WO2019232885A1

WO2019232885A1 - Method for testing capabilities of salesman, apparatus, computer device, and storage medium

Info

Publication number: WO2019232885A1
Application number: PCT/CN2018/095025
Authority: WO
Inventors: 金戈; 徐亮; 肖京
Original assignee: 平安科技（深圳）有限公司
Priority date: 2018-06-06
Filing date: 2018-07-09
Publication date: 2019-12-12
Also published as: CN108764753A

Abstract

Disclosed in the present application are a method for testing the capabilities of a salesman, an apparatus, a computer device, and a storage medium, the method comprising: inputting answer text of a salesman responding to preset question text into an emotion analysis model for calculation so as to obtain service mood data of the salesman, thereby determining the capability level of the salesman. The present application simulates a conversation between a customer and a salesman, and according to response content of the salesman, more objectively evaluates the capabilities of the salesman.

Description

Method, device, computer equipment and storage medium for testing salesman ability

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on June 6, 2018, with application number 2018105750106, and the invention name is "Methods, Devices, Computer Equipment, and Storage Media for Testing Salesman Competence", and its entire contents Incorporated by reference in this application.

Technical field

The present application relates to the field of computer technology, and in particular, to a method, an apparatus, a computer device, and a storage medium for testing the ability of a salesman.

Background technique

At present, most of the business ability evaluations of salespeople are based on the feedback obtained by the salesperson after the customer and the sales performance of the salesperson are evaluated. In this way, the ability evaluation is based on the results of the salesperson.

However, in the process of communication with the customer, the salesperson's ability to solve problems and the emotions expressed to the customer through the communication with the customer can all reflect the businessperson's level of business ability.

There is currently no way for the basic salesperson to chat with the customer to judge the ability of the salesperson.

technical problem

The main purpose of this application is to provide a method, a device, a computer device, and a storage medium for testing the ability of a salesperson by simulating a customer chatting with the salesperson.

Technical solutions

In order to achieve the above-mentioned object of the invention, a method for testing a salesman's ability is provided in this application, including:

Controlling the output device to output a preset question text;

Obtaining the answer text of the salesperson's reply to the question text;

Inputting the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and performing calculation to obtain the business sentiment data of the salesman for the question text;

According to the business mood data, the ability level of the salesman is determined.

This application also provides a device for testing the ability of a salesperson, including:

An output module for controlling an output device to output a preset question text;

An obtaining module, configured to obtain an answer text in which a salesperson responds to the question text;

A obtaining module, configured to input the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and perform calculation to obtain the business sentiment data of the salesman for the question text;

A determining module, configured to determine the ability level of a salesperson according to the business mood data.

The present application further provides a computer device including a memory and a processor, where the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, implements the steps of any of the foregoing methods.

The present application further provides a computer non-volatile readable storage medium having computer-readable instructions stored thereon, which when executed by a processor, implement the steps of the method described in any one of the above.

Beneficial effect

The method, device, computer equipment, and storage medium for testing the ability of a salesperson in this application judge the ability of a salesperson by simulating the chat between the customer and the salesperson, obtaining the questions posed by the customer and the response content of the salesperson, by understanding The salesman's business process is used to judge the salesman's ability. When judging the ability of the salesman, the problems are classified according to the emotion of the problem, and the processing ability of the salesman when facing different problems is correspondingly understood, and the ability of the salesman is more comprehensively judged.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic flowchart of a method for testing a salesman's ability according to an embodiment of the present application; FIG.

FIG. 2 is a schematic flowchart of step S4 in the method for testing a salesman's ability according to an embodiment of the present application; FIG.

FIG. 3 is a schematic diagram of a specific process for obtaining emotional text data of a question in the method for testing the ability of a salesman according to an embodiment of the present application; FIG.

FIG. 4 is a schematic flowchart of step S42 in the above step S4 according to an embodiment of the present application;

5 is a schematic flowchart of a method for testing a salesman's ability according to an embodiment of the present application;

6 is a schematic flowchart of a method for testing a salesman's ability according to an embodiment of the present application;

7 is a schematic flowchart of a method for testing a salesman's ability according to an embodiment of the present application;

8 is a schematic block diagram of a structure of an apparatus for testing a salesman's ability according to an embodiment of the present application;

9 is a schematic block diagram of a structure of a determination module of a device for testing a salesman's ability according to an embodiment of the present application;

FIG. 10 is a schematic structural block diagram of an acquisition unit of a device for testing a salesman's ability according to an embodiment of the present application; FIG.

11 is a schematic block diagram of a structure of a determining unit of a device for testing a salesman's ability according to an embodiment of the present application;

FIG. 12 is a schematic block diagram of a structure of an apparatus for testing a salesman's ability according to an embodiment of the present application; FIG.

