Artificial intelligence-based dialogue recommendation method, device, equipment and storage medium

Abstract

The embodiment of the application belongs to the field of artificial intelligence, and relates to an artificial intelligence-based dialogue recommendation method, which comprises the following steps: obtaining call audio data of a seat and a target customer; converting the call audio data into text data; generating an intention prediction result corresponding to the text data based on an intention prediction model; extracting an audio spectrum of the target customer from the call audio data; generating a sentiment analysis result corresponding to the audio spectrum based on a sentiment analysis model; querying a dialogue library based on the text data, the intention prediction result and the sentiment analysis result to obtain a target dialogue; and pushing the target dialogue to the target customer. The application also provides an artificial intelligence-based dialogue recommendation device, a computer device and a storage medium. In addition, the application also relates to blockchain technology, and the target dialogue can be stored in the blockchain. The application improves the efficiency and intelligence of target dialogue acquisition and improves the accuracy of dialogue recommendation in a call scene.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to artificial intelligence-based speech recommendation methods, devices, computer equipment, and storage media. Background Technology

[0002] Online sales of insurance are a current and likely long-term industry trend, and a mainstream form of data-driven transformation for many insurance companies. Therefore, under this new trend, insurance companies have higher requirements for agents' communication skills, professional competence, and adaptability. However, high agent turnover and insufficient training have posed challenges to insurance company performance, leading to the emergence of new online sales systems. Current systems provide agents with "golden scripts" to facilitate better communication and timely handling of customer requests. However, this approach requires agents to spend significant time selecting the appropriate scripts from a pool of experienced agents, and the accuracy of the selected scripts cannot be guaranteed. Therefore, accurately recommending precise scripts to agents during customer calls has become a pressing technical challenge. Summary of the Invention

[0003] The purpose of this application is to propose a method, apparatus, computer device, and storage medium for recommending dialogue based on artificial intelligence, so as to solve the technical problem that there is a lack of means to accurately recommend precise dialogue to agents when they are communicating with customers.

[0004] To address the aforementioned technical problems, this application provides an artificial intelligence-based speech recommendation method, employing the following technical solution:

[0005] Acquire audio data of conversations between agents and target customers;

[0006] Convert the call audio data into text data;

[0007] The text data is processed based on a preset intent prediction model to generate intent prediction results corresponding to the text data.

[0008] Extract the audio profile of the target customer from the call audio data;

[0009] The audio graph is analyzed and processed based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph.

[0010] Based on the text data, the intent prediction results, and the sentiment analysis results, a preset script database is queried to find the corresponding target script.

[0011] The target message is pushed to the target customer.

[0012] Furthermore, the step of performing prediction processing on the text data based on a preset intent prediction model to generate an intent prediction result corresponding to the text data specifically includes:

[0013] The text data is input into the input layer of the intent prediction model, and the text data in the input layer is transformed to obtain the corresponding transformed data.

[0014] The transformed data is processed by vector feature transformation through the embedding layer within the intent prediction model to obtain the corresponding word vectors.

[0015] The word vectors are convolved by the first convolutional layer in the intent prediction model to extract feature data corresponding to the word vectors.

[0016] The intent prediction model classifies the feature data by performing intent classification on the first classification layer, thereby obtaining the intent prediction result corresponding to the text data.

[0017] Furthermore, the step of extracting the target customer's audio profile from the call audio data specifically includes:

[0018] Extract customer audio data corresponding to the target customer from the call audio data;

[0019] The customer audio data is preprocessed to obtain processed customer audio data;

[0020] Feature extraction is performed on the processed customer audio data to obtain the audio spectrum of the target customer.

[0021] Furthermore, the step of analyzing and processing the audio spectrogram based on a preset sentiment analysis model to generate a sentiment analysis result corresponding to the audio spectrogram specifically includes:

[0022] The audio graph is input into the second convolutional layer of the sentiment analysis model, and the audio graph is convolved by the second convolutional layer to generate a sentiment feature vector corresponding to the audio graph.

[0023] The emotion feature vector is normalized to obtain the target emotion feature vector;

[0024] The target sentiment feature vector is input into the long short-term memory neural network in the sentiment analysis model. The long short-term memory neural network processes the target sentiment feature vector to generate the corresponding long-term domain sentiment features.

[0025] The long-term sentiment features are input into a fully connected layer within the sentiment analysis model. The long-term sentiment features are then processed by the fully connected layer to obtain the corresponding target long-term sentiment features.

[0026] The target long-term sentiment features are input into the second classification layer of the sentiment analysis model. The second classification layer performs prediction processing on the target long-term sentiment features and outputs the sentiment feature category corresponding to the target long-term sentiment features.

