CN113780610A

CN113780610A - Customer service portrait construction method and device

Info

Publication number: CN113780610A
Application number: CN202011401450.3A
Authority: CN
Inventors: 何峰; 何刚; 肖翔
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2020-12-02
Filing date: 2020-12-02
Publication date: 2021-12-10
Anticipated expiration: 2040-12-02
Also published as: CN113780610B

Abstract

The invention discloses a customer service portrait construction method and device, and relates to the technical field of computers. One embodiment of the method comprises: determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set, calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm according to the service capability index set, and constructing and outputting a customer service portrait based on the comprehensive evaluation value, wherein the customer service portrait comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene. The method and the system can comprehensively evaluate the real service level of the customer service, overcome the defect of low confidence caused by that only a small amount of session data participates in statistics and a large amount of session customers do not give evaluation, accurately construct a customer service portrait and avoid the problem of inconsistent evaluation standards.

Description

Customer service portrait construction method and device

Technical Field

The invention relates to the technical field of computers, in particular to a customer service portrait construction method and device.

Background

An online customer service system becomes an important component of an e-commerce website, and generally, a merchant manages a series of artificial customer services, how to finely manage the artificial customer services, how to evaluate the quality of the customer service, how to depict the capability of each customer service, and how to distribute the customer service to the most appropriate current scene and customer, which are problems to be solved by customer service portrayal.

The existing customer service portrait construction scheme describes customer service capacity based on a single index, namely, the service capacity of customer service is evaluated only by satisfaction. When the customer service portrayal capability is described through the satisfaction evaluation given by the customer, the customer service system is limited by the requirement of a superior service index in an actual customer service system, if the satisfaction score needs to reach a certain value, the customer service can pertinently send the invitation to the customer in the service process, for example, the customer service perceives that the chat state in the home communication session is harmonious, the customer emotion is positive, the customer probably gives a good comment, the invitation is sent to the customer, if the customer emotion is negative, the customer service avoids without sending the invitation, and some customers usually feel troublesome and do not specially give bad comments to the customer service. Therefore, the service level of the customer service cannot be reflected really based on the evaluation data given by the customer.

In the process of implementing the invention, the inventor finds that at least the following problems exist in the prior art:

the single index cannot comprehensively evaluate the real service level of customer service, has small reference significance for evaluating the customer service ability, only a small amount (for example, 10%) of session data participate in statistics, and a large amount (for example, 90%) of sessions are not evaluated by customers, have low confidence coefficient, cannot accurately construct customer service portraits, and cannot distinguish the problem of inconsistent evaluation standards caused by factors such as scenes, categories and the like.

Disclosure of Invention

In view of this, embodiments of the present invention provide a method and an apparatus for building a customer service profile, which can comprehensively evaluate a real service level of a customer service, overcome a defect of low confidence caused by that only a small amount of session data participates in statistics and a large amount of session customers do not give an evaluation, accurately build a customer service profile, and avoid a problem of inconsistent evaluation criteria.

To achieve the above object, according to an aspect of the embodiments of the present invention, a method for constructing a customer service portrait is provided.

A customer service portrait construction method comprises the following steps: determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set; according to the service capacity index set, calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm; and constructing and outputting a customer service portrait based on the comprehensive evaluation value, wherein the customer service portrait comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene.

Optionally, the determining a session scenario of the customer service session data and a set of service capability indicators of the customer service in the session scenario includes: marking the conversation according to a plurality of dimension information in the customer service conversation data, wherein the marking result comprises the conversation scene and a plurality of category marking values; and taking the session scene as a statistic dimension, and respectively counting each class marking value of the session corresponding to the customer service to obtain a service capacity index set of the customer service in the session scene, wherein each service capacity index value in the service capacity index set corresponds to one class marking value.

Optionally, before marking the session according to the multiple dimension information in the customer service session data, the method includes: the method comprises the steps of obtaining multiple pieces of dimension information of each session from customer service session data, and completing missing dimension information of the sessions which are not successfully obtained by model prediction, wherein the missing dimension information comprises one or more of satisfaction degree grade, solution result, customer or customer service emotion tendency and session scene, the satisfaction degree grade is obtained by session satisfaction degree model prediction, the solution result is obtained by judging whether the session is solved by the model prediction, the customer or customer service emotion tendency is obtained by customer and customer service emotion model prediction, and the session scene is obtained by session scene manual summary model prediction.

Optionally, the session scene artificial summary model, the session satisfaction model and the session resolution model are combined models obtained by combining network structures of a multilayer perceptron, a first recurrent neural network and a second recurrent neural network, and the customer and customer service emotion model is realized based on a long-short term memory neural network.

Optionally, in the combined model, the input of the multi-layer perceptron is a session-level structured feature, and the input of the first recurrent neural network is a customer chat text vector; the input of the second recurrent neural network is a customer service chat text vector; when the satisfaction degree grades are predicted, the output of the combined model is the probability of each satisfaction degree grade corresponding to the conversation; when the solution result is predicted, the output of the combined model is the probability of various solution results corresponding to the session; and when the conversation scene is predicted, the output of the combined model is the probability of the conversation corresponding to various conversation scenes.

Optionally, the session-level structured feature is constructed based on one or more session-level features of response time, session duration, number of customer messages, number of customer service messages, average response duration, whether to switch, type of session consultation, response times and total response duration in the customer service session data; the customer chat text vector is generated based on the customer chat text in the customer service session data; the customer service chat text vector is generated based on the customer service chat text in the customer service session data.

Optionally, the plurality of category marking values include one or more category marking values selected from a group consisting of a session satisfaction evaluation value, a solution evaluation value, a session emotion evaluation value, a receptivity, an average response time and a first resolution rate, wherein the session satisfaction evaluation value is calculated based on the satisfaction level, the solution evaluation value is calculated based on the solution, the session emotion evaluation value is calculated based on the customer or customer service emotion tendency, and the receptivity, the average response time and the first resolution rate are respectively obtained statistically based on one or more types of statistical information in the customer service session data.

Optionally, the taking the session scene as a statistical dimension, respectively performing statistics on each category marking value of the session corresponding to the customer service to obtain a service capability index set of the customer service in the session scene includes: counting one or more service capacity index values of the average satisfaction degree, the average solution rate, the average response time mean, the total reception capacity, the average negative emotion value and the average first solution rate of each session corresponding to the customer service in the session scene to obtain a service capacity index set of the customer service, wherein the average satisfaction is the average value of session satisfaction evaluation values of all sessions corresponding to the customer service, the average resolution is the average of the evaluation values of the resolution result of each session corresponding to the customer service, the average response time length is the average of the average response time lengths of each session corresponding to the customer service, the average first solution rate is the average value of the first solution rates of all the sessions corresponding to the customer service, the average negative emotion value is the average value of negative emotion evaluation values in the session emotion evaluation values of all the sessions corresponding to the customer service, and the total receiving capacity is the sum of the receiving capacities of all the sessions corresponding to the customer service.

