CN112632268B

CN112632268B - Complaint work order detection processing method, complaint work order detection processing device, computer equipment and storage medium

Info

Publication number: CN112632268B
Application number: CN202011601408.6A
Authority: CN
Inventors: 陈杭; 赖众程; 李骁; 王亮; 海洋
Original assignee: Ping An Bank Co Ltd
Current assignee: Ping An Bank Co Ltd
Priority date: 2020-12-29
Filing date: 2020-12-29
Publication date: 2024-07-09
Anticipated expiration: 2040-12-29
Also published as: CN112632268A

Abstract

The invention discloses a complaint work order detection processing method, a complaint work order detection processing device, computer equipment and a storage medium. The method comprises the following steps: acquiring a to-be-measured complaint work order and work order additional data corresponding to the to-be-measured complaint work order; identifying a processing result in the work order additional data, and screening the complaint work order to be detected according to the work order additional data to obtain a potential false work order; identifying processing comments in the potential false work order and problem summaries in the work order additional data, and carrying out theme consistency judgment on the processing comments and the problem summaries corresponding to the potential false work order to obtain a consistency judgment result; if the consistency judgment result is that the topics are consistent, acquiring a comparative complaint work order corresponding to the potential false work order from the target corpus; and carrying out repeated checking processing on the processing comments in the potential false work orders and the processing comments in the comparison complaint work orders, obtaining the work order repetition rate, and determining whether the potential false work orders are false complaint work orders according to the work order repetition rate. The method can improve the detection efficiency of the false complaint worksheets.

Description

Complaint work order detection processing method, complaint work order detection processing device, computer equipment and storage medium

Technical Field

The present invention relates to the field of text detection processing technologies, and in particular, to a method and apparatus for detecting and processing a complaint work order, a computer device, and a storage medium.

Background

In the process of providing goods or services for clients, enterprises are required to be provided with complaint treatment persons to treat the client complaints. Generally, during the process of processing customer complaints, complaint contents of the customer complaints, processing comments in a specific negotiation processing process in a telephone return visit process and comments on processing results by a short message return visit need to be recorded in real time by a complaint processor, so that complaint work order data are formed. In actual work, complaint handler can appear tired out lazy condition of slackening, adopts to skip the customer and revisions the solution flow of negotiating back, directly revises the work order of complaining of history, forms false complaint work order, leads to customer complaint unable to obtain in time to handle, when harm customer interests, also leads to the enterprise image impaired, leads to the customer to run off even.

In the prior art, false complaint worksheets are detected by a paper check duplicate detection method, such as a character string comparison method or a word frequency statistics-based method. The detection method adopting paper duplicate checking has the following defects: firstly, the paper duplicate checking detection method does not distinguish whether the detection objects are suspected or not, but defaults that all the detection objects are plagiarism, and all the detection objects need to be processed, so that the detection workload is large. Secondly, the detection method of the paper check is free of prior information, all character strings in the object to be detected are compared equally, and therefore detection efficiency is low. In addition, the method for detecting the paper duplicate detection is to match the detection object with a massive corpus, which also results in low detection efficiency and long detection time. Therefore, how to quickly detect and identify false complaint worksheets is a problem to be solved.

Disclosure of Invention

The embodiment of the invention provides a method, a device, computer equipment and a storage medium for detecting and processing a complaint work order, which are used for solving the problem of low detection and processing efficiency of the complaint work order.

A complaint work order detection processing method comprises the following steps:

Acquiring a to-be-measured complaint work order and work order additional data corresponding to the to-be-measured complaint work order;

identifying a processing result in the work order additional data, and screening the complaint work order to be tested according to the work order additional data to obtain a potential false work order;

Identifying processing comments in the potential false work order and problem summaries in the work order additional data, and carrying out theme consistency judgment on the processing comments and the problem summaries corresponding to the potential false work order to obtain a consistency judgment result;

If the consistency judging result is that the topics are consistent, acquiring a comparative complaint work order corresponding to the potential false work order from a target corpus;

And carrying out duplicate checking processing on the processing opinion in the potential false work order and the processing opinion in the comparative complaint work order, obtaining the work order repetition rate, and determining whether the potential false work order is the false complaint work order according to the work order repetition rate.

A complaint work order detection processing device includes:

The system comprises a to-be-measured work order data acquisition module, a to-be-measured complaint work order acquisition module and a data processing module, wherein the to-be-measured work order data acquisition module is used for acquiring to-be-measured complaint work orders and work order additional data corresponding to the to-be-measured complaint work orders;

The potential false work order acquisition module is used for identifying a processing result in the work order additional data, screening the complaint work order to be tested according to the work order additional data and acquiring a potential false work order;

The consistency judging module is used for identifying the processing opinion in the potential false work order and the problem summary in the work order additional data, carrying out theme consistency judgment on the processing opinion and the problem summary corresponding to the potential false work order, and obtaining a consistency judging result;

The comparative complaint work order acquisition module is used for acquiring a comparative complaint work order corresponding to the potential false work order from a target corpus if the consistency judgment result is that the theme is consistent;

And the false complaint work order judging module is used for carrying out repeated checking processing on the processing comments in the potential false work order and the processing comments in the comparison complaint work order, obtaining the work order repetition rate and determining whether the potential false work order is the false complaint work order according to the work order repetition rate.

A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the complaint work order detection processing method described above when executing the computer program.

A computer-readable storage medium storing a computer program which, when executed by a processor, implements the complaint work order detection processing method described above.

