Telephone channel auxiliary answering method, system and terminal

Abstract

The present application relates to the technical field of customer service management, and discloses a telephone channel assisted answering method, system and terminal. The present invention collects corresponding personal data while improving the professional skills of new employees through interactive training for the operators, and constructs corresponding basic vocabulary and basic sentence libraries to facilitate the statistics of the operators' habit data. The operator's conversation quality is then scored based on the content of the operator's conversation to help the operator improve in real time at work. The application also makes intelligent judgments on the call processes of all operators to screen out some call processes with a high probability of multiple meanings, changing the way managers check the problem by random sampling, and can help managers manage the operators better and more efficiently.

Description

Technical Field

[0001] The present application relates to the technical field of customer service management, and in particular to a telephone channel assisted answering method, system and terminal. Background Art

[0002] Customer service via telephone channels has always been an essential component of most industries. With the development of intelligent technology, new intelligent voice products that combine telephone customer service with AI are significantly promoting the development of customer service via telephone channels.

[0003] Currently, AI-powered voice recognition can be used to intelligently review call content and provide intelligent training for newly hired operators, thereby comprehensively improving the service quality of service providers' answering staff. However, the current method of reviewing call content through voice recognition still has flaws. Since the current intelligent review uses voice recognition keywords to determine whether the call content is compliant, some operators or answering staff, if they are familiar with the rules, can express meanings that do not comply with company regulations by changing the tone of voice, thereby circumventing some issues with high complaint rates. For example, when answering a call with multiple nouns, if the tone of voice appears on different nouns, some customers will misunderstand it differently.

[0004] Although current intelligent voice recognition can determine the meaning of a sentence by identifying ambiguous words, there is still a lack of a good method to eliminate or identify ambiguity caused by changes in tone and stress. This leads to the problem that the combination of intelligent voice products and customer service through telephone channels still has the problem that customer service calls contain content that is easy to be misunderstood, and the content of customer service calls is too long to be carefully reviewed. In view of this, the present invention proposes a telephone channel assisted answering method, system and terminal. Summary of the Invention

[0005] In order to solve the problem that customer service calls are prone to misunderstanding and cannot be recognized when using intelligent voice products for assistance, and that the content of customer service calls is too long to be carefully reviewed, this application provides a telephone channel assisted answering method, system and terminal.

[0006] In the first aspect, the present application provides a telephone channel auxiliary answering system, which adopts the following technical solutions:

[0007] A telephone channel auxiliary answering system, comprising:

[0008] A training module records preset training content and generates training plans. Specifically, it collects training needs before training, builds training plans, and sets training items. It provides interactive and intelligent training during training, and tracks training results after training. Training plans are generated based on preset templates, requiring managers to select the corresponding templates. Training data generated during the training process is also stored.

[0009] A personal data module, which obtains training data and the conversation content of the operator's telephone calls, and establishes a basic vocabulary and basic sentence library based on the obtained training data and the conversation content of daily calls;

[0010] A quality inspection module includes a first quality inspection unit and a second quality inspection unit. The first quality inspection unit performs a quality score on the current call process based on the conversation content of the current call process. The second quality inspection unit identifies accented characters based on the conversation content of the current call process and calculates a semantic function based on the number and position of accented characters. The quality inspection result is obtained according to the calculated semantic function to determine whether there is multiple semantics.

[0011] Through the above technical solution: an auxiliary answering system for telephone customer service answering is provided. Specifically, the present invention collects corresponding personal data while improving the professional skills of new employees through interactive training for the operators, and constructs corresponding basic vocabulary and basic sentence libraries to facilitate the statistics of the operator's habit data. Then, the operator's conversation quality is scored based on the content of the operator's conversation to help the operator improve in real time at work, and intelligent judgment is made on the call processes of all operators to screen out some call processes with a high probability of multiple meanings, which changes the way managers check the problem by random inspections and can help managers manage the operators better and more efficiently.