Optionally, the calculating, according to the service capability index set, a comprehensive evaluation value of each customer service in the session scenario by using a comprehensive evaluation algorithm includes: for each service capability index, mapping the service capability index value of each customer service to a discrete value in a discrete value set corresponding to the service capability index, wherein the discrete value set is calculated and obtained according to a specific calculation rule based on the maximum service capability index value and the minimum service capability index value in the service capability index of each customer service; sorting the discrete values of the service capacity index values of the customer services aiming at each service capacity index, wherein each service capacity index value of each customer service has a sorting ranking rank; and adding the ranking ranks of various service capability indexes of the customer service, wherein the obtained rank sum value is a comprehensive evaluation value of the customer service in the session scene.

According to another aspect of the embodiment of the invention, a customer service portrait constructing device is provided.

A customer service representation construction apparatus comprising: the index determining module is used for determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set; the evaluation value calculation module is used for calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm according to the service capability index set; and the portrait construction module is used for constructing and outputting customer service portraits based on the comprehensive evaluation values, and the customer service portraits comprise sequencing results of the comprehensive evaluation values of the customer services in the session scene.

Optionally, the index determination module is further configured to: marking the conversation according to a plurality of dimension information in the customer service conversation data, wherein the marking result comprises the conversation scene and a plurality of category marking values; and taking the session scene as a statistic dimension, and respectively counting each class marking value of the session corresponding to the customer service to obtain a service capacity index set of the customer service in the session scene, wherein each service capacity index value in the service capacity index set corresponds to one class marking value.

Optionally, the system further comprises an information obtaining and supplementing module, configured to: the method comprises the steps of obtaining multiple pieces of dimension information of each session from customer service session data, and completing missing dimension information of the sessions which are not successfully obtained by model prediction, wherein the missing dimension information comprises one or more of satisfaction degree grade, solution result, customer or customer service emotion tendency and session scene, the satisfaction degree grade is obtained by session satisfaction degree model prediction, the solution result is obtained by judging whether the session is solved by the model prediction, the customer or customer service emotion tendency is obtained by customer and customer service emotion model prediction, and the session scene is obtained by session scene manual summary model prediction.

Optionally, the index determination module is further configured to: counting one or more service capacity index values of the average satisfaction degree, the average solution rate, the average response time mean, the total reception capacity, the average negative emotion value and the average first solution rate of each session corresponding to the customer service in the session scene to obtain a service capacity index set of the customer service, wherein the average satisfaction is the average value of session satisfaction evaluation values of all sessions corresponding to the customer service, the average resolution is the average of the evaluation values of the resolution result of each session corresponding to the customer service, the average response time length is the average of the average response time lengths of each session corresponding to the customer service, the average first solution rate is the average value of the first solution rates of all the sessions corresponding to the customer service, the average negative emotion value is the average value of negative emotion evaluation values in the session emotion evaluation values of all the sessions corresponding to the customer service, and the total receiving capacity is the sum of the receiving capacities of all the sessions corresponding to the customer service.

Optionally, the evaluation value calculation module is further configured to: for each service capability index, mapping the service capability index value of each customer service to a discrete value in a discrete value set corresponding to the service capability index, wherein the discrete value set is calculated and obtained according to a specific calculation rule based on the maximum service capability index value and the minimum service capability index value in the service capability index of each customer service; sorting the discrete values of the service capacity index values of the customer services aiming at each service capacity index, wherein each service capacity index value of each customer service has a sorting ranking rank; and adding the ranking ranks of various service capability indexes of the customer service, wherein the obtained rank sum value is a comprehensive evaluation value of the customer service in the session scene.

According to yet another aspect of an embodiment of the present invention, an electronic device is provided.

An electronic device, comprising: one or more processors; a memory for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the service portrait construction method provided by the embodiments of the present invention.

According to yet another aspect of an embodiment of the present invention, a computer-readable medium is provided.

A computer readable medium, on which a computer program is stored, the program, when executed by a processor, implements the customer service representation construction method provided by the embodiments of the present invention.

One embodiment of the above invention has the following advantages or benefits: determining a session scene of the customer service session data and a service capability index set of the customer service in the session scene, wherein each customer service corresponds to one service capability index set, the number of the session scenes is one or more, calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm according to the service capability index sets, constructing and outputting a customer service picture based on the comprehensive evaluation value, and the customer service picture comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene. The method can comprehensively evaluate the real service level of the customer service, overcomes the defect of low confidence caused by that only a small amount of session data participates in statistics and a large amount of session customers do not give evaluation, can accurately construct a customer service portrait and avoids the problem of inconsistent evaluation standards.

Further effects of the above-mentioned non-conventional alternatives will be described below in connection with the embodiments.

Drawings

The drawings are included to provide a better understanding of the invention and are not to be construed as unduly limiting the invention. Wherein:

FIG. 1 is a schematic diagram of the main steps of a customer service representation construction method according to one embodiment of the present invention;

FIG. 2 is a schematic diagram of a customer service representation construction flow according to one embodiment of the invention;

FIG. 3 is a schematic diagram of a model architecture of a multi-tier perceptron according to one embodiment of the present invention;

FIG. 4 is a schematic diagram of a model structure of a recurrent neural network, in accordance with one embodiment of the present invention;

FIG. 5 is a schematic diagram of the overall structure of a combined model of MLP and LSTM according to one embodiment of the invention;

FIG. 6 is a schematic diagram of the main blocks of a customer service representation construction apparatus according to one embodiment of the present invention;

FIG. 7 is an exemplary system architecture diagram in which embodiments of the present invention may be employed;

fig. 8 is a schematic structural diagram of a computer system suitable for implementing a terminal device or a server according to an embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present invention are described below with reference to the accompanying drawings, in which various details of embodiments of the invention are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

FIG. 1 is a schematic diagram of the main steps of a customer service portrait construction method according to an embodiment of the present invention.

As shown in fig. 1, the customer service portrait building method according to an embodiment of the present invention includes steps S101 to S103 as follows.

Step S101: determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set.

Step S102: and calculating the comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm according to the service capacity index set.

Step S103: and constructing and outputting a customer service portrait based on the comprehensive evaluation value, wherein the customer service portrait comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene.

