Customer service information processing method and device

Abstract

The invention provides a customer service information processing method and device, and the method comprises the steps: obtaining historical corpus data, training a natural language understanding initial model, and obtaining a natural language understanding model; training the session management initial model by using the identified intention and entity and the corresponding historical session data to obtain a session management model; and inputting to-be-processed data into the natural language understanding model, identifying a user intention and an entity, and matching historical scene data by the session management model according to the identified user intention and the entity and corresponding historical session data and outputting corresponding operation. According to the invention, the reply time of the customer service staff to a single customer and a single customer message can be reduced, the customer response speed of the customer service staff is improved, the number of customers received in unit time is increased, and the service efficiency is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a customer service information processing method and device. Background Art

[0002] With the development of the Internet, online communication has become an effective way for express delivery, merchants and other companies to communicate with customers. In order to save costs, one customer service representative often has to communicate with multiple customers. Because customers' demands are complicated, the customer service staff themselves have different levels, and during the peak season, one customer has to face many customers, which leads to low service quality and efficiency.

[0003] In the related technology, there are currently some document products for auxiliary queries that can cover most scenarios and facilitate customer service to make inquiries when they are not sure how to respond. However, such products have problems such as low query efficiency, incomplete scenario coverage, and low frequency of script updates. Summary of the invention

[0004] The present disclosure provides a customer service information processing method and device to at least solve the above technical problems existing in the prior art.

[0005] According to a first aspect of the present disclosure, there is provided a method for processing customer information, the method comprising:

[0006] Obtaining historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical session data corresponding to the intent and entity;

[0007] The historical corpus data is used as training data to train the natural language understanding initial model to obtain a natural language understanding model; the intent and entity identified by the natural language understanding model and the corresponding historical session data are used to train the session management initial model to obtain a session management model; the natural language understanding model is used to identify the client user intent and entity information, and the session management model is used to determine the next operation according to the identified intent, entity information and corresponding historical session data;

[0008] The data to be processed is input into the natural language understanding model to identify user intentions and entities. The session management model matches historical scenario data according to the identified user intentions and entities and the corresponding historical session data, and outputs corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the session management model is the scenario data stored during training.

[0009] In one possible implementation, it also includes:

[0010] Iteratively updating the natural language understanding model and the conversation management model; including:

[0011] Input the data to be processed into the natural language understanding model, identify the user intent and entity, determine whether the user intent is an existing intent, and if so, input the user intent and entity and corresponding session data into the session management model; otherwise, after making a basic reply, add or modify the intent or entity based on the data to be processed to update and train the natural language understanding model to obtain an updated natural language understanding model;

[0012] The conversation management model obtains corresponding conversation data according to the intentions and entities identified by the natural language understanding model and determines whether there is matching historical scenario data in the conversation management model. If so, the corresponding operation is output according to the historical scenario data; if not, after making a basic reply, the corresponding process is added or modified and the conversation management model is updated and trained to obtain an updated conversation management model.

[0013] In one possible implementation manner, determining whether the user intention is an existing intention includes:

[0014] Calculate the confidence ranking of the recognized intent as an existing intent, and determine whether the confidence of the first ranked intent is greater than a first threshold;

[0015] When the confidence of the first ranking is greater than the first threshold, the difference between the confidence of the first ranking and the confidence of the second ranking is calculated, and if the difference is higher than the second threshold, it is determined to be an existing intention, otherwise a basic response is performed.

[0016] In one possible implementation, the outputting corresponding operations according to the operations in the historical scenario data includes:

[0017] If it is determined that the user's intention requires a query, the historical scenario data corresponding to the user's intention is replied to the session result, and the session data is saved;

[0018] If it is determined that the user's intention requires the creation of a work order, a work order is created based on the historical scenario data corresponding to the user's intention, and the work order information is filled in based on the user's intention and the entity, and the work order is submitted when the work order needs to be submitted.

[0019] In one possible implementation, it further includes:

[0020] The auxiliary adoption rate, speech adoption rate and recommendation rate of the conversation within the preset time period are counted and displayed.

[0021] In one possible implementation manner, obtaining historical corpus data includes:

[0022] Determine the corresponding intent and entity based on the business scenario;

[0023] Acquire historical conversation messages with the customer from a preset database based on the intent and entity;

[0024] The historical conversation messages are filtered to obtain historical corpus data.

