WO2021003930A1 - 客服录音的质检方法、装置、设备及计算机可读存储介质 - Google Patents

客服录音的质检方法、装置、设备及计算机可读存储介质 Download PDF

Info

Publication number
WO2021003930A1
WO2021003930A1 PCT/CN2019/117539 CN2019117539W WO2021003930A1 WO 2021003930 A1 WO2021003930 A1 WO 2021003930A1 CN 2019117539 W CN2019117539 W CN 2019117539W WO 2021003930 A1 WO2021003930 A1 WO 2021003930A1
Authority
WO
WIPO (PCT)
Prior art keywords
text
quality inspection
preset
inspected
search
Prior art date
Application number
PCT/CN2019/117539
Other languages
English (en)
French (fr)
Inventor
张超
汤耀华
Original Assignee
深圳前海微众银行股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳前海微众银行股份有限公司 filed Critical 深圳前海微众银行股份有限公司
Publication of WO2021003930A1 publication Critical patent/WO2021003930A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/60Information retrieval; Database structures therefor; File system structures therefor of audio data
    • G06F16/63Querying
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/60Information retrieval; Database structures therefor; File system structures therefor of audio data
    • G06F16/68Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/686Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/205Parsing
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04MTELEPHONIC COMMUNICATION
    • H04M3/00Automatic or semi-automatic exchanges
    • H04M3/42Systems providing special services or facilities to subscribers
    • H04M3/50Centralised arrangements for answering calls; Centralised arrangements for recording messages for absent or busy subscribers ; Centralised arrangements for recording messages
    • H04M3/51Centralised call answering arrangements requiring operator intervention, e.g. call or contact centers for telemarketing
    • H04M3/5175Call or contact centers supervision arrangements
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02PCLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30Computing systems specially adapted for manufacturing