[0027] The sentiment analysis results are generated based on the sentiment feature categories.

[0028] Furthermore, the step of generating the sentiment analysis result based on the sentiment feature category specifically includes:

[0029] Obtain the probability of each of the stated emotion feature categories;

[0030] Obtain the target sentiment feature category with the highest probability from all the stated sentiment feature categories;

[0031] The target sentiment feature category is used as the sentiment analysis result.

[0032] Furthermore, the step of querying a preset script database based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script from the script database specifically includes:

[0033] Multiple first dialogues corresponding to the intent prediction result are retrieved from the dialogue script library to form a first dialogue script set;

[0034] Multiple second dialogues corresponding to the sentiment analysis results are retrieved from the first dialogue set to form a second dialogue set;

[0035] Find the third script corresponding to the text data from the second script set;

[0036] The target script is generated based on the third script.

[0037] Furthermore, the step of searching for the third script corresponding to the text data from the second script set specifically includes:

[0038] Based on a preset similarity algorithm, the similarity between the text data and each of the second dialogues in the second dialogue set is calculated;

[0039] Each of the aforementioned similarities is compared with a preset similarity threshold, and a specified dialogue with a similarity greater than or equal to the similarity threshold is selected from all the second dialogues;

[0040] The specified script is used as the third script.

[0041] To address the aforementioned technical problems, this application also provides an artificial intelligence-based speech recommendation device, which employs the following technical solution:

[0042] The acquisition module is used to acquire audio data of conversations between agents and target customers;

[0043] The conversion module is used to convert the call audio data into text data;

[0044] The prediction module is used to perform prediction processing on the text data based on a preset intent prediction model, and generate intent prediction results corresponding to the text data.

[0045] The extraction module is used to extract the audio spectrum of the target customer from the call audio data;

[0046] The analysis module is used to analyze and process the audio graph based on a preset sentiment analysis model, and generate sentiment analysis results corresponding to the audio graph.

[0047] The query module is used to perform query processing on the preset script database based on the text data, the intent prediction result and the sentiment analysis result, and to find the corresponding target script from the script database;

[0048] The push module is used to push the target message to the target customer.

[0049] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0050] Acquire audio data of conversations between agents and target customers;

[0051] Convert the call audio data into text data;

[0052] The text data is processed based on a preset intent prediction model to generate intent prediction results corresponding to the text data.

[0053] Extract the audio profile of the target customer from the call audio data;

[0054] The audio graph is analyzed and processed based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph.

[0055] Based on the text data, the intent prediction results, and the sentiment analysis results, a preset script database is queried to find the corresponding target script.

[0056] The target message is pushed to the target customer.

[0057] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0058] Acquire audio data of conversations between agents and target customers;

[0059] Convert the call audio data into text data;

[0060] The text data is processed based on a preset intent prediction model to generate intent prediction results corresponding to the text data.

[0061] Extract the audio profile of the target customer from the call audio data;

[0062] The audio graph is analyzed and processed based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph.

[0063] Based on the text data, the intent prediction results, and the sentiment analysis results, a preset script database is queried to find the corresponding target script.

[0064] The target message is pushed to the target customer.

[0065] Compared with the prior art, the embodiments of this application have the following main advantages:

[0066] This embodiment first acquires the call audio data between the agent and the target customer; then, it converts the call audio data into text data; next, it performs prediction processing on the text data based on a preset intent prediction model to generate an intent prediction result corresponding to the text data; subsequently, it extracts the target customer's audio spectrum from the call audio data, and analyzes the audio spectrum based on a preset sentiment analysis model to generate a sentiment analysis result corresponding to the audio spectrum; finally, it queries a preset script library based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script and pushes the target script to the target customer. This embodiment, by performing multi-dimensional information mining on the call audio data between the agent and the target customer, and then processing the mined information based on the intent prediction model, sentiment analysis model, and script library, can quickly and accurately find the required target script related to the call audio data from the script library, improving the efficiency and intelligence of target script acquisition and enhancing the accuracy of script recommendation in call scenarios. Attached Figure Description

[0067] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0069] Figure 2 A flowchart of an embodiment of the AI-based speech recommendation method according to this application;

[0070] Figure 3 This is a schematic diagram of a structure of an embodiment of the AI-based speech recommendation device according to this application;

[0071] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0073] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0074] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0075] like Figure 1As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0076] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0077] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0078] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0079] It should be noted that the AI-based speech recommendation method provided in this application is generally executed by a server / terminal device, and correspondingly, the AI-based speech recommendation device is generally set in the server / terminal device.

