Abnormal complaint behavior identification method and device, electronic equipment and storage medium

Abstract

The invention discloses an abnormal complaint behavior recognition method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring to-be-recognized audio data, and determining a first audio clip of a target object in the to-be-recognized audio data; extracting a first voiceprint feature of the first audio clip, and determining a similarity parameter between the first voiceprint feature and the second voiceprint feature; extracting semantic verbal skill features of the client communication text corresponding to the first audio clip, and determining behavior parameters corresponding to the to-be-identified audio according to the semantic verbal skill features; and according to the similarity parameter and the behavior parameter, determining an identification result corresponding to the to-be-identified audio data, the identification result being used for representing whether a dialogue corresponding to the to-be-identified audio data has a behavior of replacing a customer to make a complaint. According to the method and the device, the technical problem of low recognition accuracy of the complaint behavior of the valet in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the technical field of customer service quality management, and specifically, to a method, device, electronic device and storage medium for identifying abnormal complaint behavior. Background Art

[0002] In the field of customer service quality management, companies are increasingly in need of accurately and efficiently identifying and handling abnormal complaints, especially malicious complaints on behalf of customers.

[0003] Complaints on behalf of customers typically involve experts or representatives filing complaints or claims against service providers in the name of non-direct customers. With intensified market competition, these complaints, often without the knowledge of customers, have gradually become a key means of suppressing competitors. Malicious complaints on behalf of customers are filed in the name of customers in order to gain undue advantage or harm competitors. This not only complicates complaint handling for businesses but can also damage their reputation and profits. However, relevant technologies for identifying complaints on behalf of customers have technical limitations, such as low accuracy.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, electronic device and storage medium for identifying abnormal complaint behavior, so as to at least solve the technical problem that the related technology has a low accuracy rate in identifying customer complaint behavior.

[0006] According to one aspect of an embodiment of the present application, a method for identifying abnormal complaint behavior is provided, including: obtaining audio data to be identified, and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service representative; extracting a first voiceprint feature of the first audio segment, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time staff who makes complaints on behalf of customers and is stored in a voiceprint library of complaint experts; extracting semantic speech features of the customer communication text corresponding to the first audio segment, and determining behavior parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavior parameters are used to characterize the degree of consistency between the semantic speech features and the behavior pattern of making complaints on behalf of customers; determining a recognition result corresponding to the audio data to be identified based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains behavior of making complaints on behalf of customers.

[0007] Optionally, determining the first audio segment of the target object in the audio data to be identified includes: separating the audio segment of the first object and the audio segment of the second object in the audio data to be identified, and respectively obtaining the longest audio segment among multiple audio segments of the first object and the second object as the target audio segment; extracting the first voiceprint vector of the target audio segment corresponding to the first object and the second voiceprint vector of the target audio segment corresponding to the second object, and determining the highest similarity between the first voiceprint vector and each voiceprint vector in the voiceprint library of customer service personnel, and the highest similarity between the second voiceprint vector and each voiceprint vector in the voiceprint library of customer service personnel; when the highest similarity corresponding to the first voiceprint vector is greater than the highest similarity corresponding to the second voiceprint vector, the first object is corresponded to the customer service personnel, the second object is determined as the target object, and the audio segment of the second object is determined as the first audio segment; when the highest similarity corresponding to the second voiceprint vector is greater than the highest similarity corresponding to the first voiceprint vector, the second object is corresponded to the customer service personnel, the first object is determined as the target object, and the audio segment of the first object is determined as the first audio segment.

[0008] Optionally, the method also includes: obtaining historical valet complaint audio data, and determining the audio clips of the valet complaint specialists in the historical valet complaint audio data, wherein the historical valet complaint audio data is the audio data historically identified as containing conversations between the valet complaint specialists and the customer service staff; extracting the second voiceprint feature of the audio clips of the valet complaint specialists, and constructing a complaint expert voiceprint library based on the second voiceprint feature.

[0009] Optionally, the method also includes: performing speech recognition on audio clips of the customer complaint specialist to obtain customer complaint text; determining similar semantic expression sentences between multiple customer complaint texts whose semantic similarity is less than a preset semantic similarity threshold, and storing the similar semantic expression sentences as professional complaint scripts in a script database.

[0010] Optionally, the method also includes: converting the customer complaint text into a vector form according to a preset feature dimension, and obtaining semantic speech features corresponding to the customer complaint text, wherein the preset feature dimension includes multiple first-level features, each first-level feature corresponds to multiple second-level features, and the semantic speech features are used to characterize the number of times the semantic content corresponding to each second-level feature in the customer complaint text is mentioned, and the first-level features include at least one of the following: emotional expression, complaint logic, abnormal process, ambiguous identity, legal risk, and public opinion exposure; using the semantic speech features corresponding to the customer complaint text, the initial model is trained to obtain a semantic analysis and recognition model, wherein the semantic analysis and recognition model is used to predict the probability that the customer communication text is a customer complaint.

[0011] Optionally, the behavior parameters include: a first sub-parameter and a second sub-parameter; determining the behavior parameters corresponding to the audio to be identified based on the semantic speech features includes: determining the speech status of the customer communication text corresponding to the first audio clip, and determining the first sub-parameter based on the speech status, wherein the speech status is used to characterize the frequency of the customer communication text containing professional complaint speech in the speech database, and the higher the frequency, the larger the corresponding first sub-parameter; according to the preset feature dimension, the customer communication text is converted into a vector form to obtain the semantic speech features corresponding to the customer communication text; using a semantic analysis and recognition model, the semantic speech features corresponding to the customer communication text are analyzed to obtain the second sub-parameter, wherein the second sub-parameter is used to characterize the probability that the customer communication text is a complaint on behalf of the customer; based on the first sub-parameter and the second sub-parameter, the behavior parameters corresponding to the audio to be identified are determined.