There may be a plurality of session scenarios for the customer service session data, and the session scenario in the above steps refers to any session scenario.

In one embodiment, determining a session context for customer service session data and a set of service capability indicators for customer services in the session context includes: marking the conversation according to a plurality of dimension information in the customer service conversation data, wherein the marking result comprises a conversation scene and a plurality of category marking values; and taking the session scene as a statistic dimension, and respectively counting each class marking value of the session corresponding to the customer service to obtain a service capability index set of the customer service in the session scene, wherein each service capability index value in the service capability index set corresponds to one class marking value.

In one embodiment, marking the session according to the dimension information in the customer service session data comprises: the method comprises the steps of obtaining multiple pieces of dimension information of each conversation from customer service conversation data, and for the conversations of which the multiple pieces of dimension information are not successfully obtained, completing missing dimension information in a model prediction mode, wherein the missing dimension information comprises one or more of satisfaction degree grades, solving results, customer or customer service emotional tendencies and conversation scenes, the satisfaction degree grades are obtained through conversation satisfaction degree model prediction, the solving results are obtained through whether the conversation is solved or not, the customer or customer service emotional tendencies are obtained through customer and customer service emotional model prediction, and the conversation scenes are obtained through conversation scene manual summary model prediction.

In one embodiment, the session scene artificial summary model, the session satisfaction model and the session resolution model are based on a combined model obtained by combining network structures of the multilayer perceptron, the first recurrent neural network and the second recurrent neural network, and the customer and customer service emotion model is realized based on the long-short term memory neural network.

In the combined model of the embodiment of the invention, the input of the multilayer perceptron is a session-level structural feature, and the input of the first recurrent neural network is a chat text vector of a customer; the input of the second recurrent neural network is a customer service chat text vector; when the satisfaction degree grades are predicted, the output of the combined model is the probability that the conversation corresponds to each satisfaction degree grade; when the solution result is predicted, the output of the combined model is the probability of the session corresponding to various solution results; when the conversation scene is predicted, the output of the combined model is the probability that the conversation corresponds to various conversation scenes.

The session-level structured features can be constructed based on one or more session-level features of response time, session duration, customer message number, average response duration, whether to switch, session consultation type, response times and total response duration in the customer service session data; the customer chat text vector is generated based on the customer chat text in the customer service session data; the service chat text vector is generated based on the service chat text in the service session data.

The multiple category marking values may include one or more of a session satisfaction evaluation value, a solution result evaluation value, a session emotion evaluation value, a receptivity, an average response time and a first solution rate, wherein the session satisfaction evaluation value is calculated based on a satisfaction grade, the solution result evaluation value is calculated based on a solution result, the session emotion evaluation value is calculated based on a customer or customer service emotion tendency, and the receptivity, the average response time and the first solution rate are respectively obtained by statistics based on one or more types of statistical information in customer service session data.

In one embodiment, taking a session scene as a statistical dimension, respectively counting the marking values of each category of the session corresponding to the customer service to obtain a service capability index set of the customer service in the session scene, including: counting one or more service capacity index values of an average satisfaction degree, an average solution rate, an average response time length mean value, a total reception capacity, an average negative emotion value and an average first solution rate of each session corresponding to customer service in a session scene to obtain a service capacity index set of the customer service, wherein the average satisfaction degree is the mean of session satisfaction degree evaluation values of each session corresponding to the customer service, the average solution rate is the mean of solution result evaluation values of each session corresponding to the customer service, the average response time length mean value is the mean of average response time lengths of each session corresponding to the customer service, the average first solution rate is the mean of first solution rates of each session corresponding to the customer service, the average negative emotion value is the mean of negative emotion evaluation values in the session emotion evaluation values of each session corresponding to the customer service, and the total reception capacity is the sum of the reception capacities of each session corresponding to the customer service.

In one embodiment, calculating a comprehensive evaluation value of each customer service in a session scene through a comprehensive evaluation algorithm according to the service capability index set includes: for each service capability index, mapping the service capability index value of each customer service to a discrete value in a discrete value set corresponding to the service capability index, wherein the discrete value set is calculated and obtained according to a specific calculation rule based on the maximum service capability index value and the minimum service capability index value in the service capability index of each customer service; sorting the discrete values of the service capacity index values of the customer services aiming at each service capacity index, wherein each service capacity index value of each customer service has a sorting ranking rank; and adding the ranking ranks of various service capability indexes of the customer service, wherein the obtained rank sum value is a comprehensive evaluation value of the customer service in a session scene.

The service portrayal refers to a model which can depict the service capability of the service through a series of data indexes, and in the embodiment, specifically, the ranking of the comprehensive evaluation values of each service in a session scene. In a e-commerce service system, a customer service system needs to manage thousands of customer service personnel, and although many departments can be divided by service lines, the relative capacity between the customer services can be evaluated absolutely only by a supervisor, and the service level of the customer service personnel cannot be described more objectively through real data. The customer service portrait can help a supervisor to know the service capability of customer service staff more truly and give more objective performance scores, and on the other hand, the customer service is allocated in a targeted manner through an intelligent scheduling technology, the customer service with high capability value and good service is allocated preferentially, meanwhile, the customer service under each scene is ranked in capability value through division of the scene, and is allocated in sequence from good level to bad level, so that the customer service level and the customer consultation experience are integrally improved. The embodiment of the invention supplements the probability value that the conversation which is not evaluated by the customer is possibly well evaluated through a deep learning technology based on the given evaluation data, and more truly depicts the customer service capability value.

FIG. 2 is a schematic diagram of a customer service representation construction flow according to an embodiment of the invention. The method comprises four stages of model training, session marking, weighting sorting and image output. Described separately below.

In the model training phase, the embodiment of the present invention needs to train four models, which are respectively: a conversation satisfaction degree model, a conversation solution model, a customer and customer service emotion model and a conversation scene manual summary model.

The conversation satisfaction model is used for predicting the probability value of the customer possibly giving a rating grade in the conversation without giving evaluation, and the rating grade is the satisfaction degree grade so as to truly depict the satisfaction degree evaluation level of the actual customer to the customer service.

The historical customer service session data is extracted first, for example, all the customer service session data of the last month can be filtered, the customer service session data may include session data of a plurality of customer services, each customer service corresponds to one or more sessions, so that the session data of each customer service is one or more (one session data corresponds to one-pass session), and each session data may include, but is not limited to, a plurality of dimensional information such as session duration, message number, answer number, chat text message, and the like. Some dimension information in the customer service session data can be counted, and the dimension information is referred to as statistics information in the customer service session data, such as the number of sessions corresponding to a single customer service, the time interval of the customer service in the session for answering the last customer consultation sentence, the number of sessions solved at one time, the total number of sessions, and the like. Wherein, the one-time solution means that the customer does not consult repeatedly for 72 hours in the one-pass session, and then the session is considered as the one-time solution.