According to the complaint work order detection processing method, the complaint work order detection processing device, the computer equipment and the storage medium, firstly, the to-be-detected complaint work orders are screened according to the processing result in the additional data of the work orders, and the potential false work orders are determined, so that whether the potential false work orders are false complaint work orders or not can be determined through subsequent detection, the detection workload can be reduced, the detection efficiency of the false complaint work orders can be improved, and the detection time consumption can be reduced. And then, carrying out theme consistency judgment on the processing opinion in the potential false work order and the problem summary in the additional data of the work order, obtaining a consistency judgment result, and directly recognizing the potential false work order as the false complaint work order when the consistency judgment result is that the theme is inconsistent, thereby being beneficial to improving the detection efficiency of the false complaint work order and reducing the detection time consumption. And when the consistency judgment result is that the topics are consistent, the comparison complaint work orders are determined from the target corpus, so that the matching range of the duplicate checking process is reduced, the detection efficiency is improved, and the detection time is reduced. And finally, carrying out repeated checking processing on the potential false work orders and the processing comments in the comparison complaint work orders, determining the repetition rate of the work orders, and further determining whether the potential false work orders are false complaint work orders. In the scheme, the matching range of the potential false work order for duplicate checking detection is reduced, so that the detection efficiency is improved, and the detection time consumption is reduced.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of an application environment of a method for detecting and processing a complaint work order according to an embodiment of the present invention;

FIG. 2 is a flow chart of a method for complaint work order detection processing in accordance with an embodiment of the present invention;

FIG. 3 is another flow chart of a complaint work order detection processing method in an embodiment of the invention;

FIG. 4 is another flow chart of a complaint work order detection processing method according to an embodiment of the present invention;

FIG. 5 is another flow chart of a complaint work order detection processing method in an embodiment of the invention;

FIG. 6 is another flow chart of a complaint work order detection processing method in an embodiment of the invention;

FIG. 7 is another flow chart of a complaint work order detection processing method in an embodiment of the invention;

FIG. 8 is a schematic diagram of a complaint work order detection processing device according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of a computer device in accordance with an embodiment of the invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The complaint work order detection processing method provided by the embodiment of the invention can be applied to an application environment shown in figure 1. Specifically, the complaint work order detection processing method is applied to a complaint work order detection processing system, the complaint work order detection processing system comprises a client and a server as shown in fig. 1, and the client and the server are communicated through a network and are used for realizing rapid detection of whether a to-be-detected complaint work order is a false complaint work order or not so as to improve detection efficiency and reduce detection time consumption. The client is also called a client, and refers to a program corresponding to the server for providing local service for the client. The client may be installed on, but is not limited to, various personal computers, notebook computers, smartphones, tablet computers, and portable wearable devices. The server may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.

In one embodiment, as shown in fig. 2, a complaint work order detection processing method is provided, and the method is applied to the server in fig. 1 for illustration, and includes the following steps:

s201: acquiring a to-be-measured complaint work order and work order additional data corresponding to the to-be-measured complaint work order.

The to-be-detected complaint work orders refer to complaint work orders to be detected, namely detection objects. Generally, user information, date and time information corresponding to a complaint handler and processing comments formed in a return visit process are recorded on a to-be-measured complaint work order. The user information corresponding to the complaint handler includes name, job number and contact phone. The date-time information is information for recording in which time period the complaint handler contacted with the customer on which day. The processing opinion is related information for recording a specific negotiation process of the client return visit procedure. Generally speaking, complaint treating person is tired and lazy idle hours, skips the return visit negotiation solution flow, directly modifies the user information, the date and time information in the history complaint work list and the treatment opinion formed in the return visit process, and forms a false complaint work list, so that customer complaints cannot be treated in time, and the customer benefits are damaged, and the enterprise image is damaged and even the customer loss is caused. The historical complaint work orders refer to real complaint work orders formed before the current time of the system.

Wherein the work order additional data is additional data related to customer complaints except for the complaint work order to be measured. In this example, the work order additional data is used to record additional data such as business category, problem summary, complaint cause, processing result, and affiliated department. The business category is a specific category for reflecting customer complaints, and may include at least one level of business category, for example, in the process of credit card complaints, the business category may be in order: bank card-credit card-debit card marketing campaign, value added service-marketing campaign gift issuing-other, etc. The problem overview is an overview that reflects the problem for which customer complaints are directed, and may be a specific activity of goods or services offered by an enterprise. The reason for complaints is a specific reason for reflecting customer complaints. The processing results are used to reflect the results of the client after the return visit. The affiliated department refers to a department aimed at by customer complaints or a department to which a complaint processor belongs.

S202: and identifying a processing result in the work order additional data, and screening the complaint work order to be detected according to the work order additional data to obtain a potential false work order.

The processing result is used for reflecting the result after the client makes a return visit, and the processing result can be understood as the result of whether the client is satisfied with the return visit negotiation processing process of the complaint processing personnel. A potentially spurious worksheet refers to a worksheet that is potentially a spurious complaint worksheet.

Generally, depending on whether the customer is satisfied with the return visit negotiation process, the complaint worksheet to be measured can be divided into the following two feedback types: first, the type of forward feedback, i.e., the processing result in the work order attachment data reflects the customer's satisfactory feedback for the return visit negotiation process, e.g., forward feedback words of acceptance, satisfaction, approval, agreement, presentation understanding, disagreement, known, no longer complaint, successful pacifying, withdrawal complaint, presentation clarity, nolle prosequi, documented and resolved are recorded in the work order attachment data. Second, the negative feedback type, that is, the processing result in the worksheet additional data, is feedback reflecting dissatisfaction of the customer in the return visit negotiation processing process, for example, negative feedback words of unacceptable, dissatisfaction, disapproval, disagreement, failing to meet the requirements, unaccepted, unintelligible, continuing complaint, intention upgrading, and unacceptable pacifying are recorded in the worksheet additional data.

As an example, the server first identifies the content corresponding to the processing result field in the work order additional data, and identifies the processing result in the work order additional data; matching the processing result with positive feedback words in a positive feedback word bank and negative feedback words in a negative feedback word bank by adopting a character string matching algorithm, and determining a result type corresponding to the processing result, wherein the result type comprises a positive feedback type and a negative feedback type; if the result type is a forward feedback type, determining the complaint work order to be measured as a real complaint work order; and if the result type is a negative feedback type, determining the complaint work order to be measured as a potential false work order.