[0012] Optionally, based on the similarity parameter and the behavior parameter, determining the recognition result corresponding to the audio data to be identified includes: when the similarity parameter between the first voiceprint feature and the second voiceprint feature is greater than a preset voiceprint similarity threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of the customer; when the similarity parameter is not greater than the preset voiceprint similarity threshold, judging whether the behavior parameter is greater than the preset behavior parameter threshold; when the behavior parameter is greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of the customer, and when the behavior parameter is not greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the absence of an act of making a complaint on behalf of the customer.

[0013] According to another aspect of an embodiment of the present application, another method for identifying abnormal complaint behavior is provided, including: obtaining audio data to be identified, and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service representative; extracting a first voiceprint feature of the first audio segment, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of an abnormal complaint object stored in a preset database; extracting semantic speech features corresponding to the first audio segment, and determining behavioral parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavioral parameters are used to characterize the degree of consistency between the semantic speech features and the behavioral features used to represent abnormal complaints; based on the similarity parameters and the behavioral parameters, jointly determining a recognition result corresponding to the audio data to be identified, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains abnormal complaint behavior.

[0014] According to another aspect of the embodiment of the present application, an abnormal complaint behavior identification device is also provided, including: an audio segmentation module for obtaining audio data to be identified and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service staff; an audio feature extraction module for extracting a first voiceprint feature of the first audio segment and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time staff who makes complaints on behalf of customers and is stored in a complaint expert voiceprint library; a semantic feature extraction module for extracting semantic speech features of the customer communication text corresponding to the first audio segment, and determining behavior parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavior parameters are used to characterize the degree of consistency between the semantic speech features and the behavior pattern of making complaints on behalf of customers; a customer service behavior identification module for determining a recognition result corresponding to the audio data to be identified based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains behavior of making complaints on behalf of customers.

[0015] According to another aspect of the embodiments of the present application, an electronic device is provided, including: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the abnormal complaint behavior identification method is executed when the program is running.

[0016] According to another aspect of an embodiment of the present application, a non-volatile storage medium is further provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the abnormal complaint behavior identification method by running the computer program.

[0017] According to another aspect of the embodiments of the present application, a computer program product is also provided, including a computer program, which implements the steps of the abnormal complaint behavior identification method when the computer program is executed by a processor.

[0018] In an embodiment of the present application, the method comprises obtaining audio data to be identified and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service representative; extracting a first voiceprint feature of the first audio segment, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time staff member who makes complaints on behalf of customers and is stored in a voiceprint library of complaint experts; extracting semantic speech features of the customer communication text corresponding to the first audio segment, and determining behavioral parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavioral parameters are used to characterize the degree of consistency between the semantic speech features and the behavioral pattern of making complaints on behalf of customers; determining a recognition result corresponding to the audio data to be identified based on the similarity parameters and the behavioral parameters, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains a behavior of making complaints on behalf of customers. By combining voiceprint recognition and semantic analysis to jointly identify abnormal behaviors such as making complaints on behalf of customers, the purpose of improving the accuracy of identifying abnormal behaviors is achieved, thereby solving the technical problem that the related technology has a low accuracy rate in identifying behaviors of making complaints on behalf of customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 This is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for identifying abnormal complaint behavior according to an embodiment of the present application;

[0021] Figure 2 This is a schematic diagram of a method flow for identifying abnormal complaint behavior provided in an embodiment of the present application;

[0022] Figure 3 This is a schematic diagram of a method flow for constructing a complaint expert voiceprint database and a semantic analysis and recognition model according to an embodiment of the present application;

[0023] Figure 4 This is a schematic diagram of a method flow for identifying customer complaint behavior based on voiceprint and semantic analysis according to an embodiment of the present application;

[0024] Figure 5 This is a schematic diagram of another method flow for identifying abnormal complaint behavior provided in an embodiment of the present application;

[0025] Figure 6 It is a structural diagram of an abnormal complaint behavior identification device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] To facilitate those skilled in the art to better understand the embodiments of the present application, some technical terms or nouns involved in the embodiments of the present application are explained as follows:

[0029] LLM (Large Language Model): A large-scale natural language processing model based on deep learning technology. It is usually trained with massive corpus and can generate, understand, and process natural language text for tasks such as answering questions, generating content, and text analysis.

[0030] ASR (Automatic Speech Recognition): An intelligent speech processing system based on deep learning and signal processing technologies. It is usually trained with large amounts of speech data and can convert human speech into editable text content in real time. It is used for tasks such as speech transcription, voice command recognition, and real-time subtitle generation.

[0031] Among related technologies, while traditional voiceprint recognition technology can make a preliminary judgment on the speaker's identity based on voiceprint features, its recognition accuracy is greatly reduced in the context of customer complaint handling, when faced with complex situations such as physiological changes in the speaker and deliberate imitation of other people's voice tones. Related technologies that use automatic speech recognition (ASR) technology to transcribe complaint audio into text and then analyze the text's semantic features through methods based on keyword matching or simple machine learning models can initially identify keywords in the complaint text, but lack in-depth and accurate mining of the professional speech features of customer complaint experts. In particular, it is difficult to effectively identify and analyze complex semantic expressions, hidden complaint intentions, and diverse complaint logic. In other words, when handling customer complaint handling issues, related technologies, whether in the single voiceprint recognition or semantic analysis stages, have certain shortcomings and cannot meet the actual needs of enterprises for accurate and efficient identification of customer complaint handling behaviors.

