Artificial intelligence-based intention recognition method and device, equipment and medium
The joint intent recognition model, combining advanced neural networks and natural speech processing, addresses the limitations of current robots by enhancing intent recognition accuracy and robustness.
Patent Information
- Application Number
- CN202411481345.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2024-12-31
AI Technical Summary
Current intelligent customer service robots have limited comprehension capabilities, particularly in understanding complex semantics, leading to inaccurate intent recognition.
A method involving cascading a first convolutional neural network with a bidirectional gated recurrent unit and a second convolutional neural network with a self-attention module and bidirectional long short-term memory network to create a joint intent recognition model, combined with natural speech processing to convert user voice to text and perform intent recognition.
Enhances intent recognition accuracy by performing secondary classification and recognition, improving the model's robustness and ability to understand user intentions in various scenarios.
Smart Images
Figure CN119226484A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of speech recognition technology, and in particular to an intention recognition method, device, equipment and medium based on artificial intelligence. Background Art
[0002] At present, the company has many insurance products, such as auto insurance and non-auto insurance, etc. There are also a large number of users, but the number of seats is not proportional to the number of users. The following problems exist: 1. The agents have many lists to contact every day, and each list requires the agents to contact, which is time-consuming and labor-intensive; 2. The agents sometimes have too many lists to contact.
[0003] Many companies use intelligent customer service robots, which use artificial intelligence technology to simulate human customer service to communicate and solve problems. They can be online 24 hours a day: providing services to customers at any time without time restrictions; quick response: able to answer customer questions in a short time to improve customer satisfaction; multi-channel support: can be deployed on multiple channels such as websites, apps, WeChat, etc., so that customers can consult anytime and anywhere. Personalized service: by analyzing customer history and behavior, provide customers with personalized services.
[0004] However, current customer service robots have limited comprehension capabilities, which are mainly manifested in the following aspects: difficulty in accurately understanding complex semantics: for some scenarios with ambiguous meanings, complex expressions, or questions with specific contexts, they may not be able to understand the user’s best intentions and give inaccurate answers. Summary of the invention
[0005] The purpose of the embodiments of the present application is to propose an artificial intelligence-based intent recognition method, device, equipment and medium to solve the problem that the current intelligent customer service robots have limited understanding ability and cannot obtain the user's optimal intent.
[0006] In order to solve the above technical problems, the embodiment of the present application provides an intention recognition method based on artificial intelligence, including:
[0007] Cascading the first convolutional neural network with the bidirectional gated recurrent unit to obtain a first intent recognition model, setting a self-attention module on the encoder module of the second convolutional neural network, and cascading the output end of the second convolutional neural network and the output end of the self-attention module with the bidirectional long short-term memory network module to obtain a second intent recognition model;
[0008] Concatenating the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model;
[0009] Acquire user voice, convert the user voice into user text using natural speech processing technology, perform intent recognition on the user text using the joint intent recognition model, and generate an intent recognition result corresponding to the user voice.
[0010] Furthermore, the step of acquiring user voice and converting the user voice into user text using natural speech processing technology includes:
[0011] Acquiring user voice, preprocessing the user voice, and extracting acoustic features from the preprocessed user voice;
[0012] The extracted acoustic features are recognized using natural speech processing technology to obtain the user text.
[0013] Further, the acquiring of user voice, preprocessing of the user voice, and extraction of acoustic features of the preprocessed user voice include:
[0014] Filtering the user's voice to remove noise from the user's voice;
[0015] The acoustic features of the filtered user voice are extracted to obtain the frequency features, duration features and amplitude features corresponding to the user voice.
[0016] Furthermore, the step of using natural speech processing technology to recognize the extracted acoustic features to obtain the user text includes:
[0017] The frequency feature, duration feature and amplitude feature are converted into factor representation feature and syllable representation feature by using the acoustic model in the natural speech processing technology;
[0018] The factor representation features and syllable representation features are decoded using the speech model in the natural speech processing technology to obtain a preliminary user text, and the preliminary user text is corrected to obtain the user text.