As another example, the server first identifies the content corresponding to the processing result field in the work order additional data, and identifies the processing result in the work order additional data; classifying the processing results in the work order additional data by adopting a pre-trained Word2vec classification model; if the positive feedback Word in the positive feedback Word bank appears in the processing result in the work order additional data, outputting 0 by the Word2vec classification model, and determining the to-be-measured complaint work order as a real complaint work order; if negative feedback words in a negative feedback Word library appear in the processing result in the work order additional data, outputting 1 by a Word2vec classification model, and determining the to-be-measured complaint work order as a potential false work order. In this example, the Word2vec classification model is a classification model determined by training using the Word2vec network with classification training data carrying classification labels 0 or 1. Because the Word2vec network is a lightweight classification algorithm, the Word2vec classification model formed by training is high in running speed, even if the number of complaint worksheets to be tested is huge and reaches millions, whether the complaint worksheets are true complaint worksheets or potential false worksheets can be quickly identified, and the detection efficiency of the false complaint worksheets can be improved.

For example, if the result type of the processing result is a forward feedback type, it is indicated that the complaint handler has negotiated with the customer about the complaint content, and the negotiation process obtains acceptance of the customer; if the result type of the processing result is a negative feedback type, the negotiation between the complaint handler and the client cannot obtain the acceptance of the client; after the complaint handler carries out telephone return visit to determine the complaint opinion, the short message return visit is needed to confirm the processing result, so that the possibility that the processing result is true is high. Generally, if the result type of the processing result is a forward feedback type, which indicates that the client has accepted the result type, the probability of the result type being a false complaint work order is smaller; and when the result type of the processing result is a forward feedback type, the client is not acknowledged, and the probability of the processing result being a false complaint work order is high. Therefore, the to-be-detected complaint worksheets are screened based on the processing results in the worksheets additional data, so that the to-be-detected complaint worksheets are subjected to preliminary screening, detection of all to-be-detected complaint worksheets is avoided, subsequent detection workload is reduced, detection efficiency of false complaint worksheets is improved, and detection time is reduced.

S203: and identifying the processing opinion in the potential false work order and the problem summary in the work order additional data, and carrying out theme consistency judgment on the processing opinion and the problem summary corresponding to the potential false work order to obtain a consistency judgment result.

Wherein the processing opinion is related information for recording specific negotiation processing of the client return visit process. The problem overview is an overview that reflects the problem for which customer complaints are directed, and may be a specific activity of goods or services offered by an enterprise.

As an example, the server first identifies processing ideas in a potentially false work order and a summary of problems in the work order attachment data; and comparing whether the theme of the treatment opinion is consistent with the theme of the question summary, and acquiring a consistency judgment result, wherein the consistency judgment result comprises two types of consistent theme and inconsistent theme. For example, when the processing opinion is an opinion of activity a, if the problem summary is a summary of activity B, a consistency judgment result of inconsistent subjects is obtained; and if the problem summary is the summary of the activity A, acquiring a consistency judgment result of the consistency of the theme.

S204: and if the consistency judgment result is that the topics are consistent, acquiring a comparative complaint work order corresponding to the potential false work order from the target corpus.

The target corpus is a database for storing the comparative complaint worksheets. The comparative complaint worksheets are formed prior to the current time of the system for use as a comparative control for potential false worksheets.

As an example, when the consistency determination result is identified as the consistent subject, the server cannot directly determine whether the potential false work order is a false complaint work order, and the comparison complaint work order needs to be selected from the target corpus to be used as a comparison control object for performing the duplicate checking processing on the potential false work order.

As an example, all the historical complaint worksheets formed before the current time of the system are stored in the target corpus, the historical complaint worksheets are used as comparison complaint worksheets for performing repeated checking processing on potential false worksheets, compared with the comparison of the existing documents of a global database or other special databases in a paper repeated checking detection method, the number of the comparison complaint worksheets in the target corpus is small, the pertinence is high, the detection efficiency of repeated checking processing is improved, and the detection time consumption of repeated checking processing is reduced.

As a further improvement, the additional data of the worksheet corresponding to each complaint worksheet comprises information such as business category, affiliated departments and the like, the server can identify the business category and/or affiliated departments in the additional data of the worksheet corresponding to the potential false worksheet, and the historical complaint worksheets which are the same as the business category and/or affiliated departments of the potential false worksheets are used as comparison complaint worksheets for carrying out the duplicate checking treatment on the potential false worksheets from the additional data of the worksheets corresponding to all the historical complaint worksheets in the target corpus, so that the number of the comparison complaint worksheets is further reduced, namely the duplicate checking matching range is reduced; and the processing opinion of the comparative complaint worksheets formed by the same business category and/or the departments to which the business category belongs is more similar to the processing opinion of the potential false worksheets, thereby being beneficial to improving the pertinence of the duplicate checking processing, further improving the detection efficiency of the duplicate checking processing and reducing the detection time consumption of the duplicate checking processing.

S205: and carrying out repeated checking processing on the processing comments in the potential false work orders and the processing comments in the comparison complaint work orders, obtaining the work order repetition rate, and determining whether the potential false work orders are false complaint work orders according to the work order repetition rate.

The work order repetition rate refers to the repetition rate determined by adopting a duplicate checking detection technology to check and detect the potential false work order and the contrast complaint work order.

As an example, the server uses a duplication checking technology to check and duplication process the processing opinion in the potential false work order and the processing opinion in the comparative complaint work order, and obtains the work order repetition rate. In this example, the duplicate detection technique may be a conventional method of detecting duplicate paper, including but not limited to a string comparison method or a word frequency statistics-based method.