And performing data cleaning on the extracted historical customer service session data, wherein the data cleaning specifically comprises information such as filtering system information, order information, name and address, webpage link and the like. The historical customer service session data after data cleaning can be used as a training set to train the session satisfaction model of the embodiment of the invention.

And generating a satisfaction label for the historical customer service session data after data cleaning, specifically, screening only the historical customer service sessions that the customer has made evaluation, wherein the satisfaction grade can be divided into a plurality of grades, for example, 5 grades: the score is 0, 25, 50, 75 and 100, and specifically, the satisfaction degree grade can be defined as required.

And constructing structured features and unstructured features based on historical customer service session data after data cleaning.

The structured features are features which can only be represented by specific numbers, and comprise session-level features such as response time, session duration, customer message number, customer service message number, average response duration, whether to switch, session consultation type, response times, total response duration and the like.

The constructed structural features are subjected to Z-score (Z value) standardization, and the Z-score standardization process can comprise the following steps by taking response time as an example: calculating the average value and standard deviation of each response time population, for any response time, subtracting the calculated average value from the response time to obtain the difference between the response time and the average value, and dividing the difference between the response time and the average value by the calculated standard deviation to obtain the response time after standardization.

The unstructured features specifically refer to text content information of a customer and a customer service chat, namely text features, and the specific method for constructing the unstructured features is as follows: splitting all sentences of the training set according to characters, counting the frequencies of all the characters, numbering from 1 to 9999 at most according to the character frequency in sequence from high to low, wherein the numbering is the encoding of the text characters. For each session data, the customer chat text and the customer service chat text are respectively connected and combined into a long text, namely: for each session data of each customer service, two long texts are generated: one of the texts is a long text of a chat text of a customer (a long text formed by connecting and combining all chat sentences of the customer), and the other text is a long text of a customer service chat text (a long text formed by connecting and combining all chat sentences of the customer service). And performing text conversion on each long text, namely converting each character in the text into the code (one of 1-9999) of the text character. Defining a vector with the length of N1, if the length of text characters in the vector exceeds N1, intercepting the first N1 characters, and supplementing 0 if the length of text characters is less than N1.

The conversation satisfaction model of the embodiment of the invention is a multi-input single-output model, and a plurality of inputs are respectively marked as input 1, input 2 and input 3, wherein:

input 1 is a session-level structural feature, the session-level structural feature can be a structural feature or a structural feature after z-score standardization processing, the dimensionality is (N2, M), N2 is the number of training samples in a training set, each session data is a training sample, and M is a feature number;

input 2 is a customer chat text vector with dimensions of (N2, K), N2 is the number of training samples in the training set, K is the text encoding length, and K is equal to the above vector length N1;

input 3 is a customer service chat text vector, the dimensionality is (N2, J), N2 is the number of training samples in a training set, J is the text encoding length, and J is equal to the vector length N1 above;

and (3) outputting: the customer is rated and classified into 5 rating categories (i.e., 5 satisfaction levels), and is subjected to one-hot processing.

The customer chat text vector and the customer service chat text vector are unstructured, and are generated based on the customer chat text in the customer service session data, namely, the long text of the customer chat text is converted into a vector with the length of N1 (K) after being coded as described above; the service chat text vector is generated based on the service chat text in the service session data, that is, the long text of the service chat text introduced above is encoded and then converted into a vector with the length of N1 (J).

The conversation satisfaction model of the embodiment of the invention is a combined model constructed by utilizing a multilayer perceptron and a recurrent neural network. Wherein, the multilayer perceptron uses a multilayer perceptron (MLP) algorithm to construct a model for the session-level structured features of the input 1. Recurrent neural network for the above-described input 2 and input 3, a model is constructed using a recurrent neural network (specifically, an LSTM (long short term memory neural network)) algorithm. The method comprises the steps of performing network connection on a multi-layer perceptron and a recurrent neural network, for example, combining MLP and LSTM network structures to form a combined model, inputting customer chat text vectors (namely customer text characteristics) into a first recurrent neural network (first LSTM), inputting customer service chat text vectors (namely customer service text characteristics) into a second recurrent neural network (second LSTM), inputting session-level structured characteristics into the multi-layer perceptron, and outputting the outputs of the first recurrent neural network, the second recurrent neural network and the multi-layer perceptron through a connecting layer and a hidden layer and finally through an output layer, wherein when the combined model is used as a session satisfaction model, the output is the probability that a session corresponds to each satisfaction level.

The model structure of the multi-layer perceptron of one embodiment of the invention is shown in fig. 3, and comprises an input layer, a hidden layer and an output layer, wherein the layers of the multi-layer perceptron are all connected, that is, any neuron in the upper layer is connected with all neurons in the lower layer. The model structure of the recurrent neural network according to an embodiment of the present invention is shown in fig. 4, and may include an input layer, an embedded layer, a first bidirectional LSTM, a second bidirectional LSTM, a max pooling layer, an average pooling layer, a connection layer, a hidden layer, a normalization layer, a drop layer (Dropout layer), and an output layer, where the first bidirectional LSTM is cascaded with the second bidirectional LSTM. The overall structure of the composite model of one embodiment of the present invention is shown in fig. 5.

For a session without giving satisfaction evaluation to a customer (namely, a corresponding satisfaction level is not selected), calling the session satisfaction model to predict the satisfaction level of the session, inputting session related features (namely, session level structured features) and conversation text contents (namely, a customer chat text vector and a customer service chat text vector), outputting the probability of the evaluation level (namely, the evaluation category, namely, the satisfaction level) which is possibly given by the customer through the session satisfaction model, and taking the satisfaction level corresponding to the maximum probability as the evaluation level (satisfaction level) given by the customer to the session. The accuracy rate is up to 95% verified by using the data of the last week (not included in the training data set).

The model of whether the session is solved or not is used for predicting the probability value that the customer may give the solution result in the session without giving the solution result evaluation, the solution result is mainly solved and not solved, the problem that the customer requests the customer to solve or consults the customer service is solved, and otherwise, the solution result is not solved. The resolved probability value is the resolution. Whether the model data acquisition, the characteristic construction and the model structure of the session solution model are consistent with the session satisfaction model or not is determined, namely: the model structure also uses the combined model described above, with inputs also being input 1 (session level structured feature), input 2 (customer chat text vector), input 3 (customer service chat text vector) as described above. The training labels are of two types, 0 and 1, representing the customer rating as resolved and the customer rating as unresolved, respectively. Whether the output of the model is solved or not is the probability value of the solving result, and the accuracy of the model can reach 95% by adopting data verification of a week.