[0019] Furthermore, the step of using the joint intent recognition model to perform intent recognition on the user text and generating an intent recognition result corresponding to the user voice includes:
[0020] Using the first intention recognition model to perform intent recognition on the user text to obtain a first recognition result;
[0021] Using the second intent recognition model to perform intent recognition on the user text to obtain a second recognition result;
[0022] Inputting the first recognition result and the second recognition result into a joint classifier, performing secondary classification and recognition on the first recognition result and the second recognition result using confidence, and obtaining corresponding first confidence and second confidence respectively;
[0023] comparing the first confidence level with the second confidence level;
[0024] The recognition result with a higher confidence level is used as the intention recognition result corresponding to the user voice.
[0025] Furthermore, the step of using the first intent recognition model to perform intent recognition on the user text to obtain a first recognition result includes:
[0026] Using the first convolutional neural network to perform feature extraction on the input user text to obtain a first feature sequence;
[0027] Using the bidirectional gated recurrent unit to perform bidirectional context information extraction on the first feature sequence, synchronously updating the current state at each time step, and using the current hidden state feature as the first output feature sequence;
[0028] The first output feature sequence is sequentially input into the first fully connected layer and the first softmax layer to obtain a first recognition result.
[0029] Furthermore, the step of using the second intent recognition model to perform intent recognition on the user text to obtain a second recognition result includes:
[0030] Using the second convolutional neural network to perform feature extraction on the input user text to obtain a second feature sequence;
[0031] Calculate the attention weight of each element of the second feature sequence and other elements using the self-attention module;
[0032] Using the attention weight, weighted sum is performed on each element of the second feature sequence to obtain a third feature sequence;
[0033] Extracting the overall features of the third feature sequence using the bidirectional long short-term memory network module to obtain a second output feature sequence;
[0034] The second output feature sequence is sequentially input into the second fully connected layer and the second softmax layer to obtain the second recognition result.
[0035] In order to solve the above problems, an optimal intention recognition device based on a joint model is provided, which adopts the above-mentioned intention recognition method based on artificial intelligence, and is characterized by comprising:
[0036] A first acquisition module is used to cascade the first convolutional neural network with a bidirectional gated recurrent unit to obtain a first intent recognition model, set a self-attention module on the encoder module of the second convolutional neural network, and cascade the output end of the second convolutional neural network and the output end of the self-attention module with a bidirectional long short-term memory network module to obtain a second intent recognition model;
[0037] A second acquisition module, used to splice the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model;
[0038] The recognition module is used to obtain user voice, convert the user voice into user text using natural speech processing technology, perform intent recognition on the user text using the joint intent recognition model, and generate an intent recognition result corresponding to the user voice.
[0039] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based intention recognition method as described in the first aspect are implemented.
[0040] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of the artificial intelligence-based intention recognition method as described in the first aspect are implemented.
[0041] Compared with the prior art, the embodiments of the present application mainly have the following technical effects: a first intention recognition model is constructed by cascading a first convolutional neural network with a bidirectional gated recurrent unit, a second intention recognition model is constructed by cascading a second convolutional neural network and a self-attention module with a bidirectional long short-term memory network module, and a joint intention recognition model is constructed to perform secondary classification and recognition on the classification recognition results to obtain the final classification and recognition results, thereby enhancing the recognition effect of multi-scenario intentions, solving the problem that the current intelligent customer service robot has limited understanding ability and cannot obtain the user's optimal intentions, and improving the robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0043] Figure 1A flowchart of an embodiment of an intention recognition method based on artificial intelligence according to the present application;
[0044] Figure 2 yes Figure 1 A flowchart of a specific implementation of step 300;
[0045] Figure 3 yes Figure 2 A flowchart of a specific implementation of step 310;
[0046] Figure 4 yes Figure 2 A flowchart of a specific implementation of step 320;
[0047] Figure 5 yes Figure 2 A flowchart of a specific implementation method after step 320;
[0048] Figure 6 yes Figure 5 A flowchart of a specific implementation of step 330;
[0049] Figure 7 yes Figure 5 A flowchart of a specific implementation of step 340;
[0050] Figure 8 is a schematic diagram of the structure of the joint intent recognition model in this application;
[0051] Fig. 9 is a structural schematic diagram of an embodiment of an optimal intention recognition device based on a joint model according to the present application;
[0052] Fig.10 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION
[0053] The following will be combined with the accompanying drawings in the invention to clearly and completely describe the technical solutions in 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, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.
[0054] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.