[0032] In order to solve the above problems, the embodiments of the present application provide relevant solutions. First, a voiceprint vector library of customer service personnel is constructed by manual separation and voiceprint model calculation. The voiceprint vectors of customer service complaint experts are accurately extracted from the confirmed customer complaint audio to form a complaint expert voiceprint vector library. Then, automatic speech recognition technology and open source large language models are used to extract similar semantic expression sentences from the speech texts of customer complaint experts and manually verify them. A customer complaint speech feature system containing multiple main features and sub-features is constructed, and the corresponding model is trained. Finally, when identifying customer complaint behavior, voiceprint similarity calculation and semantic feature analysis are comprehensively used. Through specific rules and threshold judgments, efficient and accurate identification of customer complaint behavior is achieved, effectively solving the shortcomings of existing technologies in customer complaint identification and improving the efficiency of enterprises' supervision and processing capabilities of customer complaint behavior. The following is a detailed description.

[0033] According to an embodiment of the present application, an embodiment of a method for identifying abnormal complaint behavior is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The following is a hardware block diagram of a computer terminal (or electronic device) for implementing a method for identifying abnormal complaint behavior. Figure 1As shown, the computer terminal 10 (or electronic device) may include one or more (illustrated as 102a, 102b, ..., 102n in the figure) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0035] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10 (or electronic device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the abnormal complaint behavior identification method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned abnormal complaint behavior identification method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0037] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0038] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or electronic device).

[0039] In the above operating environment, the embodiment of the present application provides a method for identifying abnormal complaint behavior. Figure 2 This is a schematic diagram of a method flow for identifying abnormal complaint behavior provided in an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0040] Step S202: Acquire audio data to be recognized and determine a first audio segment of a target subject in the audio data to be recognized, wherein the audio data to be recognized is audio data containing a conversation between the target subject and a customer service representative;

[0041] Step S204: extracting a first voiceprint feature of the first audio clip and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time person who files complaints on behalf of customers and is stored in a complaint expert voiceprint database;

[0042] Step S206: extracting semantic and technical features of the customer communication text corresponding to the first audio clip, and determining behavioral parameters corresponding to the audio to be recognized based on the semantic and technical features, wherein the behavioral parameters are used to represent the degree of consistency between the semantic and technical features and the behavioral pattern of making complaints on behalf of the customer;

[0043] Step S208 : determining a recognition result corresponding to the audio data to be recognized based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be recognized contains behavior of making a complaint on behalf of the customer.

[0044] Through the above steps, by combining voiceprint recognition and semantic analysis to jointly identify abnormal behaviors such as customer complaints, the purpose of accurately identifying customer complaints is achieved, thereby solving the technical problem that the relevant technology has a low accuracy rate in identifying customer complaints.

[0045] The following further introduces the method for identifying abnormal complaint behavior in steps S202 to S208 of the embodiment of the present application.

[0046] First, the establishment process of the complaint expert voiceprint database and the training steps of the semantic analysis and recognition model used to analyze semantic speech are introduced in detail. Figure 3 As shown, in this embodiment of the application, the voiceprint features of valet complaint experts are accurately extracted by separating audio clips and calculating the mean of voiceprint vectors, and a valet complaint expert voiceprint database is established. Automatic speech recognition technology and open source large language models are used to extract the valet complaint expert's speech from the valet complaint expert text, and a valet complaint speech feature system is constructed. The model is trained using the valet complaint speech feature system. The specific steps for establishing the valet complaint expert voiceprint database are as follows.

[0047] In some embodiments of the present application, the method further includes the following steps: obtaining historical valet complaint audio data, and determining audio segments of valet complaint specialists in the historical valet complaint audio data, wherein the historical valet complaint audio data is historically identified audio data containing conversations between valet complaint specialists and customer service personnel; extracting a second voiceprint feature of the audio segment of the valet complaint specialists, and constructing a complaint expert voiceprint library based on the second voiceprint feature.

[0048] Specifically, we first obtain n real conversation audios of all customer service representatives in the enterprise with customers, separate the customer service and customer speech segments in the audio, use the voiceprint model to calculate the voiceprint vector of each customer service representative's n speech segments, and calculate the mean of the n voiceprint vectors as the voiceprint vector of the customer service representative to form the customer service representative voiceprint vector library DS sv ;

[0049] After that, the confirmed valet complaint audio file (i.e., historical valet complaint audio data) is obtained, and then the audio file is separated into audio segments of two people (e.g., the first subject and the second subject) speaking. The audio segments with the longest speaking time, voice1 and voice2, are respectively selected. That is, the longest audio segment among the multiple audio segments of the first subject and the second subject is obtained as the target audio segment. The calculation rule can be expressed as shown in the following formula:

[0050]

[0051] Calculate the voiceprint vectors vector1 and vector2 of voice1 and voice2 respectively, and compare vector1 and vector2 with the customer service voiceprint vector library DS sv Calculate the cosine similarity of the vectors in each and obtain their respective maximum similarity MaxS sv1 and MaxS sv2 .

[0052] If MaxS sv1 Greater than or equal to MaxS sv2 , then the first person (first object) is identified as a customer service representative, and the second person (second object) is identified as a customer complaint expert. sv1 Smaller than MaxS sv2 , the first person (first object) is identified as the customer service expert, and the second person (second object) is identified as the customer service representative; according to the above steps, all the confirmed customer service complaint audio files can be traversed to extract the voiceprint vectors of all confirmed customer service complaint experts, and the customer service complaint expert voiceprint vector library DS can be constructed. ev .