As an example, the server may compare the work order repetition rate to a repetition rate threshold after detecting the work order repetition rate of the treatment opinion in the potentially false work order and the comparative complaint work order; if the work order repetition rate is larger than the repetition rate threshold, the potential false work order is determined to be a false complaint work order forged by a complaint handler; if the repetition rate of the work orders is not greater than the repetition rate threshold, the potential false work orders are determined to be real complaint work orders, so that whether the potential false work orders are false complaint work orders forged by complaint handling persons or not is rapidly determined, the detection efficiency of the false complaint work orders is improved, and the detection time is shortened. The repetition rate threshold is a threshold for evaluating whether the repetition rate reaches the criterion of being the same as the repetition rate.

S206: if the consistency judgment result is that the theme is inconsistent, the potential false work order is determined to be the false complaint work order.

As an example, when the consistency judgment result is identified as inconsistent theme, the process opinion or the problem summary of the potential false work order is indicated to be inconsistent with the corresponding theme, so that the process opinion or the problem summary is most likely to be modified by a complaint handler in the process of forging the potential false work order, and therefore, the server can directly determine the potential false work order as the false complaint work order.

In the complaint work order detection processing method provided by the embodiment, firstly, the work order to be detected is screened according to the processing result in the additional data of the work order, and the potential false work order is determined, so that whether the potential false work order is the false complaint work order or not can be determined through subsequent detection, the detection workload can be reduced, the detection efficiency of the false complaint work order can be improved, and the detection time consumption can be reduced. And then, carrying out theme consistency judgment on the processing opinion in the potential false work order and the problem summary in the work order additional data, obtaining a consistency judgment result, and directly recognizing the potential false work order as the false complaint work order when the consistency judgment result is that the theme is inconsistent, thereby being beneficial to improving the detection efficiency of the false complaint work order and reducing the detection time consumption. And when the consistency judgment result is that the topics are consistent, the comparison complaint work orders are determined from the target corpus, so that the matching range of the duplicate checking process is reduced, the detection efficiency is improved, and the detection time is reduced. And finally, carrying out repeated checking processing on the potential false work orders and the processing comments in the comparison complaint work orders, determining the repetition rate of the work orders, and further determining whether the potential false work orders are false complaint work orders. In the scheme, the matching range of the potential false work order for duplicate checking detection is reduced, so that the detection efficiency is improved, and the detection time consumption is reduced.

In one embodiment, as shown in fig. 3, before step S201, that is, before acquiring the to-be-measured complaint work order and the work order additional data corresponding to the to-be-measured complaint work order, the complaint work order detection processing method includes:

s301: and acquiring the historical complaint worksheets and worksheet additional data corresponding to the historical complaint worksheets.

S302: and identifying a processing result in the work order additional data, and acquiring a result label corresponding to the historical complaint work order.

S303: and determining the customer satisfaction worksheet from the historical complaint worksheets according to the result labels corresponding to the historical complaint worksheets.

S304: and obtaining the subject training data according to the processing opinion and the problem summary corresponding to the customer satisfaction worksheet.

S305: model training is carried out based on the theme training data, and a target theme model is obtained.

The historical complaint work orders refer to real complaint work orders formed before the current time of the system. In this example, user information, date and time information, and processing comments formed in the return visit process corresponding to the complaint handler are also recorded on the history complaint work order. The work order additional data corresponding to the historic complaint work order is additional data related to customer complaints and is recorded in addition to the historic complaint work order, and the work order additional data is used for recording additional data such as business categories, problem summaries, complaint reasons, processing results, affiliated departments and the like. The result label corresponding to the history complaint work order is a label for reflecting whether the customer is satisfied after the return visit processing.

As an example, in step S302, the server may obtain all the history complaint worksheets before the current time of the system and the worksheet additional data corresponding to each history complaint worksheet from the target corpus; classifying the processing results in the work order additional data by adopting a pre-trained Word2vec classification model; if the positive feedback Word in the positive feedback Word bank appears in the processing result in the work order additional data, the result label output by the Word2vec classification model is 0; if the negative feedback Word in the negative feedback Word bank appears in the processing result in the work order additional data, the result label output by the Word2vec classification model is 1. In this example, the Word2vec classification model is a classification model determined by training using the Word2vec network with classification training data of class tag 0/1. Because the Word2vec network is a lightweight classification algorithm, the Word2vec classification model formed by training has high running speed, and can rapidly output a result label corresponding to each historical complaint work order.

As an example, in step S303, if the result label corresponding to the history complaint work order is 0, it is indicated that the processing result in the work order additional data of the history complaint work order is a forward feedback word, and it is indicated that the customer is satisfied with the complaint processing result, so that the history complaint work order can be determined as the customer satisfaction work order. If the result label corresponding to the historical complaint work order is 1, the processing result in the work order additional data of the historical complaint work order is a negative feedback word, and the customer is dissatisfied with the complaint processing result, so that the historical complaint work order can be determined as the customer dissatisfied work order.

As an example, in step S304, the server obtains the topic training data according to the processing opinion and the problem summary corresponding to the customer satisfaction worksheet, and specifically may extract the topic label according to the problem summary, and use the processing opinion and the topic label in the customer satisfaction worksheet as the topic training data, so as to perform model training by using the topic training data, thereby determining the target topic model. The target topic model can process the processing opinion and determine the topic corresponding to the processing opinion so as to improve the efficiency of topic consistency judgment.

As an example, in step S305, the server may train the topic model network using topic training data to obtain a target topic model. The topic model network is a model network for training a target topic model, and may be, but is not limited to LSA (LATENT SEMANTIC ANALYSIS ), PLSA (probabilice LATENT SEMANTIC ANALYSIS, probabilistic latent semantic analysis), LDA (LATENT DIRICHLET Allocation, implicit dirichlet Allocation), HDP (HIERARCHICAL DIRICHLET Process, layer-to-dirichlet procedure), and the like. In the example, the topic training data formed by the processing opinions and the problem summaries in the customer satisfaction worksheet is adopted, the model parameters of the topic model network are adjusted to form a target topic model, so that the accuracy of topic identification of the processing opinions in the complaint worksheet by the target topic model is guaranteed, and the accuracy of topic consistency judgment results is improved when the follow-up topic consistency judgment is guaranteed.