FIG. 13 is a schematic block diagram of a structure of an apparatus for testing a salesman's ability according to an embodiment of the present application; FIG.

FIG. 14 is a schematic block diagram of a computer device according to an embodiment of the present application.

Best Mode of the Invention

Referring to FIG. 1, an embodiment of the present application provides a method for testing a salesman's ability, including steps:

S1. Control the output device to output a preset question text;

S2. Obtain the answer text of the salesperson's reply to the question text;

S3. Input the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and perform calculations to obtain the salesman's business sentiment data for the question text;

S4. Determine the ability level of the salesman according to the business mood data.

As in step S1 above, the main body of the method execution may be a robot, and the output device of the robot includes a display screen and a speaker. The output device displays a preset question text, that is, when the robot starts to test the salesman, the robot displays the question to the salesman through the display screen, or plays the question to the salesman through the speaker. The question text is some questions set by the administrator in advance, or simulated questions asked by customers to understand the company's products, organized into text form, stored in the robot, a memory card is set inside the robot, or stored in the server The robot accesses the server to obtain the question text. When the salesman needs to test, these question texts are called for display, either by directly displaying the text on the robot's display, or by converting the question text into an audio file and playing it through a speaker. Different question texts represent different questions.

As in step S2 above, the salesman sees the question text, and responds to the question expressed by the question text through input devices such as a keyboard and a touch screen. The robot records the answer text entered by the salesman. After receiving the confirmation from the salesperson, the robot saves the answer text in the memory or server. In another specific embodiment, after the robot receives the voice signal input by the salesperson, the robot converts the voice information into text through speech recognition to obtain the answer text returned by the salesperson.

As in the above S3 step, LSTM (Long Short-Term Memory) is a long-short-term memory network, which is a kind of time-recurrent neural network, which is suitable for processing and predicting important events with relatively long intervals and delays in the time series. LSTM is an effective technique for solving long-order dependence problems. The sentiment analysis model is a model after a large amount of training of the LSTM-CNN model, which is used to obtain a data after a series of calculations on the input text information. The data is business sentiment data, that is, data that expresses the attitudes and feelings of the salesperson. Negative and positive are two opposing feelings. The larger the data, the more positive; the smaller the data, the more negative. The sentiment analysis model judges emotions based on the words and words of the answer text. When training the sentiment analysis model, the staff first defined a large amount of text with different business sentiment data, and then input a large amount of text and the corresponding business sentiment data into the LSTM-CNN model, so that the LSTM-CNN model was formed. An sentiment analysis model that performs sentiment analysis based on text. After the robot enters the answer text of the salesman into the sentiment analysis model, the business sentiment data corresponding to the answer text of the salesman can be output.

As in the above step S4, if the business sentiment data is larger, it means that the salesperson's mood is more positive and the corresponding ability level is higher. Business sentiment data is directly proportional to ability levels. Through the preset mapping rules of business sentiment data and business levels, the ability level of the salesman is determined.

Referring to FIG. 2, further, the above step of determining the ability level of a salesperson based on the business sentiment data includes:

S41. Obtain question mood data of the question text, and determine corresponding question attributes according to the question mood data, where the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

S42. Determine the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data.

In this embodiment, the question text is a simulation of a question posed by a customer, and also has a certain emotion, and the emotion of the question is quantified by the question emotion data. According to the quantified results, each problem is defined as a different type, and each type corresponds to an attribute. The question mood data corresponding to the question text and the attributes corresponding to the question mood data are stored in the memory card in advance by the robot. The robot calls the data in the memory card to obtain question mood data of the question text. Some customers are more prone to emotional excitement and higher requirements. The corresponding speaking emotions will carry some negative emotions. The problem emotional data of the question text is relatively low. The corresponding attributes are negative attributes. Further, the negative attributes include general negative emotions. Attributes and very negative attributes; some customers are easy-going and have little emotional fluctuations, and the corresponding speaking emotions will tend to be flat. The question emotional data of this question text is medium, and the corresponding attributes are neutral attributes; some customers are more active and active, The corresponding speaking emotions will have positive and positive emotions. The problem mood data of the question text is relatively high. The corresponding attributes are positive attributes. Further, the positive attributes include general positive attributes and very positive attributes. Determining the level of competence of a salesperson refers to the fact that the same salesperson has different levels of competence when facing different attribute problems. The following table is a question attribute corresponding to question mood data in a specific embodiment:

During the test, the robot displays at least one question of each question attribute to the salesman, and then receives the reply text of the salesman's reply to all the question property questions. Then calculate the business sentiment data of each reply text, calculate the average value of the business sentiment data of the reply text corresponding to the same question attribute, and then calculate the sum of the average values of the respective question attributes to obtain the corresponding question attributes of the salesman. Ultimate business sentiment data. According to the level corresponding to the final business sentiment data, the salesman is evaluated to the salesman's ability level. Each ability level corresponds to different final business sentiment data. The following table is the result of a robot test on a salesman.