[0080] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0081] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of the AI-based conversation recommendation method according to this application. The AI-based conversation recommendation method includes the following steps:

[0082] Step S201: Obtain the audio data of the call between the agent and the target customer.

[0083] In this embodiment, the AI-based speech recommendation method runs on an electronic device (e.g., Figure 1 The server / terminal device shown can acquire call audio data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future known wireless connection methods. The aforementioned call audio data may refer to the conversation audio data collected by the agent and the target customer during the call through recording.

[0084] Step S202: Convert the call audio data into text data.

[0085] In this embodiment, the call audio data is converted into text data using ASR technology, and the agent audio corresponding to the agent's speech portion and the customer audio corresponding to the customer's speech portion can be separated from the call audio data.

[0086] Step S203: Based on a preset intent prediction model, perform prediction processing on the text data to generate an intent prediction result corresponding to the text data.

[0087] In this embodiment, the specific implementation process of predicting the text data based on the preset intent prediction model and generating the intent prediction result corresponding to the text data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0088] Step S204: Extract the audio spectrum of the target customer from the call audio data.

[0089] In this embodiment, the specific implementation process of extracting the target customer's audio profile from the call audio data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0090] Step S205: Analyze and process the audio graph based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph.

[0091] In this embodiment, the specific implementation process of analyzing and processing the audio graph based on the preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0092] Step S206: Based on the text data, the intent prediction result, and the sentiment analysis result, perform query processing on the preset script database to find the corresponding target script from the script database.

[0093] In this embodiment, the above-mentioned process of querying the preset script database based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script from the script database is described in more detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0094] Step S207: Push the target message to the target customer.

[0095] In this embodiment, the target customer's terminal information can be obtained, and then the target script can be pushed to the target customer based on this information. After the conversation ends, the text and audio data are added to a script database for script analysis. Simultaneously, industry expert experience is incorporated to summarize new agent script practices, helping agents better demonstrate professionalism and improve problem-solving efficiency when facing customer issues.

[0096] This application first acquires the audio data of a call between an agent and a target customer; then, it converts the audio data into text data; next, it performs predictive processing on the text data based on a preset intent prediction model to generate an intent prediction result corresponding to the text data; subsequently, it extracts the target customer's audio graph from the audio data, and analyzes the audio graph based on a preset sentiment analysis model to generate a sentiment analysis result corresponding to the audio graph; finally, it queries a preset script library based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script and pushes the target script to the target customer. This application, by performing multi-dimensional information mining on the audio data of the call between the agent and the target customer, and then processing the mined information based on the intent prediction model, sentiment analysis model, and script library, can quickly and accurately find the required target script related to the call audio data from the script library, improving the efficiency and intelligence of target script acquisition and enhancing the accuracy of script recommendation in call scenarios.

[0097] In some alternative implementations, step S203 includes the following steps:

[0098] The text data is input into the input layer of the intent prediction model, and the text data in the input layer is transformed to obtain the corresponding transformed data.

[0099] In this embodiment, the aforementioned conversion process refers to the process of converting text data into corresponding numerical data by matching a dictionary. This conversion process may further include word segmentation of the text data to obtain corresponding word segmentation data.

[0100] The transformed data is processed by vector feature transformation through the embedding layer within the intent prediction model to obtain the corresponding word vectors.

[0101] In this embodiment, the aforementioned embedding layer may specifically refer to the embedding layer within the intention prediction model.

[0102] The word vectors are convolved by the first convolutional layer within the intent prediction model to extract feature data corresponding to the word vectors.

[0103] In this embodiment, the aforementioned first convolutional layer may specifically refer to a CNN layer within the intended prediction model.

[0104] The intent prediction model classifies the feature data by performing intent classification on the first classification layer, thereby obtaining the intent prediction result corresponding to the text data.

[0105] In this embodiment, the training process of the above-mentioned intent prediction model may include: obtaining training data from a preset historical call corpus; wherein the training data is text data corresponding to the voice data generated by historical users and agents during calls; randomly allocating the training data to obtain a training sample set and a test sample set; wherein the training data is pre-labeled with intent tags; training a preset initial intent prediction model based on the training sample set, and validating the trained initial intent prediction model using the validation dataset, and using the validated intent prediction model as the intent prediction model. Specifically, the initial intent prediction model may employ a convolutional neural network, which includes an input layer, an embedding layer, a convolutional layer, and a classification layer (i.e., an output layer). After inputting the training sample set into the initial intent prediction model, the model first performs word segmentation and vector feature transformation on the training samples in the input layer to obtain the word vector corresponding to each word in the training sample set. Then, the word vector corresponding to each word in the training sample set is input into the convolutional layer of the initial intent prediction model for feature extraction to obtain the feature data for each word. Finally, the similarity between the feature data and the preset intent label is calculated in the first classification layer of the initial intent prediction model, and the recognition result with the highest similarity is output as the intent prediction result corresponding to the training sample. Based on the intent prediction result and the preset standard result, a backpropagation algorithm is used for fitting calculation to obtain the recognition error. The recognition error is compared with a preset error threshold. If the recognition error is greater than the preset error threshold, the trained initial intent prediction model is iteratively updated based on the loss function of the intent prediction model until the recognition error is less than or equal to the preset error threshold, thus obtaining a validated call prediction model. The intent prediction result is output through the softmax function to achieve intent classification. When building the initial intent prediction model, a corresponding loss function is set, where the loss function is the cross-entropy loss function. During the training of the intent prediction model, the fitted call intent model is obtained by iteratively updating the trained call intent model. Furthermore, the building and training of the intent prediction model can both be completed using the TensorFlow library in Python.