[0056] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0057] The embodiment of the present application provides an intention recognition method based on artificial intelligence, which can be applied to a computer device with the ability to run instructions or programs. Among them, the computer device can be but is not limited to various servers, workstations, personal computers, laptops, etc. Running on different computing devices is only a difference in the execution subject of the scheme. People in this field can foresee that running on different computing devices can produce the same technical effect. The present invention is described in detail below through specific embodiments.
[0058] The purpose of the embodiments of the present application is to propose an artificial intelligence-based intent recognition method, device, equipment and medium to solve the problem that the current intelligent customer service robots have limited understanding ability and cannot obtain the user's optimal intent.
[0059] In order to solve the above technical problems, the embodiment of the present application provides an intention recognition method based on artificial intelligence, which adopts the following technical solutions: Figure 1 , Figure 1 A flowchart of an embodiment of an intention recognition method based on artificial intelligence according to the present application includes:
[0060] S100, cascading the first convolutional neural network with the bidirectional gated recurrent unit to obtain a first intent recognition model, setting a self-attention module on the encoder module of the second convolutional neural network, and cascading the output end of the second convolutional neural network and the output end of the self-attention module with the bidirectional long short-term memory network module to obtain a second intent recognition model;
[0061] The first intent recognition model CNN-BGRU (Convolutional Neural Network-Bidirectional Gated Recurrent Unit) combines convolutional neural network (CNN) and bidirectional gated recurrent unit (BGRU) for user text recognition.
[0062] CNN is a type of feedforward neural network that includes convolution calculations and has a deep structure. It is one of the representative algorithms of deep learning. By stacking multiple convolutional layers, activation function layers, pooling layers (optional) and fully connected layers (usually in the last few layers), a powerful feature extraction and classification network is formed. In the embodiment of the present application, a convolutional neural network (CNN) is used to automatically extract the local features of the input user text data. Through the combination of multiple convolutional layers and pooling layers, higher-level abstract features can be gradually extracted to distinguish different text characters.
[0063] BGRU is a variant of recurrent neural network (RNN), specifically a bidirectional extended version of gated recurrent unit (GRU). BGRU combines the advantages of bidirectional RNN and GRU to better capture the contextual information in sequence data. BGRU consists of two GRUs: Forward GRU: accepts a forward input sequence, and processes and transmits information in order from left to right (or from front to back). Reverse GRU: accepts a reverse input sequence, that is, the reverse order of the original sequence, and processes and transmits information in order from right to left (or from back to front). In an embodiment of the present application, a bidirectional gated recurrent unit (BGRU) is used to process sequence data. BGRU is a bidirectional GRU, which can simultaneously consider the forward and reverse information of the feature sequence, so as to better capture the contextual relationship.
[0064] The second intent recognition model uses a bidirectional long short-term memory network module (BLSTM) and introduces a self-attention module. The BLSTM consists of two LSTMs (Long Short-Term Memory) in the forward and backward directions. The LSTM model in the bidirectional long short-term memory network module (BLSTM) is used to solve the long-term dependency problem of the traditional RNN (Recurrent Neural Network). The forward LSTM processes each element from the beginning to the end of the sequence, while the backward LSTM processes each element from the end to the beginning of the sequence. The bidirectional long short-term memory network module (BLSTM) can simultaneously use past and future contextual information to better capture the overall characteristics of the sequence. The self-attention module allows the second intent recognition model to adaptively focus on different parts of the second feature sequence when processing the second feature sequence data of the user text, thereby capturing the complex relationship between the elements in the second feature sequence.
[0065] S200, concatenating the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model;
[0066] like Figure 8 , Figure 8 This is a schematic diagram of the structure of the joint intent recognition model in this application; the first intent recognition model includes a first convolutional neural network, BGRU, a first fully connected layer and a first softmax layer, and the second intent recognition model includes a second convolutional neural network, BLSTM, a self-attention module, a second fully connected layer and a second softmax layer. The two intent recognition models are spliced through a joint classifier. Among them, the joint classifier is used to perform secondary classification and recognition on the first recognition result and the second recognition result of the first intent recognition model and the second intent recognition model. The joint intent recognition model obtained by splicing can obtain an intent recognition result with a higher confidence and can better obtain the user's optimal intention.
[0067] S300, obtaining user voice, converting the user voice into user text using natural speech processing technology, performing intent recognition on the user text using a joint intent recognition model, and generating an intent recognition result corresponding to the user voice.
