058] The complaint number of the complaint sheet that has been marked and the corresponding label are added to the database, and visual display is performed through query, screening, statistics and the like. Meanwhile, the complaint summary and label data are stored through a distributed search and analysis engine such as ElasticSearch, so that the complaint data can be conveniently and quickly searched and counted, overall situation analysis of complaints of the whole bank, complaint situation analysis of each management department, complaint situation analysis of key areas, complaint situation analysis of each branch, statistical analysis of different dimensions of labels including customers, channels, products, behaviors and complaint reasons, quality and efficiency analysis of complaint handling, and the like are realized, so as to improve the evaluation and management of the management department on the quality of complaint handling, and meanwhile, provide a basis for improving products, service behaviors and the like.

[0059] Embodiment 2

[0060] As Figure 2 The application also provides a bank customer complaint intelligent shunting and handling system based on NLP technology, which can be realized by executing the process steps of the bank customer complaint intelligent shunting and handling method based on NLP technology, that is, the bank customer complaint intelligent shunting and handling method based on NLP technology can be understood by those skilled in the art as the preferred implementation manner of the bank customer complaint intelligent shunting and handling system based on NLP technology.

[0061] The bank customer complaint intelligent shunting and handling system based on NLP technology provided by the application comprises the following modules: module M1, establishing a bank customer complaint label system; module M2, establishing a complaint text intelligent classification rule library and a model library; module M3, classifying and assisting in classifying complaint texts through the classification rule library and the model library; module M4, intelligently shunting and handling according to the labels corresponding to the complaint handling departments obtained after automatic or assisted label classification; module M5, optimizing and iteratively adding labels to the complaint label system; and module M6, storing and displaying complaint data.

[0062] The rules library in the form of keyword-label is formed for keywords involved in different categories; different complaint business types adopt different complaint business handling channels, and are classified into different complaint handling departments according to the complaint reasons and demands of bank customers, so as to form a label and complaint handling department dispatching and handling logic, which is solidified as a label dispatching logic rule library; historical complaint data are used to perform sample labeling according to the established complaint label system, and NLP model training is performed, the model input is complaint text data in the form of Chinese string, and the model output is the probability of different complaint labels, so as to form an intelligent label classification model library.

[0063] The complaint text is labeled and classified by using the trained NLP model, the customer service personnel answers the customer complaint and completes the complaint text summary, the intelligent classification rule and the NLP model are called to make a prediction, the predicted label and probability are returned, and the result of the single classification label prediction probability is higher than that of other labels; when the prediction exists, the top three labels are returned, and the customer service personnel manually assists in label selection; according to different types of labels, the label dispatching logic rule library is used to automatically dispatch the complaint to the corresponding complaint handling department for disposal and feedback.

[0064] The newly added complaint business label and product label with a quantity lower than a preset threshold are automatically added to the rule library, and if the complaint summary contains related keywords, automatic matching and labeling are performed, when a new label and a new keyword are added, based on the corresponding relationship between the existing keywords and categories, if there is a conflict, the conflicting keywords are modified accordingly, and a semantic similarity model is added for similarity prediction, and then semantic indexing is used for auxiliary intelligent label classification.

[0065] The complaint number and corresponding label of the complaint that has been labeled are added to the database, and visual display is performed through query, screening and statistical operation, and the complaint summary and label data are stored through the distributed search and analysis engine ElasticSearch, to realize overall complaint analysis of the whole line, complaint analysis of each management department, complaint analysis of key areas, complaint analysis of each branch, statistical analysis of different dimension labels, and complaint processing quality and efficiency analysis.

[0066] Those skilled in the art know that, in addition to implementing the system, device and each module thereof provided by the present application in the form of pure computer readable program code, the same program can be realized by logically programming the method steps to make the system, device and each module thereof provided by the present application in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers. Therefore, the system, device and each module thereof provided by the present application can be considered as a hardware component, and the modules included therein for realizing various programs can also be considered as structures within the hardware component; the modules for realizing various functions can also be considered as both software programs for realizing methods and structures within the hardware component.

[0067] The specific embodiments of the present application are described above. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essential content of the present application. The embodiments of the present application and the features in the embodiments can be arbitrarily combined with each other without conflict.