[0053] The following is a detailed introduction to the construction and training steps of the semantic analysis and recognition model.

[0054] On the one hand, we can use the audio clips of full-time complaint handling staff to build a database of professional complaint scripts. The specific steps are as follows.

[0055] In some embodiments of the present application, the method also includes: performing speech recognition on audio clips of the customer complaint specialist to obtain customer complaint text; determining similar semantic expression sentences between multiple customer complaint texts whose semantic similarity is less than a preset semantic similarity threshold, and storing the similar semantic expression sentences as professional complaint scripts in a script database.

[0056] Specifically, the automatic speech recognition technology can be used to extract the text of each audio clip (customer complaint text) from the audio clip of the customer complaint expert (customer complaint professional), and obtain the text set Text of all customer complaint experts. ts Then, based on the open source large language model and prompt word engineering, similar semantic expression sentences can be extracted from the collection of customer complaint texts. repeat , that is, to determine the similar semantic expression sentences between multiple customer complaint texts whose semantic similarity is less than the preset semantic similarity threshold, as shown in the following formula:

[0057] Text repat =LLM(Text ts ,prompt)

[0058] Where LLM is the open source large language model, prompt is the prompt word of the large language model;

[0059] In addition, you can also express similar semantics in sentences Text repeatManual verification is performed to remove sentences that are not related to complaints or business processing, such as greetings, insults, and intonation. Finally, the similar semantic expression texts after manual verification can be used as professional complaint scripts on behalf of customers and stored in the script database, as shown in the following formula:

[0060] DKTS text ={text1,text2,…,text n}

[0061] For example, suppose the text collection Text of the customer complaint expert ts For example, ["I am not satisfied with the refund of my order. If the issue is not resolved, I will file a complaint with the regulatory authorities.", "I am very dissatisfied with the service and request an immediate refund, otherwise we will take further action.", "If you do not handle it, we will demand full compensation.", "I am not satisfied with the processing speed. If the issue is not resolved, we will request termination of the contract."]; when inputting it into the large language model for extracting similar semantic expression sentences, the corresponding prompt words can be: [Extract sentences with similar semantic expressions from the following text set: {}, please list all similar semantic expression sentences in the form of a list:]; finally, the similar semantic expression sentence Text output by the large model is obtained. repat For [("If it is not resolved, I will complain to the regulatory authorities.", "Otherwise I will complain.")].

[0062] On the other hand, we can also use the preset feature dimensions of the complaint language to train the model through the semantic language features of the complaint text. The specific steps are as follows.

[0063] In some embodiments of the present application, the method also includes the following steps: according to a preset feature dimension, converting the customer complaint text into a vector form to obtain the semantic speech features corresponding to the customer complaint text, wherein the preset feature dimension includes multiple first-level features, each first-level feature corresponds to multiple second-level features, and the semantic speech features are used to characterize the number of times the semantic content corresponding to each second-level feature in the customer complaint text is mentioned, and the first-level features include at least one of the following: emotional expression, complaint logic, abnormal process, fuzzy identity, legal risk, and public opinion exposure; using the semantic speech features corresponding to the customer complaint text, the initial model is trained to obtain a semantic analysis and recognition model, wherein the semantic analysis and recognition model is used to predict the probability that the customer communication text is a customer complaint.

[0064] Specifically, in this embodiment, six main features of customer complaint speech consisting of emotional expression, complaint logic, abnormal process, ambiguous identity, legal risk, and public opinion exposure, as well as corresponding preset feature dimensions of sub-features of customer complaint speech can be constructed, as shown in the following table.

[0065]

[0066]

[0067] Then, we can use the large language model to judge the text from the valet complaint ts Determine whether the sub-customer complaint speech feature exists one by one, and record the number of times each sub-speech feature exists; ts Each text is represented as a feature vector, where each element represents the number of times a sub-feature appears, as shown in the following formula:

[0068] F i =[x i1 ,x i2 ,…,x in ]

[0069] Among them, F i Represents Text ts The feature vector of the i-th text in x ij is the number of times the j-th sub-feature appears in the i-th text;

[0070] For example, suppose the current valet complaint text is ts For example, if the sentence is ["I am dissatisfied with the order refund. If the issue is not resolved, I will file a complaint with the regulatory authorities."], the following prompt word is input into the large language model: [Please analyze the following complaint text to determine whether it contains the following sub-speech features: 1. Dissatisfaction; 2. Threatening language; 3. Sense of urgency; 4. Opening remarks; 5. Problem description; 6. Closing remarks; 7. Request for immediate action; 8. Bypassing normal communication channels; 9. Not the person (representing multiple customers); 10. Use of legal terms; 11. Exposure on social media / online forums; Please determine each one and count the number of occurrences.], then the semantic speech feature F corresponding to the customer complaint text is obtained. i is [1,1,1,0,1,0,0,1,0,0,0].

[0071] After obtaining the semantic speech features corresponding to the customer complaint text, the weight vector of the sub-features can be initialized as W = [w1, w2, ..., w n ], where the weight of each sub-feature is w j The initial value is set to 1. At the same time, set y i The actual label of the i-th text: 0.7 represents a valet complaint, 0 represents a normal complaint, and the objective function J is established. w And use this objective function to train the model. For example, you can use Text ts The dataset is divided into training set and validation set in the ratio of 80% and 20%, and the weights are updated using the gradient descent method to minimize the objective function J. w Get the trained model, as shown in the following formula:

[0072]

[0073] Where m is the number of samples.

