Intelligent quality inspection method and device based on intention recognition and readable storage medium
By adopting intelligent quality inspection methods based on intention recognition in the call center, using deep learning algorithms and multi-pattern matching technology, the problems of low efficiency and insufficient accuracy of traditional artificial quality inspection are solved, and more efficient and accurate quality inspection results are achieved .
Patent Information
- Application Number
- CN202411081666.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-22
AI Technical Summary
Traditional artificial quality inspections have problems in call centers with low coverage, low efficiency and susceptible to subjective factors, and it is difficult to meet the high efficiency and high-quality service needs of modern call centers.
Using intelligent quality inspection method based on intent recognition, keyword matching is carried out by obtaining voice text, preset keyword matching, and combining deep learning algorithms to build a natural language understanding intent recognition model, combining multi-pattern matching technology, integrating keyword rules and semantics Information to realize accurate scoring of quality inspection rules and weighted calculation of quality inspection results.
It significantly improves the accuracy of quality inspection and the accuracy of complex data identification, and improves the accuracy and efficiency of quality inspection results.
Smart Images

Figure CN119005202A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to an intelligent quality inspection method, device and readable storage medium based on intent recognition. Background Art
[0002] In the rapidly developing digital age, call centers serve as an important bridge for communication between enterprises and customers. The service quality of call centers directly affects the customer satisfaction of enterprises. Among them, quality inspection is one of the indispensable operation processes. Call center quality inspection not only directly supervises the agents and promotes the improvement of the agent's business capabilities, but also optimizes the enterprise's operation process and enhances the company's external image.
[0003] The traditional quality inspection method uses manual sampling inspection, which has low coverage, low efficiency and is easily affected by subjective factors. With the surge in business volume and the diversification of customer needs, manual quality inspection can no longer meet the high efficiency and high-quality service needs of modern call centers.
[0004] Therefore, how to realize the intelligence of the quality inspection process and improve the accuracy of quality inspection results has become a problem that needs to be solved. Summary of the invention
[0005] The technical problem to be solved by the present application is to provide an intelligent quality inspection method, device and readable storage medium based on intent recognition to solve the problems existing in the prior art in view of the above-mentioned deficiencies in the prior art.
[0006] In a first aspect, the present application provides an intelligent quality inspection method based on intent recognition, the method comprising:
[0007] S1. Obtain the voice text to be inspected;
[0008] S2. Based on preset keywords, keyword matching is performed on the voice text to obtain a first quality inspection score;
[0009] S3, performing intent recognition on the speech text, and obtaining a second quality inspection score according to the intent recognition result;
[0010] S4. Obtain a quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
[0011] In some embodiments, S2 includes:
[0012] S21. Based on the preset quality inspection rules, obtain corresponding preset keywords;
[0013] S22, establishing a dictionary tree according to the preset keywords, and constructing a failure pointer;
[0014] S23, performing keyword matching on the voice text through the dictionary tree and the failure pointer to obtain a matching result;
[0015] S24. Obtain a first quality inspection score according to the matching result.
[0016] In some embodiments, S22 includes:
[0017] Establishing a dictionary tree according to the preset keywords, wherein each node in the dictionary tree represents a character;
[0018] A failure pointer is constructed for each node in the dictionary tree, and the failure pointer points to the next node to be matched after the current node fails to match.
[0019] In some embodiments, S23 includes:
[0020] The voice text is matched with the dictionary tree character by character. If the current node fails to match, jump to the next node according to the failure pointer to continue matching.
[0021] In some embodiments, S3 includes:
[0022] Calling a regular matching module to match the speech text;
[0023] If a preset mode is matched, the intent category corresponding to the preset mode is output;
[0024] If the preset pattern is not matched, the binary classification model is called to determine whether it is an invalid intention. If so, the invalid intention is output; if not, the multi-classification model is called to predict the intention and obtain the intention prediction result.