[0025] In one possible implementation, the using the historical corpus data as training data to train the initial model for natural language understanding includes:

[0026] Performing word segmentation processing on the historical corpus data to obtain input words;

[0027] Convert the input word into a word vector using a word embedding language model;

[0028] The word vectors are classified into corresponding intent classes using a classifier.

[0029] In one possible implementation, the natural language understanding model and the conversation management model are both trained using a Rasa framework.

[0030] In one possible implementation, inputting the data to be processed into the natural language understanding model includes:

[0031] After the initial model of natural language understanding is trained, a dictionary is generated in which the key is the string corresponding to the corpus text and the value is the intent name corresponding to the historical corpus data;

[0032] The dictionary is introduced into the intention recognition part in the Rasa framework for judgment. If the data to be processed is in the dictionary, the corresponding dictionary value is directly returned as the result of user intention recognition;

[0033] The data to be processed that obtains the user intent recognition result is input into the natural language understanding model to identify the entity.

[0034] According to a second aspect of the present disclosure, a customer service information processing device is provided, the device comprising:

[0035] An acquisition module is used to obtain historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical session data corresponding to the intent and entity;

[0036] A training module is used to train the natural language understanding initial model using the historical corpus data as training data to obtain a natural language understanding model; and to train the session management initial model using the intent and entity identified by the natural language understanding model and the corresponding historical session data to obtain a session management model; the natural language understanding model is used to identify the client user's intent and entity information, and the session management model is used to determine the next operation based on the identified intent, entity information and the corresponding historical session data;

[0037] The output module is used to input the data to be processed into the natural language understanding model, identify user intentions and entities, and the conversation management model matches historical scenario data according to the identified user intentions and entities and the corresponding historical conversation data, and outputs corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the conversation management model is the scenario data stored during training.

[0038] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0039] at least one processor; and

[0040] a memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the present disclosure.

[0042] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method described in the present disclosure.

[0043] The customer service information processing method and device disclosed in the present invention can reduce the customer service's response time to a single customer and a single customer message, increase the speed at which customer service responds to customers, increase the number of customers received per unit time, and improve service efficiency.

[0044] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, in which:

[0046] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0047] Figure 1 The following is a schematic diagram showing the implementation process of the customer service information processing method of the present disclosure. Figure 1 ;

[0048] Figure 2 The following is a schematic diagram showing the implementation process of the customer service information processing method of the present disclosure. Figure 2 ;

[0049] Figure 3 The following is a schematic diagram showing the implementation process of the customer service information processing method of the present disclosure. Figure 3 ;

[0050] Figure 4 The following is a schematic diagram showing the implementation process of the customer service information processing method of the present disclosure. Figure 4 ;

[0051] Figure 5 A schematic diagram showing the structure of a customer service information processing device according to an embodiment of the present disclosure is shown;

[0052] Figure 6 A schematic diagram showing the structure of a customer service information processing system according to an embodiment of the present disclosure is shown;

[0053] Figure 7 A schematic diagram showing the structure of the client terminal of an embodiment of the present disclosure is shown;

[0054] Figure 8 A schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0055] In order to make the purpose, features, and advantages of the present disclosure more obvious and easy to understand, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0056] In the actual service process of customer service, there are many aspects that affect the quality and efficiency of customer service. Specifically, due to different customer service levels and experiences, the quality of customer service replies varies; the traditional sharing of experience by excellent customer service has limited effect on improving tail customer service; for less common customer intentions, since customer service rarely encounters them, customer service cannot respond in a timely and effective manner, which reduces customer service response time and reply quality;) For information such as express logistics, customer service needs to call other interfaces to copy the waybill number for inquiry, and also needs to copy and paste the query results and make certain additions and deletions before sending them to customers, which is time-consuming and laborious; for the conversation summary in the work order, customer service usually needs to review all conversations with the customer, and then use the mouse to copy and paste the keywords of the customer's message into the conversation summary, and then click Submit to finally submit the corresponding work order; for filling in information such as name and phone number in the work order, customer service needs to find the customer's name and phone number, and then manually copy and paste the name and phone number into the work order for submission.