问题文本Question text	问题情绪数据Problem mood data	问题属性Problem attributes	业务情绪数据Business sentiment data
问题文本1Question text 1	-0.8-0.8	非常消极Very negative	-0.3-0.3
问题文本2Question text 2	-0.4-0.4	一般消极Generally negative	0.40.4
问题文本3 Question text 3	0.30.3	中性neutral	0.80.8
问题文本4 Question text 4	0.90.9	非常积极Very positive	0.90.9
问题文本5 Question text 5	0.40.4	一般积极Generally positive	0.80.8
问题文本6 Question text 6	00	中性neutral	00
问题文本7Question text 7	-0.4-0.4	一般消极Generally negative	0.40.4

In the above embodiment, according to different problem attributes, the same business mood data may correspond to different ability levels. In one embodiment, as shown in the above table, the average value of the business sentiment data corresponding to the five question attributes can be calculated respectively. For example, if the problem attribute is very negative and the business sentiment data is -0.3, then the business sentiment data corresponding to the problem attribute is -0.3. There are two business sentiment data corresponding to the general negative attribute, which is 0.4 and 0.4, then the business sentiment data corresponding to the problem attribute is 0.4. According to this calculation method, the business sentiment data corresponding to the five question attributes are -0.3, 0.4, 0.4, 0.8, 0.9. These five values are the salesman ability levels corresponding to the problem attributes. In a specific embodiment, the ability level is divided into three levels of low, medium and high. The business mood data corresponding to the low ability level is -0.5 or less (excluding -0.5), and the business mood data corresponding to the medium ability level is -0.5 to 0.5, the business sentiment data corresponding to the high ability level is above 0.5 (excluding 0.5). Then the corresponding level of competence of the five problem attributes of the salesman is medium, medium, medium, high, and high.

In another specific embodiment, the five numbers are added to obtain 2.2, that is, 2.2 is the final business sentiment data. Then according to 2.2 which specific business level, that is, you can determine which level the salesman belongs to. The final business sentiment data and business level have a set of preset mapping rules.

The above is a calculation method for calculating the final business sentiment data, and other calculation methods may also be used to calculate the final business sentiment data.

Referring to FIG. 3, further, the step of obtaining question mood data of the question text includes:

S411. The question text is converted into an encoded unstructured vector Z by an automatic encoder, and a structural variable C is added to the vector Z, and a tag sequence is generated using the LSTM-RNN method.

S412. Pass the marker sequence through a discriminator.

Turn into problem mood data.

In this embodiment, the question text is a question text extracted from a question set of a question frequently asked by a customer, that is, a text sentence in the following figure. The problem text is converted into a vector by the encoder, that is, the text is vectorized. The vectorization process can use the one-hot Representation model. One-hot Representation is to use a very long vector to represent a word, the length of the vector is the size of the dictionary N, each vector has only one dimension is 1, and the remaining dimensions are all 0, the position of 1 indicates the position of the word in the dictionary . This One-hot Representation is stored in a sparse manner, and the vectorization process is very simple. After the problem text is vectorized by the encoder, a vector Z is obtained, and then a structural variable C is added on the basis of Z. The purpose of increasing C is to make the structure of the vector Z consistent with the subsequent LSTM model, so that the vector Z can be input to Inside the LSTM-RNN model. After entering the LSTM-RNN model, the labeled sequence is obtained

Then

Input to discriminator, discriminator recognizes

Mood, get mood data for question text. Among them, the training process of the discriminator is to train X _L = {(X _L , C _L )} with labeled sentence samples, and obtain the parameter θ _{D of the} discriminator:

In the above formula, D represents the training sample space. After training a large number of trained samples through the discriminator, the generation parameters of the discriminator's emotional data for the question text are obtained. When obtaining the question mood data of the question text, the question text is input to the discriminator, and the question mood data of the question text can be obtained through the formula obtained after training.

Referring to FIG. 4, further, the above-mentioned step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data includes:

S421. Obtain multiple business emotion data corresponding to multiple question texts of the same question attribute;

S422. Use a regularization formula to regularly calculate a plurality of the business mood data and a plurality of the question mood data for the same question attribute;

S423. Obtain the capability level of the salesman corresponding to the same problem attribute according to the calculation result.