[0025] In some embodiments, calling a multi-classification model to perform intent prediction and obtaining an intent prediction result includes:
[0026] Input the speech text into a multi-classification model, and obtain word vectors and character vectors through word and character segmentation;
[0027] Based on the word vector and the character vector, obtaining a first text feature through a convolutional neural network (CNN) module;
[0028] Based on the word vector and character vector, a second text feature is obtained by fusing a bidirectional long short-term memory neural network BILSTM module based on an attention mechanism;
[0029] The first text feature and the second text feature are concatenated and classified through a fully connected layer to obtain an intent prediction result.
[0030] In some embodiments, S4 includes:
[0031] A weighted sum is performed based on the first quality inspection score and the second quality inspection score to obtain a quality inspection result of the speech text.
[0032] In a second aspect, the present application provides an intelligent quality inspection device based on intent recognition, the device comprising:
[0033] A voice text acquisition module, which is configured to acquire the voice text to be inspected;
[0034] A keyword matching module, which is configured to perform keyword matching on the voice text based on preset keywords to obtain a first quality inspection score;
[0035] An intention recognition module, configured to perform intention recognition on the speech text and obtain a second quality inspection score according to the intention recognition result;
[0036] A quality inspection result determination module is configured to obtain a quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
[0037] In a third aspect, the present application provides an intelligent quality inspection device based on intent recognition, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the intelligent quality inspection method based on intent recognition described in the first aspect above.
[0038] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the intelligent quality inspection method based on intent recognition described in the first aspect above is implemented.
[0039] The present application provides an intelligent quality inspection method, device and readable storage medium based on intent recognition, the method comprising: obtaining a voice text to be quality inspected; performing keyword matching on the voice text based on preset keywords to obtain a first quality inspection score; performing intent recognition on the voice text, and obtaining a second quality inspection score according to the intent recognition result; obtaining the quality inspection result of the voice text according to the first quality inspection score and the second quality inspection score. The present application provides an intelligent quality inspection method based on intent recognition, which uses a deep learning algorithm to build a natural language understanding intent recognition model, combines multi-mode matching technology, and integrates keyword rules and semantic information to achieve accurate scoring and quantification of each quality inspection rule, and obtains the final quality inspection score by weighting. The present application uses an intelligent quality inspection method based on intent recognition, and uses a deep learning algorithm to build an NLU intent recognition model, which greatly improves the accuracy of quality inspection, and significantly improves the recognition accuracy of complex data. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0041] Figure 1 A flowchart of an intelligent quality inspection method based on intent recognition provided in an embodiment of the present application;
[0042] Figure 2 A flowchart of another intelligent quality inspection method based on intent recognition provided in an embodiment of the present application;
[0043] Figure 3 Schematic diagram of speech-to-text conversion provided in the embodiment of the present application
[0044] Figure 4 A schematic diagram of intent recognition classification provided in an embodiment of the present application,
[0045] Figure 5 A schematic diagram of the Skip-Gram model provided in the embodiment of the present application,
[0046] Figure 6 Schematic diagram of a multi-classification model provided in an embodiment of the present application
[0047] Figure 7 Schematic diagram of determining the final quality inspection result provided in the embodiment of the present application
[0048] Figure 8 A schematic diagram of the structure of an intelligent quality inspection device based on intent recognition provided in an embodiment of the present application;
[0049] Fig. 9 A schematic diagram of the structure of another intelligent quality inspection device based on intent recognition provided in an embodiment of the present application.
[0050] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0051] In order to enable those skilled in the art to better understand the technical solution of the present application, the implementation mode of the present application will be further described in detail below with reference to the accompanying drawings.
[0052] It should be understood that the specific embodiments and drawings described herein are only used to explain the present application, rather than to limit the present application.
[0053] It can be understood that, in the absence of conflict, the various embodiments in the present application and the various features in the embodiments can be combined with each other.
[0054] It can be understood that, for the convenience of description, the drawings of the present application only show the parts related to the present application, while the parts unrelated to the present application are not shown in the drawings.
[0055] It can be understood that each unit and module involved in the embodiments of the present application may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.