[0074] This embodiment of the application utilizes an open-source large language model and prompt word engineering to extract professional valet complaint scripts. It also constructs a semantic feature system encompassing six main features and their sub-features: emotional expression, complaint logic, abnormal process, ambiguous identity, legal risk, and public opinion exposure. The large language model is used to determine the presence of sub-features and record their occurrence, representing the text as a feature vector. The model is then trained using gradient descent. This allows for in-depth and comprehensive analysis of the semantic features of valet complaint texts, accurately identifying various complex valet complaint scripts and intent. This effectively improves the accuracy of valet complaint identification and addresses the shortcomings of related technologies in mining semantic features of valet complaints.

[0075] After completing the establishment of the complaint expert voiceprint database and the training of the semantic analysis recognition model, the audio data to be recognized can be recognized. Specifically, Figure 4 As shown, in this embodiment of the present application, voiceprint similarity can be used for judgment. If the voiceprint similarity is lower than a threshold, semantic feature analysis is performed. A threshold is determined in the semantic feature analysis. If the threshold is exceeded, it is considered that a customer complaint has occurred, thereby achieving accurate identification of customer complaint behavior, as described below.

[0076] In some embodiments of the present application, determining the recognition result corresponding to the audio data to be identified based on the similarity parameter and the behavior parameter includes: when the similarity parameter between the first voiceprint feature and the second voiceprint feature is greater than the preset voiceprint similarity threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of a customer; when the similarity parameter is not greater than the preset voiceprint similarity threshold, judging whether the behavior parameter is greater than the preset behavior parameter threshold; when the behavior parameter is greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of a customer, and when the behavior parameter is not greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the absence of an act of making a complaint on behalf of a customer.

[0077] The following is a detailed introduction to the recognition process of the audio data to be recognized.

[0078] First, obtain the audio data to be identified, and determine the first audio segment of the target object in the audio data to be identified, extract the first voiceprint feature of the first audio segment, and determine the similarity parameter between the first voiceprint feature and the second voiceprint feature in the voiceprint library of the complaint expert. The specific steps are as follows.

[0079] In some embodiments of the present application, determining the first audio segment of the target object in the audio data to be identified includes: separating the audio segment of the first object and the audio segment of the second object in the audio data to be identified, and respectively obtaining the longest audio segment among multiple audio segments of the first object and the second object as the target audio segment; extracting a first voiceprint vector of the target audio segment corresponding to the first object and a second voiceprint vector of the target audio segment corresponding to the second object, and determining the highest similarity between the first voiceprint vector and each voiceprint vector in a voiceprint library of customer service personnel, and the highest similarity between the second voiceprint vector and each voiceprint vector in the voiceprint library of customer service personnel; if the highest similarity corresponding to the first voiceprint vector is greater than the highest similarity corresponding to the second voiceprint vector, corresponding to the first object as the customer service personnel, determining the second object as the target object, and determining the audio segment of the second object as the first audio segment; if the highest similarity corresponding to the second voiceprint vector is greater than the highest similarity corresponding to the first voiceprint vector, corresponding to the second object as the customer service personnel, determining the first object as the target object, and determining the audio segment of the first object as the first audio segment.

[0080] Specifically, the audio clips of the customer service staff and the target object (which may be a customer or a complaint expert) are separated from the audio data to be identified. The process steps for separating the first audio clip of the target object from the audio data to be identified are similar to the process steps for separating the audio clip of the complaint expert voiceprint database in the above process, and will not be repeated here. After that, the voiceprint features of the target object are extracted and compared with the complaint expert voiceprint database DS ev The off-chord similarity (i.e., similarity parameter) of the second voiceprint vector in the audio is calculated one by one. If it exceeds the threshold α, it is considered that the audio involves a valet complaint behavior. If it does not exceed the threshold α, subsequent semantic analysis processing is performed.

[0081] When the similarity parameter is not greater than the preset voiceprint similarity threshold α, the semantic speech features of the customer communication text corresponding to the first audio segment are extracted, and based on the semantic speech features, the behavioral parameters corresponding to the audio to be identified are determined. The specific steps are as follows.

[0082] In some embodiments of the present application, the behavioral parameters include: a first sub-parameter and a second sub-parameter; determining the behavioral parameters corresponding to the audio to be identified based on the semantic speech features includes: determining the speech status of the customer communication text corresponding to the first audio clip, and determining the first sub-parameter based on the speech status, wherein the speech status is used to characterize the frequency of the customer communication text containing professional complaint speech in the speech database, and the higher the frequency, the larger the corresponding first sub-parameter; according to the preset feature dimension, the customer communication text is converted into a vector form to obtain the semantic speech features corresponding to the customer communication text; using a semantic analysis and recognition model, the semantic speech features corresponding to the customer communication text are analyzed to obtain the second sub-parameter, wherein the second sub-parameter is used to characterize the probability that the customer communication text is a complaint on behalf of the customer; based on the first sub-parameter and the second sub-parameter, the behavioral parameters corresponding to the audio to be identified are determined.

[0083] Specifically, the target object's speech text is extracted from the first audio segment using automatic speech recognition technology. user (i.e., customer communication text), and then, the large model can be used to determine whether the customer communication text contains the complaint professional language in the language database, that is, to determine the language status of the customer communication text corresponding to the first audio clip, as shown in the following formula:

[0084] LLM(text user ,prompt,Text ts )

[0085] If it exists, assign β1=0.3, otherwise β1=0, where β1 is the first sub-parameter mentioned above.