[0057] In the related art, document products are used for auxiliary query, but they have the following disadvantages:

[0058] 1) Low query efficiency. First, customer service needs to clarify the category of the customer's question, then find more detailed question types under the corresponding category, and then manually copy and paste the corresponding reply into the reply box, and then manually add or delete some fields to reply according to the current conversation scenario.

[0059] 2) The scenario coverage is incomplete. It only covers most common scenarios, but not niche scenarios. As a result, even if the customer service makes an inquiry, they still cannot find the right response words.

[0060] 3) The update frequency of the script is low, and it is impossible to make targeted updates according to different festivals, seasons or even weather.

[0061] A customer service information processing method and device provided by the present application are described below in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, the present application provides a customer service information processing method, the method comprising:

[0063] S101, obtaining historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical conversation data corresponding to the intent and entity;

[0064] It can be understood that the business scenarios in this application may be, for example, express delivery, merchants, etc., the user intention may be, for example, querying express delivery, establishing express delivery work orders, etc.; the entities may be, for example, objects, obligations, personnel, etc.

[0065] In the present application, for example, the historical corpus data is when a user wants to query express logistics, and the historical session data related to the query of the express number can be obtained as the historical corpus data based on entities such as express and order number.

[0066] In some embodiments, obtaining historical corpus data includes:

[0067] Determine the corresponding intent and entity based on the business scenario;

[0068] Acquire historical conversation messages with the customer from a preset database based on the intent and entity;

[0069] The historical conversation messages are filtered to obtain historical corpus data.

[0070] Specifically, in this application, the corresponding intent and entity are determined based on the business scenarios sorted out from previous business, and then the customer's historical conversation messages are collected from the MySQL database through keywords and other methods. After screening by business personnel, they are used as training corpus for the subsequent natural language understanding initial model.

[0071] For example, after determining the intent and entity, the business and product need to determine how the intelligent customer service needs to respond based on the specific intent. For example, if the customer expresses the intention to "urge the delivery" but does not provide the waybill number, the intelligent customer service will then need to respond with a line asking for the waybill number. If the customer does not provide the waybill number but continues to express the intention to "urge the delivery", in addition to continuing to respond with the line asking for the waybill number, the intelligent customer service will also need to count the number of times the customer does not provide the waybill number. When the number of times the customer refuses to provide the waybill number reaches the set value, the set value can be set according to actual needs, for example, it can be set to 3 times, and the intelligent customer service will need to respond with a line that it cannot help you urge the delivery if you do not provide the waybill number.

[0072] S102, using the historical corpus data as training data to train the natural language understanding initial model to obtain a natural language understanding model; using the intent and entity identified by the natural language understanding model and the corresponding historical session data to train the session management initial model to obtain a session management model; the natural language understanding model is used to identify the client user's intent and entity information, and the session management model is used to determine the next operation according to the identified intent, entity information and the corresponding historical session data;

[0073] It should be noted that the natural language understanding model and conversation management model in this application adopt the Rasa framework. Rasa is an open source machine learning framework specifically used to build context-aware AI assistants and chatbots. Therefore, in this application, before training the model, for intent, the sorted intent and corresponding corpus can be stored in an Excel spreadsheet in advance, and the intent and corpus stored in the Excel spreadsheet can be converted into a yml file that can be recognized by Rasa through a script. For entities, according to the requirements of Rasa, the entities are marked in the Rasa format at the entity of the corpus. For entities that need to be identified using regular expressions, the corresponding regular expressions need to be configured. The above file can generally be named nlu.yml. In this way, the processed data can be input into the Rasa framework for training.

[0074] After the business and product processes are sorted out, this application needs to organize historical scenario data and convert the process into historical scenario data (story) that Rasa can recognize. This file can be named story.yml.

[0075] This application includes two models, namely the natural language understanding model and the session management model. The function of the natural language understanding model is mainly to identify the intent of the user message and the entity information therein. The main function of the session management model is to determine what action the intelligent customer service should perform based on the intent and entity information identified in the previous step, combined with the historical session information. The action here can be a direct reply to a message or the execution of a piece of code, etc. The action to be performed depends on the session process combed in the previous step. After the session management model executes the corresponding action, it generally replies to the customer with one or more messages. In the process of training the session management model, the historical scenario data generated by the process combing and conversation will be stored for subsequent session management model prediction.