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a quality inspection method, device, equipment and computer-readable storage medium for customer service recording.
  • the traditional quality inspection method is to manually listen to customer service recordings to conduct random inspections and evaluations. This quality inspection method not only consumes a lot of manpower, has low quality inspection efficiency, but also has low coverage of random inspections and poor quality inspection results.
  • a quality inspection system based on artificial intelligence technology for intelligent quality inspection of customer service recording has been proposed.
  • the process of analyzing the recorded text data takes into account the complexity of the actual recorded text, and in order to improve the accuracy of the quality inspection, a time-complex analysis method is adopted to make the entire quality The time complexity of the inspection system is increased.
  • the main purpose of this application is to provide a quality inspection method, device, equipment, and computer-readable storage medium for customer service recording, which aims to solve the analysis caused by the current customer service quality inspection system to analyze the recorded text to ensure the accuracy of the quality inspection Technical problem with high process time complexity.
  • the quality inspection method for customer service recording includes the steps:
  • the quality inspection result of the text to be inspected is obtained according to the in-depth search result.
  • the preset preliminary search model is a pattern matching model
  • the step of searching for QC elements in the text to be QC according to the preset preliminary search model includes:
  • the target text that matches the preset text mode is matched, it is determined that the quality inspection element corresponding to the preset text mode is found.
  • the step of determining whether to perform an in-depth search for the quality inspection element according to the preliminary search result includes:
  • the number of elements is not greater than the preset number, it is determined not to perform in-depth search on the quality inspection elements.
  • the preset depth search model includes a text matching model
  • the step of searching for a target element not found in the preliminary search result in the text to be QC according to the preset depth search model includes:
  • the step of obtaining the text fragments to be QC in the text to be QC includes:
  • text segmentation is performed on the text to be quality-inspected to obtain text fragments to be quality-inspected.
  • the preset depth search model further includes a reading comprehension model
  • the method further includes:
  • the target element If the target element is located, it is determined that the target element is found.
  • the step of obtaining the quality inspection result of the text to be inspected according to the in-depth search result includes:
  • the score of each quality inspection item is used as the quality inspection result of the text to be inspected.
  • this application also provides a quality inspection device for customer service recording, and the quality inspection device for customer service recording includes:
  • the preliminary search module is configured to search for quality inspection elements in the text to be inspected according to the preset preliminary search model after obtaining the text to be inspected converted from the customer service recording, and obtain the preliminary search result;
  • a determining module configured to determine whether to perform an in-depth search for the quality inspection element according to the preliminary search result
  • the in-depth search module is configured to search for target elements not found in the preliminary search results in the text to be inspected according to the preset depth search model when it is determined to perform in-depth search on the quality inspection elements to obtain the depth Search result
  • the quality inspection result generation module is configured to obtain the quality inspection result of the text to be inspected according to the deep search result.
  • the present application also provides a quality inspection device for customer service recording.
  • the quality inspection device for customer service recording includes a memory, a processor, and a device that is stored in the memory and can run on the processor.
  • a quality inspection program for customer service recording when the quality inspection program for customer service recording is executed by the processor, implements the steps of the quality inspection method for customer service recording as described above.
  • this application also provides a computer-readable storage medium that stores a quality inspection program for customer service recording, which is implemented when the customer service recording quality inspection program is executed by a processor The steps of the quality inspection method for customer service recording as described above.
  • a preliminary search for quality inspection elements in the text to be inspected according to the preset preliminary search model is carried out, and the preliminary search results are obtained; according to the preliminary search results, determine whether to check the quality In-depth search for elements; if it is determined to perform in-depth search for quality inspection elements, according to the preset depth search model, the target elements that are not found in the preliminary search results are searched for in the text to be inspected, and the in-depth search results are obtained; according to the depth search results , Obtain the quality inspection results of the text to be inspected, and realize that while ensuring the accuracy of quality inspection, reduce the time complexity of the entire quality inspection system and improve the efficiency of quality inspection.
  • Figure 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a first embodiment of a quality inspection method for applying for customer service recording
  • Figure 3 is a schematic diagram of a quality inspection process involved in an embodiment of the application.
  • Fig. 4 is a functional schematic block diagram of a preferred embodiment of a quality inspection device for customer service recording according to this application.
  • FIG. 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of this application.
  • Fig. 1 can be a structural diagram of the hardware operating environment of the quality inspection equipment for customer service recording.
  • the quality inspection equipment for customer service recording in the embodiment of this application may be a PC, or a terminal device with a display function, such as a smart phone, a smart TV, a tablet computer, and a portable computer.
  • the quality inspection equipment for customer service recording may include a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile memory), such as disk storage.
  • the memory 1005 may also be a storage device independent of the foregoing processor 1001.
  • the quality inspection equipment for customer service recording may also include a camera, RF (Radio Frequency (radio frequency) circuits, sensors, audio circuits, WiFi modules, etc.
  • RF Radio Frequency (radio frequency) circuits
  • sensors e.g., a camera
  • audio circuits e.g., a Wi-Fi module
  • WiFi modules e.g., a Wi-Fi module
  • FIG. 1 the structure of the quality inspection equipment for customer service recording shown in FIG. 1 does not constitute a limitation on the quality inspection equipment for customer service recording, and may include more or less components than shown in the figure, or a combination of some Components, or different component arrangements.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a quality inspection program for customer service recording.
  • the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server;
  • the user interface 1003 is mainly used to connect to the client (user side) and conduct data with the client Communication; and the processor 1001 can be used to call the quality inspection program of the customer service recording stored in the memory 1005, and perform the following operations:
  • the quality inspection result of the text to be inspected is obtained according to the in-depth search result.
  • the preset preliminary search model is a pattern matching model
  • the step of searching for QC elements in the text to be QC according to the preset preliminary search model includes:
  • the target text that matches the preset text mode is matched, it is determined that the quality inspection element corresponding to the preset text mode is found.
  • step of determining whether to perform an in-depth search for the quality inspection element according to the preliminary search result includes:
  • the number of elements is not greater than the preset number, it is determined not to perform in-depth search on the quality inspection elements.
  • the preset depth search model includes a text matching model
  • the step of searching for a target element not found in the preliminary search result in the text to be QC according to the preset depth search model includes:
  • the step of obtaining the text fragments to be QC in the text to be QC includes:
  • text segmentation is performed on the text to be quality-inspected to obtain text fragments to be quality-inspected.
  • the preset depth search model further includes a reading comprehension model.
  • the processor 1001 may call the quality inspection program of the customer service recording stored in the memory 1005 , Also do the following:
  • the target element If the target element is located, it is determined that the target element is found.
  • the step of obtaining the quality inspection result of the text to be inspected according to the in-depth search result includes:
  • the score of each quality inspection item is used as the quality inspection result of the text to be inspected.
  • the first embodiment of the quality inspection method for customer service recording of this application provides a quality inspection method for customer service recording. It should be noted that although the logical sequence is shown in the flowchart, in some cases, Perform the steps shown or described in a different order than here. For ease of description, the execution subject is omitted in the following embodiments for description.
  • the quality inspection method of the customer service recording includes:
  • Step S10 after obtaining the text to be inspected by the customer service recording, search for quality inspection elements in the text to be inspected according to a preset preliminary search model, and obtain a preliminary search result;
  • the telephone service process of the customer service in advance, and save the recording file in the database.
  • customer service recordings such as receiving a quality inspection instruction
  • obtain customer service recording files from the database use voice recognition technology to recognize the recording files, and convert voice data into text data.
  • the text to be inspected is the text converted from the customer’s voice in the customer service recording file, not the text converted from the user’s voice.
  • the customer service in the recording data can be based on the speaker’s voice characteristics (such as voiceprint characteristics).
  • the voice of the user is distinguished from the voice of the user.
  • Quality inspection items and quality inspection elements can be preset as quality inspection indicators.
  • the quality inspection items refer to the phraseology texts such as the opening words, security reminders, ID verification or identity confirmation that should be included in the text to be inspected, for example: the opening words of the words operation text is "Hello! This is the work of WeBank The staff is calling you. Thank you for using the micro loan product! This call is mainly to confirm the loan-related information with you. Is it convenient for you to call now? To ensure the quality of service, this call may be recorded. Please understand.”
  • different quality inspection items can be set, for example, 4 quality inspection items can be set: opening sentence, security reminder, identity verification and identity verification.
  • the quality inspection element is one or several fragments in the text of the quality inspection item.
  • the opening phrase when used as the quality inspection item, it can include "WeBank staff”, "Confirm the loan with you” and "This call is possible Will be recorded” these three quality inspection elements. Whether each quality inspection element appears in the text to be inspected, and how much it appears is an evaluation standard that reflects the quality of customer service.
  • the preset preliminary search model can be a preset model that can perform a preliminary search for quality inspection elements.
  • a preset pattern matching model is used as a preliminary search model.
  • the preliminary search takes less time, and is less complicated for customer service Recording can find most or all of the quality inspection elements.
  • Finding quality inspection elements in the text to be inspected refers to finding quality inspection elements under all quality inspection items. According to whether each quality inspection element is found, the preliminary search result is obtained.