The emotion model of the customer and the customer service (emotion model for short) is used for judging whether the customer or the customer service shows negative emotion in one-pass conversation (namely one conversation), generally speaking, the customer shows the negative emotion, which indicates that the customer service is not in place, and indirectly reflects the service level of the customer service; the customer service shows negative emotion, the behavior is absolutely avoided in the customer service industry, the emotion model monitors negative emotion change of the customer service, a management layer is facilitated to master customer service dynamics, and meanwhile the service capability value of the customer service is directly reflected. The emotion model of the embodiment of the invention adopts the LSTM technology which is mature in the industry, short texts are input, namely a sentence in a text of chatting between a customer and a customer service, and probability values of positive emotions and negative emotions of the text are output, namely the probability values of the positive emotions and the negative emotions of the sentence (or called customer or customer service emotional tendencies, wherein if the sentence is from the customer, the probability value represents the emotional tendency of the customer, and if the sentence is from the customer service, the probability value represents the emotional tendency of the customer service). The LSTM employed by the customer and customer service emotion models may use open source algorithms and the implementation techniques will not be described in detail here.

The session scene artificial summary model predicts the probability of a specific consultation scene (i.e. session scene) in which the session of an unreduced scene is likely to be located based on the sessions generalized to the scene. Generally, after a session is ended, a customer service summarizes the session into a specific scene, but statistics is carried out until 40% of the sessions are still not summarized, and if only the summarized sessions are counted, data loss is caused, and a counting result is distorted. Therefore, the embodiment of the invention carries out scene prediction on the conversation of the unreduced scene by training the artificial summary model of the conversation scene, and can avoid data loss and statistic result distortion. The model data acquisition, the feature construction and the model structure of the session scene manual summary model are consistent with the session satisfaction model, namely: namely: the model structure also uses the above combined model, and the inputs are also input 1 (session level structured feature), input 2 (customer chat text vector), input 3 (customer service chat text vector) above. The tags can be divided into 25 session scene categories such as after-sale processing, commodity consultation, commodity price guarantee, session consultation and dispute arbitration, and the session scene categories can be defined according to business needs, and can be not limited to the session scene categories, but can be defined as another session scene category according to needs. The output of the session scene artificial summary model is the probability value of each session scene category corresponding to the session, and the session scene with the highest probability value is taken as the session scene category induced by the session. The model accuracy can reach 75% by adopting data verification of a week.

Marking customer service session data, wherein the marking is mainly used for marking model prediction indexes and statistical indexes for the customer service session data, so that missing dimension information of the session can be supplemented. The marking is done separately for each session.

For part of sessions, multiple pieces of dimension information in customer service session data may be missing, mainly because dimension information such as satisfaction levels and solution results is missing due to customer non-evaluation, or because customer service does not summarize scenes, session scene dimension information is missing, and in addition, customer or customer service emotional tendency dimension information is also missing.

And marking the customer service session data for each session by using the trained models (the session satisfaction model, the session solution model, the customer and customer service emotion model and the session scene manual summary model), so that the dimension information of each session in the customer service session data is complete. In an actual service scene, the customer satisfaction and the solution ratio of the customer is very low, about 10%, the embodiment of the invention is based on a deep learning technology, based on the existing 10% evaluation data training model, supplements the satisfaction and the solution ratio of 90% without evaluation, and more truly describes the customer service capability.

Specifically, model output information such as the probability of the satisfaction level of each session of the customer service session data, the resolution (i.e., the resolved probability value), the probability value of the positive/negative emotion, the probability value of the session scene category, and the like is obtained based on the trained models, and the following model prediction indexes are obtained by using the obtained model output information:

session satisfaction scoring (i.e., session satisfaction evaluation value): the session satisfaction evaluation value is calculated based on the satisfaction level, specifically, the session satisfaction evaluation value is equal to the maximum probability x the category score of the satisfaction level corresponding to the session predicted by the session satisfaction model, the category score, that is, the score corresponding to the satisfaction level corresponding to the maximum probability, is divided into 5 grades, for example, 0 grade, 25 grade, 50 grade, 75 grade and 100 grade, if the probability that a certain session corresponds to the maximum satisfaction level is 75 grade, the probability of the satisfaction level is multiplied by 75 grade, and the session satisfaction evaluation value of the session is obtained;

whether to solve the score (i.e., solution result evaluation value): the solution result evaluation value is calculated based on the solution result, specifically, the solution result evaluation value is equal to the solved probability value (i.e., solution rate) × 100 of whether the session is solved by the model prediction;

conversational emotion scoring (i.e., conversational emotion rating value): the conversation emotion evaluation value is calculated and obtained on the basis of the emotional tendency of the customer or the customer service, specifically, the average value of negative emotion probability values of all sentences in the one-way conversation predicted by the emotional model of the customer and the customer service;

and (3) conversation scene classification: namely, the conversation scene corresponding to the category with the highest probability predicted by the conversation scene artificial summary model.

In addition, a statable index for generating the customer service representation is defined for statable information in the customer service session data, wherein the statable index comprises:

receiving quantity: the number of sessions that the customer service receives in a certain period of time;

average response time length: namely, the customer service answers the average value of the time intervals of the last customer consultation sentence in the conversation;

first solution rate: if the customer does not consult repeatedly within a preset time length (for example, 72 hours) in the one-time session, the session is considered as one-time solution, and the ratio of the number of the one-time solution sessions to the total number of sessions is counted, namely the first solution rate.

The number of sessions received by the customer service in a certain period of time, the time interval for the customer service to answer the last customer consultation sentence in the session, the number of sessions solved at one time, the total number of sessions and the like can be obtained through statistics according to a plurality of dimension information of the customer service session data, and the number of sessions, the average response time and the first solution rate can be obtained through corresponding statistical operation for the statistical information in the customer service session data. For a single session, the throughput is 1 and the first resolution rate is 1 or 0.

And obtaining the marking result of each session based on the model prediction index and the statistical index so as to be used for the next grouping weighted sequencing.

Besides the classification of the conversation scene (namely the conversation scene), the marking result comprises a conversation satisfaction evaluation value, a solution result evaluation value, a conversation emotion evaluation value, a reception quantity, an average response time length and a first resolution rate which are marking values of a plurality of categories.