Claims

1. A bank customer complaint intelligent shunting and handling method based on NLP technology, characterized in that, Comprise: Step 1: Establish a bank customer complaint tag system; Step 2: Establish a complaint text intelligent classification rule base and model base; Step 3: Through the classification rule base and model base, the complaint text is classified and the auxiliary label classification is classified; Step 4: According to the label corresponding to the complaint handling department obtained after automatic or auxiliary label classification, intelligent shunt disposal is carried out; Step 5: The complaint tag system is optimized and iterated; Step 6: The complaint data is stored and displayed; Using the NLP model that has been trained, the complaint text is classified and labeled, the customer service personnel answer the customer complaint and complete the complaint text summary, the intelligent classification rule and NLP model are called to predict, the predicted label and probability are returned, and the single classification label prediction probability is higher than that of other labels. The result is directly automatically labeled; When the prediction exists, the top three labels are returned, and the customer service personnel manually assist in label selection; According to different types of labels, according to the label dispatching logic rule base, the program is used to automatically dispatch the complaint to the corresponding complaint handling department to handle and feedback the complaint; The newly added complaint business tags and product tags with a number lower than the preset threshold are automatically added to the rule base, and if the complaint summary contains related keywords, the automatic matching and labeling are carried out. When new tags and new keywords are added, based on the corresponding relationship of the existing keywords and categories, if there is a conflict, the conflicting keywords are modified accordingly, and a semantic similarity model is added to predict the similarity, and then the semantic index is used for auxiliary intelligent label classification.

2. The method for intelligent diversion and handling of bank customer complaints based on NLP technology according to claim 1, characterized in that, The keywords related to different categories form a rule base in the form of keyword-label; Different complaint business types adopt different complaint business handling channels, and according to the bank customer's complaint reason and complaint demand, it is summarized to different complaint handling departments, forming a label and complaint handling department dispatching processing logic, which is solidified as a label dispatching logic rule base; Using historical complaint data, sample labeling is carried out according to the established complaint tag system, and NLP model training is carried out. The model input is Chinese string type complaint text data, and the model output is the probability of different complaint labels, forming an intelligent label classification model base. 3.The NLP technology-based intelligent shunt handling method for bank customer complaints according to claim 1, characterized in that, The complaint number and corresponding label of the complaint that has been labeled are added to the database, and visual display is realized through query, screening and statistical operation, and at the same time, the complaint summary and label data are stored through distributed search and analysis engine ElasticSearch, realizing overall complaint analysis of the whole bank, complaint analysis of each management department, key area complaint analysis, complaint analysis of each branch, statistical analysis of different dimension labels, and complaint handling quality and efficiency analysis.

4. A bank customer complaint intelligent shunting and handling system based on NLP technology, characterized in that, Comprise: Module M1: Establish a bank customer complaint tag system; Module M2: Establish a complaint text intelligent classification rule base and model base; Module M3: Through the classification rule base and model base, the complaint text is classified and the auxiliary label classification is classified; Module M4: According to the label corresponding to the complaint handling department obtained after automatic or auxiliary label classification, intelligent shunt disposal is carried out; Module M5: The complaint tag system is optimized and iterated; Module M6: store and display the complaint data; Using the trained NLP model, the complaint text is labeled and classified. The customer service personnel answer the customer complaints and complete the complaint text summary. The intelligent classification rules and NLP model are called to make predictions, and the predicted labels and probabilities are returned. The single classification label prediction probability is higher than that of other labels, and the result is directly and automatically labeled. When the prediction has the first few items with close probability, the top 3 labels are returned, and the customer service personnel manually assist in label selection. According to different types of labels, according to the label dispatching logic rule library, the program is used to automatically dispatch complaints to the corresponding complaint handling department for disposal and feedback. The newly added complaint business labels and product labels with a number lower than the preset threshold are automatically added to the rule library. If the complaint summary contains related keywords, it will be automatically matched and labeled. When new labels and new keywords are added, based on the corresponding relationship of existing keywords and categories, if there is a conflict, the conflicting keywords will be modified accordingly. At the same time, a semantic similarity model is used to predict the similarity, and then semantic indexing is used to assist intelligent label classification. 5.The NLP technology-based intelligent bank customer complaint triage and handling system according to claim 4, characterized in that, For different categories of keywords, form a keyword-label rule library; Different complaint business types use different complaint business handling channels. According to the bank customer's complaint reason and complaint demand, it is concluded to different complaint handling departments, forming a label and complaint handling department dispatching logic, which is solidified as a label dispatching logic rule library. Using historical complaint data, sample labeling is performed according to the established complaint label system, and NLP model training is performed. The model input is Chinese string type complaint text data, and the model output is the probability of different complaint labels, forming an intelligent label classification model library. 6.The NLP technology-based intelligent bank customer complaint triage and handling system according to claim 4, characterized in that, The complaint number and corresponding label of the complaint that has been labeled are added to the database, and visual display is performed through query, screening and statistical operation. At the same time, complaint summary and label data are stored through distributed search and analysis engine ElasticSearch, realizing overall complaint analysis of the whole bank, complaint analysis of each management department, key area complaint analysis, branch complaint analysis, different dimension label statistical analysis, and complaint handling quality and efficiency analysis.

Images