[0106] This application uses an intent prediction model to predict the text data, which can quickly and accurately generate intent prediction results corresponding to the text data, improve the prediction accuracy of the intent prediction results, and facilitate the subsequent intelligent determination of the required target dialogue related to the call audio data from the dialogue database based on the obtained intent prediction results.

[0107] In some optional implementations of this embodiment, step S204 includes the following steps:

[0108] Extract customer audio data corresponding to the target customer from the call audio data.

[0109] In this embodiment, the agent audio data corresponding to the agent's speech portion and the customer audio data corresponding to the customer's speech portion can be separated from the call audio data.

[0110] The customer audio data is preprocessed to obtain processed customer audio data.

[0111] In this embodiment, the preprocessing includes pre-emphasis, framing, and windowing.

[0112] Feature extraction is performed on the processed customer audio data to obtain the audio spectrum of the target customer.

[0113] In this embodiment, a short-time Fourier transform can be performed on the processed customer audio data, and the data can be stacked frame by frame to obtain the corresponding spectrogram; then, a Mel-scale filter can be used to filter the spectrogram to obtain the audio spectrum of the target customer.

[0114] This application extracts customer audio data corresponding to the target customer from the call audio data; then preprocesses the customer audio data to obtain processed customer audio data; and finally extracts features from the processed customer audio data to obtain the audio spectrum of the target customer. Based on a Mel-scale filter, this application can quickly and accurately extract the audio spectrum of the target customer from the call audio data, improving the efficiency of audio spectrum acquisition and facilitating subsequent analysis of the audio spectrum to accurately obtain the sentiment analysis results corresponding to the target user.

[0115] In some alternative implementations, step S205 includes the following steps:

[0116] The audio graph is input into the second convolutional layer of the sentiment analysis model, and the audio graph is convolved by the second convolutional layer to generate a sentiment feature vector corresponding to the audio graph.

[0117] In this embodiment, the aforementioned sentiment analysis model is a neural network model composed of a convolutional neural network (CNN) and a long short-term memory neural network (LSTM). By employing a sentiment analysis model that combines CNN and LSTM, abstract emotional features in speech signals can be extracted more fully, resulting in more accurate sentiment classification. The sentiment analysis model includes at least convolutional layers, a long short-term memory neural network, fully connected layers, and a classification layer. The number of the aforementioned second convolutional layers may be multiple.

[0118] The emotional feature vector is normalized to obtain the target emotional feature vector.

[0119] The target sentiment feature vector is input into the long short-term memory neural network within the sentiment analysis model. The long short-term memory neural network processes the target sentiment feature vector to generate corresponding long-term sentiment features.

[0120] The long-term sentiment features are input into a fully connected layer within the sentiment analysis model. The long-term sentiment features are then processed by the fully connected layer to obtain the corresponding target long-term sentiment features.

[0121] The target long-term sentiment features are input into the second classification layer of the sentiment analysis model. The second classification layer performs prediction processing on the target long-term sentiment features and outputs the sentiment feature category corresponding to the target long-term sentiment features.

[0122] In this embodiment, the aforementioned classification layer specifically refers to the Softmax classification layer. The training and generation process of the aforementioned sentiment analysis model may include: obtaining voice training data from a preset historical call database; preprocessing and feature extraction of the voice training data to obtain a corresponding training audio graph; inputting the training audio graph into an initial sentiment analysis model composed of a CNN network and LSTM for forward propagation training to obtain a predicted sentiment feature category y(z); using a preset predicted sentiment category y'(z) as input to perform backpropagation training on the initial sentiment analysis model to obtain the closest true sentiment feature category Y(z), and adjusting the network parameters of the initial sentiment analysis model according to the gradient descent algorithm to make the error between the closest true sentiment feature category Y(z) and the predicted sentiment feature category y(z) less than a preset value, thereby obtaining a trained initial sentiment analysis model, and using the trained initial sentiment analysis model as the final sentiment analysis model.