In this embodiment, after the salesman answers multiple questions, the question mood data of the multiple questions may be classified according to the question data. Through clustering, the problem sentiment data corresponding to multiple question texts of the same problem attribute and the business sentiment data corresponding to one to one are collected, and the business sentiment data corresponding to multiple same problem attributes is calculated using a normalized formula . For example, through the communication between the robot and the salesman, the emotional data table of the salesman-robot problem text is obtained as follows:

问题情绪数据/MmProblem mood data / Mm	-0.8-0.8	-0.4-0.4	0.30.3	0.90.9	0.40.4	00	-0.4-0.4
业务情绪数据/YmBusiness sentiment data / Ym	-0.3-0.3	0.40.4	0.80.8	0.90.9	0.80.8	00	0.40.4

Then use the normalized formula,

Where i∈j judgment mode:

The round function table is rounded off, | Ym _j | ≤1.

The normalization obtains the following five dimensions of normalized problem sentiment data and business sentiment data:

问题情绪数据/MmProblem mood data / Mm	-1-1	-0.5-0.5	00	0.50.5	11
业务情绪数据/YmBusiness sentiment data / Ym	-0.500-0.500	0.2750.275	0.50.5	0.9370.937	0.9440.944

In this way, the business sentiment data obtained from the salesman's response to the machine sentiment data corresponding to different question texts is calculated, and the ability level of the salesman is determined for different problem attributes. The problem sentiment data is classified after the regularization, and the corresponding business sentiment data corresponding to the classified problem sentiment data is regularly calculated.

Referring to FIG. 5, further, after the step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data, the method includes:

S5. Obtain the highest capability level among the capability levels corresponding to the multiple problem attributes of the salesman;

S6. According to the target problem attribute corresponding to the highest ability level, generating position information corresponding to the target problem attribute as the most suitable position information of the salesman.

As described in step S5 above, the robot sends the problem texts of all the problem attributes to the salesman, and then obtains the capability levels corresponding to all the problem attributes, and each problem attribute corresponds to a capability level. Compare multiple ability levels to get the highest ability level. The highest ability level can be multiple.

As described in step S6 above, the target question attribute refers to the question attribute corresponding to the highest ability level. Each question attribute represents a type of customer. The higher the capability level corresponding to the question attribute, the better the salesperson is at communicating with the person corresponding to the question attribute. Different question attributes correspond to question texts presented by customers at different business stages. Therefore, through the salesman's ability level corresponding to different problem attributes, a suitable job position for the salesman can be generated. For example, if the highest ability level of the salesman A is the corresponding negative attribute, the most suitable job information for the salesman A is the position related to handling complaints. The highest ability level of Salesperson B is the corresponding neutral attribute. Therefore, the most suitable position information for Salesperson B is related to the front desk consultation.

Referring to FIG. 6, further, the steps after determining the ability level of the salesman include:

S7. Adjust the performance score of the salesperson according to the ability level.

As described in step S7 above, performance, from a management perspective, is the desired result of the organization, and is the effective output of the organization at different levels in order to achieve its goals. Performance can represent a person's business ability. The higher the mood data, the higher the enthusiasm for the work, and the higher the corresponding performance score. When the ability level is higher than a certain level, the performance score of the salesperson is improved; when the ability level is lower than a certain level, the performance score of the salesperson is reduced.

Referring to FIG. 7, further, the above-mentioned salesman includes a plurality; the above steps after determining the ability level of the salesman include:

S8. Obtain personal information of a target salesperson with the highest capability level among the respective capability levels corresponding to multiple salespersons, where the personal information includes multiple characteristic information;

S9. Count the number of characteristic information specified in the personal information corresponding to the multiple salespersons.

As described in step S8 above, the robot obtains personal information of a plurality of salesmen with the highest level of capability. Personal information includes multiple characteristics, such as gender characteristics, age characteristics, blood type characteristics, highest academic qualification characteristics, and hometown characteristics. Before the salesman performs the ability test, he needs to enter personal account information. The account information includes the personal information of the salesman, and the robot calls the personal information in the account information of the salesman. In another specific embodiment, the robot pops up a dialog box for filling in personal information before or after the test starts, and then receives the personal information entered by the salesperson; after the test is over, if the salesperson's ability level is advanced, the robot obtains the Personal information entered by the salesperson; if the ability level is not advanced after the test is completed, the personal information entered by the salesperson is not obtained.

As described in step S9 above, the personal information includes multiple features, and each feature includes one feature information. For example, the gender characteristics include two characteristics information of male and female; the age characteristics include four characteristic information of post-70s, 80s, 90s, and other; the blood type characteristics include four characteristic information of O, A, B, and AB. Count the number of specified feature information, so as to find common features with the highest level of competence, so that when the staff recruits staff in the later stage, it will be preferred to recruit those with a higher proportion.

Further, after calculating the number of designated feature information, divide by the number of the most senior salesperson to obtain the proportion of each feature information. In a specific embodiment, a total of 10,000 salespeople participate in the test, and the output result is that there are 2,000 salespeople with the highest rank. Then obtain the personal information of these two thousand salespersons, and finally output a form as follows:

After the robot calculates this ratio, it is convenient for the staff to give priority to those with a higher ratio when recruiting staff in the later period. Priority will be given to male applicants with post 90 blood type O blood.