[0076] It should be noted that in this application, after the initial model of natural language understanding is trained, the model needs to be tested; for the test of the natural language understanding model, a test set of about 5,000 items is first obtained. The corpus in the test set does not appear in the training set, and the proportion of the corpus corresponding to each intent remains consistent with the training set. The requirement is that the weighted accuracy needs to be higher than 90%. The weighted accuracy here refers to the accuracy of each intent multiplied by the proportion of the number of corpora of the intent to all test set corpora (that is, the weight). For regular entities, all recognition is required. The so-called regular entities are entities identified by regular expressions.

[0077] If the test fails and the accuracy of the model intent recognition in the natural language understanding model does not meet expectations, the corresponding reasons can be analyzed by checking the corpus with intent recognition errors. For example, it is possible that the recognition accuracy of intents with less corpus is poor, so the corpus of such intents can be appropriately increased; it is also possible that the corpus of two intents is similar, so it is necessary to confirm with the business and product whether the intents can be merged; if there are too many corpora under one intent, it is necessary to communicate with the business and product whether the intent can be split. If the process in the session management model does not proceed as expected, it is necessary to check whether the files produced in the historical scenario data arrangement are correct. If the executed action contains code, the code needs to be debugged, etc. If the test passes, the natural language understanding model can be deployed on the server to provide services externally.

[0078] S103, inputting the data to be processed into the natural language understanding model, identifying user intentions and entities, the conversation management model matches historical scenario data according to the identified user intentions and entities and the corresponding historical conversation data, and outputting corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the conversation management model is the scenario data stored during training.

[0079] Specifically, after obtaining the trained natural language understanding model and session management model, the model can be used to reply to the user. The user's message is the data to be processed. First, the natural language understanding model understands the user's intention and entity in the data to be processed. The session management model matches the historical scenario data according to the identified user's intention and entity and the corresponding historical session data, and outputs the corresponding operation according to the operation in the historical scenario data. The operation can be a reply to the user's inquiry or a work order can be established according to the user's needs.

[0080] The customer service information processing method provided in this application can reduce the customer service's response time to a single customer and a single customer message, increase the speed at which customer service responds to customers, increase the number of customers received per unit time, and improve service efficiency.

[0081] In some embodiments, Figure 2 As shown, the customer service information processing method provided by this application also includes:

[0082] Iteratively updating the natural language understanding model and the conversation management model; including:

[0083] S201, inputting the data to be processed into the natural language understanding model, identifying the user intent and entity, determining whether the user intent is an existing intent, and if so, inputting the user intent and entity and corresponding session data into the session management model; otherwise, after making a basic reply, adding or modifying the intent or entity based on the data to be processed to update and train the natural language understanding model, and obtaining an updated natural language understanding model;

[0084] S202, the conversation management model obtains corresponding conversation data according to the intention and entity identified by the natural language understanding model and determines whether there is matching historical scenario data in the conversation management model. If so, the corresponding operation is output according to the historical scenario data; if not, after making a basic reply, the corresponding process is added or modified and the conversation management model is updated and trained to obtain an updated conversation management model.

[0085] In some embodiments, determining whether the user intention is an existing intention includes:

[0086] Calculate the confidence ranking of the recognized intent as an existing intent, and determine whether the confidence of the first ranked intent is greater than a first threshold;

[0087] When the confidence of the first ranking is greater than the first threshold, the difference between the confidence of the first ranking and the confidence of the second ranking is calculated, and if the difference is higher than the second threshold, it is determined to be an existing intention, otherwise a basic response is performed.