  • the preliminary search result may include the quality inspection elements not found and the found quality inspection elements, or all the quality inspection elements may be found, or each quality inspection element may be found. None of the inspection elements have been found.
  • Step S20 determining whether to perform an in-depth search for the quality inspection element according to the preliminary search result
  • the ratio is not greater than the preset ratio, it means that the number of quality inspection elements in the text to be inspected is not enough to confirm that the text to be inspected meets the quality requirements, but because the preliminary search model is used, the text to be inspected There may be other quality inspection elements that have not been found by the preliminary search model.
  • the depth of the quality inspection elements can be determined to further search the quality inspection elements through the subsequent in-depth search model to determine the preliminary search Whether the quality inspection elements that are not found do not appear in the text to be inspected, so as to improve the accuracy of the quality inspection system.
  • step S20 includes:
  • Step S201 detecting whether the number of elements of the target element not found in the preliminary search result is greater than a preset number
  • the quality inspection elements not found in the preliminary search results are used as target elements, and whether the number of elements of the target elements is greater than the preset number is detected.
  • the preset number can be set according to specific needs. When the accuracy rate of the quality inspection is high, the preset number can be set to be smaller.
  • Step S202 If the number of elements is greater than the preset number, it is determined to perform an in-depth search for the quality inspection elements
  • the number of elements detected is greater than the preset number, it is determined to conduct an in-depth search of the quality inspection elements.
  • the number of elements is greater than the preset number, it means that there are more quality inspection elements that have not been found.
  • other quality inspection elements may appear in the text to be inspected, but they have not been found by the preliminary search model and cannot be found. It is directly determined that the text to be inspected does not meet the quality requirements, so it can be determined to conduct an in-depth search of the quality inspection elements.
  • Step S203 If the number of elements is not greater than the preset number, it is determined not to perform in-depth search on the quality inspection elements.
  • the number of elements detected is not greater than the preset number, it is determined not to perform in-depth search of quality inspection elements, that is, because the number of quality inspection elements not found is not large, it can be determined that there are more quality inspection texts in the text. Inspection elements can determine that the text to be inspected meets the quality requirements. Therefore, there is no need to continue searching, thereby saving text analysis time and reducing the time complexity of the quality inspection system.
  • the preset quantity can be set to zero. That is, check whether the preliminary search result has undiscovered quality inspection elements. When it is greater than zero, it means there are undiscovered quality inspection elements. As long as there are undiscovered quality inspection elements, the subsequent deep search model is used for in-depth search , To more accurately determine whether the target element appears in the text to be inspected.
  • Step S30 when it is determined to perform a deep search for the quality inspection element, search for a target element not found in the preliminary search result in the text to be quality inspection according to a preset depth search model to obtain a deep search result;
  • the preset depth search model can be a preset model that can accurately search the quality inspection elements in the quality inspection text, such as setting a text matching model or a machine reading comprehension model as the depth search model, and the time required for the depth search is longer. Long, but for the more complex customer service voice, it can more accurately determine whether there are quality inspection elements in the text to be inspected.
  • the in-depth search result can include only whether each target element is found or not, or it can include the search conditions of all quality inspection elements.
  • the preliminary search results are found Quality inspection elements, if 5 quality inspection elements are not found, perform an in-depth search on the 5 quality inspection elements that are not found, and find 3 quality inspection elements.
  • the obtained in-depth search result may include only these 3 quality inspection elements , It can also include a total of 8 quality inspection elements found.
  • the deep search model is used Will waste more time. If only the preliminary search model is used to search for quality inspection elements, although the time complexity can be reduced, the accuracy of the quality inspection is low. In this embodiment, by first performing a preliminary search according to a preset preliminary search model, most of the quality inspection elements are found. If the preliminary search results can meet the requirements for quality evaluation of the text to be inspected, no subsequent follow-up is required. In-depth search.
  • the quality evaluation conditions if most of the quality inspection elements are not found, then the default depth search model is used for the deep search for the quality inspection elements that are not found, because not all the quality inspection elements are carried out by the deep search model The search ensures the accuracy of the quality inspection and reduces the time complexity, thereby improving the efficiency of the quality inspection.
  • Step S40 Obtain the quality inspection result of the text to be inspected according to the deep search result.
  • the quality inspection result of the text to be inspected can be obtained according to the preliminary search result.
  • the quality inspection result of the text to be inspected can be obtained according to the in-depth search result, or according to In-depth search results and preliminary search results get the quality inspection results of the text to be inspected.
  • whether each quality inspection element is found in the preliminary search result can be recorded in the form of a report, and the report is used as the quality inspection result.
  • the quality inspection results corresponding to each customer service recording can be summarized and pushed to the quality inspection personnel together, so that the quality inspection personnel can know the search status of the quality inspection elements in each customer service recording.
  • a preliminary search is performed on the quality inspection elements in the text to be inspected according to the preset preliminary search model, and the preliminary search result is obtained; the preliminary search result is determined Whether to perform an in-depth search for quality inspection elements; if it is determined to perform an in-depth search for quality inspection elements, according to the preset depth search model, search for target elements that are not found in the preliminary search results in the text to be inspected to obtain the in-depth search results; According to the in-depth search result, the quality inspection result of the text to be inspected is obtained.
  • the second embodiment of the quality inspection method for customer service recording of this application provides a quality inspection method for customer service recording.
  • the preset preliminary search model is a pattern matching model
  • the step of searching for QC elements in the text to be QC according to the preset preliminary search model includes:
  • Step A10 obtaining the preset text mode of the quality inspection element
  • the pattern matching model refers to a model that matches the target text that conforms to the syntax rule in the text by describing the text pattern of a certain syntax rule.
  • the text pattern can be a regular expression.
  • the regular expression uses a single string to describe and match a series of strings that conform to a certain syntax rule.
  • the text mode corresponding to each quality inspection element can be set in advance. For example, the quality inspection element of "WeBank Staff", you can set the corresponding regular expression: r"(.*)(here)(.*)( ⁇ Public Bank Staff
  • the preset text mode of each quality inspection element can be saved in the database in advance.
  • the preliminary search process is: obtaining the preset text model of the quality inspection element, for example, the preset text mode corresponding to each quality inspection element can be obtained from the database.
  • Step A20 matching the preset text mode with the text to be inspected
  • this mode has a certain generalization and can cover a variety of expressions, for example, it can cover texts such as "Here WeBank is calling you”. If there is a text that meets this syntax rule in the text to be inspected, The match is successful. If there is no text that meets this syntax rule, the match fails.
  • Step A30 If the target text that matches the preset text mode is matched, it is determined that the quality inspection element corresponding to the preset text mode is found.
  • the target text that matches the preset text mode is matched, it is determined to find the quality inspection element corresponding to the preset text mode. If the above regular expression is used for matching and the target text of "Here WeBank is calling", it can be determined that the quality inspection element "WeBank staff" appears in the text to be inspected. The quality inspection element was found.
  • a preliminary search for quality inspection elements is performed through a pattern matching mode with low time complexity, which can cover most of the quality inspection elements, thereby reducing the time complexity of the entire quality inspection system and improving the efficiency of quality inspection.
  • the preset depth search model includes a text matching model
  • the step of searching for a target element not found in the preliminary search result in the text to be QC according to the preset depth search model includes:
  • Step B10 Obtain a text fragment to be inspected in the text to be inspected, and obtain a preset standard speech text of the target element;
  • the text matching model is a model that calculates the similarity between text and text.
  • the text matching model can calculate the similarity with multiple algorithms, such as logistic regression algorithm and bag-of-words algorithm.
  • the process of the deep search may be: obtaining a text segment to be quality-inspected in the text to be quality-inspected.
  • the sentence can be segmented in the text to be inspected according to the segmentation of the voice of the customer service through the preset sentence breaker. If the sentence is segmented by ",", the text to be inspected will be obtained The fragments can be divided into text to be inspected according to sentence breaks, and divided into multiple text fragments to be inspected.
  • the text matching model requires that the lengths of the two texts for calculating the similarity are not too large. Therefore, dividing the text to be quality-tested into segments of the text to be quality-tested can make the calculation result of the similarity more accurate.
  • the corresponding standard language text can be set in advance.
  • the quality inspection element of "WeBank staff” can set the corresponding standard language text: "Here is WeBank staff calling you” , You can also save the standard phonetics text in the database.
  • the target element that is not found in the preliminary search result is determined, and the preset standard speech text of the target element is obtained from the database.
  • the step of obtaining the text fragments to be QC in the text to be QC includes:
  • Step B101 according to the preset unsupervised segmentation model or the preset supervised segmentation model, perform text segmentation on the text to be quality-inspected to obtain text fragments to be quality-inspected.
  • the preset unsupervised segmentation model can use the edit distance algorithm, which can determine the correspondence between the standard quality inspection text and the text to be inspected, and segment the text to be inspected according to the corresponding relationship.
  • the preset supervised segmentation model can be a deep learning model that is trained in advance by training manually labeled data, and the deep learning model is used to segment the quality inspection text, for example, "Hello, this is WeBank Calling you, are you Mr. *”, divided into “Hello”, “WeBank is calling you”, and "Are you Mr. *” 3 text fragments pending quality inspection.
  • Step B20 Calculate the text similarity between the preset standard phonetic text and the text fragment to be QC according to the text matching model
  • the text similarity between the preset standard speech text and the text segment to be quality-checked is calculated.
  • the text similarity between the text fragments to be quality-checked "Here is the call from the staff of WeBank” and "Here is the call from WeBank” is calculated according to the logistic regression algorithm.