[0068] In this embodiment, after the user actively calls in, the telephone platform is responsible for real-time transfer, and then converts the user's voice into user text in real time through the natural speech processing module (NLP), and then uses the joint intention recognition model to recognize the user's intention. After recognizing the user's intention, the corresponding response words are generated, and finally the response voice is synthesized to respond to the user.
[0069] The present application constructs a first intent recognition model by cascading a first convolutional neural network with a bidirectional gated recurrent unit, constructs a second intent recognition model by cascading a second convolutional neural network and a self-attention module with a bidirectional long short-term memory network module, and constructs a joint intent recognition model to perform secondary classification and recognition on the classification and recognition results to obtain the final classification and recognition results, thereby enhancing the recognition effect of multi-scenario intentions, solving the problem that the current intelligent customer service robots have limited understanding capabilities and cannot obtain the optimal intentions of users, and improving the robustness of the model.
[0070] In some optional embodiments, such as Figure 2 , Figure 2 yes Figure 1 A flowchart of a specific implementation method of S300; S300 includes S310, obtaining user voice, preprocessing the user voice, and extracting acoustic features of the preprocessed user voice; S320, using natural speech processing technology to recognize the extracted acoustic features to obtain user text.
[0071] The main thing obtained by extracting acoustic features from speech is a series of factors that can describe the essential information of speech signals, such as frequency features, duration features, amplitude features, etc. These features play an important role in subsequent speech recognition, speaker recognition or other speech processing tasks.
[0072] This application acoustically extracts the user's voice and recognizes it through natural speech processing technology, which can accurately identify the voice content and convert it into text, reducing the errors that may occur in manual transcription, such as mishearing, omissions, spelling errors, etc., and ensuring the integrity and accuracy of the information.
[0073] For example, in a customer service scenario about insurance, a user called the customer service hotline to report that the insurance product he purchased had quality problems. The customer service robot used speech recognition technology and the NLP model to convert the customer's voice into text in real time.
[0074] The customer service robot quickly and accurately converts the speech into text and analyzes and understands the text content. In this process, the speech-to-text function of natural speech processing technology enables the customer service robot to quickly understand the user's question and accurately convert the speech into the corresponding text, preparing for the subsequent accurate recognition of the user's intention.
[0075] In some optional embodiments, such as Figure 3 , Figure 3 yes Figure 2A flowchart of a specific implementation method of S310 in the embodiment; S310 includes: S311, filtering the user voice to remove noise in the user voice; S312, extracting acoustic features of the filtered user voice to obtain frequency features, duration features and amplitude features of the corresponding user voice.
[0076] The speech processing model in natural speech processing (NLP) analyzes the input speech signal and extracts various acoustic features that reflect the physical characteristics of the speech, such as frequency features, amplitude features, and time features.
[0077] By preprocessing the speech signal, including noise removal, signal enhancement and other operations, the accuracy of subsequent conversion of user speech into user text can be improved.
[0078] In some preferred embodiments, Figure 4 , Figure 4 yes Figure 2 A flowchart of a specific implementation method of S320, S320 includes: S321, using the acoustic model in natural speech processing technology to convert frequency features, duration features and amplitude features into factor representation features and syllable representation features; S322, using the speech model in natural speech processing technology to decode the factor representation features and syllable representation features, obtain preliminary user text, and correct the preliminary user text to obtain user text.
[0079] The extracted acoustic features are processed using a pre-trained speech processing model in natural speech processing (NLP). The speech processing model usually consists of two parts: an acoustic model and a language model. The acoustic model is responsible for converting acoustic features into intermediate representations such as phonemes or syllables, while the language model further decodes these intermediate representations based on the statistical laws and grammatical structure of the language to determine the most likely text content.
[0080] After obtaining the preliminary text results, language understanding and error correction are still required. For example, using technologies such as part-of-speech tagging and named entity recognition to determine the grammatical correctness and semantic rationality of the text, and correct possible errors.
[0081] By correcting the converted user text, the accuracy of the user text input into the joint intent text model is improved, which is conducive to identifying the user's optimal intent.