[0088] Specifically, Figure 3As shown, in this application, the natural language understanding model and the conversation management model can be updated after a preset time period, and the natural language model can also be iteratively updated when it is determined that the recognition accuracy of the natural language understanding model is low or the user's intention cannot be recognized. Specifically, the data to be processed is input into the natural language understanding model, and the model will identify the user's message and give a confidence ranking of the intention recognition. First, the natural language understanding model will determine whether it is an existing intention, and calculate the confidence ranking of the user's intention recognized by the trained natural language understanding initial model. It can be understood that the confidence includes multiple, and it is determined whether the confidence ranked first is higher than the first threshold. If it is higher, the difference between the confidence ranked first and the confidence ranked second is calculated, and it is determined whether the difference is higher than the second threshold. If so, it is determined to be an existing intention, otherwise the intelligent customer service will make a basic reply, and then collect such corpus for screening, communicate with the business and products, whether it is necessary to add or modify the intention, and then train a new natural language understanding model. When the confidence of the first ranking is greater than the first threshold, and the difference between the confidences of the first and second rankings is greater than the second threshold, the user intent and entity are identified, and then the session management model is entered. The session management model first determines whether it is an existing process. The specific judgment logic is to retrieve whether there is matching historical scenario data from the historical scenario data configured in the session management model based on the currently identified intent and entity, and the historical session obtained from the session management model. If there is, the corresponding process will be followed, and if not, the basic response will be replied. Then, the staff can determine whether a new or modified process needs to be added, and then a new session management model can be trained.

[0089] This application can not only improve the accuracy of user intent recognition but also better handle user inquiries by timely updating the natural language understanding model and conversation management model.

[0090] In some embodiments, Figure 4 As shown, the outputting corresponding operations according to the operations in the historical scenario data includes:

[0091] If it is determined that the user's intention requires a query, the historical scenario data corresponding to the user's intention is replied to the session result, and the session data is saved;

[0092] If it is determined that the user's intention requires the creation of a work order, a work order is created based on the historical scenario data corresponding to the user's intention, and the work order information is filled in based on the user's intention and the entity, and the work order is submitted when the work order needs to be submitted.

[0093] Specifically, in this application, when the conversation is in progress, the information about the customer in the work order will be updated in real time, for example, the customer's name, phone number, waybill number, size and category of the work order, etc.; specifically, whenever the natural language understanding model recognizes the intent and the entity on the work order, it will also determine whether it is the same as the previous entity value. If different, the entity on the work order will be updated. When the work order needs to be submitted, the customer service clicks to generate a conversation summary. Among them, the Qwen-32B model can be used to generate whether the entity value is the same as the previous one, so that the work order can be submitted.

[0094] In some embodiments, it also includes:

[0095] The auxiliary adoption rate, speech adoption rate and recommendation rate of the conversation within the preset time period are counted and displayed.

[0096] It is understandable that in this application, after the conversation ends, this application has a statistical function, which can count the corpus recommended by the customer service system, mainly counting the use of natural language understanding models and conversation management models by various outlets and customer service individuals, and counting indicators such as auxiliary adoption rate, speech adoption rate and recommendation rate. For each customer consultation, the conversation management model in this application may recommend multiple speech. The auxiliary adoption volume is for the customer consultation volume. As long as the intelligent customer service recommended speech corresponding to the current customer consultation message has been adopted by the customer service, the count is increased by 1. For example, for a single-sentence customer consultation message, the customer service has three recommended speech. As long as the customer service adopts 1, the auxiliary adoption volume is increased by 1. If 2 or 3 are adopted, the value will not be accumulated. The auxiliary adoption volume is the same as the customer service adopting 1. The auxiliary adoption rate is the auxiliary adoption rate divided by the customer consultation volume. The speech adoption volume is for the intelligent customer service recommendation. Still in the above example, if the customer service adopts 1, the speech adoption volume is increased by 1, if the customer service adopts 2, the speech adoption volume is increased by 2, and so on. The speech adoption rate is the speech adoption volume divided by the total recommendation volume of the intelligent customer service. The recommendation rate refers to whether the intelligent customer service has made recommendations for the current customer consultation message, which is equal to the number of customer consultations with recommendations divided by the total number of customer consultations. The above indicators can well understand the customer service's use of the natural language understanding model and the conversation management model, and then make targeted improvements to the two models.

[0097] In some embodiments, the step of using the historical corpus data as training data to train the initial model for natural language understanding includes:

[0098] Performing word segmentation processing on the historical corpus data to obtain input words;

[0099] Convert the input word into a word vector using a word embedding language model;

[0100] The word vectors are classified into corresponding intent classes using a classifier.