In one embodiment, as shown in fig. 4, step S305, that is, performing model training based on the topic training data, obtains a target topic model, includes:

S401: and training the topic model network by adopting external training data to obtain an original topic model.

S402: and training the original topic model by using topic training data to obtain a target topic model.

The external training data refers to training data which belong to different application scenes with the theme training data. For example, if the subject training data is training data formed based on a customer satisfaction worksheet corresponding to an application scenario of credit card complaints, the external training data is training data formed based on a customer satisfaction worksheet other than the application scenario of credit card complaints.

As an example, in step S401, the topic model network may be trained using a large amount of external training data corresponding to application scenarios other than the topic training data, so as to initialize model parameters of the topic model network, thereby obtaining an original topic model. Understandably, a large amount of external training data is adopted to perform model training to form an original topic model, so that the recognition accuracy of the original topic model can be ensured.

As an example, in step S402, after the original topic model is obtained by training with the external training data, the topic training data formed by the processing opinion and the problem summary corresponding to the customer satisfaction worksheet may be used to perform model training on the original topic model, fine-tune the model parameters in the original topic model, and obtain the target topic model, so that the base recognition result is more accurate when the target topic model performs topic recognition on the complaint worksheet formed in the application scenario corresponding to the topic training data. The target topic model herein is a pre-trained model for achieving topic identification.

In one embodiment, as shown in fig. 5, in step S203, performing a subject consistency determination on the processing opinion and the problem summary corresponding to the potential false work order, to obtain a consistency determination result, including:

S501: and processing the processing opinion corresponding to the potential false work order by adopting the target theme model to acquire the generated theme corresponding to the potential false work order.

S502: and acquiring a real theme corresponding to the potential false work order according to the problem outline corresponding to the potential false work order.

S503: and carrying out consistency judgment on the generated theme corresponding to the potential false work order and the real theme, and obtaining a consistency judgment result.

The target topic model is a model which is trained in advance and is used for realizing topic identification. Generating topics refers to generating topics by identifying processing opinions by using a target topic model. The real theme refers to a theme obtained by identifying a problem summary.

As an example, in step S501, the server may process the processing opinion corresponding to the potential false work order by using the target theme model trained in advance, so as to quickly determine the generating theme corresponding to the processing opinion, obtain the generating theme corresponding to the potential false work order, and ensure the obtaining efficiency of the generating theme.

As an example, in step S502, the server may word and identify the question summary according to the question summary in the potential false work order, and determine therefrom the real subject matter described by the question summary, which is a subject matter directly determined from the records in the question summary.

As an example, in step 503, the server may use a string matching algorithm to perform matching processing on the generated theme and the real theme corresponding to the potential false work order, so as to determine whether the generated theme and the real theme are consistent, thereby obtaining a consistency determination result, and the determination process is simpler and more convenient.

As another example, in step S503, the server may further use a conventional BERT (Bidirectional Encoder Representations from Transformers, based on the bi-directional coding representation of the converter) model to perform matching processing on the generated theme and the real theme corresponding to the potential false work order, so as to determine whether the generated theme and the real theme are consistent, thereby obtaining a consistency determination result, and the determination process is simpler and more convenient.

In an embodiment, in step S503, the server may further use an improved BERT model to perform matching processing on the generated theme and the real theme corresponding to the potential false work order, so as to determine whether the generated theme and the real theme are consistent, thereby obtaining a consistency determination result. The improved BERT model is a model formed by improving a conventional BERT model, and can improve the accuracy of consistency judgment results. As shown in fig. 6, performing consistency judgment on a generated theme and a real theme corresponding to a potential false work order, and obtaining a consistency judgment result includes:

S601: and inputting the generated theme and the real theme corresponding to the potential false work order into an improved BERT model, extracting original codes corresponding to the last four layers of the improved BERT model, and obtaining a three-dimensional coding matrix.

S602: and expanding the three-dimensional coding matrix into a four-dimensional coding matrix, and carrying out average pooling on the four-dimensional coding matrix by adopting an average pooling method to obtain pooled codes.

S603: and inputting the pooled code into a Softmax layer of the improved BERT model, and obtaining a consistency judgment result output by the Softmax layer.

As an example, in step S601, the server inputs the generated theme and the real theme corresponding to the potential false work order into an improved BERT model, extracts the original codes corresponding to the last four layers in the weight of the improved BERT model, and obtains a three-dimensional coding matrix with the shape of (batch_size, 4, hidden_size); compared with the mode that only the last layer is extracted in the conventional BERT model, the three-dimensional coding matrix obtained by improving the BERT model can obtain richer semantic information, and is beneficial to guaranteeing the accuracy of consistency judgment results. The original code refers to a code extracted at an implicit layer of the BERT model after the generated theme and the real theme corresponding to the potential false work order are input into the BERT model, and the code can be CLS (Common Language Specification) codes.

As an example, in step S602, the server may perform dimension expansion on the three-dimensional coding matrix with the shape of (batch_size, 4, hidden_size) to obtain the four-dimensional coding matrix with the shape of (batch_size, 1,4, hidden_size), so that the data dimension of the four-dimensional coding matrix is the same as the data dimension of the grayscale image, and therefore, the average pooling method of grayscale image processing may be used to reduce the dimension of the four-dimensional coding matrix, which is helpful for improving the processing efficiency and guaranteeing the feasibility of the processing. And then, carrying out average pooling on the four-dimensional coding matrix by adopting an average pooling method so as to realize dimension reduction on the four-dimensional coding matrix and form pooled coding. Understandably, compared with the method of directly averaging the last four layers of original codes by using the conventional BERT model, the four-dimensional coding matrix is subjected to average pooling by using an average pooling method, and the four-dimensional coding matrix can be deeply advanced on the last four layers of original codes to be fully fused, so that pooled codes containing more local information are extracted.