[0056] It can be understood that the terms "first", "second", etc. in the embodiments of the present application are used to distinguish different objects, or to distinguish different processing of the same object, rather than to describe a specific order of objects.
[0057] It is understandable that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present application may occur in an order different from that marked in the drawings.
[0058] It is understood that the flowcharts and block diagrams of the present application illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present application. Among them, each box in the flowchart or block diagram may represent a unit, module, program segment, code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or by a combination of hardware and computer instructions.
[0059] It can be understood that the units and modules involved in the embodiments of the present application can be implemented by software or hardware, for example, the units and modules can be located in a processor.
[0060] At present, in order to improve the efficiency of quality inspection, intelligent quality inspection systems have emerged. Common intelligent quality inspection systems use natural language processing (NLP), automatic speech recognition (ASR) and big data analysis technology to automatically monitor and evaluate the service process. This technology can not only analyze the content of the call, but also accurately capture the tone, speed and customer response of the service staff, thereby comprehensively improving the service quality.
[0061] However, the accuracy of quality inspection results obtained by existing technologies still needs to be improved, especially when processing complex and emotionally variable speech data.
[0062] In order to improve the accuracy of quality inspection results, this application proposes an intelligent quality inspection method based on intent recognition.
[0063] The main ideas of this application are:
[0064] This application uses natural language processing (NLP) and deep learning technology to build a user conversation intention recognition model. On the basis of quality inspection rules, it expands semantic information to understand and parse user intentions in the conversation, thereby realizing automatic evaluation of service quality and helping to improve the accuracy of quality inspection results.
[0065] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0066] The present application provides an intelligent quality inspection method based on intent recognition. The working process of the method can be implemented by electronic devices, such as computers, handheld smart terminals, etc. For the convenience of explanation, the embodiments of the present application are described with the method execution subject as a computer.
[0067] Figure 1 A schematic diagram of an intelligent quality inspection method based on intent recognition provided in an embodiment of the present application, Figure 2 Another schematic diagram of the intelligent quality inspection method based on intent recognition provided in the embodiment of the present application is as follows: Figure 1 as well as Figure 2 As shown, the present application provides an intelligent quality inspection method based on intent recognition, the method includes S1-S4, which are as follows:
[0068] S1. Obtain the voice text to be inspected;
[0069] The voice text may be converted based on the voice recording to be inspected. The voice text obtained in this step may be directly obtained by converting the converted voice text, or may be obtained by converting the voice recording to obtain the voice text.
[0070] Optional, Figure 3 A schematic diagram of speech-to-text conversion provided in an embodiment of the present application, such as Figure 3 As shown in the figure, after the quality inspection process is started, the voice recording to be inspected is first obtained, and speech recognition (ASR) technology is used to convert the voice signal into text or commands that can be understood by the machine. It mainly includes two parts: training and recognition.
[0071] Among them, training is usually completed offline, performing signal processing and knowledge mining on the massive speech and language databases collected in advance to obtain the "acoustic model" and "language model" required by the speech recognition system.
[0072] In addition, the recognition process is usually completed online and can be divided into two modules: the front-end and the back-end. The front-end mainly performs endpoint detection to remove noise, noise reduction, and feature extraction; the back-end mainly uses the trained acoustic model and language model to decode the feature vector. The former converts the feature vector into phoneme information, and uses the dictionary to match the phoneme with the word, and then uses the language model to obtain the direct relationship between the single word and the word, and finally outputs the corresponding text information.
[0073] S2. Based on preset keywords, keyword matching is performed on the voice text to obtain a first quality inspection score;
[0074] In some embodiments, S2 includes:
[0075] S21. Based on the preset quality inspection rules, obtain corresponding preset keywords;
[0076] S22, establishing a dictionary tree according to the preset keywords, and constructing a failure pointer;
[0077] S23, performing keyword matching on the voice text through the dictionary tree and the failure pointer to obtain a matching result;
[0078] S24. Obtain a first quality inspection score according to the matching result.