[0123] The sentiment analysis results are generated based on the sentiment feature categories.

[0124] In this embodiment, the specific implementation process of generating the sentiment analysis result based on the sentiment feature category will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0125] This application uses a sentiment analysis model to analyze and process the audio graph to perform accurate sentiment recognition and classification, and quickly and accurately generates sentiment analysis results corresponding to the audio graph. This improves the accuracy of the sentiment analysis results and facilitates the subsequent intelligent determination of the required target dialogue related to the call audio data from the dialogue database based on the obtained sentiment analysis results.

[0126] In some optional implementations, generating the sentiment analysis result based on the sentiment feature category includes the following steps:

[0127] Obtain the probability of each of the stated emotion feature categories.

[0128] In this embodiment, after the target long-term sentiment feature is input into the second classification layer of the sentiment analysis model, the second classification layer will output several sentiment feature categories with non-zero probabilities corresponding to the target long-term sentiment feature.

[0129] Obtain the target sentiment feature category with the highest probability from all the stated sentiment feature categories.

[0130] The target sentiment feature category is used as the sentiment analysis result.

[0131] In this embodiment, by selecting the target emotional feature category with the highest probability as the emotional analysis result, accurate identification of the emotion in the audio spectrum of the target customer can be achieved.

[0132] This application obtains the probability of each of the stated emotion feature categories; then, it selects the target emotion feature category with the highest probability from all the stated emotion feature categories, and uses the target emotion feature category as the emotion analysis result. By selecting the target emotion feature category with the highest probability as the emotion analysis result, this application can achieve accurate identification of the emotion in the audio graph of the target customer, which is beneficial for subsequently determining the required target dialogue related to the call audio data from the dialogue database based on the obtained intent prediction results.

[0133] In some optional implementations of this embodiment, step S206 includes the following steps:

[0134] Multiple first scripts corresponding to the intent prediction results are retrieved from the script library to form a first script set.

[0135] In this embodiment, the aforementioned script library is a database of preset scripts created based on actual usage needs. Each emotional state is associated with at least one script, and each user intent is also associated with at least one script. Once a customer's emotional state is known, the corresponding script can be found in the script library based on that emotional state; similarly, once a customer's user intent is known, the corresponding script can be found in the script library based on that user intent. The scripts in the script library are derived from experience. For example, when a customer experiences extreme negative emotions, the scripts used to effectively calm the customer are summarized by agents and psychologists. Similarly, when a customer intends to purchase insurance, the scripts used to effectively help the customer successfully purchase insurance are summarized by agents and psychologists.

[0136] Multiple second-level statements corresponding to the sentiment analysis results are retrieved from the first set of statements to form a second set of statements.

[0137] Find the third script corresponding to the text data from the second script set.

[0138] In this embodiment, the specific implementation process of generating the target script based on the third script will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0139] The target script is generated based on the third script.

[0140] In this embodiment, the number of third-level dialogues can be obtained. If the number of third-level dialogues is less than or equal to a preset number, then the third-level dialogue is directly used as the target dialogue. If the number of third-level dialogues is greater than the preset number, then a preset number of fourth-level dialogues with the highest similarity to the text information are selected from all third-level dialogues based on a preset similarity algorithm, and these fourth-level dialogues are used as the target dialogue.

[0141] This application forms a first set of dialogues by searching the dialogue database for multiple first dialogues corresponding to the intent prediction result; then, it forms a second set of dialogues by searching the first set of dialogues for multiple second dialogues corresponding to the sentiment analysis result; furthermore, it forms a third set of dialogues by searching the second set of dialogues for a third dialogue corresponding to the text data; finally, it generates the target dialogue based on the third dialogue. This application achieves rapid and accurate retrieval of the required target dialogue related to the call audio data from the dialogue database by querying the dialogue database from three dimensions: intent prediction result, sentiment analysis result, and text data, thus improving the efficiency and intelligence of target dialogue acquisition.

[0142] In some optional implementations of this embodiment, the step of finding the third script corresponding to the text data from the second script set includes the following steps:

[0143] Based on a preset similarity algorithm, the similarity between the text data and each of the second dialogues in the second dialogue set is calculated.

[0144] In this embodiment, the above similarity algorithm can be an open-source similarity algorithm available online.

[0145] Each similarity is compared with a preset similarity threshold, and a specified dialogue with a similarity greater than or equal to the similarity threshold is selected from all the second dialogues.