  • Step B30 detecting whether the text similarity is greater than a preset similarity
  • the preset similarity is set according to specific needs. For example, the range of the similarity is 0 to 1, and the preset similarity can be set to 0.8.
  • Step B40 If the text similarity is greater than the preset similarity, it is determined that the target element is found.
  • the text similarity is greater than the preset similarity, it indicates that the similarity between the text segment to be inspected and the preset standard speech text is high, and it can be determined that the target element corresponding to the preset standard speech text has been found.
  • the accuracy of quality inspection can be improved, and the false detection rate and false recall rate can be reduced; and the remaining Searching for unfound target elements also reduces the time complexity of the quality inspection system and improves the efficiency of quality inspection.
  • the in-depth search result can be obtained after the in-depth search is performed according to the text matching model.
  • the in-depth search result can be used as the final search result, or it can be determined according to the search condition of the quality inspection elements in the in-depth search result Whether to continue the further in-depth search. If all the quality inspection elements are found, no further in-depth search can be performed. If the number of quality inspection elements found is small and insufficient for service quality evaluation of the quality inspection text, you can Continue the subsequent deep search.
  • the preset depth search model further includes a reading comprehension model, and after the step B30, it further includes:
  • Step B50 If the text similarity is not greater than the preset similarity, locate the target element in the text to be QC according to the reading comprehension model;
  • the preset depth search model also includes a reading comprehension model, which is a model that detects whether a short paragraph of text appears in an article, such as the R-NET model.
  • the reading comprehension model can be used for in-depth search. For example, in this embodiment, if it is detected that the text similarity is not greater than the preset similarity, it means that the similarity between the text segment to be inspected and the preset standard speech text is not high, and it means that the text matching model is not found To the target element corresponding to the preset standard speech text.
  • the target element can be further searched using the reading comprehension model, and the target element can be located in the text to be inspected according to the reading comprehension model.
  • Step B60 If the target element is located, it is determined that the target element is found.
  • the reading comprehension model can find out the position of the target element in the text to be inspected, and give a classification label 0 or 1, where 0 means that the quality inspection element does not exist in the text to be inspected, and 1 means that the quality inspection element exists in the text to be inspected.
  • Quality inspection text For example, it is found that "WeBank staff" appears in a short text of length 8 starting from the 6th character of the text to be inspected, that is, the position is (6, 14), and label 1 is given.
  • the classification label output by the reading comprehension model it can be determined whether the target element is located. If the target element is located, the target element is determined to be found. At this time, the depth search result can be updated, and the updated depth search result is used as the final search result.
  • the target elements that are not found by the text matching model are further searched according to the reading comprehension model. Since the search accuracy of the reading comprehension model is better than that of the text matching model and the pattern matching model, the quality inspection The accuracy rate is further improved, and the false detection rate and false recall rate are further reduced; in addition, since the reading comprehension model is to find the target elements that have not been found, therefore, the time complexity of the entire quality inspection system is reduced and the Quality inspection efficiency.
  • the third embodiment of the quality inspection method for customer service recording of this application provides a quality inspection method for customer service recording.
  • the step S40 includes:
  • Step S401 scoring each of the quality inspection elements according to the in-depth search result and preset scoring rules
  • the quality inspection result of the text to be inspected is obtained according to the preliminary search result.
  • the quality inspection result of the text to be inspected can be obtained according to the in-depth search result, or according to the depth
  • the search results and preliminary search results obtain the quality inspection results of the text to be inspected.
  • the preliminary search results and the deep search results are collectively referred to as search results.
  • score each quality inspection element is obtained.
  • the search result includes the result of whether each quality inspection element is found.
  • the preset scoring rule can be set according to needs. For example, for each quality inspection element, if it is found, the quality inspection element will be scored 1 point. If it arrives, the quality inspection element is scored 0, and the score corresponding to each quality inspection element can also be different. For example, for an important quality inspection element, the score can be higher when it is found.
  • Step S402 Count the score of each quality inspection item according to the score of each quality inspection element, wherein the quality inspection item includes at least one of the quality inspection elements;
  • the score of each quality inspection item is counted according to the score of each quality inspection element, wherein the quality inspection item includes at least one quality inspection element.
  • the score of each quality inspection element under the quality inspection item can be added to obtain the score value of the quality inspection item, or it can be weighted average of the quality inspection elements under the quality inspection item to obtain the quality inspection item.
  • the score of the inspection item and the weight of each quality inspection element can be set in advance according to the importance of the quality inspection element.
  • step S403 the score of each quality inspection item is used as the quality inspection result of the text to be inspected.
  • the score of each quality inspection item may be saved in the form of a table, and as the quality inspection result, the table is pushed to the quality inspection personnel or the customer service personnel corresponding to the text to be inspected.
  • the score of each quality inspection item is obtained, and the score of the quality inspection item is used as the final quality inspection result, making the quality inspection result more intuitive and easy Understand, it is convenient for the quality inspection personnel to manage the quality inspection results.
  • a preferred quality inspection process provided by this embodiment of the application, according to the preferred quality inspection process to perform quality inspection on the text to be inspected: a preliminary search for quality inspection elements is performed through a pattern matching model; if If there is a quality inspection element that is not found, the text to be inspected is segmented to obtain the text fragment to be inspected, and the text similarity between the text fragment to be inspected and the standard speech text is calculated according to the text matching model. Determine the search for quality inspection elements; if there are still quality inspection elements that have not been found, follow the reading comprehension model to further search for the quality inspection elements that have not been found; finally score according to the search results to obtain the text to be inspected The score of each quality inspection item is used as the quality inspection result.
  • the quality inspection text is analyzed and scored, which can adapt to customer service recordings of varying degrees of complexity. While ensuring the accuracy of quality inspections, it reduces the time complexity of the quality inspection system and obtains intuitive quality inspection results. To facilitate the unified management of quality inspection results by quality inspectors.
  • an embodiment of the present application also proposes a quality inspection device for customer service recording.
  • the quality inspection device for customer service recording includes:
  • the preliminary search module 10 is configured to search for quality inspection elements in the text to be inspected according to a preset preliminary search model after obtaining the text to be inspected converted from the customer service recording, and obtain a preliminary search result;
  • the determining module 20 is configured to determine whether to perform an in-depth search for the quality inspection element according to the preliminary search result
  • the in-depth search module 30 is configured to, when it is determined to perform a deep search on the quality inspection element, in the text to be inspected according to a preset depth search model, search for target elements that are not found in the preliminary search result to obtain In-depth search results;
  • the quality inspection result generating module 40 is configured to obtain the quality inspection result of the text to be inspected according to the in-depth search result.
  • the preset preliminary search model is a pattern matching model
  • the preliminary search module 10 includes:
  • the first acquiring unit is configured to acquire the preset text mode of the quality inspection element
  • a matching unit configured to match the preset text mode with the text to be inspected
  • the first determining unit is configured to determine that the quality inspection element corresponding to the preset text mode is found if the target text that matches the preset text mode is matched.
  • the determining module 20 includes:
  • the first detection unit is configured to detect whether the number of elements of the target element not found in the preliminary search result is greater than a preset number
  • the second determining unit is configured to determine to perform an in-depth search on the quality inspection element if the number of elements is greater than the preset number; if the number of elements is not greater than the preset number, determine not to perform the quality inspection. Check elements for in-depth search.
  • the preset depth search model includes a text matching model
  • the depth search module 30 includes:
  • the second acquiring unit is configured to acquire a fragment of the text to be inspected in the text to be inspected, and to acquire the preset standard speech text of the target element;
  • a calculation unit configured to calculate the text similarity between the preset standard verbal text and the text fragment to be quality-checked according to the text matching model
  • the second detection unit is configured to detect whether the text similarity is greater than a preset similarity
  • the third determining unit is configured to determine that the target element is found if the text similarity is greater than the preset similarity.
  • the second acquiring unit includes:
  • the molecule segmentation unit is set to perform text segmentation on the text to be inspected according to a preset unsupervised segmentation model or a preset supervised segmentation model to obtain text fragments for quality inspection.
  • the preset depth search model further includes a reading comprehension model
  • the depth search module 30 further includes:
  • a positioning unit configured to locate the target element in the text to be inspected according to the reading comprehension model if the text similarity is not greater than the preset similarity
  • the fourth determining unit is configured to determine that the target element is found if the target element is located.
  • the quality inspection result generating module 40 includes:
  • the scoring unit is configured to score each of the quality inspection elements according to the in-depth search result and preset scoring rules
  • a statistical unit configured to count the score of each quality inspection item according to the score of each of the quality inspection elements, wherein the quality inspection item includes at least one of the quality inspection elements;
  • the result generating unit is configured to use the score of each of the quality inspection items as the quality inspection result of the text to be inspected.
  • an embodiment of the present application also proposes a computer-readable storage medium, the computer-readable storage medium stores a quality inspection program for customer service recording, and when the quality inspection program for customer service recording is executed by a processor, the implementation is as described above The steps of the quality inspection method for customer service recording.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Signal Processing (AREA)
  • Library & Information Science (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Machine Translation (AREA)