[0086] Then, extract the customer communication text user The semantic feature vector w user Then, the semantic analysis and recognition model trained previously is used to analyze the semantic speech feature vector to obtain the prediction score β2 (i.e., the second sub-parameter mentioned above); finally, the sum of β1 and β2 is calculated to obtain the behavior parameter corresponding to the audio to be identified. If the behavior parameter is greater than the preset behavior parameter threshold ε, it is considered that there is a complaint behavior on behalf of the customer. In this embodiment, 0.3<ε≤1.

[0087] This application scheme significantly improves the accuracy and efficiency of recognition by combining voiceprint recognition and semantic analysis. Among them, voiceprint recognition can quickly screen out suspected full-time complainers, while semantic analysis deeply explores the subtle features of the complaint text. The two complement each other and effectively avoid the misjudgment that may be caused by a single technical means. In addition, by using large language models and prompt word engineering to extract professional speech, a comprehensive semantic speech feature system is constructed, which enables the system to recognize complex and changeable complaint logic and intentions. This method improves the company's ability to monitor abnormal complaint behavior, can more comprehensively and accurately identify customer complaint behavior, reduces the risk of misjudgment and missed judgment, and improves the company's supervision efficiency and handling capabilities for customer complaint behavior.

[0088] According to an embodiment of the present application, another method for identifying abnormal complaint behavior is provided. Figure 5 This is a schematic diagram of a method flow for identifying abnormal complaint behavior provided in an embodiment of the present application. Figure 5 As shown, the method includes the following steps:

[0089] Step S502: Acquire audio data to be recognized and determine a first audio segment of a target subject in the audio data to be recognized, wherein the audio data to be recognized is audio data containing a conversation between the target subject and a customer service representative;

[0090] Step S504: extracting a first voiceprint feature of the first audio clip, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of an abnormal complaint subject stored in a preset database;

[0091] Step S506: extracting semantic speech features corresponding to the first audio clip, and determining behavioral parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavioral parameters are used to represent the degree of consistency between the semantic speech features and the behavioral features used to represent abnormal complaints;

[0092] Step S508 : Determine the recognition result corresponding to the audio data to be recognized based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be recognized contains abnormal complaint behavior.

[0093] It should be noted that the abnormal complaint behavior identification method provided in this embodiment is Figure 2 The method embodiment corresponding to the abnormal complaint behavior identification method shown is, therefore, the relevant explanations and descriptions of the above-mentioned abnormal complaint behavior identification method are also applicable to the embodiments of this application and will not be repeated here.

[0094] According to an embodiment of the present application, an embodiment of a device for identifying abnormal complaint behavior is also provided. Figure 6This is a schematic diagram of the structure of an abnormal complaint behavior identification device provided according to an embodiment of the present application. Figure 6 As shown, the device includes:

[0095] An audio segment segmentation module 60 is configured to obtain audio data to be identified and determine a first audio segment of a target subject in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target subject and a customer service representative;

[0096] an audio feature extraction module 62 for extracting a first voiceprint feature from the first audio clip and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a professional who files complaints on behalf of a customer and is stored in a complaint expert voiceprint database;

[0097] Semantic feature extraction module 64 is configured to extract semantic speech features of the customer communication text corresponding to the first audio clip, and determine behavioral parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavioral parameters are used to represent the degree of consistency between the semantic speech features and the behavioral pattern of making complaints on behalf of the customer;

[0098] The customer-on-behavior recognition module 66 is used to determine the recognition result corresponding to the audio data to be recognized based on the similarity parameter and the behavior parameter, wherein the recognition result is used to indicate whether the conversation corresponding to the audio data to be recognized involves the behavior of making complaints on behalf of customers.

[0099] Optionally, determining the first audio segment of the target object in the audio data to be identified includes: separating the audio segment of the first object and the audio segment of the second object in the audio data to be identified, and respectively obtaining the longest audio segment among multiple audio segments of the first object and the second object as the target audio segment; extracting the first voiceprint vector of the target audio segment corresponding to the first object and the second voiceprint vector of the target audio segment corresponding to the second object, and determining the highest similarity between the first voiceprint vector and each voiceprint vector in the voiceprint library of customer service personnel, and the highest similarity between the second voiceprint vector and each voiceprint vector in the voiceprint library of customer service personnel; when the highest similarity corresponding to the first voiceprint vector is greater than the highest similarity corresponding to the second voiceprint vector, the first object is corresponded to the customer service personnel, the second object is determined as the target object, and the audio segment of the second object is determined as the first audio segment; when the highest similarity corresponding to the second voiceprint vector is greater than the highest similarity corresponding to the first voiceprint vector, the second object is corresponded to the customer service personnel, the first object is determined as the target object, and the audio segment of the first object is determined as the first audio segment.

[0100] Optionally, the abnormal complaint behavior identification device is also used to: obtain historical valet complaint audio data, and determine the audio clips of the valet complaint specialists in the historical valet complaint audio data, wherein the historical valet complaint audio data is the audio data historically identified as containing conversations between the valet complaint specialists and customer service personnel; extract the second voiceprint feature of the audio clips of the valet complaint specialists, and construct a complaint expert voiceprint library based on the second voiceprint feature.

[0101] Optionally, the abnormal complaint behavior identification device is also used to: perform voice recognition on audio clips of dedicated customer complaint personnel to obtain customer complaint texts; determine similar semantic expression sentences between multiple customer complaint texts whose semantic similarity is less than a preset semantic similarity threshold, and store the similar semantic expression sentences as professional complaint scripts in the script database.