The grouping weighted sorting of the embodiment of the invention adopts an improved RSR (rank sum ratio comprehensive evaluation method) to carry out comprehensive weighted sorting on the calculated multiple category marking values to obtain a final sorting result of the customer service portrait. The specific implementation method comprises the following steps:

taking a session manual summary scene (i.e., a session scene) as a statistical dimension, calculating a comprehensive capacity score value (i.e., a comprehensive evaluation value) of each customer service in the session scene, and specifically, based on a service capacity index set of each customer service in the session scene (which may be any session scene), calculating a comprehensive evaluation value of each customer service in the session scene by using an improved RSR algorithm (an example of the comprehensive evaluation algorithm in the embodiment of the present invention) in the embodiment of the present invention.

The service capability index set (i.e., multiple targets) of the embodiment of the invention comprises: the method comprises the following steps of average satisfaction, average solution rate, average response time average, total reception capacity, average negative emotion value and average first solution rate, wherein each is a service capability index value. And calculating the service capacity index values according to the marking values of various categories, wherein the service capacity index values are calculated by customer service dimensions, namely the calculated service capacity index values comprise the average satisfaction, the average solution rate, the average response time length average, the total reception capacity, the average negative emotion value, the average first solution rate and the like of each customer service according to the dimension of a session scene (namely under a certain session scene). The specific calculation method is as follows, in a certain session scenario:

the average satisfaction is the average value of session satisfaction evaluation values of all sessions corresponding to customer service;

the average resolution rate is the average value of the evaluation values of the resolution results of each session corresponding to the customer service;

the average response time length mean value is the mean value of the average response time length of each session corresponding to the customer service;

the average first solution rate is the average value of the first solution rates of all the sessions corresponding to the customer service;

the average negative emotion value is the mean value of negative emotion evaluation values in the session emotion evaluation values of all the sessions corresponding to the customer service;

the total reception capacity is the sum of the reception capacities of all the sessions corresponding to the customer service.

Discretizing each service capability index into 100 values based on the service capability index values of all customer services in the session scene, wherein the discretization process specifically comprises the following steps: taking the maximum service ability index value and the minimum service ability index value in the service ability index, dividing the difference value of the maximum service ability index value and the minimum service ability index value by 100, and mapping all the values of the service ability index to the 100 values according to the equidistant value. One example of the 100 values, i.e., sets of discrete values, includes 100 discrete values.

The difference between the maximum service capability index value and the minimum service capability index value is divided by 100, and the 100 values (discrete value set) are obtained by calculation according to the equidistant values, namely the process of calculating the discrete value set according to a specific calculation rule.

The following describes a process of discretization of a service capability index by way of example, taking an average satisfaction degree as an example, assuming that 10000 customer services are counted in total, then 10000 average satisfaction degrees are counted in total, assuming that a maximum value of the average satisfaction degrees is 100 and a minimum value is 0, then the maximum value is subtracted from the minimum value, a difference value is equal to 100, and the difference value is divided by 100, that is, a result is 1, 10000 satisfaction degrees are mapped to 0,2,3,. Then, assuming that one average satisfaction is 66.1, this value is mapped to 66, and another average satisfaction is 66.9, this value is mapped to 67. The mapping rule of the embodiment of the invention can be flexibly set according to actual needs.

In the discrete value set, the minimum discrete value is determined according to the minimum service capability index value, and the maximum discrete value is related to the minimum discrete value, the distance between two adjacent discrete values and the number of the discrete values. For example, in the above example, according to the average satisfaction maximum value of 100 and the minimum value of 0, the distance between two adjacent discrete values in the discrete value set may be determined, and the discrete values are 0,2, 3.

And sequencing the discrete values of the service capacity index values of the customer services aiming at each service capacity index to obtain a discrete value sequence of the service capacity index. Then, each service capability index value of each customer service has a ranking, i.e. the arrangement position of the discrete value of the service capability index value in the discrete value sequence of the service capability index.

When the discrete value sequence is generated, the discrete values of the cost-type indexes (average response time length mean value and average negative emotion value) can be sorted from small to large, and the discrete values of the benefit-type indexes (average satisfaction degree, average solution rate, total receptivity and average first solution rate) can be sorted from large to small.

And for each customer service, summing the ranking of various service capacity index values of the customer service to obtain a rank sum value. For example, the ranking ranks of the discrete values of the service capability indicators, such as the average satisfaction, the average resolution, the average response duration mean, the total receptivity, the average negative emotion value, the average first solution rate, etc., in the respective discrete value sequences are 1, 2,3, 4, 5, and 6, respectively, so that the rank sum value is 1+2+3+4+5+6, which is the comprehensive evaluation value of the customer service in the session scenario, is 21.

And sequencing according to the rank and the value corresponding to each customer service from small to large to obtain a sequencing result of the comprehensive evaluation value of the customer service under the session scene, namely the customer service portrait constructed by the embodiment of the invention.

When the portrait is output, the embodiment of the invention can finally output the ranking results (i.e. comprehensive customer service score ranking) of the comprehensive evaluation values of all the customer services under the first-level scenes (i.e. the scenes of all the conversations) of the manual summary of each conversation. For example:

scene A: customer service A, customer service B, customer service C, … … and customer service N

Scene B: customer service C, customer service A, customer service B, … … and customer service N

Preconditions may also be set, e.g., customer service throughput should be greater than 100 to participate in the ranking statistics.

The customer service portrait constructed based on the embodiment of the invention improves the management efficiency of customer service personnel on one hand, and can match the best customers through the comprehensive evaluation value labels, such as VIP (VIP) customers, to distribute customer services with higher comprehensive evaluation values, thereby improving the customer experience. And moreover, the customer service portrait is constructed in a conversation scene, and the problem of inconsistent evaluation standards caused by the fact that the scene, the category and other factors cannot be distinguished is solved.

FIG. 6 is a schematic diagram of the main modules of a customer service representation construction apparatus according to an embodiment of the present invention.

As shown in fig. 6, the customer service image construction apparatus 600 according to an embodiment of the present invention mainly includes: an index determination module 601, an evaluation value calculation module 602, and a portrait construction module 603.

An index determining module 601, configured to determine a session scenario of the customer service session data and a service capability index set of the customer service in the session scenario, where each customer service corresponds to one service capability index set;

an evaluation value calculation module 602, configured to calculate, according to the service capability index set, a comprehensive evaluation value of each customer service in the session scenario through a comprehensive evaluation algorithm;

and the figure constructing module 603 is configured to construct and output a customer service figure based on the comprehensive evaluation value, where the customer service figure includes a ranking result of the comprehensive evaluation value of each customer service in the session scene.