[0101] It is understandable that before formal training, this application first processes the historical corpus data to improve the accuracy of the natural language understanding model, specifically by segmenting the historical corpus data. After Jieba segmentation, it becomes a more fine-grained input word, and then is converted into a word vector through the word embedding language model. Finally, the transformer model is used as a classifier to classify the word vector generated in the second step into a certain intent.

[0102] In some embodiments, inputting the data to be processed into the natural language understanding model includes:

[0103] After the initial model of natural language understanding is trained, a dictionary is generated in which the key is the string corresponding to the corpus text and the value is the intent name corresponding to the historical corpus data;

[0104] The dictionary is introduced into the intention recognition part in the Rasa framework for judgment. If the data to be processed is in the dictionary, the corresponding dictionary value is directly returned as the result of user intention recognition;

[0105] The data to be processed that obtains the user intent recognition result is input into the natural language understanding model to identify the entity.

[0106] It is understandable that in order to improve the recognition accuracy of the natural language understanding model, this application replaces the word embedding model Bert-Base with Roberta-WWM; 2) The above-mentioned Roberta-WWM model is pre-trained using the corpus in the preset database, and the number of pre-trained corpuses is about 5 million. After experimental comparison, replacing the word embedding model and pre-training increased the accuracy of model intent recognition by 1.5% and 7% respectively.

[0107] In this application, in order to make up for the uncertainty of model intent recognition, for example, the corpus appears in the training corpus, but the confidence of the model recognition does not reach the threshold, in order to achieve accurate recognition of the model. After model training, this application generates a dictionary with the key as the string corresponding to the corpus text and the value as the intent name corresponding to the corpus; then this dictionary is introduced in the intent recognition part of the Rasa source code for judgment. If the user message is in the dictionary, the corresponding dictionary value is directly returned as the result of the model intent recognition. In this way, the natural language understanding model no longer needs to identify the user intent in the data to be processed, but only needs to identify the entities in the data to be processed, which saves time and is more accurate in recognition.

[0108] In actual scenarios, there are situations where intent switching occurs. If the process switching is written in the traditional way, it is not only inflexible but also difficult to expand. This application uses historical scenario data, marking core intents and process nodes to make process switching more flexible and meet business needs. When facing high concurrency, this application uses a redis cluster to store conversation history and sets the number of threads so that the Rasa service can provide a certain degree of concurrency. And by optimizing the redis storage content, while reducing the information stored in redis, it can ensure that the main information of multiple rounds of conversations is stored, thereby improving the response speed of the Rasa service. In order to achieve flexible configuration of the speech, the reply speech is configured separately in Apollo to achieve rapid replacement and modification of the reply speech, that is, when only the speech is modified, the model does not need to be retrained, thereby achieving rapid iteration.

[0109] The customer service information processing method provided in this application builds a natural language understanding initial model and a conversation management model based on the Rasa framework. The two models are used to provide high-quality responses to the client, which improves the customer service quality. Most of the work of the customer service staff when responding has changed from manual to automatic, helping the customer service staff to improve service efficiency. This application uses high-quality response scripts, and also makes the scripts take effect immediately after configuration, so that the scripts recommended by the system are more targeted, improving the quality of the customer service response scripts.

[0110] In addition, the technical solution provided by this application covers almost all scenarios of express customer service. For rare user intentions, corresponding speech recommendations can be made, thereby reducing the customer service search time, allowing the customer service to respond faster, and lowering the learning threshold for new customer service. This application can automatically fill in work orders, improving the efficiency of customer service when submitting work orders.

[0111] like Figure 5 As shown, the present application provides a customer service information processing device, the device comprising:

[0112] The acquisition module 501 is used to obtain historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical session data corresponding to the intent and entity;

[0113] The training module 502 is used to train the natural language understanding initial model using the historical corpus data as training data to obtain a natural language understanding model; the session management initial model is trained using the intent and entity identified by the natural language understanding model and the corresponding historical session data to obtain a session management model; the natural language understanding model is used to identify the client user's intent and entity information, and the session management model is used to determine the next operation according to the identified intent, entity information and the corresponding historical session data;

[0114] The output module 503 is used to input the data to be processed into the natural language understanding model, identify user intentions and entities, and the conversation management model matches the historical scenario data according to the identified user intentions and entities and the corresponding historical conversation data, and outputs corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the conversation management model is the scenario data stored during training.

