CN114942975A

CN114942975A - Customer service monitoring method and device

Info

Publication number: CN114942975A
Application number: CN202210597460.1A
Authority: CN
Inventors: 刘宇琦
Original assignee: Bank of China Ltd
Current assignee: Bank of China Ltd
Priority date: 2022-05-30
Filing date: 2022-05-30
Publication date: 2022-08-26

Abstract

The invention provides a method and a device for monitoring customer service, in particular to the technical field of artificial intelligence, wherein the method comprises the following steps: constructing a training sample according to historical dialogue texts of customer service and users and a preset emotion dictionary; training a preset father neural network model by using the training sample, extracting training parameters of the trained father neural network model, and training a preset son neural network model by using the training parameters and the training sample; obtaining the current satisfaction degree of the user to the customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is the first satisfaction or not, and if so, giving a first alarm to the staff. The invention can automatically and accurately determine the current satisfaction degree of the user to the customer service, and timely give an alarm to the staff when the user satisfaction degree is lower, so that the staff can strengthen the management on the customer service, thereby improving the quality of the customer service and further improving the user experience.

Description

Customer service monitoring method and device

Technical Field

The invention relates to the field of customer service monitoring, in particular to the technical field of artificial intelligence, and particularly relates to a customer service monitoring method and a customer service monitoring device.

Background

In the prior art, the satisfaction evaluation of a customer service is often performed by a requesting user, and the service quality of the customer service is determined according to the satisfaction evaluation condition of the user, so that the customer service is monitored. Because users often work or live with insufficient time, the users are reluctant to evaluate the satisfaction degree of the customer service, or the users fill evaluation contents randomly during the evaluation of the satisfaction degree, so that the satisfaction degree data is insufficient and inaccurate when the customer service is monitored, and the monitoring accuracy of the customer service is low. Moreover, because the judgment of the service quality of the customer service according to the satisfaction evaluation condition of the user is realized manually, the monitoring efficiency of the customer service is low, and the automation degree is poor.

Disclosure of Invention

The invention aims to provide a customer service monitoring method to solve the problems of low accuracy, low efficiency and poor automation degree of monitoring customer service. Another object of the present invention is to provide a customer service monitoring device. It is a further object of this invention to provide such a computer apparatus. It is a further object of the invention to provide a readable medium. It is a further object of the invention to provide a computer program product.

In order to achieve the above object, an aspect of the present invention discloses a customer service monitoring method, including:

constructing a training sample according to historical dialogue texts of customer service and users and a preset emotion dictionary;

training a preset father neural network model by using the training sample, extracting training parameters of the trained father neural network model, and training a preset son neural network model by using the training parameters and the training sample;

obtaining the current satisfaction degree of the user to the customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is the first satisfaction or not, and if so, giving a first alarm to the staff.

Optionally, further comprising:

before constructing a training sample according to the historical dialogue text of the customer service and the user and a preset emotion dictionary, translating the historical dialogue voice of the customer service and the user to obtain an initial historical dialogue text;

and carrying out data cleaning processing on the initial historical dialogue text to obtain the historical dialogue text.

Optionally, the constructing a training sample according to the historical dialogue text of the customer service and the user and a preset emotion dictionary includes:

obtaining a historical user speaking text according to the historical conversation text;

obtaining a speech vocabulary vector of the historical user according to the speech text of the historical user;

obtaining historical satisfaction corresponding to the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary;

and constructing the training sample according to the historical user speaking vocabulary vector and the historical satisfaction.

Optionally, the obtaining of the historical satisfaction corresponding to the speech vocabulary vector of the historical user according to the speech vocabulary vector of the historical user and a preset emotion dictionary includes:

obtaining a satisfaction score of the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary;

and determining the historical satisfaction according to the satisfaction score.

Optionally, the obtaining a satisfaction score of the historical user speech vocabulary vector according to the historical user speech vocabulary vector and a preset emotion dictionary includes:

according to the historical user speaking vocabulary vector and a preset emotion dictionary, obtaining positive vocabularies, positive weight values corresponding to the positive vocabularies, negative vocabularies and negative weight values corresponding to the negative vocabularies in the historical user speaking vocabulary vector;

and accumulating the positive weight values corresponding to all the positive words and the negative weight values corresponding to all the negative words in the historical user speaking word vector to obtain the satisfaction scores of the historical user speaking word vector.

Optionally, the obtaining, according to the historical user speech vocabulary vector and a preset emotion dictionary, a positive vocabulary, a positive weight corresponding to each positive vocabulary, a negative vocabulary, and a negative weight corresponding to each negative vocabulary in the historical user speech vocabulary vector includes:

according to the historical user speaking vocabulary vector and a preset emotion dictionary, obtaining positive vocabularies, initial positive weight values corresponding to the positive vocabularies, negative vocabularies and initial negative weight values corresponding to the negative vocabularies in the historical user speaking vocabulary vector;

judging whether each front word has a corresponding first degree word in the historical user speech word vector, if so, obtaining a first degree coefficient according to the first degree word, and obtaining a front weight corresponding to the front word according to the first degree coefficient and the initial positive weight; if not, taking the initial positive weight value as a positive weight value corresponding to the positive vocabulary;

judging whether each negative vocabulary has a corresponding second degree word in the historical user speaking vocabulary vector, if so, obtaining a second degree coefficient according to the second degree word, and obtaining a negative weight corresponding to the negative vocabulary according to the second degree coefficient and the initial negative weight; if not, the initial negative weight value is used as a negative weight value corresponding to the negative vocabulary.