[0012] Optionally, the process of scoring the quality of the current call process based on the conversation content of the current call process includes:

[0013] Obtain the conversation content of the operator, including the operator's answer and the customer's question;

[0014] Extract keywords based on the answer statement and question statement respectively, and lock the customer's question type from the preset question library through the keywords extracted from the question statement;

[0015] Obtain the preset scoring items for each question type, match the scoring items based on the keywords of the answer statement, and obtain the basic score of each scoring item based on the matching results. The quality score is the sum of all basic scores.

[0016] Optionally, the process of recognizing accented characters includes:

[0017] Split the answer sentence into independent Chinese characters and obtain the pronunciation audio of the Chinese characters;

[0018] The corresponding pronunciation audio of the Chinese characters obtained in the answer sentence is compared with the pronunciation audio of the same Chinese characters pre-stored in the basic vocabulary, and whether the Chinese characters in the answer sentence are stressed is determined based on the pronunciation volume and pronunciation duration of the pronunciation audio of both parties.

[0019] Through the above technical solution: it is given to judge whether the pronunciation habits of the current Chinese characters are consistent with their pronunciation habits based on the basic vocabulary of the current operator, and some Chinese characters that do not conform to the pronunciation habits are marked as stressed, so as to facilitate subsequent quality inspection and judgment.

[0020] Optionally, the answer statement is divided into N complete sentences, and the sentence sequence of each sentence is obtained. The semantic function is represented by Mc and is calculated as follows:

[0021]

[0022] in, It is the basic value preset according to the question type. is the number of stressed characters in the i-th sentence of the answer, is the average number of occurrences of accented characters in a sentence in the statistics of the personal data module corresponding to the sentence sequence of the current sentence, and J is Based on the numerical judgment function, i is a non-zero integer not greater than N, is the coefficient of the number of sentences i, , It is the basic value preset based on the sentence sequence corresponding to the current sentence. is the inverse of the number of successful quality inspections by the current operator in the previous quality inspection cycle, P is the probability coefficient, ,in, is the probability that the jth stressed word in the ith sentence appears in the same position in the corresponding sentence sequence, M is the total number of sentence sequences in the answer sentence, j is a non-zero integer not greater than M, and e is a natural constant.

[0023] Optionally, the sentence sequence is a coding combination of parts of speech, which is obtained by dividing a complete sentence into several phrases, determining the part of speech of each phrase, and arranging them in order. The sentence sequence and its corresponding training data are stored in a basic sentence library.

[0024] The above technical solution provides a calculation process for a semantic function. The semantic function of the present invention is calculated based on the number of accents in a sentence and the probability of the accent appearing at a certain position in a sentence sequence. Operators generally use trained speech techniques to respond to customers, combined with the operators' own habitual sentence patterns in real time. Therefore, some sentence sequences are reused a lot. The positions and number of accented characters in these reused sentence sequences have relatively stable habitual data. When the two change, especially an increase in the number of accented characters or an inverse probability change in the positions of accented characters, it indicates that emotional speech exists in the current conversation. However, because the speech content and keywords meet the standards, conventional keyword detection cannot detect it. A single judgment based on audio data such as speech speed and tone is prone to many misjudgments in a large amount of call data, resulting in excessive workload in the subsequent manual verification process, which deviates from the original intention of intelligent assistance. The present invention can more accurately distinguish possible multiple-semantic call sentences, thereby assisting managers in more standardized and efficient management.

[0025] Optionally, the process of determining whether multiple meanings exist based on the quality inspection result obtained by the calculated meaning function includes:

[0026] Set safety thresholds for the semantic function based on historical data. It should be understood that different companies or different types of companies have different corresponding safety thresholds. Users can set an initial value and adjust the safety threshold based on the success rate and number of successes in each quality inspection cycle;

[0027] Compare the calculated semantic function value with the safety threshold. If the calculated value is greater than the safety threshold, the quality inspection result is to mark the current call process; otherwise, the quality inspection result is not to mark it.

[0028] The marked call process is determined to have multiple semantics, and the marked call process is uploaded and manually inspected, and whether the quality inspection is successful is output based on the manual quality inspection result.