The index determining module 601 may specifically be configured to: marking the conversation according to a plurality of dimension information in the customer service conversation data, wherein the marking result comprises the conversation scene and a plurality of category marking values; and taking the session scene as a statistic dimension, and respectively counting each class marking value of the session corresponding to the customer service to obtain a service capacity index set of the customer service in the session scene, wherein each service capacity index value in the service capacity index set corresponds to one class marking value.

The customer service representation construction apparatus 600 may further include an information acquisition and supplement module for: the method comprises the steps of obtaining multiple pieces of dimension information of each session from customer service session data, and completing missing dimension information of the sessions which are not successfully obtained by model prediction, wherein the missing dimension information comprises one or more of satisfaction degree grade, solution result, customer or customer service emotion tendency and session scene, the satisfaction degree grade is obtained by session satisfaction degree model prediction, the solution result is obtained by judging whether the session is solved by the model prediction, the customer or customer service emotion tendency is obtained by customer and customer service emotion model prediction, and the session scene is obtained by session scene manual summary model prediction.

The session scene artificial summary model, the session satisfaction degree model and the session solution model are combined models obtained by combining network structures of the multilayer perceptron, the first cyclic neural network and the second cyclic neural network, and the customer and customer service emotion model is realized based on the long-short term memory neural network.

In the combined model, the input of the multilayer perceptron is a session-level structural feature, and the input of the first recurrent neural network is a customer chat text vector; the input of the second recurrent neural network is a customer service chat text vector; when the satisfaction degree grades are predicted, the output of the combined model is the probability of each satisfaction degree grade corresponding to the conversation; when the solution result is predicted, the output of the combined model is the probability of various solution results corresponding to the session; and when the conversation scene is predicted, the output of the combined model is the probability of the conversation corresponding to various conversation scenes.

The session-level structured features are constructed based on one or more session-level features of response time, session duration, customer message number, customer service message number, average response duration, whether to switch, session consultation type, response times and total response duration in the customer service session data; the customer chat text vector is generated based on the customer chat text in the customer service session data; the customer service chat text vector is generated based on the customer service chat text in the customer service session data.

The marking values of multiple categories comprise marking values of one or more categories of a session satisfaction evaluation value, a solution result evaluation value, a session emotion evaluation value, a reception quantity, an average response time and a first resolution rate, wherein the session satisfaction evaluation value is obtained through calculation based on the satisfaction grade, the solution result evaluation value is obtained through calculation based on the solution result, the session emotion evaluation value is obtained through calculation based on the emotion tendency of the customer or the customer service, and the reception quantity, the average response time and the first resolution rate are respectively obtained through statistics based on one or more kinds of statistical information in the customer service session data.

The index determining module 601 may be further specifically configured to: counting one or more service capacity index values of the average satisfaction degree, the average solution rate, the average response time mean, the total reception capacity, the average negative emotion value and the average first solution rate of each session corresponding to the customer service in the session scene to obtain a service capacity index set of the customer service, wherein the average satisfaction is the average value of session satisfaction evaluation values of all sessions corresponding to the customer service, the average resolution is the average of the evaluation values of the resolution result of each session corresponding to the customer service, the average response time length is the average of the average response time lengths of each session corresponding to the customer service, the average first solution rate is the average value of the first solution rates of all the sessions corresponding to the customer service, the average negative emotion value is the average value of negative emotion evaluation values in the session emotion evaluation values of all the sessions corresponding to the customer service, and the total receiving capacity is the sum of the receiving capacities of all the sessions corresponding to the customer service.

The evaluation value calculation module 602 may be further configured to: for each service capability index, mapping the service capability index value of each customer service to a discrete value in a discrete value set corresponding to the service capability index, wherein the discrete value set is calculated and obtained according to a specific calculation rule based on the maximum service capability index value and the minimum service capability index value in the service capability index of each customer service; sorting the discrete values of the service capacity index values of the customer services aiming at each service capacity index, wherein each service capacity index value of each customer service has a sorting ranking rank; and adding the ranking ranks of various service capability indexes of the customer service, wherein the obtained rank sum value is a comprehensive evaluation value of the customer service in the session scene.

In addition, the detailed implementation of the customer service image construction apparatus in the embodiment of the present invention is already described in detail in the above customer service image construction method, and therefore, the repeated description is not repeated here.

FIG. 7 illustrates an exemplary system architecture 700 of a customer service portrait construction method or apparatus to which embodiments of the invention may be applied.

As shown in fig. 7, the system architecture 700 may include

terminal devices

701, 702, 703, a network 704, and a server 705. The network 704 serves to provide a medium for communication links between the

terminal devices

701, 702, 703 and the server 705. Network 704 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

A user may use the

terminal devices

701, 702, 703 to interact with a server 705 over a network 704, to receive or send messages or the like. The

terminal devices

701, 702, 703 may have installed thereon various communication client applications, such as a shopping-like application, a web browser application, a search-like application, an instant messaging tool, a mailbox client, social platform software, etc. (by way of example only).

The

terminal devices

701, 702, 703 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.

The server 705 may be a server providing various services, such as a background management server (for example only) providing support for shopping websites browsed by users using the

terminal devices

701, 702, 703. The backend management server may analyze and perform other processing on the received data such as the product information query request, and feed back a processing result (for example, target push information, product information — just an example) to the terminal device.

It should be noted that the customer service representation constructing method provided by the embodiment of the present invention is generally executed by the server 705, and accordingly, the customer service representation constructing apparatus is generally installed in the server 705.

It should be understood that the number of terminal devices, networks, and servers in fig. 7 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

Referring now to FIG. 8, shown is a block diagram of a computer system 800 suitable for use in implementing a terminal device or server of an embodiment of the present application. The terminal device or the server shown in fig. 8 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.

As shown in fig. 8, the computer system 800 includes a Central Processing Unit (CPU)801 that can perform various appropriate actions and processes in accordance with a program stored in a Read Only Memory (ROM)802 or a program loaded from a storage section 808 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data necessary for the operation of the system 800 are also stored. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input/output (I/O) interface 805 is also connected to bus 804.

The following components are connected to the I/O interface 805: an input portion 806 including a keyboard, a mouse, and the like; an output section 807 including a signal such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 808 including a hard disk and the like; and a communication section 809 including a network interface card such as a LAN card, a modem, or the like. The communication section 809 performs communication processing via a network such as the internet. A drive 810 is also connected to the I/O interface 805 as necessary. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as necessary, so that a computer program read out therefrom is mounted on the storage section 808 as necessary.

In particular, according to the embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer 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 809 and/or installed from the removable medium 811. The computer program executes the above-described functions defined in the system of the present application when executed by the Central Processing Unit (CPU) 801.