Optionally, the training a preset sub-neural network model by using the training parameters and the training samples includes:

obtaining a speaking vocabulary vector of the historical user according to the training sample;

obtaining the standard satisfaction degree of the trained father neural network model output according to the training parameters;

training the sub-neural network model with the historical user speech vocabulary vectors and the standard satisfaction.

Optionally, further comprising:

and when the current satisfaction is judged not to be the first satisfaction, judging whether the current satisfaction is the second satisfaction or not, and if so, giving a second alarm to the staff.

Optionally, further comprising:

before the current satisfaction degree of the user to the customer service is obtained according to the current conversation text of the customer service and the user and the trained sub-neural network model, translating the current conversation voice of the customer service and the user to obtain an initial current conversation text;

and carrying out data cleaning processing on the initial current dialog text to obtain the current dialog text.

Optionally, further comprising:

after the standard satisfaction degree output by the trained father neural network model is obtained according to the training parameters, obtaining the temperature parameters of the trained father neural network model according to the training parameters;

correspondingly, before the training the sub-neural network model with the historical user speaking vocabulary vectors and the standard satisfaction degree, the method further comprises the following steps:

configuring the temperature parameter into the sub-neural network model.

In order to achieve the above object, another aspect of the present invention discloses a customer service monitoring apparatus, comprising:

the training sample construction module is used for constructing a training sample according to the historical dialogue texts of the customer service and the user and a preset emotion dictionary;

the training module is used for training a preset father neural network model by using the training sample, extracting training parameters of the trained father neural network model, and training a preset son neural network model by using the training parameters and the training sample;

the alarm module is used for obtaining the current satisfaction degree of the user to the customer service according to the current conversation text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is the first satisfaction or not, and if so, giving a first alarm to the staff.

The invention also discloses a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method when executing the program.

The invention also discloses a computer-readable medium, on which a computer program is stored which, when executed by a processor, implements a method as described above.

The invention also discloses a computer program product comprising a computer program which, when executed by a processor, implements the method as described above.

According to the customer service monitoring method and the customer service monitoring device, the training sample is constructed according to the historical dialogue text of the customer service and the user and the preset emotion dictionary, the training sample can be constructed on the basis of the actual dialogue content of the customer service and the user and the actual emotion of the user to the customer service in the dialogue, so that the training sample is more consistent with the actual situation and is sufficient, the accuracy of the follow-up step on the training of the father neural network is improved, and the accuracy of the follow-up step on the training of the son neural network is indirectly improved; the preset father neural network model is trained by the training sample, the training parameters of the trained father neural network model are extracted, the preset son neural network model is trained by the training parameters and the training sample, and the advantages of high operation accuracy of the father neural network model and high operation speed of the son neural network model can be combined on the basis of the knowledge distillation principle, so that the reliability of the son neural network is improved, and the speed and the accuracy of obtaining the current satisfaction degree of a user to customer service by using the son neural network in the subsequent step are improved. The current satisfaction degree of the user to the customer service is obtained according to the current conversation text of the customer service and the user and the trained sub-neural network model, whether the current satisfaction degree is the first satisfaction degree is judged, if yes, a first alarm is given to a worker, the current satisfaction degree of the user to the customer service can be automatically confirmed, and the worker can be automatically and timely given an alarm when the current satisfaction degree is at the preset low satisfaction degree, so that the customer service can be rapidly and accurately monitored on the basis of not requiring the user to evaluate the satisfaction degree, the efficiency and the automation degree of customer service monitoring are improved, and the user experience is improved. In summary, the method and the device for monitoring the customer service provided by the invention can automatically, quickly and accurately determine the current satisfaction degree of the user to the customer service, and timely give an alarm to the staff when the satisfaction degree of the user is low, so that the staff can strengthen the management of the customer service to improve the quality of the customer service, thereby improving the accuracy, speed and efficiency of monitoring the customer service on the basis of not requiring the user to evaluate the satisfaction degree, and further improving the user experience.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart illustrating a customer service monitoring method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram illustrating an alternative procedure for constructing training samples according to an embodiment of the present invention;

FIG. 3 is a diagram illustrating an alternative step of obtaining historical satisfaction corresponding to a historical user utterance vocabulary vector according to an embodiment of the present invention;

FIG. 4 is a schematic diagram illustrating an alternative procedure for training a predetermined sub-neural network model according to an embodiment of the present invention;

FIG. 5 is a diagram illustrating an alternative process of obtaining current dialog text, in accordance with an embodiment of the present invention;

FIG. 6 is a block diagram of a customer service monitoring device according to an embodiment of the present invention;

FIG. 7 illustrates a schematic diagram of a computer device suitable for use in implementing embodiments of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As used herein, the terms "first," "second," … …, etc. do not denote any order or order, nor are they used to limit the invention, but rather are used to distinguish one element from another element or operation described by the same technical terms.