[0029] Optionally, the auxiliary answering system includes:

[0030] A mid-speech assistance module, which includes a conversation navigation unit and a customer emotion detection unit. The conversation navigation unit locks the corresponding scoring items based on the customer's question type and marks and displays unfinished scoring items;

[0031] The customer emotion detection unit determines the customer emotion based on the customer's speaking speed and tone, and the customer emotion includes positive and negative emotions.

[0032] Through the above technical solution: a module is provided to assist the operator during the call. The in-call assistance module of the present invention is set based on the quality scoring process, and the quality scoring process is made public and visualized, so that the operator can make up for any errors in a timely manner.

[0033] Optionally, the auxiliary answering system includes:

[0034] An optimization module, which includes a tagging process to improve the assisted answering system's ability to judge the current industry customer's intentions. During the tagging process, annotators tag user questions onto standard FAQs to enhance the assisted answering system's generalized understanding capabilities.

[0035] During the data annotation process, the optimization module recommends at least three possible intentions for the annotator to select, saving the annotator's working time and achieving rapid annotation and effect iteration.

[0036] In a second aspect, the present application provides a telephone channel auxiliary answering method, which adopts the following technical solution:

[0037] A telephone channel auxiliary answering method includes the following steps:

[0038] Record the preset training content and generate training plans, as well as store the training data generated during the training process;

[0039] Establish a basic vocabulary and sentence library based on the acquired training data and daily conversation content;

[0040] Score the quality of the current call process based on the conversation content of the current call process;

[0041] Identify accented characters based on the conversation content of the current call and calculate the semantic function based on the number and position of accented characters;

[0042] A quality inspection result is obtained based on the calculated semantic function to determine whether there is multiple semantics.

[0043] In a third aspect, the present application provides a telephone channel auxiliary answering terminal, which adopts the following technical solutions:

[0044] It includes a microphone array module for receiving sound and a mobile exhibition terminal for displaying auxiliary content.

[0045] In summary, this application includes at least one of the following beneficial technical effects:

[0046] The present invention provides interactive training for operators, collects corresponding personal data while improving the professional skills of new employees, and constructs corresponding basic vocabulary and basic sentence libraries to facilitate the statistics of operator habit data. The operator's conversation quality is then scored based on the content of the operator's conversation to help the operator improve in real time at work. The invention also makes intelligent judgments on the call processes of all operators to filter out some call processes with a high probability of multiple meanings, changing the way managers check this problem by random sampling, and can help managers manage the operators better and more efficiently.

[0047] The present invention compares the position and number of stressed characters with the customary data of the position and number of stressed characters through a semantic function. Specifically, the calculation is based on the number of stresses in a sentence and the probability of stresses appearing at a certain position in a sentence sequence. This can more accurately distinguish possible multi-semantic conversation sentences, thereby assisting managers to manage more standardized and efficient methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the auxiliary answering system module.

[0049] Figure 2 It is a flowchart of the steps of the assisted answering method. DETAILED DESCRIPTION

[0050] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0051] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0052] The present application embodiment discloses a telephone channel auxiliary answering system, referring to Figure 1 include:

[0053] The training module records the preset training content and generates training plans. Specifically, it collects training needs before training, builds training plans and sets training items. It provides interactive and intelligent training during training, and tracks training results after training. The training plan is generated based on preset templates, requiring managers to select the corresponding templates. The training data generated during the training process is also stored.

[0054] The personal data module obtains training data and the conversation content of the operator's calls, and builds a basic vocabulary and basic sentence library based on the obtained training data and the conversation content of daily calls;

[0055] The quality inspection module includes a first quality inspection unit and a second quality inspection unit. The first quality inspection unit scores the quality of the current call process based on the conversation content of the current call process. The second quality inspection unit identifies accented characters based on the conversation content of the current call process and calculates a semantic function based on the number and position of accented characters. The quality inspection result is obtained according to the calculated semantic function to determine whether there is multiple semantics.