Abstract

本申请公开了一种客服录音的质检方法、装置、设备及计算机可读存储介质,所述方法包括:当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;根据所述初步查找结果确定是否对所述质检要素进行深度查找;当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;根据所述深度查找结果得到所述待质检文本的质检结果。本申请实现了在保证质检准确率同时,降低整个质检系统的时间复杂度,提高质检效率。

Description

客服录音的质检方法、装置、设备及计算机可读存储介质
本申请要求于2019年7月10日提交中国专利局、申请号为201910620601.5、发明名称为“客服录音的质检方法、装置、设备及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及一种客服录音的质检方法、装置、设备及计算机可读存储介质。
背景技术
随着企业对客服服务质量要求的提高,对客服录音进行质检的需求越来越大。传统的质检方式是靠人工听客服录音,进行抽查和评估,这种质检方式不仅耗费大量的人力、质检效率低,抽查的覆盖率也低,质检效果不佳。为解决人工质检效果不佳的问题,目前提出了基于人工智能技术对客服录音进行智能质检的质检系统。但是,目前的质检系统中对录音文本数据进行分析的过程,考虑到实际录音文本的复杂程度,以及为了提高质检的准确率,采用了时间复杂度较高的分析方式,从而使得整个质检系统的时间复杂度提高。
发明内容
本申请的主要目的在于提供一种客服录音的质检方法、装置、设备及计算机可读存储介质,旨在解决目前客服质检系统对录音文本进行分析,为保证质检准确率而导致的分析过程时间复杂度高的技术问题。
为实现上述目的,本申请提供一种客服录音的质检方法,所述客服录音的质检方法包括步骤:
当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
根据所述初步查找结果确定是否对所述质检要素进行深度查找;
当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;
根据所述深度查找结果得到所述待质检文本的质检结果。
可选地,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
获取所述质检要素的预设文本模式;
将所述预设文本模式与所述待质检文本进行匹配;
若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
可选地,所述根据所述初步查找结果确定是否对所述质检要素进行深度查找的步骤包括:
检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;
若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
可选地,所述预设深度查找模型包括文本匹配模型,所述按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素的步骤包括:
获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
检测所述文本相似度是否大于预设相似度;
若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
可选地,所述获取所述待质检文本中的待质检文本片段的步骤包括:
按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
可选地,所述预设深度查找模型还包括阅读理解模型,所述检测所述文本相似度是否大于预设相似度的步骤之后,还包括:
若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
若定位到所述目标要素,则确定查找到所述目标要素。
可选地,所述根据所述深度查找结果得到所述待质检文本的质检结果的步骤包括:
根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
将各所述质检项的分值作为所述待质检文本的质检结果。
此外,为实现上述目的,本申请还提供一种客服录音的质检装置,所述客服录音的质检装置包括:
初步查找模块,设置为当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
确定模块,设置为根据所述初步查找结果确定是否对所述质检要素进行深度查找;
深度查找模块,设置为当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;
质检结果生成模块,设置为根据所述深度查找结果得到所述待质检文本的质检结果。
此外,为实现上述目的,本申请还提供一种客服录音的质检设备,所述客服录音的质检设备包括存储器、处理器和存储在所述存储器上并可在所述处理器上运行的客服录音的质检程序,所述客服录音的质检程序被所述处理器执行时实现如上所述的客服录音的质检方法的步骤。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有客服录音的质检程序,所述客服录音的质检程序被处理器执行时实现如上所述的客服录音的质检方法的步骤。
本申请通过当获取到客服录音转化的待质检文本后,按照预设初步查找模型在待质检文本中对质检要素进行初步查找,得到初步查找结果;根据初步查找结果确定是否对质检要素进行深度查找;若确定对质检要素进行深度查找,则按照预设深度查找模型,在待质检文本中查找初步查找结果中未查找到的目标要素,得到深度查找结果;根据深度查找结果,得到待质检文本的质检结果,实现了在保证质检准确率同时,降低整个质检系统的时间复杂度,提高质检效率。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的结构示意图;
图2为本申请客服录音的质检方法第一实施例的流程示意图;
图3为本申请实施例涉及的一种质检流程示意图;
图4本申请客服录音的质检装置较佳实施例的功能示意图模块图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供了一种客服录音的质检设备,参照图1,图1是本申请实施例方案涉及的硬件运行环境的结构示意图。
需要说明的是,图1即可为客服录音的质检设备的硬件运行环境的结构示意图。本申请实施例客服录音的质检设备可以是PC,也可以是智能手机、智能电视机、平板电脑、便携计算机等具有显示功能的终端设备。
如图1所示,该客服录音的质检设备可以包括:处理器1001,例如CPU,网络接口1004,用户接口1003,存储器1005,通信总线1002。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
可选地,客服录音的质检设备还可以包括摄像头、RF(Radio Frequency,射频)电路,传感器、音频电路、WiFi模块等等。本领域技术人员可以理解,图1中示出的客服录音的质检设备结构并不构成对客服录音的质检设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及客服录音的质检程序。
在图1所示的客服录音的质检设备中,网络接口1004主要用于连接后台服务器,与后台服务器进行数据通信;用户接口1003主要用于连接客户端(用户端),与客户端进行数据通信;而处理器1001可以用于调用存储器1005中存储的客服录音的质检程序,并执行以下操作:
当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
根据所述初步查找结果 确定是否对所述质检要素进行深度查找;
当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;
根据所述所述深度查找结果得到所述待质检文本的质检结果。
进一步地,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
获取所述质检要素的预设文本模式;
将所述预设文本模式与所述待质检文本进行匹配;
若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
进一步地,所述根据所述初步查找结果确定是否对所述质检要素进行深度查找的步骤包括:
检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;
若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
进一步地,所述预设深度查找模型包括文本匹配模型,所述按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素的步骤包括:
获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
检测所述文本相似度是否大于预设相似度;
若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
进一步地,所述获取所述待质检文本中的待质检文本片段的步骤包括:
按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
进一步地,所述预设深度查找模型还包括阅读理解模型,所述检测所述文本相似度是否大于预设相似度的步骤之后,处理器1001可以调用存储器1005中存储的客服录音的质检程序,还执行以下操作:
若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
若定位到所述目标要素,则确定查找到所述目标要素。
进一步地,所述根据所述深度查找结果得到所述待质检文本的质检结果的步骤包括:
根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
将各所述质检项的分值作为所述待质检文本的质检结果。
基于上述的硬件结构,提出本申请客服录音的质检方法的各个实施例。
参照图2,本申请客服录音的质检方法第一实施例提供一种客服录音的质检方法,需要说明的是,虽然在流程图中示出了逻辑顺序,但是在某些情况下,可以以不同于此处的顺序执行所示出或描述的步骤。为便于描述,在以下各实施例中省略执行主体进行阐述。所述客服录音的质检方法包括:
步骤S10,当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
在本实施例中,预先可以对客服的电话服务过程进行录音,并将录音文件保存在数据库中。在需要对客服录音进行质检时,如接收到质检指令时,从数据库中获取客服的录音文件,对录音文件采用语音识别技术进行识别,将语音数据转化为文本数据。待质检文本是客服录音文件中,客服的语音转成的文本,而非用户的语音转成的文本,具体地,可以根据说话人的语音特征(如声纹特征),对录音数据中客服的语音和用户的语音进行区分。
可以预先设置质检项和质检要素作为质检指标。其中,质检项是指待质检文本中应当包括的开头语、安全提醒、身份证核实或确认本人等话术文本,例如:开头语的话术文本是“您好!这里是微众银行工作人员给您致电。感谢您使用微粒贷产品!本次来电主要是与您确认借款的相关信息,请问您现在通话方便吗?为了保证服务质量,本次通话可能会被录音,请您了解”。对于不同服务对象或服务类型的客服录音,可设置不同的质检项,如可设置4个质检项:开头语、安全提醒、身份核实和确认本人。质检要素是质检项的话说文本中的一个或几个片段,例如,开头语作为质检项时,可以包括“微众银行工作人员”、“与您确认借款”和“本次通话可能会被录音”这三个质检要素。待质检文本中是否出现各个质检要素,以及出现的多少是体现客服的服务质量好坏的评价标准。
当获取到客服录音转化的待质检文本后,按照预设初步查找模型在待质检文本中查找质检要素,得到初步查找结果。其中,预设初步查找模型可以是预先设置的能够对质检要素进行初步查找的模型,如预先设置模式匹配模型作为初步查找模型,初步查找所耗费的时间较少,对于复杂程度较低的客服录音,能够查找到大部分或全部的质检要素。在待质检文本中查找质检要素是指查找所有质检项下的质检要素。根据是否查找到各质检要素,得到初步查找结果,初步查找结果可能是包括未查找到的质检要素和查找到的质检要素,也可能是查找到所有的质检要素,也可能各质检要素都未查找到。
步骤S20,根据所述初步查找结果确定是否对所述质检要素进行深度查找;
根据初步查找结果确定是否对质检要素进行深度查找。具体地,可以是计算初步查找结果中,查找到的质检要素的数量占总质检要素数量的比例;当比例大于一个预设比例时,确定不进行深度查找,因为,当对待质检文本进行评分时,若该比例大于预设比例,说明待质检文本出现的质检要素很多,在此基础上,已经可以确定该待质检文本符合质量要求,即可不进行后续的深度查找,从而减少文本分析的时间,进而降低了整个质检系统的时间复杂度,提高了质检效率。当该比例不大于该预设比例时,说明待质检文本中出现的质检要素的个数不够确定该待质检文本符合质量要求,但是由于采用的是初步查找模型,待质检文本中可能还出现了其他质检要素,但是未被初步查找模型查找出来,此时可确定对质检要素进行深度深度,以通过后续的深度查找模型对质检要素进行进一步的查找,以确定初步查找未查找到的质检要素是否确实未出现在待质检文本中,从而提高质检系统的准确率。
进一步地,步骤S20包括:
步骤S201,检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
在得到初步查找结果后,将初步查找结果中未查找到的质检要素作为目标要素,检测目标要素的要素数量是否大于预设数量。其中,预设数量可根据具体需要进行设置,当对质检的准确率要求较高时,可将预设数量设置得较小。
步骤S202,若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;