As used herein, the terms "comprising," "including," "having," "containing," and the like are open-ended terms that mean including, but not limited to.

As used herein with respect to "and/or" any or all combinations.

It should be noted that, in the technical solution of the present invention, the acquisition, storage, use, processing, etc. of the data all conform to the relevant regulations of the national laws and regulations.

The embodiment of the invention discloses a customer service monitoring method, which specifically comprises the following steps as shown in figure 1:

s101: and constructing a training sample according to the historical dialogue texts of the customer service and the user and a preset emotion dictionary.

S102: and training a preset father neural network model by using the training sample, extracting training parameters of the trained father neural network model, and training a preset son neural network model by using the training parameters and the training sample.

S103: obtaining the current satisfaction degree of the user to the customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is a first satisfaction, and if so, giving a first alarm to the staff.

The child neural network model is a lightweight model derived based on the father neural network model, the father neural network model is high in operation accuracy rate and low in operation speed, and the child neural network model is high in operation speed but lower in operation accuracy rate than the father neural network model. Therefore, the preset father neural network model is trained by the training sample, the training parameters of the trained father neural network model are extracted, and the preset son neural network model is trained by the training parameters and the training sample, so that the operation accuracy of the son neural network can be improved to a greater extent, and the current satisfaction can be obtained through rapid and accurate operation. And the training parameters and the training samples of the parent neural network model can be used for training a plurality of child neural network models, so that the method has higher generalization capability and expansion capability, can monitor a plurality of customer service services by utilizing a plurality of child neural network models, and further improves the monitoring efficiency.

For example, the parent neural network model may be, but is not limited to, a RoBERTa model, and correspondingly, the child neural network model may be, but is not limited to, a TinyBERT model.

Illustratively, the obtaining of the current satisfaction degree of the user on the customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model may be, but is not limited to, inputting the current dialog text into the trained sub-neural network model, so that the sub-neural network model performs an operation based on the input to obtain the current satisfaction degree of the user on the customer service.

Illustratively, the first satisfaction level may be, but is not limited to, a preset satisfaction level interval corresponding to the user's extreme dissatisfaction feeling, for example, but not limited to (— ∞, -3), (∞, -5) or [ -5, -3], etc. it should be noted that the first satisfaction level can be determined by those skilled in the art according to practical situations, and the above description is only an example and is not limited thereto.

For example, the first warning may be, but is not limited to, sending information to the staff, such as "the user is extremely unsatisfied with the customer service, please strengthen the management of the customer service to improve the quality of service of the customer service", or sending the current satisfaction together to the staff. It should be noted that, the specific implementation manner of performing the first alarm may be determined by those skilled in the art according to actual situations, and the above description is only an example, and is not limited thereto.

In a preferred embodiment, the current satisfaction degree of the user to the customer service is obtained according to the current dialog text of the customer service and the user and the trained sub-neural network model, and the current satisfaction degree of the user to the customer service may also be obtained by inputting the current dialog text into the trained sub-neural network model, so that the sub-neural network model performs an operation based on the input to obtain a first current satisfaction degree component of the user to the customer service, receiving the manually input satisfaction degree of the user as a second current satisfaction degree component, multiplying the first current satisfaction degree component by a preset first component coefficient to obtain a first sub-satisfaction degree, multiplying the second current satisfaction degree component by a preset second component coefficient to obtain a second sub-satisfaction degree, and overlapping the first sub-satisfaction degree and the second sub-satisfaction degree to obtain the current satisfaction degree. Illustratively, the first component coefficient and the second component coefficient may be determined by those skilled in the art according to practical situations, and the embodiment of the present invention is not limited to this, but the first component coefficient and the second component coefficient need to be equal to 1 after being superimposed. For example, if the first component coefficient is 0.6, the first current satisfaction component is 5, the second component coefficient is 0.4, and the second current satisfaction component is-5, then the first sub-satisfaction coefficient is 3, the second sub-satisfaction coefficient is-2, and then the current satisfaction is 1. It should be noted that the above description is only an example, and no limitation is made to a specific implementation manner of obtaining the current satisfaction degree of the customer service of the user according to the current dialog text of the customer service and the user and the trained sub-neural network model. The optimal implementation mode can obtain the current satisfaction degree by combining the input satisfaction degree of the user and the satisfaction degree calculated by the neural network, so that the current satisfaction degree can better meet the actual satisfaction condition of the user, the accuracy of obtaining the current satisfaction degree is further improved, and the accuracy and the efficiency of monitoring the customer service are further improved.