[0115] It is understandable that the customer service information processing method provided in this application can also be applied to a customer service information processing system, such as Figure 6 As shown, the customer service information processing system includes a client, a customer service end and a server end.

[0116] Among them, the client side includes Rasa module, statistics module and work order module, and the server side includes MySQL database, redis database, business query interface and large model interface.

[0117] Among them, Figure 7 As shown in the figure, the Rasa module is used to identify the intent and entity of the customer message, and then predict what action the system should perform next based on the historical conversation; specifically, it includes:

[0118] Natural language understanding, identifying the intent and entity values ​​of user messages.

[0119] Dialogue management, which is divided into dialogue state tracking and dialogue strategy; specifically includes:

[0120] Dialogue state tracking: The main function of dialogue state tracking is to record the dialogue history. It will record key information in the dialogue history, such as intent, entities, and actions performed by the Rasa system, for use by the dialogue strategy. After obtaining the intent and entities recognized by natural language understanding, the dialogue strategy combines the dialogue history recorded by dialogue state tracking and compares it with the dialogue process stored in the dialogue strategy during the training phase to predict what action the Rasa module needs to perform next.

[0121] The statistics module is used to count the auxiliary adoption volume, the speech adoption volume, the recommendation volume, the customer satisfaction volume and other indicators, and then calculate the auxiliary adoption rate, speech adoption rate, recommendation rate and satisfaction rate, etc., to provide data support for the subsequent improvement of the auxiliary intelligent customer service system. Specifically including:

[0122] Adoption statistics are used to count the adoption of the scripts recommended by the auxiliary intelligent customer service within a certain period of time by the customer service staff, and then the usage indicators such as the auxiliary adoption volume (rate) and the script adoption volume (rate) are counted; recommendation statistics are used to count the recommendations of the auxiliary intelligent customer service in customer messages within a certain period of time, and then the recommendation rate is calculated; satisfaction statistics are used to count the satisfaction of each branch organization and each customer service staff within a certain period of time, and then the satisfaction rate is calculated.

[0123] The work order module is used to generate the information that needs to be filled in the work order, such as name, phone number, waybill number, work order size category and work order summary, etc. Specifically including:

[0124] For entity work orders, the latest results of natural language understanding model recognition are used to update the entity values ​​in the work order, thereby realizing automatic filling of entity work orders; for summary work orders, the system will call the Qwen-32B large model to summarize the current time point and all previous conversation contents and generate a conversation summary.

[0125] The server provided by this application,

[0126] The MySQL database is used to store customer messages, messages recommended by assisted intelligent customer service, and messages actually sent by customer service. It also stores data on customer service adoption of the assisted intelligent customer service system, such as the amount of assisted adoption, the amount of script adoption, etc. It also stores the intent and entity information of each customer message identified by the natural language model.

[0127] The redis database is used to store historical conversation records of the current conversation. These historical conversation records mainly contain information such as the intentions and entities of customer messages recognized by the natural language understanding model and the action names predicted by the dialogue management model.

[0128] The big model interface is used to assist the intelligent customer service system. When it confirms that a conversation summary needs to be generated, it will send the preset prompt words and the complete conversation to the big model, and then return the conversation summary, which will be automatically filled in the work order of the current conversation.

[0129] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0130] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0131] like Figure 8As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0132] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0133] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as the customer service information processing method. For example, in some embodiments, the customer service information processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the customer service information processing method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the customer service information processing method in any other appropriate manner (e.g., by means of firmware).

[0134] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0135] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0136] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0137] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, language input, or tactile input).

[0138] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0139] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0140] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0141] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0142] The above is only a specific embodiment of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present disclosure, which should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A customer service information processing method, characterized in that: The method comprises: Obtaining historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical session data corresponding to the intent and entity; The historical corpus data is used as training data to train the natural language understanding initial model to obtain a natural language understanding model; the intent and entity identified by the natural language understanding model and the corresponding historical session data are used to train the session management initial model to obtain a session management model; the natural language understanding model is used to identify the client user intent and entity information, and the session management model is used to determine the next operation according to the identified intent, entity information and corresponding historical session data; The data to be processed is input into the natural language understanding model to identify user intentions and entities. The session management model matches historical scenario data according to the identified user intentions and entities and the corresponding historical session data, and outputs corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the session management model is the scenario data stored during training.