[0146] In this embodiment, the value of the above similarity threshold is not specifically limited and can be set according to the actual business needs, for example, it can be set to 0.88.

[0147] The specified script is used as the third script.

[0148] This application calculates the similarity between the text data and each of the second dialogues in the second dialogue set based on a preset similarity algorithm. Then, it compares each similarity with a preset similarity threshold, selecting designated dialogues from all the second dialogues whose similarity is greater than or equal to the threshold, and using these designated dialogues as the third dialogue. After finding the third dialogue corresponding to the text data from the second dialogue set, this application further calls the similarity algorithm to calculate the similarity between the text data and each of the second dialogues in the second dialogue set. This is to determine the semantic relevance between the text data and the second dialogues, and to remove dialogues with low relevance to the text data from the second dialogues, thereby obtaining the third dialogue. This ensures the accuracy and intelligence of obtaining the third dialogue, guaranteeing the accuracy of the target dialogue data when the third dialogue is used to generate the target dialogue.

[0149] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned target scripts, the target scripts can also be stored in a blockchain node.

[0150] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0151] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0152] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0154] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0155] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of an artificial intelligence-based speech recommendation device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0156] like Figure 3As shown, the AI-based speech recommendation device 300 described in this embodiment includes: an acquisition module 301, a conversion module 302, a prediction module 303, an extraction module 304, an analysis module 305, a query module 306, and a push module 307. Wherein:

[0157] The acquisition module 301 is used to acquire the audio data of the call between the agent and the target customer;

[0158] Conversion module 302 is used to convert the call audio data into text data;

[0159] The prediction module 303 is used to perform prediction processing on the text data based on a preset intent prediction model, and generate an intent prediction result corresponding to the text data.

[0160] Extraction module 304 is used to extract the audio spectrum of the target customer from the call audio data;

[0161] The analysis module 305 is used to analyze and process the audio graph based on a preset sentiment analysis model, and generate sentiment analysis results corresponding to the audio graph.

[0162] The query module 306 is used to perform query processing on the preset script library based on the text data, the intent prediction result and the sentiment analysis result, and to find the corresponding target script from the script library;

[0163] The push module 307 is used to push the target message to the target customer.

[0164] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0165] In some optional implementations of this embodiment, the prediction module 303 includes:

[0166] The first processing submodule is used to input the text data into the input layer of the intent prediction model, and to perform transformation processing on the text data in the input layer to obtain corresponding transformed data.

[0167] The second processing submodule is used to perform vector feature transformation processing on the transformed data through the embedding layer in the intent prediction model to obtain the corresponding word vectors.

[0168] The computation submodule is used to perform convolution operations on the word vectors through the first convolutional layer in the intent prediction model to extract feature data corresponding to the word vectors.

[0169] The classification submodule is used to classify the feature data by intent through the first classification layer in the intent prediction model, so as to obtain the intent prediction result corresponding to the text data.

[0170] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0171] In some optional implementations of this embodiment, the extraction module 304 includes:

[0172] The first extraction submodule is used to extract customer audio data corresponding to the target customer from the call audio data;

[0173] The third processing submodule is used to preprocess the customer audio data to obtain processed customer audio data.

[0174] The second extraction submodule is used to extract features from the processed customer audio data to obtain the audio spectrum of the target customer.

[0175] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0176] In some optional implementations of this embodiment, the analysis module 305 includes:

[0177] The fourth processing submodule is used to input the audio graph into the second convolutional layer in the sentiment analysis model, and to perform convolution processing on the audio graph through the second convolutional layer to generate a sentiment feature vector corresponding to the audio graph.

[0178] The fifth processing submodule is used to normalize the emotion feature vector to obtain the target emotion feature vector;

[0179] The sixth processing submodule is used to input the target sentiment feature vector into the long short-term memory neural network in the sentiment analysis model, and to process the target sentiment feature vector through the long short-term memory neural network to generate the corresponding long-term domain sentiment features.

[0180] The seventh processing submodule is used to input the long-term sentiment features into the fully connected layer in the sentiment analysis model, and process the long-term sentiment features through the fully connected layer to obtain the corresponding target long-term sentiment features;

[0181] The prediction submodule is used to input the target long-term sentiment features into the second classification layer in the sentiment analysis model, perform prediction processing on the target long-term sentiment features through the second classification layer, and output the sentiment feature category corresponding to the target long-term sentiment features;

[0182] The first generation submodule is used to generate the sentiment analysis result based on the sentiment feature category.

[0183] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0184] In some optional implementations of this embodiment, the first generation submodule includes:

[0185] The first acquisition unit is used to acquire the probability of each of the said emotion feature categories;

[0186] The second acquisition unit is used to acquire the target sentiment feature category with the highest probability from all the sentiment feature categories;

[0187] The first determining unit is used to take the target emotional feature category as the emotional analysis result.