Further, the personal information corresponding to the salesman with the lowest ability level can also be obtained, and the proportion of each characteristic information in the statistics can also be obtained. When facilitating the recruitment of staff, the applicants with a higher proportion of characteristic information in this program will be eliminated first.

In summary, the method for testing the salesman's ability of this application is to judge the ability of the salesman by simulating the chat between the customer and the salesman, and obtaining the questions posed by the customer and the response of the salesman. Business process to judge the competence of the salesperson. When judging the ability of the salesman, the problems are classified according to the emotion of the problem, and the processing ability of the salesman when facing different problems is correspondingly understood, and the ability of the salesman is more comprehensively judged.

Referring to FIG. 8, an embodiment of the present application further provides a device for testing a salesman's ability, including:

An output module 1 for controlling an output device to output a preset question text;

A first obtaining module 2 configured to obtain an answer text in which a salesperson responds to the question text;

The obtaining module 3 is configured to input the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and perform calculation to obtain the business sentiment data of the salesman for the question text;

A determining module 4 is configured to determine a capability level of a salesperson according to the business mood data.

The execution subject in this embodiment may be a robot. The output device of the output module 1 of the robot includes a display screen and a speaker. The output device displays a preset question text, that is, when the robot starts to test the salesman, the output module 1 displays the question to the salesman through the display screen, or the output module 1 plays the question to the salesman through the speaker. The question text is some questions set by the administrator in advance, or simulated questions asked by customers to understand the company's products, organized into text form, stored in the robot, a memory card is set inside the robot, or stored on the server, The robot accesses the server to obtain the question text. When the salesman needs to test, call these question texts for display. The output module 1 can directly display the text on the robot's display, or the output module 1 can convert the question text into an audio file and use the speaker to perform the test. Play. Different question texts represent different questions.

The salesman sees the question text, and responds to the question expressed by the question text through input devices such as the keyboard and touch screen. The robot records the answer text entered by the salesman. After receiving the confirmation information from the salesperson, the first obtaining module 2 saves the answer text in a memory or a server. In another specific embodiment, after receiving the voice signal input by the salesperson, the first acquisition module 2 converts the voice information into text through speech recognition, and obtains the answer text returned by the salesperson.

LSTM (Long Short-Term Memory) is a short-term short-term memory network, a time-recurrent neural network, suitable for processing and predicting important events with relatively long intervals and delays in time series. LSTM is an effective technique for solving long-order dependence problems. The sentiment analysis model is a model after a large amount of training of the LSTM-CNN model, which is used to obtain a data after a series of calculations on the input text information. The data is business sentiment data, that is, data that expresses the attitudes and feelings of the salesperson. Negative and positive are two opposing feelings. The larger the data, the more positive; the smaller the data, the more negative. The sentiment analysis model judges emotions based on the words and words of the answer text. When training the sentiment analysis model, the staff first defined a large amount of text with different business sentiment data, and then input a large amount of text and the corresponding business sentiment data into the LSTM-CNN model, so that the LSTM-CNN model was formed. An sentiment analysis model that performs sentiment analysis based on text, and then gets module 3 to input the salesman's answer text into the sentiment analysis model, then the business sentiment data corresponding to the salesman's answer text can be output.

If the business sentiment data is larger, it means that the salesperson's mood is more positive and the corresponding ability level is higher. Business sentiment data is directly proportional to ability levels. Based on the preset mapping rules of business sentiment data and business levels, the determination module 4 determines the ability level of the salesperson.

Referring to FIG. 9, the above determining module 4 further includes:

The obtaining unit 41 is configured to obtain question mood data of the question text, and determine question attributes based on the question mood data, where the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

A determining unit 42 is configured to determine an ability level of the salesman corresponding to the problem attribute according to the problem attribute and the business mood data.

In this embodiment, the question text is a simulation of a question posed by a customer, and also has a certain emotion, and the emotion of the question is quantified by the question emotion data. According to the quantified results, each problem is defined as a different type, and each type corresponds to an attribute. The question mood data corresponding to the question text and the attributes corresponding to the question mood data are stored in the memory card in advance by the robot. The obtaining unit 41 calls data in the memory card to obtain question mood data of the question text. Some customers are more prone to emotional excitement and higher requirements, and the corresponding speaking emotions will carry some negative emotions. The question emotional data of the question text is relatively low. The obtaining unit 41 obtains the corresponding attributes as negative attributes. Further, Negative attributes include general negative attributes and very negative attributes; some customers are easy-going and have little emotional fluctuations, and the corresponding speaking emotions will tend to be bland. The question mood data of this question text is medium, and the obtaining unit 41 obtains the corresponding attributes are medium Sexual attributes; some customers are more active and active, the corresponding speaking emotions will be more positive and positive, the problem emotional data of this question text is relatively high, the corresponding attribute obtained by the acquisition unit 41 is a positive attribute, further, positive Attributes include general positive attributes and very positive attributes. Determining the level of competence of a salesperson refers to the fact that the same salesperson has different levels of competence when facing different attribute problems. The following table is a question attribute corresponding to question mood data in a specific embodiment:

Question text

Problem mood data

Problem attributes

Business sentiment data

问题文本1Question text 1	-0.8-0.8	非常消极Very negative	-0.3-0.3
问题文本2Question text 2	-0.4-0.4	一般消极Generally negative	0.40.4
问题文本3 Question text 3	0.30.3	中性neutral	0.80.8
问题文本4 Question text 4	0.90.9	非常积极Very positive	0.90.9
问题文本5 Question text 5	0.40.4	一般积极Generally positive	0.80.8
问题文本6 Question text 6	00	中性neutral	00
问题文本7Question text 7	-0.4-0.4	一般消极Generally negative	0.40.4

In another specific embodiment, the five numbers are added to obtain 2.2, that is, 2.2 is the final business sentiment data. Based on which business level 2.2 belongs to, the determining unit 42 can determine which level the salesman belongs to. The final business sentiment data and business level have a set of preset mapping rules.

Referring to FIG. 10, further, the obtaining unit 41 includes:

A sequence subunit 411 for converting the question text into an encoded unstructured vector Z through an autoencoder, and adding a structural variable C on the basis of the vector Z, and generating a tag sequence using the LSTM-RNN method

A conversion subunit 412, configured to pass the discriminator to the marker sequence

Turn into problem mood data.

In this embodiment, the question text is a question text extracted from a question set of a question frequently asked by a customer, that is, a text sentence in the following figure. The sequence subunit 411 converts the question text into a vector through an encoder, that is, vectorizes the text. The vectorization process can use the one-hot Representation model. One-hot Representation is to use a very long vector to represent a word, the length of the vector is the size of the dictionary N, each vector has only one dimension is 1, and the remaining dimensions are all 0, the position of 1 indicates the position of the word in the dictionary . This One-hot Representation is stored in a sparse manner, and the vectorization process is very simple. The sequence subunit 411 vectorizes the problem text through the encoder to obtain a vector Z, and then adds a structural variable C based on Z. The purpose of increasing C is to make the structure of the vector Z consistent with the subsequent LSTM model and make the vector Z can be input into the LSTM-RNN model. After entering the LSTM-RNN model, the label sequence is obtained

Then the conversion subunit 412 will

Input to discriminator, discriminator recognizes

Referring to FIG. 11, further, the above determining unit 42 includes:

An obtaining subunit 421, configured to obtain multiple business sentiment data corresponding to multiple question texts of the same question attribute;

A calculation subunit 422, configured to use a normalization formula to regularly calculate a plurality of the business sentiment data and a plurality of the problem sentiment data for the same problem attribute;

The obtaining subunit 423 is configured to obtain the ability level of the salesman corresponding to the problem attribute according to the calculation result.

In this embodiment, after the salesman answers multiple questions, the question sentiment data of the multiple questions can be classified according to the question data, and the acquisition subunit 421 acquires simultaneously multiple business sentiment data corresponding to multiple question texts of the same question attribute. . Through the clustering method, the acquisition subunit 421 collates the problem sentiment data corresponding to multiple question texts of the same problem attribute and the one-to-one corresponding business sentiment data, and the calculation subunit 422 calculates through the normalization formula, Normalized calculation of business sentiment data corresponding to the same problem attribute. For example, through the communication between the robot and the salesman, the emotional data table of the salesman-robot problem text is obtained as follows:

Then the calculation subunit 422 uses the normalization formula,

Where i∈j judgment mode:

The round function table is rounded off, | Ym _j | ≤1.

In this way, the business sentiment data obtained by the salesman responding to the machine sentiment data corresponding to different question texts is calculated, and the sub-unit 423 determines the ability level of the salesman for different question attributes. The problem sentiment data is classified after the regularization, and the corresponding business sentiment data corresponding to the classified problem sentiment data is regularly calculated.

Referring to FIG. 12, further, the above apparatus for testing a salesman's ability further includes:

A second obtaining module 5 configured to obtain the highest ability level among the ability levels corresponding to the multiple problem attributes of the salesman;

A generating module 6 is configured to generate position information corresponding to the target problem attribute as the most suitable position information of the salesman according to the target problem attribute corresponding to the highest capability level.