As an example, in step S603, the pooled code formed after the four-dimensional coding matrix is averaged and pooled may be input into the Softmax layer of the improved BERT model, and the consistency determination result output by the Softmax layer may be obtained. For example, if the Softmax layer outputs 0, it indicates that the consistency determination result is that the theme is inconsistent; if the Softmax layer outputs 1, the consistency judgment result is the subject consistency, so that the consistency judgment result can be quickly obtained by using the improved BERT model, and the accuracy of the consistency judgment result is ensured.

In one embodiment, as shown in fig. 7, in step S205, performing duplication checking processing on the processing opinion in the potentially false work order and the processing opinion in the comparative complaint work order to obtain the work order repetition rate includes:

s701: and extracting date and time elements from the processing opinions in the potential false work order to obtain the date and time weight corresponding to the potential false work order.

S702: and segmenting the processing opinion in the potential false work order to obtain at least two original character strings corresponding to the potential false work order and part-of-speech weights corresponding to each original character string.

S703: and calculating the weight of each original character string by adopting an improved IF-IDF algorithm according to the date and time weight and the part-of-speech weight corresponding to the original character string, and obtaining the target weight corresponding to each original character string.

S704: and determining the target character strings corresponding to the potential false work orders according to the target weights corresponding to the at least two original character strings.

S705: and performing similarity calculation on the target character string corresponding to the potential false work order and the comparison character string in the comparison complaint work order by adopting a similarity algorithm, and obtaining the work order repetition rate.

The date and time element refers to date and time information in the processing opinion.

As an example, in step S701, if the latent false work order includes a date and time element corresponding to a specific date and time format, a regular expression matching method may be used to extract the date and time element from the processing opinion in the latent false work order to determine whether the processing opinion in the latent false work order includes date and time information, so as to obtain a date and time weight.

As another example, in step S701, if the latent false work order does not include a date and time element corresponding to the specific date and time format, a named entity recognition algorithm may be used to extract the date and time element from the processing opinion in the latent false work order to determine whether the processing opinion in the latent false work order includes date and time information, thereby acquiring the date and time weight. Understandably, named entity recognition (Named-entity recognition) algorithm may be adopted by the currently mainstream BiLSTM-CRF algorithm, or other algorithms may be adopted, so long as the function of extracting the date and time elements in the processing opinion can be realized. It should be understood that, since the named entity recognition algorithm may extract the date and time elements corresponding to all the date and time formats, the named entity recognition algorithm is preferably used to extract the date and time elements in step S701, so as to ensure the processing efficiency.

In this example, determining the date-time weight according to whether the date-time information is included in the processing opinion in the potential false work order includes: when the processing opinion of the potential false work order contains date and time information, determining the date and time weight of the potential false work order as a first time weight; and when the processing opinion of the potential false work order does not contain date and time information, determining that the date and time weight is the second time weight. I.e.Wherein weight_date (Word _i) is a Date time Weight, W _D1 is a first time Weight, W _D2 is a second time Weight, and generally, W _D1>W_D2. In this example, the first time weight W _D1 may be set to 5 and the second time weight W _D2 may be set to 0 to highlight the importance of whether or not date-time information is contained.

As an example, in step S702, the server uses a conventional word segmentation method to segment the processing opinion in the potential false work order to obtain at least two segmented original character strings, and then determines the part-of-speech weight corresponding to each original character string according to a preset part-of-speech weight configuration table.

In this example, since the nouns appearing in the processed opinion are generally the most critical information related to topic, activity, or other, the part-of-speech weight of the nouns may be set to be the highest; since it is also important to deal with the verbs that appear in the opinion as involving a specific process, the verb may be set to have a part of speech weight that is slightly lower than that of the noun, but higher than that of the other parts of speech. I.e.WeightCX (Word _i) is a part-of-speech weight of each original character string, N is a verb, V is a noun, other is other parts-of-speech than nouns and verbs, W _CX1 is a part-of-speech weight corresponding to nouns, W _CX2 is a part-of-speech weight corresponding to verbs, W _CX3 is a part-of-speech weight corresponding to other parts-of-speech than nouns and verbs, and W _CX1>W_CX2>W_CX3. For example, W _CX1＝3,W_CX2＝2,W_CX3 =1 may be set.

In general, the conventional IF-IDF algorithm calculates TF (Term Frequency) and IDF (Inverse Document Frequency, reverse file Frequency) to determine the Weight corresponding to each original string in the document, i.e., weight_score _i＝TF(Word_i)*IDF(Word_i. TF refers to the frequency with which a given word appears in the document. IDF mainly refers to the fact that if a document containing a given term is fewer, the greater the IDF value of that term, the better the document variability of that term.

As an example, in step S703, the weight calculation may be performed on each original string by using the modified IF-IDF algorithm, specifically, in addition to TF (Term Frequency) and IDF (Inverse Document Frequency ), the date-time weight and the part-of-speech weight corresponding to the original string need to be considered, so as to determine the target weight corresponding to each original string. In this example, the modified IF-IDF algorithm may specifically be Weight_Sorce_i＝TF(Word_i)*IDF(Word_i)*((1+WeightCX(Word_i)+Weight_Date(Word_i)), to determine the target weight for each original string.

The target character string refers to a plurality of original character strings which are determined to be most important from at least two original character strings corresponding to the processing opinion.

As an example, in step S704, the server may sort the target weights corresponding to all the original strings in the processing opinion of the potential false work order, and then select the N original strings ranked first to determine as the target strings, so that the target strings can reflect the meaning of the expression required for processing the opinion.