[0188] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0189] In some optional implementations of this embodiment, the query module 306 includes:

[0190] The first search submodule is used to search the script library for multiple first scripts that correspond to the intent prediction result, forming a first script set;

[0191] The second search submodule is used to search for multiple second scripts that correspond to the sentiment analysis results from the first script set, forming a second script set;

[0192] The third search submodule is used to search for the third script corresponding to the text data from the second script set;

[0193] The second generation submodule is used to generate the target script based on the third script.

[0194] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0195] In some optional implementations of this embodiment, the third search submodule includes:

[0196] The calculation unit is used to calculate the similarity between the text data and each of the second dialogues in the second dialogue set based on a preset similarity algorithm;

[0197] The comparison unit is used to compare each of the similarities with a preset similarity threshold, and to filter out the specified dialogues with a similarity greater than or equal to the similarity threshold from all the second dialogues;

[0198] The second determining unit is used to use the specified script as the third script.

[0199] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the AI-based speech recommendation method in the aforementioned implementation method, and will not be repeated here.

[0200] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0201] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0202] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0203] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for AI-based speech recommendation methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0204] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the AI-based speech recommendation method.

[0205] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0206] Compared with the prior art, the embodiments of this application have the following main advantages:

[0207] In this embodiment, the call audio data between the agent and the target customer is first acquired; then, the call audio data is converted into text data; subsequently, the text data is processed based on a preset intent prediction model to generate an intent prediction result corresponding to the text data; next, the audio spectrum of the target customer is extracted from the call audio data, and the audio spectrum is analyzed based on a preset sentiment analysis model to generate a sentiment analysis result corresponding to the audio spectrum; finally, a preset script library is queried based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script and push the target script to the target customer. This embodiment, by performing multi-dimensional information mining on the call audio data between the agent and the target customer, and then processing the mined information based on the intent prediction model, sentiment analysis model, and script library, can quickly and accurately find the required target script related to the call audio data from the script library, improving the efficiency and intelligence of target script acquisition and enhancing the accuracy of script recommendation in call scenarios.

[0208] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the AI-based speech recommendation method described above.

[0209] Compared with the prior art, the embodiments of this application have the following main advantages:

[0210] In this embodiment, the call audio data between the agent and the target customer is first acquired; then, the call audio data is converted into text data; subsequently, the text data is processed based on a preset intent prediction model to generate an intent prediction result corresponding to the text data; next, the audio spectrum of the target customer is extracted from the call audio data, and the audio spectrum is analyzed based on a preset sentiment analysis model to generate a sentiment analysis result corresponding to the audio spectrum; finally, a preset script library is queried based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script and push the target script to the target customer. This embodiment, by performing multi-dimensional information mining on the call audio data between the agent and the target customer, and then processing the mined information based on the intent prediction model, sentiment analysis model, and script library, can quickly and accurately find the required target script related to the call audio data from the script library, improving the efficiency and intelligence of target script acquisition and enhancing the accuracy of script recommendation in call scenarios.

[0211] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0212] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A speech recommendation method based on artificial intelligence, characterized in that, Includes the following steps: Obtain audio data of conversations between agents and target customers; Convert the call audio data into text data; The text data is processed based on a preset intent prediction model to generate intent prediction results corresponding to the text data. Extract the audio profile of the target customer from the call audio data; The audio graph is analyzed and processed based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio graph. Based on the text data, the intent prediction results, and the sentiment analysis results, a preset script database is queried to find the corresponding target script. The target message is pushed to the target customer; The step of querying a preset script database based on the text data, the intent prediction result, and the sentiment analysis result to find the corresponding target script from the script database specifically includes: Multiple first dialogues corresponding to the intent prediction result are retrieved from the dialogue script library to form a first dialogue script set; Multiple second dialogues corresponding to the sentiment analysis results are retrieved from the first dialogue set to form a second dialogue set; Find the third script corresponding to the text data from the second script set; The target script is generated based on the third script; The step of finding the third script corresponding to the text data from the second script set specifically includes: Based on a preset similarity algorithm, the similarity between the text data and each of the second dialogues in the second dialogue set is calculated; Each of the aforementioned similarities is compared with a preset similarity threshold, and a specified dialogue with a similarity greater than or equal to the similarity threshold is selected from all the second dialogues; Use the specified script as the third script; Specifically, by obtaining the number of third-level dialogues, if the number of third-level dialogues is less than or equal to a preset number, then the third-level dialogue is directly used as the target dialogue. If the number of third-level dialogues is greater than the preset number, then based on a preset similarity algorithm, a preset number of fourth-level dialogues with the highest similarity to the text information are selected from all third-level dialogues, and the fourth-level dialogue is used as the target dialogue.