若检测到要素数量大于预设数量,则确定对质检要素进行深度查找。当要素数量大于预设数量时,说明还有较多的质检要素没有查找到,此时,待质检文本中可能还出现了其他质检要素,但是未被初步查找模型查找出来,不能够直接确定待质检文本不符合质量要求,因此,可确定对质检要素进行深度查找。
步骤S203,若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
若检测到要素数量不大于预设数量,则确定不对质检要素进行深度查找,即由于未查找到的质检要素数量不多,此时已经可以确定待质检文本中出现了较多的质检要素,可以确定待质检文本符合质量要求,因此,不需要进行继续查找,从而节约文本分析时间,降低质检系统的时间复杂度。
进一步地,当对质检的准确率要求较高时,如当待质检文本中必须出现所有的质检要素才能够确定待质检文本符合质量要求时,可将预设数量设置为零,即检测初步查找结果是否有未查找到的质检要素,当大于零时,说明有未查找到的质检要素,只要有未查找到的质检要素,即采用后续的深度查找模型进行深度查找,以更准确地确定目标要素是否出现在待质检文本中。
步骤S30,当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;
当确定对质检要素进行深度查找时,按照预设深度查找模型在待质检文本中进行质检要素的查找,此时,查找的是初步查找结果中未查找到的目标要素,得到深度查找结果。其中,预设深度查找模型可以是预先设置的能够对待质检文本中的质检要素进行精确查找的模型,如设置文本匹配模型或机器阅读理解模型作为深度查找模型,深度查找所耗费的时间较长,但是对于复杂程度较高的客服语音,能够更准确地确定待质检文本中是否出现质检要素。深度查找结果中可以是仅包括各目标要素是否查找到的情况,也可以包括所有质检要素的查找情况,如当一共预设有10个质检要素时,初步查找结果中,查找到5个质检要素,5个质检要素未查找到,则对未查找到的5个质检要素进行深度查找,找到3个质检要素,得到的深度查找结果可以是仅包括这3个质检要素,也可以是包括一共查找到的8个质检要素。
若对于待质检文本,直接使用深度查找模型进行质检要素的查找,虽然能够保证质检的准确率,但是将耗费较长的时间,特别对于复杂程度较低的客服语音,采用深度查找模型会浪费较多时间。而若仅使用初步查找模型进行质检要素的查找,虽然能够减低时间复杂度,但是质检的准确率低。在本实施例中,通过先按照预设初步查找模型进行初步查找,查找到大部分的质检要素,若初步查找结果能够满足对待质检文本做出质量评价的条件,则不需进行后续的深度查找,如查找到大部分或者全部的质检要素时,不再进行后续的深度查找,从而减少了文本分析时间,降低了时间复杂度;若初步查找结果不能够满足对待质检文本做出质量评价的条件,如大部分质检要素都未查找到,则再对未查找到的质检要素采用预设深度查找模型进行深度查找,由于不是对全部的质检要素均采用深度查找模型进行查找,使得保证质检的准确率的同时,降低了时间复杂度,从而提高了质检效率。
步骤S40,根据所述深度查找结果得到所述待质检文本的质检结果。
当未进行后续的深度查找时,可根据初步查找结果得到待质检文本的质检结果,当进行了后续的深度查找时,可根据深度查找结果得到待质检文本的质检结果,或根据深度查找结果和初步查找结果得到待质检文本的质检结果。具体地,可将初步查找结果中各质检要素是否查找到的情况,以报表的形式进行记录,将报表作为质检结果。并可以将各客服录音对应的质检结果进行汇总,一并推送给质检人员,使得质检人员能够获知各个客服录音中质检要素的查找情况。
在本实施例中,通过当获取到客服录音转化的待质检文本后,按照预设初步查找模型在待质检文本中对质检要素进行初步查找,得到初步查找结果;根据初步查找结果确定是否对质检要素进行深度查找;若确定对质检要素进行深度查找,则按照预设深度查找模型,在待质检文本中查找初步查找结果中未查找到的目标要素,得到深度查找结果;根据深度查找结果,得到待质检文本的质检结果。本申请实现了在保证质检准确率同时,降低整个质检系统的时间复杂度,提高质检效率。
进一步的,基于上述第一实施例,本申请客服录音的质检方法第二实施例提供一种客服录音的质检方法。在本实施例中,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
步骤A10,获取所述质检要素的预设文本模式;
模式匹配模型是指通过描述某个句法规则的文本模式在文本中匹配符合该句法规则的目标文本的模型。其中,文本模式可以是正则表达式,正则表达式使用单个字符串来描述、匹配一系列符合某个句法规则的字符串。预先可以设置每个质检要素对应的文本模式,如“微众银行工作人员”这一质检要素,可以设置对应的正则表达式:r"(.*)(这里)(.*)(微众银行工作人员|微众银行)(.*)(给您致电)(.*)",一个质检要素也可以对应设置多个文本模式。预先可以将各质检要素的预设文本模式保存在数据库中。
当预设初步查找模型为模式匹配模型时,初步查找的过程为:获取质检要素的预设文本模型,如可从数据库中获取每一个质检要素对应的预设文本模式。
步骤A20,将所述预设文本模式与所述待质检文本进行匹配;
将预设文本模式与待质检文本进行匹配。如将r"(.*)(这里)(.*)(微众银行工作人员|微众银行)(.*)(给您致电)(.*)"这一正则表达式与待质检文本进行匹配,这一模式具有一定的泛化性,可以覆盖多种表达方式,如可以覆盖“这里微众银行给您致电”这样的文本,若待质检文本中出现符合这个句法规则的文本,则匹配成功,若未出现符合这个句法规则的文本,则匹配失败。
步骤A30,若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
若匹配到符合预设文本模式的目标文本,则确定查找到与预设文本模式对应的质检要素。如通过上述正则表达式进行匹配,匹配到“这里微众银行给您致电”这一目标文本,则可以确定待质检文本中出现了“微众银行工作人员”这一质检要素,也即查找到了该质检要素。
在本实施例中,通过时间复杂度低的模式匹配模式对质检要素进行初步查找,能够覆盖大部分的质检要素,从而降低了整个质检系统的时间复杂度,提高了质检效率。
进一步地,所述预设深度查找模型包括文本匹配模型,所述按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素的步骤包括:
步骤B10,获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
文本匹配模型是计算文本与文本之间相似度的模型,文本匹配模型计算相似度的算法可以有多种,如逻辑回归算法和词袋算法。当预设深度查找模型包括文本匹配模型时,深度查找的过程可以是:获取待质检文本中的待质检文本片段。其中,将识别客服录音转化为待质检文本时,可根据客服语音的断句,在待质检文本中通过预设的断句符进行断句,如通过“,”进行断句,则获取待质检文本片段可以是根据断句符对待质检文本进行切分,切分多个待质检文本片段。文本匹配模型要求计算相似度的两个文本的长度相隔不大,因此,通过待质检文本进行切分为待质检文本片段,可以使得相似度计算结果更加准确。
对于每个质检要素可预先设置对应的标准话术文本,如“微众银行工作人员”这个质检要素,可设置对应的标准话术文本:“这里是微众银行工作人员给您致电”,也可以将标准话术文本保存在数据库中。当确定进行深度查找时,确定初步查找结果中未查找到的目标要素,从数据库中获取目标要素的预设标准话术文本。
进一步地,所述获取所述待质检文本中的待质检文本片段的步骤包括:
步骤B101,按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
若待质检文本中未通过断句符进行断句,或者,通过根据断句符进行切分得到的待质检文本片段仍然很长时,可按照预设无监督切分模型或预设有监督切分模型,对待质检文本进行文本切分,得到待质检文本片段。其中,预设无监督切分模型可采用编辑距离算法,编辑距离算法可确定标准质检文本跟待质检文本之间的对应关系,根据对应关系来对待质检文本进行切分。预设有监督切分模型可以是,预先通过训练人工标注的数据,训练得到一个深度学习模型,通过该深度学习模型来对待质检文本进行切分,例如,将“您好这里是微众银行给您致电请问您是*先生吗”,切分为“您好”,“这里是微众银行给您致电”,“请问您是*先生吗”3个待质检文本片段。
步骤B20,按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
按照文本匹配模型计算预设标准话术文本和待质检文本片段之间的文本相似度。如按照逻辑回归算法计算预设标准话术文本“这里是微众银行工作人员给您致电”与“这里是微众银行给您致电”这一待质检文本片段之间的文本相似度。
步骤B30,检测所述文本相似度是否大于预设相似度;
检测计算得到的文本相似度是否大于预设相似度,其中,预设相似度根据具体需要进行设置,如相似度取值范围是0~1,则预设相似度可设置为0.8。
步骤B40,若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
若检测到文本相似度大于预设相似度,则说明待质检文本片段与预设标准话术文本之间的相似度高,则可确定查找到了预设标准话术文本对应的目标要素。
在本实施例中,通过对初步查找结果中未查找到的目标要素,按照文本匹配模型进行深度查找,可提高质检的准确率,降低误检率、误召回率;并且通过文本模型对剩余未找到的目标要素进行查找,也降低了质检系统的时间复杂度,提高了质检效率。
需要说明的是,按照文本匹配模型进行深度查找后,可得到深度查找结果,此时,可以将深度查找结果作为最终的查找结果,也可以是根据深度查找结果中质检要素的查找情况,确定是否继续进行进一步的深度查找,若质检要素全部查找到,则可以不再进行后续的深度查找,若查找到的质检要素的数量较少,不够对待质检文本作出服务质量评价时,可继续进行后续的深度查找。
进一步的,所述预设深度查找模型还包括阅读理解模型,所述步骤B30之后,还包括:
步骤B50,若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
预设深度查找模型还包括阅读理解模型,阅读理解模型是检测一小段文本是否出现在一篇文章中的一种模型,如R-NET模型。当确定需要进一步对未查找到的目标要素进行深度查找时,可采用阅读理解模型进行深度查找。如在本实施例中,若检测到文本相似度不大于预设相似度,则说明待质检文本片段与预设标准话术文本之间的相似度不高,则说明通过文本匹配模型未查找到该预设标准话术文本对应的目标要素。此时可对该目标要素采用阅读理解模型进行进一步地查找,按照阅读理解模型在待质检文本中对目标要素进行定位。如在从“您好这里是微众银行工作人员给您致电感谢您使用微粒贷产品本次来电主要是与您确认借款的相关信息请问您现在通话方便吗为了保证服务质量本次通话可能会被录音请您了解”中对“微众银行工作人员”这一目标要素进行定位。
步骤B60,若定位到所述目标要素,则确定查找到所述目标要素。
阅读理解模型可以找出目标要素在待质检文本中的位置,并给出一个分类标签0或1,其中0表示质检要素不存在于待质检文本中,1表示质检要素存在于待质检文本中。如找出“微众银行工作人员”出现在待质检文本的第6个字开始的、长度为8的小段文本中,即位置是(6,14),并给出标签1。根据阅读理解模型输出的分类标签,即可确定是否定位到目标要素。若定位到目标要素,则确定查找到目标要素,此时,可更新深度查找结果,将更新后的深度查找结果作为最终的查找结果。
在本实施例中,通过对文本匹配模型未查找到的目标要素,按照阅读理解模型进行更进一步的查找,由于阅读理解模型的查找准确率要优于文本匹配模型和模式匹配模型,使得质检的准确率进一步提高,也进一步降低了误检率和误召回率;此外,由于阅读理解模型是对未查找到的目标要素进行查找,因此,降低了整个质检系统的时间复杂度,提高了质检效率。
进一步地,也可以是在按照模式匹配模型进行初步查找后,确定需要进行后续的深度查找时,跳过按照文本匹配模型进行深度查找的过程,直接采用阅读理解模型进行深度查找。
进一步的,基于上述第一或第二实施例,本申请客服录音的质检方法第三实施例提供一种客服录音的质检方法。在本实施例中,所述步骤S40包括:
步骤S401,根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
当未进行后续的深度查找时,根据初步查找结果得到待质检文本的质检结果,当进行了后续的深度查找时,可根据深度查找结果得到待质检文本的质检结果,或根据深度查找结果和初步查找结果得到待质检文本的质检结果。以下,将初步查找结果和深度查找结果统称为查找结果。根据查找结果,以及预设打分规则,给各个质检要素进行打分。其中,查找结果包括各个质检要素的是否查找到的结果,预设打分规则可以是根据需要进行设置,如对于每个质检要素,查找到则给该质检要素打1分,若未查找到,则给该质检要素打0分,每个质检要素对应的分值也可以不同,如对于重要的质检要素,查找到时其得分可以更高。
步骤S402,根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
根据各质检要素的分值统计各质检项的分值,其中,质检项包括至少一个质检要素。具体地,可以是将质检项下的各个质检要素的分值相加,得到该质检项的分值,也可以是将质检项下的各个质检要素进行加权平均,得到该质检项的分值,各个质检要素的权重可以预先根据质检要素的重要程度进行设置。
步骤S403,将各所述质检项的分值作为所述待质检文本的质检结果。
将计算得到的各个质检项的分值作为待质检文本的质检结果。具体地,可以将各质检项的分值以表格的形式进行保存,作为质检结果,将表格推送给质检人员或者是该待质检文本对应的客服人员。各个质检项的得分越高,说明客服的服务质量越高。
在本实施例中,通过对待质检文本中各个质检要素进行打分,得到各个质检项的分值,将质检项的分值作为最终的质检结果,使得质检结果更加直观、易懂,便于质检人员的对质检结果的统一管理。