2. The method according to claim 1, characterized in that Also includes: Iteratively updating the natural language understanding model and the conversation management model; including: Input the data to be processed into the natural language understanding model, identify the user intent and entity, determine whether the user intent is an existing intent, and if so, input the user intent and entity and corresponding session data into the session management model; otherwise, after making a basic reply, add or modify the intent or entity based on the data to be processed to update and train the natural language understanding model to obtain an updated natural language understanding model; The conversation management model obtains corresponding conversation data according to the intentions and entities identified by the natural language understanding model and determines whether there is matching historical scenario data in the conversation management model. If so, the corresponding operation is output according to the historical scenario data; if not, after making a basic reply, the corresponding process is added or modified and the conversation management model is updated and trained to obtain an updated conversation management model.

3. The method according to claim 2, characterized in that The determining whether the user intention is an existing intention includes: calculating a confidence ranking of the identified intention as an existing intention, and determining whether the confidence of the first ranked intention is greater than a first threshold; When the confidence of the first ranking is greater than the first threshold, the difference between the confidence of the first ranking and the confidence of the second ranking is calculated, and if the difference is higher than the second threshold, it is determined to be an existing intention, otherwise a basic response is performed.

4. The method according to claim 1, characterized in that: The outputting corresponding operations according to the operations in the historical scenario data includes: If it is determined that the user's intention requires a query, the historical scenario data corresponding to the user's intention is replied to the session result, and the session data is saved; If it is determined that the user's intention requires the creation of a work order, a work order is created based on the historical scenario data corresponding to the user's intention, and the work order information is filled in based on the user's intention and the entity, and the work order is submitted when the work order needs to be submitted.

5. The method according to claim 4, characterized in that Also includes: The auxiliary adoption rate, speech adoption rate and recommendation rate of the conversation within the preset time period are counted and displayed.

6. The method according to claim 1, characterized in that The obtaining of historical corpus data includes: Determine the corresponding intent and entity based on the business scenario; Acquire historical conversation messages with the customer from a preset database based on the intent and entity; The historical conversation messages are filtered to obtain historical corpus data.

7. The method according to claim 1, characterized in that The step of using the historical corpus data as training data to train the initial model for natural language understanding includes: Performing word segmentation processing on the historical corpus data to obtain input words; Convert the input word into a word vector using a word embedding language model; The word vectors are classified into corresponding intent classes using a classifier.

8. The method according to claim 1, characterized in that The natural language understanding model and the conversation management model are both trained using the Rasa framework.

9. The method according to claim 8, characterized in that Inputting the data to be processed into the natural language understanding model includes: After the initial model of natural language understanding is trained, a dictionary is generated in which the key is the string corresponding to the corpus text and the value is the intent name corresponding to the historical corpus data; The dictionary is introduced into the intention recognition part in the Rasa framework for judgment. If the data to be processed is in the dictionary, the corresponding dictionary value is directly returned as the result of user intention recognition; The data to be processed that obtains the user intent recognition result is input into the natural language understanding model to identify the entity.

10. A customer service information processing device, characterized in that: The device comprises: An acquisition module is used to obtain historical corpus data; the historical corpus data includes: the intent and entity corresponding to each business scenario, and the historical session data corresponding to the intent and entity; A training module is used to train the natural language understanding initial model using the historical corpus data as training data to obtain a natural language understanding model; the intent and entity identified by the natural language understanding model and the corresponding historical session data are used to train the session management initial model to obtain a session management model; the natural language understanding model is used to identify the client user intent and entity information, and the session management model is used to determine the next operation according to the identified intent, entity information and corresponding historical session data; The output module is used to input the data to be processed into the natural language understanding model, identify user intentions and entities, and the conversation management model matches historical scenario data according to the identified user intentions and entities and the corresponding historical conversation data, and outputs corresponding operations according to the operations in the historical scenario data; wherein the historical scenario data in the conversation management model is the scenario data stored during training.

Images