2. The AI-based speech recommendation method according to claim 1, characterized in that, The step of performing prediction processing on the text data based on a preset intent prediction model to generate an intent prediction result corresponding to the text data specifically includes: The text data is input into the input layer of the intent prediction model, and the text data in the input layer is transformed to obtain the corresponding transformed data. The transformed data is processed by vector feature transformation through the embedding layer within the intent prediction model to obtain the corresponding word vectors. The word vectors are convolved by the first convolutional layer in the intent prediction model to extract feature data corresponding to the word vectors. The intent prediction model classifies the feature data by performing intent classification on the first classification layer, thereby obtaining the intent prediction result corresponding to the text data.

3. The AI-based speech recommendation method according to claim 1, characterized in that, The step of extracting the target customer's audio profile from the call audio data specifically includes: Extract customer audio data corresponding to the target customer from the call audio data; The customer audio data is preprocessed to obtain processed customer audio data; Feature extraction is performed on the processed customer audio data to obtain the audio spectrum of the target customer.

4. The AI-based speech recommendation method according to claim 1, characterized in that, The step of analyzing and processing the audio spectrogram based on a preset sentiment analysis model to generate sentiment analysis results corresponding to the audio spectrogram specifically includes: The audio graph is input into the second convolutional layer of the sentiment analysis model, and the audio graph is convolved by the second convolutional layer to generate a sentiment feature vector corresponding to the audio graph. The emotional feature vector is normalized to obtain the target emotional feature vector; The target sentiment feature vector is input into the long short-term memory neural network in the sentiment analysis model. The long short-term memory neural network processes the target sentiment feature vector to generate the corresponding long-term domain sentiment features. The long-term sentiment features are input into a fully connected layer within the sentiment analysis model. The long-term sentiment features are then processed by the fully connected layer to obtain the corresponding target long-term sentiment features. The target long-term sentiment features are input into the second classification layer of the sentiment analysis model. The second classification layer performs prediction processing on the target long-term sentiment features and outputs the sentiment feature category corresponding to the target long-term sentiment features. The sentiment analysis results are generated based on the sentiment feature categories.

5. The AI-based speech recommendation method according to claim 4, characterized in that, The step of generating the sentiment analysis result based on the sentiment feature category specifically includes: Obtain the probability of each of the stated emotion feature categories; Obtain the target sentiment feature category with the highest probability from all the stated sentiment feature categories; The target sentiment feature category is used as the sentiment analysis result.

6. A speech recommendation device based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire audio data of conversations between agents and target customers; The conversion module is used to convert the call audio data into text data; The prediction module is used to perform prediction processing on the text data based on a preset intent prediction model, and generate intent prediction results corresponding to the text data. The extraction module is used to extract the audio spectrum of the target customer from the call audio data; The analysis module is used to analyze and process the audio graph based on a preset sentiment analysis model, and generate sentiment analysis results corresponding to the audio graph. The query module is used to perform query processing on the preset script database based on the text data, the intent prediction result and the sentiment analysis result, and to find the corresponding target script from the script database; The push module is used to push the target message to the target customer; The query module includes: The first search submodule is used to search the script library for multiple first scripts that correspond to the intent prediction result, forming a first script set; The second search submodule is used to search for multiple second scripts that correspond to the sentiment analysis results from the first script set, forming a second script set; The third search submodule is used to search for the third script corresponding to the text data from the second script set; The second generation submodule is used to generate the target script based on the third script; The third search submodule includes: The calculation unit is used to calculate the similarity between the text data and each of the second dialogues in the second dialogue set based on a preset similarity algorithm; The comparison unit is used to compare each of the similarities with a preset similarity threshold, and to filter out the specified dialogues with a similarity greater than or equal to the similarity threshold from all the second dialogues; The second determining unit is used to use the designated script as the third script; Specifically, by obtaining the number of third-level dialogues, if the number of third-level dialogues is less than or equal to a preset number, then the third-level dialogue is directly used as the target dialogue. If the number of third-level dialogues is greater than the preset number, then based on a preset similarity algorithm, a preset number of fourth-level dialogues with the highest similarity to the text information are selected from all third-level dialogues, and the fourth-level dialogue is used as the target dialogue.

7. A computer device comprising a memory and a processor, the memory storing computer-readable instructions, wherein the processor, when executing the computer-readable instructions, implements the steps of the AI-based speech recommendation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the AI-based speech recommendation method as described in any one of claims 1 to 5.

Images