[0056] In this embodiment, an auxiliary answering system for telephone customer service answering is provided. Specifically, the present invention collects corresponding personal data while improving the professional skills of new employees through interactive training for the operators, and constructs corresponding basic vocabulary and basic sentence libraries to facilitate the statistics of the operators' habit data. The operator's conversation quality is then scored based on the content of the operator's conversation to help the operator improve in real time at work. In addition, the call process of all operators is intelligently judged to screen out some call processes with a high probability of multiple meanings, which changes the way managers check the problem by random inspections and can help managers manage the operators better and more efficiently.

[0057] The process of scoring the quality of the current call process based on the conversation content of the current call process includes:

[0058] Obtain the conversation content of the operator, including the operator's answer and the customer's question;

[0059] Extract keywords based on the answer statement and question statement respectively, and lock the customer's question type from the preset question library through the keywords extracted from the question statement;

[0060] Get the preset scoring items for each question type, match the scoring items based on the keywords of the answer statement, and get the basic score of each scoring item based on the matching results. The quality score is the sum of all basic scores.

[0061] The process of recognizing accented characters includes:

[0062] Split the answer sentence into independent Chinese characters and obtain the pronunciation audio of the Chinese characters;

[0063] The corresponding pronunciation audio of the Chinese characters obtained in the answer sentence is compared with the pronunciation audio of the same Chinese characters pre-stored in the basic vocabulary, and whether the Chinese characters in the answer sentence are stressed is judged based on the pronunciation volume and pronunciation duration of the pronunciation audio of both parties. As an example, the pronunciation duration and pronunciation volume are expressed in numerical terms, and the judgment conditions for whether it is stressed are given based on the numerical range, for example, the pronunciation volume is greater than A and the pronunciation duration is greater than B.

[0064] In this embodiment, a method is provided for judging whether the pronunciation habit of the current Chinese character is consistent with the pronunciation habit of the current operator based on the basic vocabulary of the current operator, and marking some Chinese characters that do not conform to the pronunciation habit as accents, so as to facilitate subsequent quality inspection and judgment.

[0065] The answer statement is divided into N complete sentences, and the sentence sequence of each sentence is obtained. The semantic function is represented by Mc and is calculated as follows:

[0066]

[0067] in, is a basic value preset according to the problem type, which is a constant. is the number of stressed characters in the i-th sentence of the answer, is the average number of occurrences of accented characters in a sentence in the statistics of the personal data module corresponding to the sentence sequence of the current sentence, and J is Based on the numerical judgment function, if > 0, then output 1, otherwise output 0, i is a non-zero integer not greater than N, is the coefficient of the number of sentences i, , It is a basic value preset based on the sentence sequence corresponding to the current sentence, which is a constant. is the inverse of the number of successful quality inspections by the current operator in the previous quality inspection cycle, P is the probability coefficient, ,in, is the probability that the jth stressed word in the ith sentence appears in the same position in the corresponding sentence sequence, M is the total number of sentence sequences in the answer sentence, j is a non-zero integer not greater than M, and e is a natural constant.

[0068] A sentence sequence is a coding combination of parts of speech. It is a sentence coding obtained by dividing a complete sentence into several phrases, determining the part of speech of each phrase, and arranging them in order. The sentence sequence and its corresponding training data are stored in the basic sentence library. In a Chinese sentence, words with parts of speech such as subject, predicate, and object are included. Different codes are set for each part of speech. Then "subject, predicate, object" or "subject, predicate, object, and determinant" is a sentence sequence, and the corresponding coding combination is the sentence code. Each sentence sequence does not correspond to only one sentence.

[0069] This embodiment provides a calculation process for a semantic function. The semantic function of the present invention is calculated based on the number of accents in a sentence and the probability of the accent appearing at a certain position in a sentence sequence. Operators generally use trained speech to respond to customers, combined with the operators' own habitual sentence patterns in real time. Therefore, some sentence sequences may be reused extensively. The positions and number of accented characters in these reused sentence sequences have relatively stable habitual data. When these two change, especially an increase in the number of accented characters or an inverse probability change in the positions of accented characters, it indicates that the current conversation contains emotional speech. However, because the speech content and keywords meet the standard, conventional keyword detection cannot detect it. A single judgment based on audio data such as speech speed and intonation is prone to many misjudgments in a large amount of call data, resulting in excessive workload in the subsequent manual verification process, which deviates from the original intention of intelligent assistance. The present invention can more accurately distinguish possible multiple semantic conversation sentences, thereby assisting managers in more standardized and efficient management.