[0079] Among them, quality inspection rules are necessary elements for quality inspection. For example, customer service must say a greeting after answering the phone. This is a basic service specification rule. The basic information of rule configuration includes the rule name, the highest and lowest score items, and the filling of the rule description. Each rule is composed of one or more conditions arranged through a certain logical relationship, and each condition is composed of different operators and inspection scopes. Operators can be understood as analysis methods. Commonly used ones include text inspection (keyword inspection, text similarity inspection, etc.), voice inspection (speech speed inspection, role judgment, etc.), and model inspection (customer, customer service detection model). The inspection scope is composed of applicable roles, preconditions, and detection scope.
[0080] The specific rules and procedures are as follows:
[0081] 1. Give the rule a clear, concise name, such as "Answer Phone Greeting Rule".
[0082] 2. Set deduction and bonus scores based on the importance of the rule and the severity of violating the rule. For example, if violating the rule will result in a serious decline in customer experience, a higher score can be set.
[0083] 3. Calculation of rule condition satisfaction: For rules composed of multiple conditions, a comprehensive score can be given based on the satisfaction of the conditions. Condition satisfaction = Σ (score of each condition satisfaction) / total number of conditions. If the condition satisfaction score is greater than 0.7, the rule condition is considered to be satisfied.
[0084] 4. According to the rule objectives, determine the conditions that need to be detected, and select appropriate operators (analysis methods) and inspection scopes for each condition.
[0085] After the quality inspection rules are configured, the quality inspection plan is configured. The plan consists of different rules, and the weight of each rule in the plan can be configured. The intelligent quality inspection model will extract the recording files within the time period configured in the plan according to the quality inspection plan, and perform quality inspection analysis based on the rules or classifications configured in the above plan.
[0086] In some embodiments, S22 includes: establishing a dictionary tree according to the preset keywords, wherein each node in the dictionary tree represents a character; constructing a failure pointer for each node in the dictionary tree, wherein the failure pointer points to the next node to be matched after the current node fails to match.
[0087] In some embodiments, S23 includes: matching the voice text with the dictionary tree character by character, and if the current node fails to match, jumping to the next node according to the failure pointer to continue matching.
[0088] Specifically, this application uses a multi-mode matching method for keyword matching. The multi-mode matching mainly constructs keywords based on the rule requirements in the quality inspection plan, and then determines whether the keywords in the voice-to-text data of customer service and user conversations meet the specifications. The detection steps are as follows:
[0089] 1) Create a dictionary tree (Trie tree):
[0090] Use the given multiple words (pattern strings) to build a dictionary tree (Trie tree).
[0091] Each node in the dictionary tree represents a character, and the path from the root node to a certain node represents the prefix of a word.
[0092] When all characters of a word are traversed, the current node is marked as the end of the word (is_over=true).
[0093] 2) Build failure pointer (Fail pointer):
[0094] A fail pointer is constructed for each node in the dictionary tree, pointing to the next node that should be tried to match after the current pattern string fails to match in the text string.
[0095] This is usually achieved through breadth-first search (BFS) and construction of the Longest Common Prefix and Suffix (LCPS).
[0096] 3) Match the text:
[0097] Starting from the beginning of the text string, match the dictionary tree character by character.
[0098] If a character does not exist in the child nodes of the current node, it jumps to the next node to continue matching according to the Fail pointer.
[0099] When a node marking the end of a word is reached, a match for the pattern has been found in the text string.
[0100] 4) Output the matching results:
[0101] Record the positions and occurrence counts of all found pattern strings in the text string. Output the matching results in a specific format or perform subsequent processing.
[0102] 5) Input the score for the above matching results according to the operator.
[0103] S3, performing intent recognition on the speech text, and obtaining a second quality inspection score according to the intent recognition result;
[0104] In this step, you can use a pre-trained network model for intent recognition. First, based on the user intent recognition results in historical call data as training data, the TextRcnn deep neural network is used as the basis, and the feature embedding of word vector information is used as the network input to build a user conversation intent recognition model. On the basis of quality inspection rules, the semantic information is expanded, and the natural language processing method is used to achieve more accurate intent recognition.