It should be noted that the computer readable medium shown in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present invention may be implemented by software or hardware. The described modules may also be provided in a processor, which may be described as: a processor includes an index determination module, an evaluation value calculation module, and a representation construction module. The names of these modules do not in some cases constitute a limitation on the module itself, for example, the index determination module may also be described as a "module for determining a session scenario for servicing session data and a set of service capability indexes for servicing a customer in the session scenario".

As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to comprise: determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set; according to the service capacity index set, calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm; and constructing and outputting a customer service portrait based on the comprehensive evaluation value, wherein the customer service portrait comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene.

According to the technical scheme of the embodiment of the invention, a session scene of customer service session data and a service capability index set of customer services in the session scene are determined, a comprehensive evaluation value of each customer service in the session scene is calculated through a comprehensive evaluation algorithm according to the service capability index set, and a customer service picture is constructed and output based on the comprehensive evaluation value. The method can comprehensively evaluate the real service level of the customer service, overcomes the defect of low confidence caused by that only a small amount of session data participates in statistics and a large amount of session customers do not give evaluation, can accurately construct a customer service portrait and avoids the problem of inconsistent evaluation standards.

The above-described embodiments should not be construed as limiting the scope of the invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions can occur, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

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

determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set;

according to the service capacity index set, calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm;

and constructing and outputting a customer service portrait based on the comprehensive evaluation value, wherein the customer service portrait comprises a sequencing result of the comprehensive evaluation value of each customer service in the session scene.

2. The method of claim 1, wherein determining a session context for the customer service session data and a set of service capability indicators for the customer service in the session context comprises:

marking the conversation according to a plurality of dimension information in the customer service conversation data, wherein the marking result comprises the conversation scene and a plurality of category marking values;

and taking the session scene as a statistic dimension, and respectively counting each class marking value of the session corresponding to the customer service to obtain a service capacity index set of the customer service in the session scene, wherein each service capacity index value in the service capacity index set corresponds to one class marking value.

3. The method of claim 2, wherein said marking sessions according to the plurality of dimensional information in the customer service session data comprises:

the method comprises the steps of obtaining multiple pieces of dimension information of each session from customer service session data, and completing missing dimension information of the sessions which are not successfully obtained by model prediction, wherein the missing dimension information comprises one or more of satisfaction degree grade, solution result, customer or customer service emotion tendency and session scene, the satisfaction degree grade is obtained by session satisfaction degree model prediction, the solution result is obtained by judging whether the session is solved by the model prediction, the customer or customer service emotion tendency is obtained by customer and customer service emotion model prediction, and the session scene is obtained by session scene manual summary model prediction.

4. The method according to claim 3, wherein the session scene artificial summary model, the session satisfaction model and the session resolution model are based on a combined model obtained by combining network structures of a multi-layer perceptron, a first recurrent neural network and a second recurrent neural network, and the customer-service emotion model is realized based on a long-short term memory neural network.

5. The method of claim 4, wherein in the combined model, the input of the multi-tier perceptron is a session-level structured feature, and the input of the first recurrent neural network is a customer chat text vector; the input of the second recurrent neural network is a customer service chat text vector; and the number of the first and second electrodes,

when the satisfaction degree grades are predicted, the output of the combined model is the probability of each satisfaction degree grade corresponding to the conversation;

when the solution result is predicted, the output of the combined model is the probability of various solution results corresponding to the session;

and when the conversation scene is predicted, the output of the combined model is the probability of the conversation corresponding to various conversation scenes.

6. The method of claim 5, wherein the session-level structured features are constructed based on one or more session-level features of response time, session duration, number of customer messages, average response duration, whether to switch, type of session consultation, number of responses, total duration of response in the customer service session data; the customer chat text vector is generated based on the customer chat text in the customer service session data; the customer service chat text vector is generated based on the customer service chat text in the customer service session data.

7. The method of claim 3, wherein the plurality of category marking values comprise one or more of a session satisfaction rating, a solution result rating, a session emotion rating, a receptiveness amount, a mean response time, and a first resolution rate, wherein the session satisfaction rating is calculated based on the satisfaction level, the solution result rating is calculated based on the solution result, the session emotion rating is calculated based on the customer or customer service emotional tendency, and the receptiveness amount, the mean response time, and the first resolution rate are respectively statistically obtained based on one or more statable information in the customer service session data.

8. The method according to claim 7, wherein the taking the session scenario as a statistical dimension, respectively counting the marking value of each category of the session corresponding to the customer service to obtain a service capability index set of the customer service in the session scenario includes:

counting one or more service capacity index values of the average satisfaction degree, the average solution rate, the average response time mean, the total reception capacity, the average negative emotion value and the average first solution rate of each session corresponding to the customer service in the session scene to obtain a service capacity index set of the customer service, wherein the average satisfaction is the average value of session satisfaction evaluation values of all sessions corresponding to the customer service, the average resolution is the average of the evaluation values of the resolution result of each session corresponding to the customer service, the average response time length is the average of the average response time lengths of each session corresponding to the customer service, the average first solution rate is the average value of the first solution rates of all the sessions corresponding to the customer service, the average negative emotion value is the average value of negative emotion evaluation values in the session emotion evaluation values of all the sessions corresponding to the customer service, and the total receiving capacity is the sum of the receiving capacities of all the sessions corresponding to the customer service.

9. The method according to claim 1, wherein calculating a comprehensive evaluation value of each customer service in the session scenario by a comprehensive evaluation algorithm according to the service capability index set comprises:

for each service capability index, mapping the service capability index value of each customer service to a discrete value in a discrete value set corresponding to the service capability index, wherein the discrete value set is calculated and obtained according to a specific calculation rule based on the maximum service capability index value and the minimum service capability index value in the service capability index of each customer service;

sorting the discrete values of the service capacity index values of the customer services aiming at each service capacity index, wherein each service capacity index value of each customer service has a sorting ranking rank;

and adding the ranking ranks of various service capability indexes of the customer service, wherein the obtained rank sum value is a comprehensive evaluation value of the customer service in the session scene.

10. A customer service portrait construction device, comprising:

the index determining module is used for determining a session scene of customer service session data and a service capability index set of customer services in the session scene, wherein each customer service corresponds to one service capability index set;

the evaluation value calculation module is used for calculating a comprehensive evaluation value of each customer service in the session scene through a comprehensive evaluation algorithm according to the service capability index set;

and the portrait construction module is used for constructing and outputting customer service portraits based on the comprehensive evaluation values, and the customer service portraits comprise sequencing results of the comprehensive evaluation values of the customer services in the session scene.

11. An electronic device, comprising:

one or more processors;

a memory for storing one or more programs,

the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-9.

12. 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 1-9.