In this embodiment, the target problem attribute refers to a problem attribute corresponding to the highest ability level. The question texts of all the question attributes are sent to the salesman, and then the ability levels corresponding to all the question attributes are obtained, and each question attribute corresponds to an ability level. After comparing multiple capability levels, the second acquisition module 5 obtains the highest capability level. The highest ability level can be multiple. Each question attribute represents a type of customer. The higher the capability level corresponding to the question attribute, the better the salesperson is at communicating with the person corresponding to the question attribute. Different question attributes correspond to question texts presented by customers at different business stages. Therefore, through the salesman's ability level corresponding to different problem attributes, the generating module 6 can generate a suitable job position for the salesman. For example, if the highest ability level of the salesperson A is the corresponding negative attribute of the problem attribute, then the generating module 6 generates the most suitable position information for the salesperson A as the position related to handling the complaint. The highest ability level of the salesperson B is the corresponding attribute attribute of the neutral attribute. Then, the generating module 6 generates the most suitable position information for the salesperson B as the front office related consulting position.

Referring to FIG. 13, further, the above apparatus for testing the ability of a salesman further includes:

An adjustment module 7 is configured to adjust the performance score of the salesperson according to the ability level.

In this embodiment, from a management perspective, performance is the desired result of the organization, and it is the effective output of the organization to achieve its goals at different levels. Performance can represent a person's business ability and salesman's emotional data. The higher the enthusiasm for the work, the higher the corresponding performance score. When the ability level is higher than a certain level, the adjustment module 7 increases the performance score of the salesperson; when the ability level is lower than a certain level, the adjustment module 7 decreases the performance score of the salesperson.

In summary, the device for testing the salesperson's ability of the present application judges the salesperson's ability by simulating the chat between the customer and the salesperson, and obtaining the questions posed by the customer and the response from the salesperson. Business process to judge the competence of the salesperson. When judging the ability of the salesman, the problems are classified according to the emotion of the problem, and the processing ability of the salesman when facing different problems is correspondingly understood, and the ability of the salesman is more comprehensively judged.

Referring to FIG. 14, an embodiment of the present application further provides a computer device. The computer device may be a server, and its internal structure may be as shown in FIG. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the computer design processor is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer-readable instructions, and a database. The memory provides an environment for operating systems and computer-readable instructions in a non-volatile storage medium. The database of the computer equipment is used to store data such as models that test the ability of the salesman. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instructions are executed, the processes of the embodiments of the methods described above are executed. Those skilled in the art can understand that the structure shown in FIG. 14 is only a block diagram of a part of the structure related to the solution of the application, and does not constitute a limitation on the computer equipment to which the solution of the application is applied.

An embodiment of the present application further provides a computer non-volatile readable storage medium having computer readable instructions stored thereon. When the computer readable instructions are executed, the processes of the embodiments of the methods described above are executed. The above is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the description of the application and the drawings, or directly or indirectly used in other related The technical fields are equally included in the patent protection scope of this application.

Claims

A method for testing the ability of a salesman, which is characterized by:

Controlling the output device to output a preset question text;

Obtaining the answer text of the salesperson's reply to the question text;

Inputting the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and performing calculation to obtain the business sentiment data of the salesman for the question text;

According to the business mood data, the ability level of the salesman is determined.
The method for testing the ability of a salesperson according to claim 1, wherein the step of determining the ability level of a salesperson based on the business sentiment data comprises:

Acquiring question emotional data of the question text, and determining corresponding question attributes according to the question emotional data, wherein the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

Determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business sentiment data.
The method for testing the ability of a salesperson according to claim 2, wherein the step of obtaining question mood data of the question text comprises:

The question text is converted into an encoded unstructured vector Z by an autoencoder, and a structural variable C is added to the vector Z, and a tag sequence is generated using the LSTM-RNN method

Mark the sequence through a discriminator
Turn into problem mood data.
The method for testing the ability of a salesperson according to claim 2, wherein the step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data comprises:

Acquiring multiple business sentiment data corresponding to multiple question texts of the same question attribute;

Using a normalization formula to regularly calculate a plurality of the business sentiment data and a plurality of the problem sentiment data for the same problem attribute;

According to the calculation result, the ability level of the salesman corresponding to the same problem attribute is obtained.
The method for testing the ability of a salesperson according to claim 2, wherein after the step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business sentiment data, comprising: :

Obtaining the highest capability level among the capability levels corresponding to the multiple problem attributes of the salesman;

According to the target problem attribute corresponding to the highest ability level, position information corresponding to the target problem attribute is generated as the most suitable position information of the salesman.
The method for testing the ability of a salesman according to claim 1, wherein the step after determining the ability level of the salesman comprises:

Adjust the performance score of the salesperson according to the ability level.
A device for testing the ability of a salesman, which is characterized by including:

An output module for controlling an output device to output a preset question text;