According to the customer service monitoring method and the customer service monitoring device, the training sample is constructed according to the historical dialogue texts of the customer service and the user and the preset emotion dictionary, the training sample can be constructed on the basis of the actual dialogue content of the customer service and the user and the actual emotion of the user to the customer service in the dialogue, so that the training sample is more consistent with the actual situation and is sufficient, the accuracy of the follow-up steps on the training of a father neural network is improved, and the accuracy of the follow-up steps on the training of a son neural network is indirectly improved; the method is characterized in that a preset father neural network model is trained by the training sample, training parameters of the trained father neural network model are extracted, and a preset son neural network model is trained by the training parameters and the training sample, so that the method can be based on the principle of knowledge distillation and combines the advantages of high accuracy of operation of the father neural network model and high speed of operation of the son neural network model, thereby improving the reliability of the son neural network, and further improving the speed and accuracy of obtaining the current satisfaction degree of a user on customer service by using the son neural network in the subsequent steps. The method comprises the steps of obtaining the current satisfaction degree of a user to customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model, judging whether the current satisfaction degree is the first satisfaction degree, if so, giving a first alarm to a worker, automatically confirming the current satisfaction degree of the user to the customer service, and automatically giving an alarm to the worker in time when the current satisfaction degree is at the preset low satisfaction degree, so that the customer service can be rapidly and accurately monitored on the basis of not requiring the user to evaluate the satisfaction degree, the efficiency and the automation degree of customer service monitoring are improved, and the user experience is improved. In summary, the method and the device for monitoring the customer service provided by the invention can automatically, quickly and accurately determine the current satisfaction degree of the user to the customer service, and timely give an alarm to the staff when the satisfaction degree of the user is low, so that the staff can strengthen the management of the customer service to improve the quality of the customer service, thereby improving the accuracy, speed and efficiency of monitoring the customer service on the basis of not requiring the user to evaluate the satisfaction degree, and further improving the user experience.

In an optional embodiment, further comprising:

Illustratively, the translation may be implemented by, but is not limited to, existing voice-to-text software or platform, and the like, for example, by, but not limited to, flyer-hearing voice-to-text software or a hundredth voice-to-text platform, and the like.

For example, the data cleaning process may be, but not limited to, repairing wrongly written or mispronounced words existing in the text, missing key words in the translation process, or the unsatisfactory degree evaluation information of the user in a manual manner, so that the historical dialogue text is text data with complete expression and correct format, or automatically cleaning the initial historical dialogue text by using an existing data cleaning method such as a spline interpolation method or an average value filling method, so as to obtain the historical dialogue text. It should be noted that, for the specific implementation manner of performing data cleaning processing on the initial historical dialog text to obtain the historical dialog text, the specific implementation manner may be determined by those skilled in the art according to actual situations, and the foregoing description is only an example, and does not limit this.

The initial historical dialogue text is obtained by translating the historical dialogue voices of the customer service and the user, the corresponding historical dialogue text can be automatically, quickly and accurately obtained on the basis of the actual historical dialogue voice, the historical dialogue text is obtained by cleaning the initial historical dialogue text, the error content in the historical dialogue text can be reduced, the accuracy of the obtained historical dialogue text is further improved, the accuracy of a training sample in the subsequent step is improved, and the neural network model can be trained more favorably.

In an alternative embodiment, as shown in fig. 2, the constructing a training sample according to the historical dialog text of the customer service and the user and the preset emotion dictionary includes the following steps:

s201: and obtaining a historical user speech text according to the historical dialogue text.

S202: and obtaining a speech vocabulary vector of the historical user according to the speech text of the historical user.

S203: and obtaining the historical satisfaction corresponding to the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary.

S204: and constructing the training sample according to the historical user speaking vocabulary vector and the historical satisfaction.

Illustratively, the historical dialog text includes but is not limited to a speech text of a user in the historical dialog and a speech text of a customer service, so that the historical user speech text can be directly obtained according to the historical dialog text. When the historical dialogue text is generated, the corresponding speaker mark of the speech text can be marked, and then the speech text of the historical user can be obtained by screening from the historical dialogue text directly according to the speaker mark. It should be noted that, for the specific implementation manner of step S201, it can be determined by those skilled in the art according to practical situations, and the above description is only an example, and is not limited thereto.

For example, the historical user speech word vector is obtained according to the historical user speech text, and may be, but is not limited to, obtaining words included in the historical user speech text by performing word segmentation on the historical user speech text through an existing word segmentation model, a word segmentation program, word segmentation software, or the like, and performing vectorization processing on the words to obtain the historical user speech word vector. It should be noted that, for the specific implementation manner of step S202, it can be determined by those skilled in the art according to practical situations, and the above description is only an example, and is not limited thereto.

For example, the training sample is constructed according to the historical user speaking vocabulary vector and the historical satisfaction, which may be, but is not limited to, constructing the training sample by using the historical user speaking vocabulary vector as an expected input sample and using the historical satisfaction as a corresponding expected output sample. The specific implementation manner of step S204 can be determined by those skilled in the art according to practical situations, and the above description is only an example and is not limited thereto.