[0105] In some embodiments, S3 includes:
[0106] Calling a regular matching module to match the speech text;
[0107] If a preset mode is matched, the intent category corresponding to the preset mode is output;
[0108] If the preset pattern is not matched, the binary classification model is called to determine whether it is an invalid intention. If so, the invalid intention is output; if not, the multi-classification model is called to predict the intention and obtain the intention prediction result.
[0109] Specifically, Figure 4 A schematic diagram of intent recognition classification provided in an embodiment of the present application, such as Figure 4As shown, first use the regular pattern to judge the call text. If the call text matches a regular pattern, the intent corresponding to the pattern is the intent of the call text; otherwise, use the binary classification model to judge. If it is an invalid intent, the call text is marked as an invalid intent; otherwise, use the multi-classification intent model to predict the intent corresponding to the text and return the result.
[0110] In some embodiments, calling a multi-classification model to perform intent prediction and obtaining an intent prediction result includes:
[0111] Input the speech text into a multi-classification model, and obtain word vectors and character vectors through word and character segmentation;
[0112] Based on the word vector and the character vector, obtaining a first text feature through a convolutional neural network (CNN) module;
[0113] Based on the word vector and character vector, a second text feature is obtained by fusing a bidirectional long short-term memory neural network BILSTM module based on an attention mechanism;
[0114] The first text feature and the second text feature are concatenated and classified through a fully connected layer to obtain an intent prediction result.
[0115] The following is an explanation of the multi-classification model used in this step. The overall modeling process of the multi-classification model includes:
[0116] (1) Data preparation
[0117] Add the "extended question" in the intent data to the 'call text' to supplement the data volume; prepare a stop word list and a word segmentation dictionary based on the actual text.
[0118] (2) Data preprocessing
[0119] After obtaining the text data through the above speech recognition technology, it needs to be preprocessed, which mainly involves the following steps:
[0120] 1) Data cleaning: Delete useless symbols, duplicate data, smooth noise data in text data, and remove data that is not related to the customer service voice quality inspection topic
[0121] 2) Outlier processing: Using the 3σ principle, values where the measured value deviates from the mean value by more than 3 standard deviations are marked as missing values and processed as missing values.
[0122] 3) Missing value processing: directly delete missing values and no longer perform data interpolation to ensure data accuracy.
[0123] 4) Chinese text segmentation: Use Jieba word segmentation to split the Chinese character sequence into individual words.
[0124] 5) Remove stop words: Use a stop word list to remove stop words from the words in the dividend.
[0125] (3) Generate word vector model
[0126] Since the amount of training data is relatively large, the Skip-Gram model, which performs well in large-scale training predictions, is used to train word vectors.
[0127] Specifically, Figure 5 A schematic diagram of the Skip-Gram model provided in the embodiment of the present application is shown in FIG. Figure 5 As shown, the steps of training using the Skip-Gram model include:
[0128] 1) Segment the text using Jieba in the preprocessing stage;
[0129] 2) All the words in the corpus are organized into a list to build word frequency statistics, dictionaries, and inverse dictionaries. Because computers cannot understand Chinese, the text needs to be converted into numerical values to replace it;
[0130] 3) Build the training data needed for the Skip-gram model and predict the context through the central word.
[0131] (4) Determine the multi-classification model structure
[0132] Since the length of call texts is usually very short, statistics show that the average length of call texts is 20 characters, so the multi-classification model structure uses BiLSTM-Attention-CNN.
[0133] Specifically, Figure 6 A schematic diagram of a multi-classification model provided in an embodiment of the present application, such as Figure 6 As shown in the figure, when calling the multi-classification model for intent prediction, the call text is input, firstly segmented into words and characters, and then the word vectors and character vectors are obtained through the Embedding layer. After dropout, the network part adopts the recurrent neural network commonly used in natural language processing. A bidirectional structure is used to capture the dependencies between semantics from two directions, and then the hidden state output and feature information (feature embedding) of the bidirectional LSTM network are fused as the input of the Attention mechanism for weight calculation, which helps the model learn important information and important features in semantics. Finally, the feature information representation of the text is obtained by weighted summation of the hidden state information. In another link, CNN is used to extract the key features of the text synchronously, and the two features obtained by BiLSTM-Attention and CNN are spliced, and then the fully connected layer is used for classification.