An obtaining module, configured to obtain an answer text in which a salesperson responds to the question text;

A obtaining module, configured to input the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and perform calculation to obtain the business sentiment data of the salesman for the question text;

A determining module, configured to determine the ability level of a salesperson according to the business mood data.
The apparatus for testing the ability of a salesperson according to claim 7, wherein the determining module comprises:

An obtaining unit, configured to obtain question mood data of the question text, and determine question attributes based on the question mood data, wherein the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

A determining unit, configured to determine an ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data.
The apparatus for testing the ability of a salesperson according to claim 8, wherein the obtaining unit comprises:

A sequence subunit, for converting the question text into an encoded unstructured vector Z through an autoencoder, and adding a structural variable C based on the vector Z, and generating a tag sequence using the LSTM-RNN method

A transformation subunit for passing the marker sequence through a discriminator
Turn into problem mood data.
The apparatus for testing a salesman's ability according to claim 8, wherein the determining unit comprises:

An acquisition subunit, configured to acquire multiple business sentiment data corresponding to multiple question texts of the same question attribute;

A calculation subunit, configured to use a regularization formula to regularly calculate a plurality of the business sentiment data and a plurality of the question sentiment data for the same question attribute;

A subunit is obtained, and is used to obtain the capability level of the salesman corresponding to the problem attribute according to the calculation result.
The device for testing a salesman's ability according to claim 8, further comprising:

A second acquisition module, configured to acquire the highest capability level among the capability levels corresponding to the multiple problem attributes of the salesman;

A generating module is configured to generate position information corresponding to the target problem attribute as the most suitable position information of the salesman according to the target problem attribute corresponding to the highest capability level.
The device for testing a salesman's ability according to claim 7, further comprising:

An adjustment module is configured to adjust a performance score of the salesperson according to the ability level.
A computer device includes a memory and a processor, the memory stores computer-readable instructions, and is characterized in that the processor implements a method for testing a salesman's ability when the processor executes the computer-readable instructions, and the test Methods, including:

Controlling the output device to output a preset question text;

Obtaining the answer text of the salesperson's reply to the question text;

Inputting the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and performing calculation to obtain the business sentiment data of the salesman for the question text;

According to the business mood data, the ability level of the salesman is determined.
The computer device according to claim 13, wherein the step of determining the ability level of a salesperson based on the business sentiment data comprises:

Acquiring question emotional data of the question text, and determining corresponding question attributes according to the question emotional data, wherein the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

Determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business sentiment data.
The computer device according to claim 14, wherein the step of obtaining question mood data of the question text comprises:

The question text is converted into an encoded unstructured vector Z by an autoencoder, and a structural variable C is added to the vector Z, and a tag sequence is generated using the LSTM-RNN method

Mark the sequence through a discriminator
Turn into problem mood data.
The computer device according to claim 14, wherein the step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data comprises:

Acquiring multiple business sentiment data corresponding to multiple question texts of the same question attribute;

Using a normalization formula to regularly calculate a plurality of the business sentiment data and a plurality of the problem sentiment data for the same problem attribute;

According to the calculation result, the ability level of the salesman corresponding to the same problem attribute is obtained.
A computer non-volatile readable storage medium having computer-readable instructions stored thereon, characterized in that when the computer-readable instructions are executed by a processor, a method for testing the ability of a salesman is tested, Methods, including:

Controlling the output device to output a preset question text;

Obtaining the answer text of the salesperson's reply to the question text;

Inputting the answer text into a preset sentiment analysis model trained based on the LSTM-CNN model and performing calculation to obtain the business sentiment data of the salesman for the question text;

According to the business mood data, the ability level of the salesman is determined.
The computer non-volatile readable storage medium according to claim 17, wherein the step of determining the ability level of a salesperson based on the business sentiment data comprises:

Acquiring question emotional data of the question text, and determining corresponding question attributes according to the question emotional data, wherein the question attributes include at least a negative attribute, a neutral attribute, and a positive attribute;

Determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business sentiment data.
The computer non-volatile readable storage medium according to claim 18, wherein the step of obtaining question mood data of the question text comprises:

The question text is converted into an encoded unstructured vector Z by an autoencoder, and a structural variable C is added to the vector Z, and a tag sequence is generated using the LSTM-RNN method

Mark the sequence through a discriminator
Turn into problem mood data.
The non-volatile computer-readable storage medium of claim 18, wherein the step of determining the ability level of the salesperson corresponding to the problem attribute according to the problem attribute and the business mood data ,include:

Acquiring multiple business sentiment data corresponding to multiple question texts of the same question attribute;

Using a normalization formula to regularly calculate a plurality of the business sentiment data and a plurality of the problem sentiment data for the same problem attribute;

According to the calculation result, the ability level of the salesman corresponding to the same problem attribute is obtained.