For example, the emotion dictionary may be, but is not limited to, an existing web emotion dictionary or a related emotion corpus, or may be an emotion dictionary obtained by performing vocabulary extension on an existing emotion dictionary based on a related service (for example, but not limited to, synonym extension or near synonym extension), or by using a statistical method to find out and extend a vocabulary related to an existing vocabulary in the emotion dictionary involved in the related service. The emotion dictionary includes, but is not limited to, emotion vocabulary, emotion weight corresponding to the emotion vocabulary, and positive and negative kinds of emotion. It should be noted that the selection of the emotion dictionary and the determination of the specific content or format of the emotion dictionary may be determined by those skilled in the art according to actual situations, and the above description is only an example and is not limited thereto.

Through the steps S201 and S202, the historical user speaking text can be refined into vocabulary vectors, and the granularity of input data during the construction of the training sample is refined, so that the granularity of the content of the training sample is smaller, and the accuracy and efficiency of training a model by using the training sample in the subsequent steps can be improved. Through the steps S203 and S204, the historical satisfaction corresponding to the speech text of the historical user can be accurately and quickly obtained by performing emotion analysis on each vocabulary vector based on the standard emotion dictionary, so that the accuracy of the training sample is improved, and the accuracy and efficiency of training a model by using the training sample in the subsequent steps are further improved.

In an alternative embodiment, as shown in fig. 3, the obtaining of the historical satisfaction corresponding to the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary includes the following steps:

s301: and obtaining a satisfaction score of the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary.

S302: and determining the historical satisfaction according to the satisfaction score.

For example, the determining the historical satisfaction degree according to the satisfaction score may be, but is not limited to, obtaining the historical satisfaction degree by multiplying, dividing, adding or subtracting the satisfaction score by a preset coefficient, or rounding the satisfaction score. It should be noted that, for the specific implementation manner of step S302, it can be determined by those skilled in the art according to practical situations, and the above description is only an example, and is not limited thereto.

Through the steps S301 and S302, the emotion analysis accuracy of each vocabulary vector can be improved, and the historical satisfaction degree is determined after the satisfaction score is processed, so that the data content or format and the like of the historical satisfaction degree better conform to the data content or format and the like of a training sample calibrated by a model, the probability of model jamming or errors in the training process caused by the fact that the training sample is not in compliance during model training is reduced, and the speed and the accuracy of the training model are improved.

In an optional embodiment, the obtaining a satisfaction score of the historical user speech vocabulary vector according to the historical user speech vocabulary vector and a preset emotion dictionary includes:

and accumulating the positive weights corresponding to all the positive words and the negative weights corresponding to all the negative words in the historical user speaking word vector to obtain a satisfaction score of the historical user speaking word vector.

Through the steps, the satisfaction score can be obtained by superposing the emotion weight corresponding to each word in the speech, and the input granularity and the calculation granularity according to which the historical satisfaction is determined in the subsequent steps are refined, so that the accuracy of the historical satisfaction corresponding to the speech word vector of the historical user can be obtained, the accuracy of the training sample can be improved, and the speed and the accuracy of the sub-neural network model for calculating the current satisfaction can be improved indirectly.

In an optional embodiment, the obtaining, according to the historical user speech vocabulary vector and a preset emotion dictionary, a positive vocabulary, a negative vocabulary and a negative weight corresponding to each of the positive vocabularies in the historical user speech vocabulary vector includes:

judging whether each negative vocabulary has a corresponding second degree word in the historical user speech vocabulary vector, if so, obtaining a second degree coefficient according to the second degree word, and obtaining a negative weight corresponding to the negative vocabulary according to the second degree coefficient and the initial negative weight; and if not, taking the initial negative weight value as a negative weight value corresponding to the negative vocabulary.

Illustratively, because the emotion dictionary comprises emotion vocabularies, emotion weight values and emotion positive and negative types corresponding to the emotion vocabularies, and the like, the method can directly obtain the front vocabularies, the initial positive weight values corresponding to the front vocabularies, the negative vocabularies and the initial negative weight values corresponding to the negative vocabularies in the speech vocabulary vector of the historical user according to the speech vocabulary vector of the historical user and the preset emotion dictionary by contrasting the speech vocabulary vector of the historical user and the emotion dictionary. For example, if the historical user utterance text is "i like the attitude of customer service but the customer service reaction is slow", and the historical user utterance vocabulary vector is "i", "like", "customer service", "the", "attitude", "but", "customer service", "reaction", or slow ", the following emotion types and weights can be obtained by referring to the emotion dictionary:

"I" emotional category: no weight: is free of

"like" emotional category: front weight: 3 (initial positive weight corresponding to the positive vocabulary)

"customer service", "of", "attitude" emotional category: no weight: is composed of

"but" emotional category: negative weight: -1 (initial negative weight value corresponding to negative vocabulary)

"customer service", "response" emotional categories: no weight: is free of

"Slow" emotion category: negative weight: -1 (initial negative weight value corresponding to negative vocabulary)

It should be noted that, according to the historical user utterance vocabulary vector and a preset emotion dictionary, a specific implementation manner of the positive vocabularies, the initial positive weight corresponding to each of the positive vocabularies, the negative vocabularies, and the initial negative weight corresponding to each of the negative vocabularies in the historical user utterance vocabulary vector is obtained, and may be determined by a person skilled in the art according to an actual situation, and the above description is only an example, and does not limit the present invention.