The comparison character string in the comparison complaint work order refers to the target character string extracted from the processing opinion corresponding to the comparison complaint work order according to the steps S701-S704. In general, when the comparison complaint worksheet is stored in the target corpus, the extracted comparison character strings and the comparison complaint worksheet can be stored in a correlated manner, so that the comparison character strings in the comparison complaint worksheet can be quickly called for similarity calculation in the follow-up process, and the detection efficiency of the false complaint worksheets is improved.

The similarity algorithm is an algorithm for judging whether character strings are similar or not.

As an example, in step S705, the server may use a cosine similarity algorithm, a hamming distance algorithm, or a Simhash algorithm to perform similarity calculation on the target string corresponding to the potential false work order and the comparison string in the comparison complaint work order, so as to obtain the work order repetition rate, so as to compare the work order repetition rate with the repetition rate threshold value; if the work order repetition rate is larger than the repetition rate threshold, the potential false work order is determined to be a false complaint work order forged by a complaint handler; if the repetition rate of the work orders is not greater than the repetition rate threshold, the potential false work orders are determined to be real complaint work orders, so that whether the potential false work orders are false complaint work orders forged by complaint handling persons or not is rapidly determined, the detection efficiency of the false complaint work orders is improved, and the detection time is shortened.

It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present invention.

In an embodiment, a complaint work order detection processing device is provided, and the complaint work order detection processing device corresponds to the complaint work order detection processing method in the embodiment one by one. As shown in fig. 8, the complaint work order detection processing device includes a work order data acquisition module 801 to be detected, a potential false work order acquisition module 802, a consistency judgment module 803, a comparative complaint work order acquisition module 804, a false complaint work order judgment module 805, and a false complaint work order determination module 806. The functional modules are described in detail as follows:

The to-be-measured worksheet data acquisition module 801 is configured to acquire to-be-measured complaint worksheets and worksheet additional data corresponding to the to-be-measured complaint worksheets.

The potential false work order obtaining module 802 is configured to identify a processing result in the work order additional data, and screen the complaint work order to be measured according to the work order additional data, so as to obtain the potential false work order.

And the consistency judging module 803 is used for identifying the processing opinion in the potential false work order and the problem summary in the work order additional data, carrying out theme consistency judgment on the processing opinion and the problem summary corresponding to the potential false work order, and obtaining a consistency judging result.

The comparative complaint work order obtaining module 804 is configured to obtain, if the consistency determination result is that the theme is consistent, a comparative complaint work order corresponding to the potential false work order from the target corpus.

The false complaint work order judging module 805 is configured to perform duplication checking processing on the processing opinion in the potential false work order and the processing opinion in the comparative complaint work order, obtain the work order repetition rate, and determine whether the potential false work order is the false complaint work order according to the work order repetition rate.

Preferably, the complaint work order detection processing device further includes:

the false complaint work order determining module 806 is configured to determine the potential false work order as the false complaint work order if the consistency determination result is that the subject is inconsistent.

Preferably, the consistency judging module includes:

The generated theme obtaining unit is used for processing the processing opinion corresponding to the potential false work order by adopting the target theme model to obtain the generated theme corresponding to the potential false work order.

The real theme obtaining unit is used for obtaining the real theme corresponding to the potential false work order according to the problem outline corresponding to the potential false work order.

And the consistency judging unit is used for carrying out consistency judgment on the generated theme and the real theme corresponding to the potential false work order and obtaining a consistency judging result.

Preferably, the consistency judging unit includes:

The three-dimensional coding matrix acquisition subunit is used for inputting the generated theme and the real theme corresponding to the potential false work order into the improved BERT model, extracting the original codes corresponding to the last four layers of the improved BERT model, and acquiring the three-dimensional coding matrix.

And the pooling code acquisition subunit is used for expanding the three-dimensional coding matrix into a four-dimensional coding matrix, and carrying out average pooling on the four-dimensional coding matrix by adopting an average pooling method to acquire pooling codes.

And the consistency judgment result output subunit is used for inputting the pooled code into the Softmax layer of the improved BERT model and obtaining a consistency judgment result output by the Softmax layer.

Preferably, the complaint work order detection processing device includes:

the historical worksheet data acquisition module is used for acquiring the historical complaint worksheets and worksheet additional data corresponding to the historical complaint worksheets.

The historical result label acquisition module is used for identifying the processing result in the work order additional data and acquiring a result label corresponding to the historical complaint work order.

And the customer satisfaction work order determining module is used for determining the customer satisfaction work order from the historical complaint work orders according to the result labels corresponding to the historical complaint work orders.

And the topic training data acquisition module is used for acquiring topic training data according to the processing opinion and the problem summary corresponding to the customer satisfaction worksheet.

The target topic model acquisition module is used for carrying out model training based on topic training data to acquire a target topic model.

Preferably, the target topic model acquisition module includes:

the original topic model acquisition unit is used for training the topic model network by adopting external training data to acquire an original topic model.

The target topic module acquisition unit is used for training the original topic model by adopting topic training data to acquire a target topic model.

Preferably, the false complaint work order judging module includes:

The date and time weight acquisition unit is used for extracting date and time elements of the processing opinions in the potential false work orders and acquiring date and time weights corresponding to the potential false work orders.

The part-of-speech weight acquisition unit is used for word segmentation of the processing opinion in the potential false work order and acquiring at least two original character strings corresponding to the potential false work order and part-of-speech weights corresponding to each original character string.

The target weight acquisition unit is used for carrying out weight calculation on each original character string by adopting an improved IF-IDF algorithm according to the date and time weight and the part-of-speech weight corresponding to the original character string, so as to acquire the target weight corresponding to each original character string.

And the target character string acquisition unit is used for determining the target character string corresponding to the potential false work order according to the target weights corresponding to the at least two original character strings.