[0102] Optionally, the abnormal complaint behavior identification device is also used to: convert the customer complaint text into a vector form according to a preset feature dimension, and obtain the semantic speech features corresponding to the customer complaint text, wherein the preset feature dimension includes multiple first-level features, each first-level feature corresponds to multiple second-level features, and the semantic speech features are used to characterize the number of times the semantic content corresponding to each second-level feature in the customer complaint text is mentioned, and the first-level features include at least one of the following: emotional expression, complaint logic, abnormal process, ambiguous identity, legal risk, and public opinion exposure; use the semantic speech features corresponding to the customer complaint text to train the initial model to obtain a semantic analysis and recognition model, wherein the semantic analysis and recognition model is used to predict the probability that the customer communication text is a customer complaint.

[0103] Optionally, the behavior parameters include: a first sub-parameter and a second sub-parameter; determining the behavior parameters corresponding to the audio to be identified based on the semantic speech features includes: determining the speech status of the customer communication text corresponding to the first audio clip, and determining the first sub-parameter based on the speech status, wherein the speech status is used to characterize the frequency of the customer communication text containing professional complaint speech in the speech database, and the higher the frequency, the larger the corresponding first sub-parameter; according to the preset feature dimension, the customer communication text is converted into a vector form to obtain the semantic speech features corresponding to the customer communication text; using a semantic analysis and recognition model, the semantic speech features corresponding to the customer communication text are analyzed to obtain the second sub-parameter, wherein the second sub-parameter is used to characterize the probability that the customer communication text is a complaint on behalf of the customer; based on the first sub-parameter and the second sub-parameter, the behavior parameters corresponding to the audio to be identified are determined.

[0104] Optionally, based on the similarity parameter and the behavior parameter, determining the recognition result corresponding to the audio data to be identified includes: when the similarity parameter between the first voiceprint feature and the second voiceprint feature is greater than a preset voiceprint similarity threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of the customer; when the similarity parameter is not greater than the preset voiceprint similarity threshold, judging whether the behavior parameter is greater than the preset behavior parameter threshold; when the behavior parameter is greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the existence of an act of making a complaint on behalf of the customer, and when the behavior parameter is not greater than the preset behavior parameter threshold, determining that the recognition result of the audio data to be identified is the absence of an act of making a complaint on behalf of the customer.

[0105] It should be noted that the various modules in the above-mentioned abnormal complaint behavior identification device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0106] It should be noted that the abnormal complaint behavior identification device provided in this embodiment can be used to perform Figure 2 The abnormal complaint behavior identification method shown, therefore, the relevant explanations and descriptions of the above-mentioned abnormal complaint behavior identification method are also applicable to the embodiments of this application and will not be repeated here.

[0107] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the following abnormal complaint behavior identification method by running the computer program: obtaining audio data to be identified, and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service representative; extracting a first voiceprint feature of the first audio segment, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time staff who makes complaints on behalf of customers and is stored in a complaint expert voiceprint library; extracting semantic speech features of the customer communication text corresponding to the first audio segment, and determining behavior parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavior parameters are used to characterize the degree of consistency between the semantic speech features and the behavior pattern of making complaints on behalf of customers; determining a recognition result corresponding to the audio data to be identified based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains behavior of making complaints on behalf of customers.

[0108] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the abnormal complaint behavior identification method described in each embodiment of the present application: obtaining audio data to be identified, and determining a first audio segment of a target object in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target object and a customer service representative; extracting a first voiceprint feature of the first audio segment, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time staff who makes complaints on behalf of customers and is stored in a complaint expert voiceprint library; extracting semantic speech features of the customer communication text corresponding to the first audio segment, and determining behavior parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavior parameters are used to characterize the degree of consistency between the semantic speech features and the behavior pattern of making complaints on behalf of customers; determining a recognition result corresponding to the audio data to be identified based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be identified contains behavior of making complaints on behalf of customers.

[0109] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0110] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0112] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0113] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0115] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying abnormal complaint behavior, characterized in that: include: Acquire audio data to be recognized, and determine a first audio segment of a target subject in the audio data to be recognized, wherein the audio data to be recognized is audio data containing a conversation between the target subject and a customer service representative; Extracting a first voiceprint feature from the first audio clip, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time person who files complaints on behalf of customers and is stored in a complaint expert voiceprint database; Extracting semantic and technical features of the customer communication text corresponding to the first audio clip, and determining, based on the semantic and technical features, behavioral parameters corresponding to the audio to be identified, wherein the behavioral parameters are used to characterize the degree of consistency between the semantic and technical features and the behavioral pattern of making complaints on behalf of the customer; Based on the similarity parameter and the behavior parameter, a recognition result corresponding to the audio data to be recognized is determined, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be recognized contains behavior of making complaints on behalf of customers.

2. The abnormal complaint behavior identification method according to claim 1 is characterized in that: Determining a first audio segment of a target object in the audio data to be recognized includes: Separating an audio segment of a first object and an audio segment of a second object in the audio data to be recognized, and obtaining the longest audio segment from among the multiple audio segments of the first object and the second object as a target audio segment; Extracting a first voiceprint vector of the target audio segment corresponding to the first object and a second voiceprint vector of the target audio segment corresponding to the second object, and determining the highest similarity between the first voiceprint vector and each voiceprint vector in a customer service personnel voiceprint library, and the highest similarity between the second voiceprint vector and each voiceprint vector in the customer service personnel voiceprint library; If the highest similarity corresponding to the first voiceprint vector is greater than the highest similarity corresponding to the second voiceprint vector, the first subject is identified as a customer service representative, the second subject is identified as a target subject, and the audio clip of the second subject is identified as the first audio clip; When the highest similarity corresponding to the second voiceprint vector is greater than the highest similarity corresponding to the first voiceprint vector, the second object is corresponded to a customer service staff, the first object is determined as the target object, and the audio segment of the first object is determined as the first audio segment.