For example, the determining whether each front vocabulary has a corresponding first degree word in the historical user speaking vocabulary vector may be implemented by, but not limited to, a semantic analysis algorithm or software. The first term may be, but is not limited to, "very", "extremely" or "extreme" and the like. The first degree word corresponding to the speech vocabulary vector is typically, but not limited to, adjacent to the speech vocabulary in the historical user's speech text.

For example, the obtaining of the first degree coefficient according to the first degree word may be implemented by, but not limited to, comparing an emotion dictionary or manually setting the first degree coefficient. For example, if the coefficient set in the emotion dictionary for the word "extraordinary" is 2, the first degree coefficient is 2.

For example, the obtaining of the front weight corresponding to the front vocabulary according to the first degree coefficient and the initial positive weight may be, but is not limited to, by multiplying the first degree coefficient by the initial positive weight or adding the first degree coefficient to the initial positive weight. For example, if the vocabulary vector "like" is preceded by the vocabulary vector "very" and "like" is a positive vocabulary and the initial positive weight is 3, and the first degree coefficient of "very" is 2, then the corresponding positive weight of "like" may be, but is not limited to, 6 (the first degree coefficient is multiplied by the initial positive weight) or 5 (the first degree coefficient is added to the initial positive weight). It should be noted that, for a specific implementation manner of obtaining the front weight corresponding to the front vocabulary according to the first degree coefficient and the initial front weight, the specific implementation manner may be determined by a person skilled in the art according to an actual situation, and the above description is only an example, and does not limit this.

For example, the determining whether each negative vocabulary has a corresponding second-degree word in the historical user speaking vocabulary vector may be implemented by, but not limited to, a semantic analysis algorithm or software. The second degree may be, but is not limited to, an adjective such as "very", "extremely" or "extremely". The second degree word corresponding to the speech vocabulary vector is typically, but not limited to, adjacent to the speech vocabulary in the historical user's speech text.

For example, the obtaining of the second degree coefficient according to the second degree word may be implemented by, but not limited to, comparing an emotion dictionary or manually setting the second degree coefficient. For example, if the coefficient set in the emotion dictionary for the word "very" is 1, the second degree coefficient is 1.

For example, the obtaining of the negative weight corresponding to the negative vocabulary according to the second degree coefficient and the initial negative weight may be, but not limited to, by multiplying the second degree coefficient by the initial negative weight or subtracting the second degree coefficient from the initial negative weight. For example, if the vocabulary vector "offence" is preceded by the vocabulary vector "very" and "offence" is a negative vocabulary and the initial negative weight is-4 and the second degree coefficient of "very" is 2, then the corresponding negative weight of "offence" may be, but is not limited to, -8 (the second degree coefficient is multiplied by the initial negative weight) or-6 (the initial negative weight is subtracted by the second degree coefficient). It should be noted that, for a specific implementation manner of obtaining the negative weight corresponding to the negative vocabulary according to the second degree coefficient and the initial negative weight, the specific implementation manner may be determined by a person skilled in the art according to an actual situation, and the above description is only an example, and does not limit this.

The influence of the degree words on the weight is also considered in the steps, and the weight of each emotion word is determined in a smaller granularity and finer mode, so that the satisfaction score obtained in the subsequent step can better meet the actual satisfaction condition of the user, the accuracy of the historical satisfaction obtained in the subsequent step can be further improved, and the accuracy of the training sample can be further improved to improve the operation accuracy of the trained neural network model.

In an alternative embodiment, as shown in fig. 4, the training of the preset sub-neural network model by using the training parameters and the training samples includes the following steps:

s401: and obtaining the speech word vector of the historical user according to the training sample.

S402: and obtaining the standard satisfaction degree of the trained father neural network model output according to the training parameters.

S403: training the sub-neural network model with the historical user speech vocabulary vectors and the standard satisfaction.

For example, since the training sample includes the historical user speech vocabulary vector, the historical user speech vocabulary vector can be obtained directly according to the training sample.

Illustratively, the training parameters include, but are not limited to, a corresponding standard satisfaction of the trained father neural network model output based on the historical user speech vocabulary vectors, a temperature parameter of the trained father neural network model, and the like, so that the standard satisfaction of the trained father neural network model output can be obtained directly according to the training parameters.

For example, the training the sub-neural network model by using the historical user speech vocabulary vector and the standard satisfaction may be, but is not limited to, training the sub-neural network model by using the historical user speech vocabulary vector as a standard input and using the standard satisfaction as a standard output. For example, for the historical user speaking vocabulary vector group of "i", "like", "customer service", "of", "attitude", "but", "customer service", "reaction", "slow", the corresponding standard satisfaction is 1, the sub-neural network model is trained by using the vector group as standard input and using the standard satisfaction as corresponding standard output, wherein there may be many sets of standard input and corresponding standard output, so as to improve the training intensity and thus improve the calculation accuracy and calculation speed of the trained sub-neural network model.