[0070] The process of determining whether there is multiple meanings based on the quality inspection results obtained by the calculated meaning function includes:

[0071] Set safety thresholds for the semantic function based on historical data. It should be understood that different companies or different types of companies have different corresponding safety thresholds. Users can set an initial value and adjust the safety threshold based on the success rate and number of successes in each quality inspection cycle;

[0072] Compare the calculated semantic function value with the safety threshold. If the calculated value is greater than the safety threshold, the quality inspection result is to mark the current call process; otherwise, the quality inspection result is not to mark it.

[0073] The marked call process is determined to have multiple semantics, and the marked call process is uploaded and manually inspected, and whether the quality inspection is successful is output based on the manual quality inspection result.

[0074] Assisted listening systems include:

[0075] The mid-conversation assistance module includes a conversation navigation unit and a customer sentiment detection unit. The conversation navigation unit locks the corresponding scoring items based on the customer's question type and marks and displays incomplete scoring items;

[0076] The customer emotion detection unit judges the customer's emotion based on the customer's speaking speed and tone. Customer emotions include positive and negative. As an example, the speaking speed and tone values ​​are set and the critical value and judgment conditions are set. When the positive condition is met, the customer emotion is judged to be positive. Similarly, negative emotions can be judged.

[0077] Through the above technical solution: In this embodiment, a module is provided to assist the operator during the call. The in-call assistance module of the present invention is set based on the quality scoring process, and the quality scoring process is made public and visualized, so that the operator can make up for any errors in a timely manner.

[0078] Assisted listening systems include:

[0079] The optimization module includes a tagging process to improve the assisted answering system's ability to judge the current industry customer's intentions. During the tagging process, annotators tag user questions onto standard FAQs to enhance the assisted answering system's generalized understanding capabilities. FAQs refer to Frequently Asked Questions.

[0080] During the data annotation process, the optimization module recommends at least three possible intentions for the annotator to select, saving the annotator's working time and achieving rapid annotation and effect iteration.

[0081] This embodiment provides a telephone channel assisted answering method, which adopts the following technical solutions:

[0082] A telephone channel auxiliary answering method, reference Figure 2 , including the following steps:

[0083] S100, recording preset training content and generating a training plan, and storing training data generated during the training process;

[0084] S200: Establish a basic vocabulary and basic sentence library based on the acquired training data and the conversation content of daily calls;

[0085] S300, scoring the quality of the current call process based on the conversation content of the current call process;

[0086] S400, identifying accented characters based on the conversation content of the current call process and calculating a semantic function based on the number and position of accented characters;

[0087] S500: Obtain a quality inspection result based on the calculated semantic function to determine whether there are multiple semantic expressions.

[0088] This embodiment also provides a telephone channel auxiliary answering terminal, which adopts the following technical solution:

[0089] It includes a microphone array module for receiving sound and a mobile exhibition terminal for displaying auxiliary content.

[0090] The embodiments of the present application also disclose a telephone channel assisted answering method, system and terminal, including a processor running a program of any one of the above-mentioned telephone channel assisted answering method, system and terminal.

[0091] The embodiment of the present application also discloses a storage medium storing a program of the telephone channel assisted answering method, system and terminal described in any one of the above.