3. The abnormal complaint behavior identification method according to claim 1 is characterized in that: The method further comprises: Acquire historical customer complaint audio data, and identify audio segments of a dedicated customer complaint officer in the historical customer complaint audio data, wherein the historical customer complaint audio data is historically identified audio data containing conversations between the dedicated customer complaint officer and customer service personnel; The second voiceprint feature of the audio clip of the complaint handling specialist is extracted, and the complaint expert voiceprint library is constructed based on the second voiceprint feature.

4. The abnormal complaint behavior identification method according to claim 3 is characterized in that: The method further comprises: Performing voice recognition on the audio clip of the customer complaint professional to obtain the customer complaint text; Determine similar semantic expression sentences between multiple customer complaint texts whose semantic similarity is less than a preset semantic similarity threshold, and store the similar semantic expression sentences as professional complaint speech in a speech database.

5. The abnormal complaint behavior identification method according to claim 4 is characterized in that: The method further comprises: According to a preset feature dimension, the customer complaint text is converted into a vector form to obtain a semantic speech feature corresponding to the customer complaint text, wherein the preset feature dimension includes multiple first-level features, each of the first-level features corresponds to multiple second-level features, and the semantic speech feature is used to characterize the number of times the semantic content corresponding to each second-level feature in the customer complaint text is mentioned, and the first-level feature includes at least one of the following: emotional expression, complaint logic, abnormal process, ambiguous identity, legal risk, and public opinion exposure; The initial model is trained using the semantic speech features corresponding to the customer complaint text to obtain a semantic analysis and recognition model, wherein the semantic analysis and recognition model is used to predict the probability that the customer communication text is a customer complaint.

6. The abnormal complaint behavior identification method according to claim 5 is characterized in that: The behavior parameter includes: a first sub-parameter and a second sub-parameter; based on the semantic speech feature, determining the behavior parameter corresponding to the audio to be recognized includes: Determine the speech status of the customer communication text corresponding to the first audio clip, and determine the first sub-parameter based on the speech status, wherein the speech status is used to represent the frequency of the customer communication text containing the professional complaint speech in the speech database, and the higher the frequency, the larger the corresponding first sub-parameter; According to the preset feature dimension, the customer communication text is converted into a vector form to obtain the semantic speech feature corresponding to the customer communication text; Using the semantic analysis and recognition model, the semantic speech features corresponding to the customer communication text are analyzed to obtain the second sub-parameter, wherein the second sub-parameter is used to represent the probability that the customer communication text is a customer complaint; The behavior parameter corresponding to the audio to be recognized is determined based on the first sub-parameter and the second sub-parameter.

7. The abnormal complaint behavior identification method according to claim 1 is characterized in that: Determining a recognition result corresponding to the audio data to be recognized based on the similarity parameter and the behavior parameter includes: If the similarity parameter between the first voiceprint feature and the second voiceprint feature is greater than a preset voiceprint similarity threshold, determining that the recognition result of the audio data to be recognized is an act of filing a complaint on behalf of a customer; If the similarity parameter is not greater than the preset voiceprint similarity threshold, determining whether the behavior parameter is greater than the preset behavior parameter threshold; When the behavior parameter is greater than the preset behavior parameter threshold, the recognition result of the audio data to be identified is determined to be the presence of a behavior of making a complaint on behalf of a customer; and when the behavior parameter is not greater than the preset behavior parameter threshold, the recognition result of the audio data to be identified is determined to be the absence of a behavior of making a complaint on behalf of a customer.

8. A method for identifying abnormal complaint behavior, characterized in that: include: Acquire audio data to be identified, and determine a first audio segment of a target subject in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target subject and a customer service representative; Extracting a first voiceprint feature of the first audio clip, and determining a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of an abnormal complaint subject stored in a preset database; Extracting semantic speech features corresponding to the first audio clip, and determining behavioral parameters corresponding to the audio to be identified based on the semantic speech features, wherein the behavioral parameters are used to represent the degree of consistency between the semantic speech features and behavioral features used to represent abnormal complaints; Based on the similarity parameter and the behavior parameter, a recognition result corresponding to the audio data to be recognized is jointly determined, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be recognized contains abnormal complaint behavior.

9. A device for identifying abnormal complaint behavior, characterized in that: include: An audio segment segmentation module is configured to obtain audio data to be identified and determine a first audio segment of a target subject in the audio data to be identified, wherein the audio data to be identified is audio data containing a conversation between the target subject and a customer service representative; an audio feature extraction module, configured to extract a first voiceprint feature from the first audio clip and determine a similarity parameter between the first voiceprint feature and a second voiceprint feature, wherein the second voiceprint feature is a voiceprint feature of a full-time person who files complaints on behalf of customers and is stored in a complaint expert voiceprint database; a semantic feature extraction module, configured to extract semantic speech features of the customer communication text corresponding to the first audio clip, and determine, based on the semantic speech features, behavioral parameters corresponding to the audio to be identified, wherein the behavioral parameters are used to characterize the degree of conformity between the semantic speech features and the behavioral pattern of making complaints on behalf of the customer; The customer-on-behavior recognition module is used to determine the recognition result corresponding to the audio data to be recognized based on the similarity parameter and the behavior parameter, wherein the recognition result is used to characterize whether the conversation corresponding to the audio data to be recognized contains the behavior of making complaints on behalf of customers.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program, when running, executes the abnormal complaint behavior identification method described in any one of claims 1 to 8.

11. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the abnormal complaint behavior identification method described in any one of claims 1 to 8 by running the computer program.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the abnormal complaint behavior identification method described in any one of claims 1 to 8 are implemented.

Images