Through the steps S401 to S403, the sub-neural network model can be trained by taking the standard output of the father neural network and the input in the training sample as new training samples more finely based on the principle of knowledge distillation, so that the high accuracy of the father neural network can be transplanted into the sub-neural network model more quickly and accurately, and the operation accuracy of the sub-neural network model is further improved.

In an optional embodiment, further comprising:

It should be noted that, for the second satisfaction degree, it can be determined by those skilled in the art according to practical situations, and the above description is only an example and not a limitation, but the second satisfaction degree interval cannot intersect with the first satisfaction degree interval, and the minimum value of the second satisfaction degree interval is greater than or equal to the maximum value of the first satisfaction degree interval.

For example, the performing of the second alarm may be, but is not limited to, sending information to the staff, such as "the user is less satisfied with the customer service, please strengthen the management of the customer service to improve the quality of service of the customer service", and the current satisfaction may also be sent to the staff together. It should be noted that, the specific implementation manner of performing the second alarm may be determined by those skilled in the art according to actual situations, and the above description is only an example, and is not limited thereto.

Through the steps, when the user is not satisfied with the customer service but not very satisfied with the customer service, the system can also give an alarm to the staff to enable the staff to strengthen the management of the customer service, so that the efficiency and the quality of monitoring the customer service can be further improved, the quality of service of the customer service can be improved, and the user experience can be better met.

In a preferred embodiment, when the current satisfaction is judged not to be the second satisfaction, whether the current satisfaction is the third satisfaction or the fourth satisfaction is judged, if yes, a more satisfactory message of the user is fed back to the staff, and if yes, a more satisfactory message of the user is fed back to the staff.

For example, the third satisfaction degree may be, but is not limited to, a preset satisfaction degree interval corresponding to a more satisfactory feeling of the user, for example, but not limited to (0, 3) or (0, 2), etc. it should be noted that, for the third satisfaction degree, it can be determined by those skilled in the art according to practical situations, and the above description is only an example and not a limitation, however, the third satisfaction degree interval cannot intersect with the second satisfaction degree interval, and the minimum value of the third satisfaction degree interval is greater than or equal to the maximum value of the second satisfaction degree interval.

It should be noted that, for the fourth satisfaction, it can be determined by those skilled in the art according to practical situations, and the above description is only an example and not a limitation, however, the fourth satisfaction cannot intersect with the third satisfaction, and the minimum value of the fourth satisfaction is greater than or equal to the maximum value of the third satisfaction.

Through the steps, when the user is satisfied or very satisfied with the customer service, the corresponding feedback can be carried out on the staff so that the staff can express or encourage the customer service, therefore, the efficiency and the quality of customer service monitoring can be further improved, the service quality of the customer service can be improved, and the user experience can be better met.

In an alternative embodiment, as shown in fig. 5, the method further comprises the following steps:

s501: and translating the current dialogue speech of the customer service and the user to obtain an initial current dialogue text before obtaining the current satisfaction degree of the user to the customer service according to the current dialogue text of the customer service and the user and the trained sub-neural network model.

S502: and carrying out data cleaning processing on the initial current dialog text to obtain the current dialog text.

For example, the data cleaning process may be, but is not limited to, repairing wrongly written or mispronounced words in the text, missing key words in the translation process, or unreal satisfaction evaluation information of the user in a manual manner, so that the initial current dialog text is text data with a complete expression and a correct format, or automatically cleaning the initial current dialog text by using an existing data cleaning method such as a spline interpolation method or a mean value filling method, so as to obtain the current dialog text. It should be noted that, for the specific implementation manner of performing data cleaning processing on the initial current dialog text to obtain the current dialog text, the specific implementation manner may be determined by those skilled in the art according to actual situations, and the above description is only an example, and does not limit this.

The method comprises the steps of translating current conversation voice of a customer service and a user to obtain an initial current conversation text, automatically, quickly and accurately obtaining a corresponding current conversation text based on the actual current conversation voice, and carrying out data cleaning processing on the initial current conversation text to obtain the current conversation text, so that error content in the current conversation text can be reduced, the accuracy of the obtained current conversation text is further improved, the accuracy of obtaining current satisfaction by taking the current conversation text as input in the subsequent steps is improved, and the accuracy of monitoring the whole customer service is further improved.

In an optional embodiment, further comprising:

configuring the temperature parameter into the sub-neural network model.

The temperature parameter is a configuration parameter in the neural network model, and can influence the operation process of the neural network model, and the temperature parameter is from the trained father neural network model, so that the temperature parameter is configured in the son neural network model, the operation accuracy of the son neural network model can be further improved, and the accuracy of the obtained current satisfaction degree is improved.

Based on the same principle, the embodiment of the present invention discloses a customer service monitoring device 600, as shown in fig. 6, the customer service monitoring device 600 includes:

the training sample construction module 601 is configured to construct a training sample according to the historical dialogue text of the customer service and the user and a preset emotion dictionary.

The training module 602 is configured to train a preset parent neural network model with the training sample, extract training parameters of the trained parent neural network model, and train a preset child neural network model with the training parameters and the training sample.