进一步地,如图3所示,为本申请实施例提供的一种优选质检流程,按照该优选质检流程对待质检文本进行质检:通过模式匹配模型对质检要素进行初步查找;若有未查找到的质检要素,则对待质检文本进行文本切分得到待质检文本片段,按照文本匹配模型计算待质检文本片段和标准话术文本之间的文本相似度,根据相似度确定质检要素的查找情况;若还有未查找到的质检要素,则按照阅读理解模型再对未查找到的质检要素进行进一步的查找;最后根据查找结果进行打分,得到待质检文本各质检项的分值,作为质检结果。按照该质检流程对待质检文本进行分析和打分,可以适应复杂程度不一的客服录音,在保证质检准确率的同时,降低质检系统的时间复杂度,同时得到直观的质检结果,方便质检人员对质检结果的统一管理。
此外,本申请实施例还提出一种客服录音的质检装置,参照图4,所述客服录音的质检装置包括:
初步查找模块10,设置为当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
确定模块20,设置为根据所述初步查找结果确定是否对所述质检要素进行深度查找;
深度查找模块30,设置为当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;
质检结果生成模块40,设置为根据所述深度查找结果得到所述待质检文本的质检结果。
进一步地,所述预设初步查找模型为模式匹配模型,所述初步查找模块10包括:
第一获取单元,设置为获取所述质检要素的预设文本模式;
匹配单元,设置为将所述预设文本模式与所述待质检文本进行匹配;
第一确定单元,设置为若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
进一步地,所述确定模块20包括:
第一检测单元,设置为检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
第二确定单元,设置为若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
进一步地,所述预设深度查找模型包括文本匹配模型,所述深度查找模块30包括:
第二获取单元,设置为获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
计算单元,设置为按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
第二检测单元,设置为检测所述文本相似度是否大于预设相似度;
第三确定单元,设置为若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
进一步地,所述第二获取单元包括:
切分子单元,设置为按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
进一步地,所述预设深度查找模型还包括阅读理解模型,所述深度查找模块30还包括:
定位单元,设置为若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
第四确定单元,设置为若定位到所述目标要素,则确定查找到所述目标要素。
进一步地,所述质检结果生成模块40包括:
打分单元,设置为根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
统计单元,设置为根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
结果生成单元,设置为将各所述质检项的分值作为所述待质检文本的质检结果。
本申请客服录音的质检装置的具体实施方式的拓展内容与上述客服录音的质检方法各实施例基本相同,在此不做赘述。
此外,本申请实施例还提出一种计算机可读存储介质,所述计算机可读存储介质上存储有客服录音的质检程序,所述客服录音的质检程序被处理器执行时实现如上所述客服录音的质检方法的步骤。
本申请客服录音的质检设备和计算机可读存储介质的具体实施方式的拓展内容与上述客服录音的质检方法各实施例基本相同,在此不做赘述。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种客服录音的质检方法,其中,所述客服录音的质检方法包括:
    当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
    根据所述初步查找结果确定是否对所述质检要素进行深度查找;
    当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;以及,
    根据所述深度查找结果得到所述待质检文本的质检结果。
  2. 如权利要求1所述的客服录音的质检方法,其中,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
    获取所述质检要素的预设文本模式;
    将所述预设文本模式与所述待质检文本进行匹配;
    若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
  3. 如权利要求1所述的客服录音的质检方法,其中,所述根据所述初步查找结果确定是否对所述质检要素进行深度查找的步骤包括:
    检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
    若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;
    若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
  4. 如权利要求1所述的客服录音的质检方法,其中,所述预设深度查找模型包括文本匹配模型,所述按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素的步骤包括:
    获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
    按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
    检测所述文本相似度是否大于预设相似度;
    若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
  5. 如权利要求4所述的客服录音的质检方法,其中,所述获取所述待质检文本中的待质检文本片段的步骤包括:
    按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
  6. 如权利要求5所述的客服录音的质检方法,其中,所述预设深度查找模型还包括阅读理解模型,所述检测所述文本相似度是否大于预设相似度的步骤之后,还包括:
    若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
    若定位到所述目标要素,则确定查找到所述目标要素。
  7. 如权利要求1所述的客服录音的质检方法,其中,所述根据所述深度查找结果得到所述待质检文本的质检结果的步骤包括:
    根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
    根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
    将各所述质检项的分值作为所述待质检文本的质检结果。
  8. 一种客服录音的质检装置,其中,所述客服录音的质检装置包括:
    初步查找模块,设置为当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
    确定模块,设置为根据所述初步查找结果确定是否对所述质检要素进行深度查找;
    深度查找模块,设置为当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;以及,
    质检结果生成模块,设置为根据所述深度查找结果得到所述待质检文本的质检结果。
  9. 如权利要求8所述的客服录音的质检装置,其中,所述预设初步查找模型为模式匹配模型,所述初步查找模块10包括:
    第一获取单元,设置为获取所述质检要素的预设文本模式;
    匹配单元,设置为将所述预设文本模式与所述待质检文本进行匹配;
    第一确定单元,设置为若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
  10. 如权利要求8所述的客服录音的质检装置,其中,所述确定模块包括:
    第一检测单元,设置为检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
    第二确定单元,设置为若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
  11. 如权利要求8所述的客服录音的质检装置,其中,所述预设深度查找模型包括文本匹配模型,所述深度查找模块包括:
    第二获取单元,设置为获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
    计算单元,设置为按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
    第二检测单元,设置为检测所述文本相似度是否大于预设相似度;
    第三确定单元,设置为若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
  12. 一种客服录音的质检设备,其中,所述客服录音的质检设备包括存储器、处理器和存储在所述存储器上并可在所述处理器上运行的客服录音的质检程序,所述客服录音的质检程序被所述处理器执行时实现如下步骤:
    当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
    根据所述初步查找结果确定是否对所述质检要素进行深度查找;
    当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;以及,
    根据所述深度查找结果得到所述待质检文本的质检结果。
  13. 如权利要求12所述的客服录音的质检设备,其中,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
    获取所述质检要素的预设文本模式;
    将所述预设文本模式与所述待质检文本进行匹配;
    若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
  14. 如权利要求12所述的客服录音的质检设备,其中,所述根据所述初步查找结果确定是否对所述质检要素进行深度查找的步骤包括:
    检测所述初步查找结果中未查找到的所述目标要素的要素数量是否大于预设数量;
    若所述要素数量大于所述预设数量,则确定对所述质检要素进行深度查找;
    若所述要素数量不大于所述预设数量,则确定不对所述质检要素进行深度查找。
  15. 如权利要求12所述的客服录音的质检设备,其中,所述预设深度查找模型包括文本匹配模型,所述按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素的步骤包括:
    获取所述待质检文本中的待质检文本片段,以及获取所述目标要素的预设标准话术文本;
    按照所述文本匹配模型计算所述预设标准话术文本和所述待质检文本片段之间的文本相似度;
    检测所述文本相似度是否大于预设相似度;
    若所述文本相似度大于所述预设相似度,则确定查找到所述目标要素。
  16. 如权利要求15所述的客服录音的质检设备,其中,所述获取所述待质检文本中的待质检文本片段的步骤包括:
    按照预设无监督切分模型或预设有监督切分模型,对所述待质检文本进行文本切分,得到待质检文本片段。
  17. 如权利要求16所述的客服录音的质检设备,其中,所述预设深度查找模型还包括阅读理解模型,所述检测所述文本相似度是否大于预设相似度的步骤之后,还包括:
    若所述文本相似度不大于所述预设相似度,则按照所述阅读理解模型在所述待质检文本中对所述目标要素进行定位;
    若定位到所述目标要素,则确定查找到所述目标要素。
  18. 如权利要求12所述的客服录音的质检设备,其中,所述根据所述深度查找结果得到所述待质检文本的质检结果的步骤包括:
    根据所述深度查找结果,以及预设打分规则给各所述质检要素进行打分;
    根据各所述质检要素的分值统计各质检项的分值,其中,所述质检项包括至少一个所述质检要素;
    将各所述质检项的分值作为所述待质检文本的质检结果。
  19. 一种计算机可读存储介质,其中,所述计算机可读存储介质上存储有客服录音的质检程序,所述客服录音的质检程序被处理器执行时实现如下步骤:
    当获取到客服录音转化的待质检文本后,按照预设初步查找模型在所述待质检文本中查找质检要素,得到初步查找结果;
    根据所述初步查找结果确定是否对所述质检要素进行深度查找;
    当确定对所述质检要素进行深度查找时,按照预设深度查找模型在所述待质检文本中,查找所述初步查找结果中未查找到的目标要素,得到深度查找结果;以及,
    根据所述深度查找结果得到所述待质检文本的质检结果。
  20. 如权利要求19所述的计算机可读存储介质,其中,所述预设初步查找模型为模式匹配模型,所述按照预设初步查找模型在所述待质检文本中查找质检要素的步骤包括:
    获取所述质检要素的预设文本模式;
    将所述预设文本模式与所述待质检文本进行匹配;
    若匹配到符合所述预设文本模式的目标文本,则确定查找到所述预设文本模式对应的所述质检要素。
PCT/CN2019/117539 2019-07-10 2019-11-12 客服录音的质检方法、装置、设备及计算机可读存储介质 WO2021003930A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910620601.5A CN110334241B (zh) 2019-07-10 2019-07-10 客服录音的质检方法、装置、设备及计算机可读存储介质
CN201910620601.5 2019-07-10