[0134] S4. Obtain a quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
[0135] In some embodiments, S4 includes:
[0136] A weighted sum is performed based on the first quality inspection score and the second quality inspection score to obtain a quality inspection result of the speech text.
[0137] Specifically, Figure 7 A schematic diagram of determining the final quality inspection result provided in the embodiment of the present application, such as Figure 7 As shown in the figure, quality inspection analysis mainly uses multi-mode matching and intent recognition technology to integrate keyword rules and semantic information to achieve accurate scoring and quantification of each quality inspection rule, and then obtains the final quality inspection score through weighted average. The calculation method is:
[0138] Final quality inspection score = α*operator score + β*label score
[0139] Among them, α is the operator score weight, β is the label weight, the operator score is the first quality inspection score, and the label score is the second quality inspection score.
[0140] In some embodiments, the method further comprises a quality control review.
[0141] Specifically, the quality inspection review is to check each deduction item. If the deduction is not consistent, it can be modified and edited. At the same time, the action is recorded and pushed to the intelligent algorithm platform to improve the accuracy of scoring through machine deep learning. The intelligent quality inspection system can analyze customer behavior through subsequent data mining, and organize the analysis results into various reports to provide decision support for the business department, and give improvement plans and suggestions. It can also take proactive actions to prevent potential problems from occurring, thereby improving overall customer satisfaction and corporate business performance.
[0142] The present application provides an intelligent quality inspection method based on intent recognition. By using a deep learning algorithm to build a natural language understanding (NLU) intent recognition model, combined with multi-mode matching technology, and integrating keyword rules and semantic information, each quality inspection rule can be accurately scored and quantified, and the final quality inspection score can be obtained by weighting. Specifically, the construction process first uses data enhancement and oversampling methods to solve the problem of imbalance between categories; secondly, R-Dropout, random weighted average (SWA), label smoothing method and other optimization methods are used to optimize the intent recognition model, so that the accuracy of NLU intent recognition reaches 93%. In addition, the system can access real-time voice to monitor the call content in real time (identify keywords, service banned words), record calls that do not meet the standards, and promptly feedback the irregular terms in the voice to the customer service representative, so as to further improve the service quality.
[0143] This application uses an intelligent quality inspection method based on intent recognition and builds an NLU intent recognition model using a deep learning algorithm, which greatly improves the accuracy of quality inspection and significantly improves the recognition accuracy of complex data.
[0144] It should be understood that, although the various steps in the flowcharts in the above-described embodiments are sequentially displayed according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and they can be executed in other orders. Moreover, at least a portion of the steps in the figure may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily to be carried out sequentially, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
[0145] Figure 8 A schematic diagram of an intelligent quality inspection device based on intent recognition provided in an embodiment of the present application, such as Figure 8 As shown, the present application provides an intelligent quality inspection device based on intent recognition, the device comprising:
[0146] A voice text acquisition module 11, which is configured to acquire the voice text to be inspected;
[0147] A keyword matching module 12, which is configured to perform keyword matching on the speech text based on preset keywords to obtain a first quality inspection score;
[0148] An intention recognition module 13 is configured to perform intention recognition on the speech text and obtain a second quality inspection score according to the intention recognition result;
[0149] The quality inspection result determination module 14 is configured to obtain the quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
[0150] Regarding the limitation of the intelligent quality inspection device based on intent recognition, please refer to the limitation of the intelligent quality inspection method based on intent recognition in the above embodiments of the present application, and this embodiment will not be repeated here.