The alarm module 603 is configured to obtain the current satisfaction degree of the user on the customer service according to the current dialog text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is the first satisfaction or not, and if so, giving a first alarm to the staff.

In an optional embodiment, the apparatus further comprises a historical dialogue speech processing module, configured to:

In an optional embodiment, the training sample construction module 601 is configured to:

In an optional embodiment, the training sample constructing module 601 is configured to:

according to the historical user speech vocabulary vector and a preset emotion dictionary, positive vocabularies, initial positive weight values corresponding to the positive vocabularies, negative vocabularies and initial negative weight values corresponding to the negative vocabularies in the historical user speech vocabulary vector are obtained;

In an optional embodiment, the training module 602 is configured to:

In an optional embodiment, the system further comprises a secondary alarm module, configured to:

In an optional embodiment, the system further comprises a current dialog speech processing module, configured to:

In an optional embodiment, the apparatus further comprises a temperature parameter configuration module, configured to:

configuring the temperature parameter into the sub-neural network model.

Since the principle of the customer service monitoring device 600 for solving the problem is similar to the above method, the implementation of the customer service monitoring device 600 can refer to the implementation of the above method, and is not described herein again.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. A typical implementation device is a computer device, which may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

In a typical example, the computer device comprises in particular a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor implements the method as described above.

Referring now to FIG. 7, shown is a schematic block diagram of a computer device 700 suitable for use in implementing embodiments of the present application.

As shown in fig. 7, the computer device 700 includes a Central Processing Unit (CPU)701, which can perform various appropriate works and processes according to a program stored in a Read Only Memory (ROM)702 or a program loaded from a storage section 708 into a Random Access Memory (RAM)) 703. In the RAM703, various programs and data necessary for the operation of the system 700 are also stored. The CPU701, the ROM702, and the RAM703 are connected to each other via a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

The following components are connected to the I/O interface 705: an input portion 706 including a keyboard, a mouse, and the like; an output section 707 including components such as a Cathode Ray Tube (CRT), a liquid crystal feedback (LCD), and the like, and a speaker; a storage section 708 including a hard disk and the like; and a communication section 709 including a network interface card such as a LAN card, a modem, or the like. The communication section 709 performs communication processing via a network such as the internet. A drive 710 is also connected to the I/O interface 705 as needed. A removable medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 710 as necessary, so that the computer program read out therefrom is mounted as necessary in the storage section 708.

In particular, according to an embodiment of the present invention, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 709, and/or installed from the removable medium 711.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Disks (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

For convenience of description, the above devices are described as being divided into various units by function, respectively. Of course, the functionality of the units may be implemented in one or more software and/or hardware when implementing the present application.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.

The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art to which the present application pertains. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

Claims

1. A customer service monitoring method is characterized by comprising the following steps:

2. The method of claim 1, further comprising:

3. The method of claim 1, wherein the constructing of the training sample according to the historical dialogue texts of the customer service and the user and the preset emotion dictionary comprises:

obtaining historical satisfaction corresponding to the historical user speech vocabulary vector according to the historical user speech vocabulary vector and a preset emotion dictionary;

and constructing the training sample according to the historical user speech vocabulary vector and the historical satisfaction.

4. The method of claim 3, wherein obtaining the historical satisfaction corresponding to the historical user speaking vocabulary vector according to the historical user speaking vocabulary vector and a preset emotion dictionary comprises:

5. The method of claim 4, wherein obtaining a satisfaction score of the historical user utterance vocabulary vector according to the historical user utterance vocabulary vector and a predetermined emotion dictionary comprises:

6. The method of claim 5, wherein the obtaining of the positive vocabulary, the positive weight corresponding to each positive vocabulary, the negative vocabulary and the negative weight corresponding to each negative vocabulary in the historical user utterance vocabulary vector according to the historical user utterance vocabulary vector and a preset emotion dictionary comprises:

judging whether each negative vocabulary has a corresponding second degree word in the historical user speaking vocabulary vector, if so, obtaining a second degree coefficient according to the second degree word, and obtaining a negative weight corresponding to the negative vocabulary according to the second degree coefficient and the initial negative weight; and if not, taking the initial negative weight value as a negative weight value corresponding to the negative vocabulary.

7. The method of claim 1, wherein the training the predetermined sub-neural network model with the training parameters and the training samples comprises:

obtaining the standard satisfaction degree output by the trained father neural network model according to the training parameters;

8. The method of claim 1, further comprising:

9. The method of claim 1, further comprising:

10. The method of claim 7, further comprising:

configuring the temperature parameter into the sub-neural network model.

11. A customer service monitoring device, comprising:

the alarm module is used for obtaining the current satisfaction degree of the user to the customer service according to the current conversation text of the customer service and the user and the trained sub-neural network model; and judging whether the current satisfaction is a first satisfaction, and if so, giving a first alarm to the staff.

12. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1-10 when executing the program.

13. A computer-readable medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-10.

14. A computer program product, characterized in that the computer program product comprises a computer program which, when being executed by a processor, carries out the method of any one of claims 1 to 10.