Publications (1)

Publication Number Publication Date
WO2021003930A1 true WO2021003930A1 (zh) 2021-01-14

Family

ID=68145904

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117539 WO2021003930A1 (zh) 2019-07-10 2019-11-12 客服录音的质检方法、装置、设备及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110334241B (zh)
WO (1) WO2021003930A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112885376A (zh) * 2021-01-23 2021-06-01 深圳通联金融网络科技服务有限公司 一种提高语音通话质检效果的方法和装置
CN113128390A (zh) * 2021-04-14 2021-07-16 北京奇艺世纪科技有限公司 抽检方法、装置、电子设备及存储介质
CN113407494A (zh) * 2021-05-27 2021-09-17 东软集团股份有限公司 一种违规文件检测方法、装置及设备
CN113434670A (zh) * 2021-06-22 2021-09-24 未鲲(上海)科技服务有限公司 话术文本生成方法、装置、计算机设备和存储介质

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110334241B (zh) * 2019-07-10 2023-08-25 深圳前海微众银行股份有限公司 客服录音的质检方法、装置、设备及计算机可读存储介质
TWI725577B (zh) * 2019-10-17 2021-04-21 中國信託商業銀行股份有限公司 智能語音資訊質檢方法及系統
CN110931014A (zh) * 2019-12-13 2020-03-27 集奥聚合(北京)人工智能科技有限公司 基于正则匹配规则的语音识别方法及装置
CN111010484A (zh) * 2019-12-16 2020-04-14 科大国创软件股份有限公司 一种通话录音自动质检方法
CN111314566A (zh) * 2020-01-20 2020-06-19 北京神州泰岳智能数据技术有限公司 一种语音质检方法、装置与系统
CN111368130A (zh) * 2020-02-26 2020-07-03 深圳前海微众银行股份有限公司 客服录音的质检方法、装置、设备及存储介质
CN111291162B (zh) * 2020-02-26 2024-04-09 深圳前海微众银行股份有限公司 质检例句挖掘方法、装置、设备及计算机可读存储介质
CN111696528B (zh) * 2020-06-20 2021-04-23 龙马智芯(珠海横琴)科技有限公司 一种语音质检方法、装置、质检设备及可读存储介质
CN112069796B (zh) * 2020-09-03 2023-08-04 阳光保险集团股份有限公司 一种语音质检方法、装置,电子设备及存储介质
CN112037819A (zh) * 2020-09-03 2020-12-04 阳光保险集团股份有限公司 一种基于语义的语音质检方法和装置
CN112329437B (zh) * 2020-10-21 2024-05-28 交通银行股份有限公司 一种智能客服语音质检评分方法、设备及存储介质
CN112699220A (zh) * 2021-01-07 2021-04-23 北京金万维科技有限公司 一种话术匹配的智能质检方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109389971A (zh) * 2018-08-17 2019-02-26 深圳壹账通智能科技有限公司 基于语音识别的保险录音质检方法、装置、设备和介质
CN109446524A (zh) * 2018-10-25 2019-03-08 第四范式(北京)技术有限公司 一种语音质检方法及装置
CN109658923A (zh) * 2018-10-19 2019-04-19 平安科技(深圳)有限公司 基于人工智能的语音质检方法、设备、存储介质及装置
CN110334241A (zh) * 2019-07-10 2019-10-15 深圳前海微众银行股份有限公司 客服录音的质检方法、装置、设备及计算机可读存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105141787A (zh) * 2015-08-14 2015-12-09 上海银天下科技有限公司 服务录音的合规检查方法及装置
CN107093431B (zh) * 2016-02-18 2020-07-07 中国移动通信集团辽宁有限公司 一种对服务质量进行质检的方法及装置
CN108737667B (zh) * 2018-05-03 2021-09-10 平安科技(深圳)有限公司 语音质检方法、装置、计算机设备及存储介质
CN109271489B (zh) * 2018-10-25 2020-12-15 第四范式(北京)技术有限公司 一种文本检测方法及装置
CN109327632A (zh) * 2018-11-23 2019-02-12 深圳前海微众银行股份有限公司 客服录音的智能质检系统、方法及计算机可读存储介质
CN109902957B (zh) * 2019-02-28 2022-12-09 腾讯科技(深圳)有限公司 一种数据处理方法和装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109389971A (zh) * 2018-08-17 2019-02-26 深圳壹账通智能科技有限公司 基于语音识别的保险录音质检方法、装置、设备和介质
CN109658923A (zh) * 2018-10-19 2019-04-19 平安科技(深圳)有限公司 基于人工智能的语音质检方法、设备、存储介质及装置
CN109446524A (zh) * 2018-10-25 2019-03-08 第四范式(北京)技术有限公司 一种语音质检方法及装置
CN110334241A (zh) * 2019-07-10 2019-10-15 深圳前海微众银行股份有限公司 客服录音的质检方法、装置、设备及计算机可读存储介质

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112885376A (zh) * 2021-01-23 2021-06-01 深圳通联金融网络科技服务有限公司 一种提高语音通话质检效果的方法和装置
CN113128390A (zh) * 2021-04-14 2021-07-16 北京奇艺世纪科技有限公司 抽检方法、装置、电子设备及存储介质
CN113407494A (zh) * 2021-05-27 2021-09-17 东软集团股份有限公司 一种违规文件检测方法、装置及设备
CN113407494B (zh) * 2021-05-27 2024-02-09 东软集团股份有限公司 一种违规文件检测方法、装置及设备
CN113434670A (zh) * 2021-06-22 2021-09-24 未鲲(上海)科技服务有限公司 话术文本生成方法、装置、计算机设备和存储介质
WO2022267174A1 (zh) * 2021-06-22 2022-12-29 未鲲(上海)科技服务有限公司 话术文本生成方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
CN110334241A (zh) 2019-10-15
CN110334241B (zh) 2023-08-25

Similar Documents

Publication Publication Date Title
WO2021003930A1 (zh) 客服录音的质检方法、装置、设备及计算机可读存储介质
WO2020034526A1 (zh) 保险录音的质检方法、装置、设备和计算机存储介质
WO2020107756A1 (zh) 信贷反欺诈方法、系统、设备及计算机可读存储介质
WO2021132927A1 (en) Computing device and method of classifying category of data
WO2017047884A1 (en) Voice recognition server and control method thereof
WO2019216529A1 (en) Electronic apparatus and controlling method thereof
WO2015178600A1 (en) Speech recognition method and apparatus using device information
WO2015068947A1 (ko) 녹취된 음성 데이터에 대한 핵심어 추출 기반 발화 내용 파악 시스템과, 이 시스템을 이용한 인덱싱 방법 및 발화 내용 파악 방법
WO2018166099A1 (zh) 信息泄露检测方法及装置、服务器及计算机可读存储介质
WO2020147385A1 (zh) 数据录入方法、装置、终端及计算机可读存储介质
WO2020168606A1 (zh) 广告视频优化方法、装置、设备及计算机可读存储介质
WO2018201638A1 (zh) 基于图像识别的信息采集方法、移动终端及存储介质
WO2013122269A1 (ko) 단말기의 사용자 인터페이스 제공 방법
WO2020082766A1 (zh) 输入法的联想方法、装置、设备及可读存储介质
WO2012130145A1 (zh) 获取和搜索相关知识信息的方法及装置
WO2021002584A1 (ko) 음성을 통한 전자문서 제공 방법, 음성을 통한 전자문서 작성 방법 및 장치
WO2021003956A1 (zh) 产品信息的管理方法、装置、设备及存储介质
WO2012165709A1 (ko) 인스턴스 경로 탐색 및 시각화 방법 및 장치
WO2020119369A1 (zh) 智能it运维故障定位方法、装置、设备及可读存储介质
EP3494498A1 (en) Electronic device and method for controlling the same
WO2021012481A1 (zh) 系统性能监控方法、装置、设备及存储介质
WO2022164192A1 (ko) 사용자의 발화 입력에 관련된 추천 문장을 제공하는 디바이스 및 방법
WO2021029643A1 (en) System and method for modifying speech recognition result
WO2021051557A1 (zh) 基于语义识别的关键词确定方法、装置和存储介质
WO2020138662A1 (ko) 전자 장치 및 그의 제어 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19936951

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19936951

Country of ref document: EP

Kind code of ref document: A1