[0151] Fig. 9 Another schematic diagram of an intelligent quality inspection device based on intent recognition provided in an embodiment of the present application is as follows Fig. 9 As shown, in some embodiments, the present application provides an intelligent quality inspection device based on intent recognition, including a memory 22 and a processor 21, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the intelligent quality inspection method based on intent recognition in the above-mentioned embodiments of the present application.
[0152] The memory is connected to the processor, the memory may be a flash memory or a read-only memory or other memory, and the processor may be a central processing unit or a single-chip microcomputer.
[0153] In some embodiments, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the intelligent quality inspection method based on intent recognition in the above-mentioned embodiments of the present application is implemented.
[0154] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.
[0155] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present application, but the present application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of the present application, and these modifications and improvements are also considered to be within the scope of protection of the present application.
Claims
1. An intelligent quality inspection method based on intent recognition, It is characterized in that The method comprises: S1. Obtain the voice text to be inspected; S2. Based on preset keywords, keyword matching is performed on the voice text to obtain a first quality inspection score; S3, performing intent recognition on the speech text, and obtaining a second quality inspection score according to the intent recognition result; S4. Obtain a quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
2. According to the intelligent quality inspection method based on intention recognition according to claim 1, It is characterized in that S2, including: S21. Based on the preset quality inspection rules, obtain corresponding preset keywords; S22, establishing a dictionary tree according to the preset keywords, and constructing a failure pointer; S23, performing keyword matching on the voice text through the dictionary tree and the failure pointer to obtain a matching result; S24. Obtain a first quality inspection score according to the matching result.
3. According to the intelligent quality inspection method based on intention recognition according to claim 2, It is characterized in that S22, including: Establishing a dictionary tree according to the preset keywords, wherein each node in the dictionary tree represents a character; A failure pointer is constructed for each node in the dictionary tree, and the failure pointer points to the next node to be matched after the current node fails to match.
4. According to the intelligent quality inspection method based on intention recognition according to claim 3, It is characterized in that S23, including: The voice text is matched with the dictionary tree character by character. If the current node fails to match, jump to the next node according to the failure pointer to continue matching.
5. According to the intelligent quality inspection method based on intention recognition according to claim 1, It is characterized in that S3, including: Calling a regular matching module to match the speech text; If a preset mode is matched, the intent category corresponding to the preset mode is output; If the preset pattern is not matched, the binary classification model is called to determine whether it is an invalid intention. If so, the invalid intention is output; if not, the multi-classification model is called to predict the intention and obtain the intention prediction result.
6. According to the intelligent quality inspection method based on intention recognition according to claim 5, It is characterized in that Call the multi-classification model to perform intent prediction and obtain intent prediction results, including: Input the speech text into a multi-classification model, and obtain word vectors and character vectors through word and character segmentation; Based on the word vector and the character vector, obtaining a first text feature through a convolutional neural network (CNN) module; Based on the word vector and character vector, a second text feature is obtained by fusing a bidirectional long short-term memory neural network BILSTM module based on an attention mechanism; The first text feature and the second text feature are concatenated and classified through a fully connected layer to obtain an intent prediction result.
7. According to the intelligent quality inspection method based on intention recognition according to claim 1, It is characterized in that S4, including: A weighted sum is performed based on the first quality inspection score and the second quality inspection score to obtain a quality inspection result of the speech text.
8. An intelligent quality inspection device based on intention recognition, It is characterized in that The device comprises: A voice text acquisition module, which is configured to acquire the voice text to be inspected; A keyword matching module, which is configured to perform keyword matching on the voice text based on preset keywords to obtain a first quality inspection score; An intention recognition module, configured to perform intention recognition on the speech text and obtain a second quality inspection score according to the intention recognition result; A quality inspection result determination module is configured to obtain a quality inspection result of the speech text according to the first quality inspection score and the second quality inspection score.
9. An intelligent quality inspection device based on intention recognition, It is characterized in that It comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the intelligent quality inspection method based on intent recognition as described in any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the intelligent quality inspection method based on intent recognition as described in any one of claims 1 to 7 is implemented.