[0092] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A telephone channel auxiliary answering system, characterized in that: include: A training module, which records preset training content and generates a training plan, and stores training data generated during the training process; A personal data module, which obtains training data and the conversation content of the operator's telephone calls, and establishes a basic vocabulary and basic sentence library based on the obtained training data and the conversation content of daily calls; a quality inspection module, the quality inspection module including a first quality inspection unit and a second quality inspection unit, wherein the first quality inspection unit performs a quality score on the current call process based on the content of the conversation during the current call process, and the second quality inspection unit identifies accented characters based on the content of the conversation during the current call process and calculates a semantic function based on the number and position of the accented characters, and obtains a quality inspection result based on the calculated semantic function to determine whether multiple semantic expressions exist; Divide the answer sentences into N complete sentences and obtain the sentence sequence of each sentence. The semantic function is used Mc Indicates that the calculation method is as follows: in, It is the basic value preset according to the question type. is the number of stressed characters in the i-th sentence of the answer, It is the average number of times the accented characters in a sentence appear in the statistics of the personal data module corresponding to the sentence sequence of the current sentence. J yes Based on the numerical judgment function, N is the total number of sentences divided by an answer statement. i is not greater than N A non-zero integer, is the coefficient of the number of sentences i, , It is the basic value preset based on the sentence sequence corresponding to the current sentence. It is the countdown of the number of successful quality inspections by the current operator in the last quality inspection cycle. P is the probability coefficient, ,in, is the first j The probability that the accented characters appear in the same position in the corresponding sentence sequence, M is the total number of sentence sequences in the answer sentence, j is not greater than M A non-zero integer, e is a natural constant; The sentence sequence is a coding combination of parts of speech, which is obtained by dividing a complete sentence into several phrases, determining the part of speech of each phrase, and arranging them in order. The sentence sequence and its corresponding training data are stored in the basic sentence library.

2. The telephone channel auxiliary answering system according to claim 1, characterized in that: The process of scoring the quality of the current call process based on the conversation content of the current call process includes: Obtain the conversation content of the operator, including the operator's answer and the customer's question; Extract keywords based on the answer statement and question statement respectively, and lock the customer's question type from the preset question library through the keywords extracted from the question statement; Obtain the preset scoring items for each question type, match the scoring items based on the keywords of the answer statement, and obtain the basic score of each scoring item based on the matching results. The quality score is the sum of all basic scores.

3. The telephone channel auxiliary answering system according to claim 2, characterized in that: The process of recognizing accented characters includes: Split the answer sentence into independent Chinese characters and obtain the pronunciation audio of the Chinese characters; The corresponding pronunciation audio of the Chinese characters obtained in the answer sentence is compared with the pronunciation audio of the same Chinese characters pre-stored in the basic vocabulary, and whether the Chinese characters in the answer sentence are stressed is determined based on the pronunciation volume and pronunciation duration of the pronunciation audio of both parties.

4. The telephone channel auxiliary answering system according to claim 1, characterized in that: The process of determining whether there is multiple meanings based on the quality inspection results obtained by the calculated meaning function includes: Setting safety thresholds for semantic functions based on historical data; Compare the calculated semantic function value with the safety threshold. If the calculated value is greater than the safety threshold, the quality inspection result is to mark the current call process; otherwise, the quality inspection result is not to mark it. The marked call process is determined to have multiple semantics, and the marked call process is uploaded and manually inspected, and whether the quality inspection is successful is output based on the manual quality inspection result.

5. The telephone channel auxiliary answering system according to claim 2, characterized in that: The auxiliary answering system includes: A mid-speech assistance module, which includes a conversation navigation unit and a customer emotion detection unit. The conversation navigation unit locks the corresponding scoring items based on the customer's question type and marks and displays unfinished scoring items; The customer emotion detection unit determines the customer emotion based on the customer's speaking speed and tone, and the customer emotion includes positive and negative emotions.

6. The telephone channel auxiliary answering system according to claim 1, characterized in that: The auxiliary answering system includes: An optimization module, which includes a labeling process in which labelers label user questions onto standard FAQs; During the data annotation process, the optimization module recommends at least three possible intents for the annotator to select.

7. A telephone channel auxiliary answering method, characterized in that: The telephone channel auxiliary answering system according to any one of claims 1 to 6 comprises the following steps: Record the preset training content and generate training plans, as well as store the training data generated during the training process; Establish a basic vocabulary and sentence library based on the acquired training data and daily conversation content; Score the quality of the current call process based on the conversation content of the current call process; Identify accented characters based on the conversation content of the current call and calculate the semantic function based on the number and position of accented characters; A quality inspection result is obtained based on the calculated semantic function to determine whether there is multiple semantics.

Images