The work order repetition rate obtaining unit is used for calculating the similarity of the target character string corresponding to the potential false work order and the comparison character string in the comparison complaint work order by adopting a similarity algorithm, and obtaining the work order repetition rate.

The specific limitation of the complaint work order detection processing device can be referred to the limitation of the complaint work order detection processing method hereinabove, and the description thereof is omitted here. The above-described respective modules in the complaint work order detection processing apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a server, and the internal structure of which may be as shown in fig. 9. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing data adopted or generated in the process of executing the complaint work order detection processing method. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements a complaint work order detection processing method.

In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor executes the computer program to implement the complaint work order detection processing method in the foregoing embodiment, such as S201-S206 shown in fig. 2, or S201-S206 shown in fig. 3-7, and is not repeated herein. Or the processor performs the functions of each module/unit in this embodiment of the complaint work order detection processing apparatus when executing the computer program, for example, the functions of the to-be-detected work order data acquisition module 801, the potential false work order acquisition module 802, the consistency judgment module 803, the comparative complaint work order acquisition module 804, the false complaint work order judgment module 805, and the false complaint work order determination module 806 shown in fig. 8, which are not repeated here.

In an embodiment, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the method for detecting and processing a complaint work order in the above embodiment is implemented, for example, S201-S206 shown in fig. 2 or S201-S206 shown in fig. 3-7, which are not repeated here. Or when executed by a processor, the computer program implements the functions of each module/unit in the embodiment of the complaint work order detection processing device, for example, the functions of the to-be-detected work order data acquisition module 801, the potential false work order acquisition module 802, the consistency judgment module 803, the comparative complaint work order acquisition module 804, the false complaint work order judgment module 805 and the false complaint work order determination module 806 shown in fig. 8, which are not repeated herein.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.

Claims

1. The complaint work order detection processing method is characterized by comprising the following steps:

identifying processing comments in the potential false work order and problem summaries in the work order additional data, and processing the processing comments corresponding to the potential false work order by adopting a target theme model to acquire a generated theme corresponding to the potential false work order;

Acquiring a real theme corresponding to the potential false work order according to the problem outline corresponding to the potential false work order;

Inputting the generated theme and the real theme corresponding to the potential false work order into an improved BERT model, extracting the original codes corresponding to the last four layers of the improved BERT model, and obtaining a three-dimensional coding matrix;

expanding the three-dimensional coding matrix into a four-dimensional coding matrix, and carrying out average pooling on the four-dimensional coding matrix by adopting an average pooling method to obtain pooling codes;

inputting the pooling code into a Softmax layer of the improved BERT model, and obtaining a consistency judgment result output by the Softmax layer;

2. The method for detecting and processing a complaint work order according to claim 1, wherein after the subject consistency judgment is performed on the processing opinion corresponding to the potential false work order and the problem summary, the method for detecting and processing a complaint work order further comprises:

and if the consistency judging result is that the theme is inconsistent, determining the potential false work order as a false complaint work order.

3. The complaint work order detection processing method as claimed in claim 2, wherein before the obtaining of the complaint work order to be measured and the work order additional data corresponding to the complaint work order to be measured, the complaint work order detection processing method includes:

Acquiring a historical complaint work order and work order additional data corresponding to the historical complaint work order;

identifying a processing result in the work order additional data, and acquiring a result label corresponding to the history complaint work order;

determining a customer satisfaction work order from the history complaint work orders according to the result labels corresponding to the history complaint work orders;

Obtaining theme training data according to the processing opinion and the problem summary corresponding to the customer satisfaction worksheet;

model training is carried out based on the theme training data, and the target theme model is obtained.

4. The complaint work order detection processing method as claimed in claim 3, wherein the model training based on the topic training data to obtain the target topic model includes:

training a topic model network by adopting external training data to obtain an original topic model;

And training the original theme model by adopting the theme training data to obtain a target theme model.

5. The complaint work order detection processing method as claimed in claim 1, wherein the performing the duplication checking processing on the processing opinion in the potential false work order and the processing opinion in the comparative complaint work order to obtain the work order repetition rate includes:

Extracting date and time elements from the processing opinions in the potential false work orders to obtain date and time weights corresponding to the potential false work orders;

The processing opinion in the potential false work order is segmented, and at least two original character strings corresponding to the potential false work order and part-of-speech weights corresponding to the original character strings are obtained;

according to the date-time weight and the part-of-speech weight corresponding to the original character strings, carrying out weight calculation on each original character string by adopting an improved IF-IDF algorithm to obtain a target weight corresponding to each original character string;

determining target character strings corresponding to the potential false work orders according to target weights corresponding to at least two original character strings;

And performing similarity calculation on the target character string corresponding to the potential false work order and the comparison character string in the comparison complaint work order by adopting a similarity algorithm, and obtaining the work order repetition rate.

6. The utility model provides a complaint work order detection processing apparatus which characterized in that includes:

The consistency judging module is used for identifying the processing opinion in the potential false work order and the problem summary in the work order additional data, adopting a target theme model to process the processing opinion corresponding to the potential false work order and obtaining a generated theme corresponding to the potential false work order; acquiring a real theme corresponding to the potential false work order according to the problem outline corresponding to the potential false work order; inputting the generated theme and the real theme corresponding to the potential false work order into an improved BERT model, extracting the original codes corresponding to the last four layers of the improved BERT model, and obtaining a three-dimensional coding matrix; expanding the three-dimensional coding matrix into a four-dimensional coding matrix, and carrying out average pooling on the four-dimensional coding matrix by adopting an average pooling method to obtain pooling codes; inputting the pooling code into a Softmax layer of the improved BERT model, and obtaining a consistency judgment result output by the Softmax layer;

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the complaint work order detection processing method of any one of claims 1 to 5 when the computer program is executed by the processor.

8. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the complaint work order detection processing method as claimed in any one of claims 1 to 5.