[0082] In some optional embodiments, such as Figure 5 , Figure 5 yes Figure 2 A flowchart of a specific implementation method after S320 in FIG. 1 ; after S320, S300 further includes:
[0083] S330, use the first intention recognition model to perform intent recognition on the user text to obtain a first recognition result; S340, use the second intention recognition model to perform intent recognition on the user text to obtain a second recognition result; S350, input the first recognition result and the second recognition result into a joint classifier, use the confidence level to perform secondary classification and recognition on the first recognition result and the second recognition result, and obtain the corresponding first confidence level and second confidence level respectively; S360, compare the first confidence level with the second confidence level, and use the recognition result with a larger confidence level as the intention recognition result corresponding to the user voice.
[0084] In this embodiment, the first confidence level and the second confidence level respectively represent the possibility that the first recognition result and the second recognition result of the predicted recognition are the true intention of the user. If the first confidence level and the second confidence level are both greater than the set confidence threshold, the one with higher confidence level is closer to the true intention. If one of them is greater than the set threshold and the other is less than the set threshold, the recognition result less than the threshold is automatically discarded. Secondary classification and recognition is performed through confidence level, thereby improving the accuracy of user intention recognition.
[0085] By performing secondary classification and recognition on the first recognition result and the second recognition result, it is helpful for the joint recognition model to recognize the user's optimal intention. Figure 6 , S330 includes: S331, using the first convolutional neural network to extract features of the input user text to obtain a first feature sequence; S332, using the bidirectional gated recurrent unit to extract bidirectional context information from the first feature sequence, synchronously updating the current state at each time step, and using the current hidden state features as the first output feature sequence; S333, inputting the first output feature sequence into the first fully connected layer and the first softmax layer in sequence to obtain a first recognition result.
[0086] Before the user text is input into the first convolutional neural network, the input text is preprocessed so that it has a uniform format and length size to facilitate processing by the first convolutional neural network.
[0087] The first convolutional neural network gradually extracts the features of the character image through the operation of convolutional layers and pooling layers. Each convolutional layer performs convolution operations with the input user text through different convolution kernels to obtain different feature maps and capture different local features. The pooling layer downsamples the feature map to reduce the dimension of the feature while retaining the main features to obtain the first feature sequence. The bidirectional gated recurrent unit (BGRU) processes the first feature sequence from the forward and reverse directions respectively to capture the contextual information in both the forward and reverse directions. At each time step, the bidirectional gated recurrent unit (BGRU) updates the current state according to the current input and the state at the previous moment, and outputs the current hidden state. The first output feature sequence output by the bidirectional gated recurrent unit (BGRU) is classified through the fully connected layer and the softmax layer. The fully connected layer maps the output of the bidirectional gated recurrent unit (BGRU) to different category spaces. The softmax layer converts the output into a probability distribution, and each category corresponds to a probability value, indicating the possibility that the input character belongs to that category. The first intent recognition model finally selects the category with the highest probability as the first recognition result.
[0088] By integrating the first convolutional neural network with a bidirectional gated recurrent unit (BGRU), the bidirectional gated recurrent unit (BGRU) can make full use of the information of the entire first feature sequence, thereby performing feature extraction more accurately.
[0089] For example, in the sentiment analysis of user text, a word expressing emotion may need to determine the specific emotional tendency based on the descriptions before and after it. BGRU can better capture this emotional change through two-way information transmission.
[0090] Further, in some optional embodiments, as shown in FIG. 7 , Figure 7 yes Figure 5 A flowchart of a specific implementation of step S340 includes: S341, using a second convolutional neural network to extract features of the input user text to obtain a second feature sequence; S342, using a self-attention module to calculate the attention weight of each element of the second feature sequence and other elements; S343, using the attention weight to perform weighted summation on each element of the second feature sequence to obtain a third feature sequence; S344, using a bidirectional long short-term memory network module to extract the overall features of the third feature sequence to obtain a second output feature sequence; S345, inputting the second output feature sequence into the second fully connected layer and the second softmax layer in sequence to obtain a second recognition result.
[0091] In this embodiment, the self-attention module is used to allow the second intent recognition model to adaptively focus on different parts of the second feature sequence when processing the second feature sequence data of the user text, thereby capturing the complex relationship between the elements in the second feature sequence. By calculating the attention weight between each element and other elements in the second feature sequence, and then performing weighted summation on the second feature sequence, a new representation of each element is obtained. This new representation incorporates the information of other elements in the second feature sequence, and focuses on the parts that are closely related to the current element in a targeted manner according to the size of the attention weight.
[0092] In this embodiment, the bidirectional long short-term memory network module (BLSTM) is composed of two LSTMs (Long Short-Term Memory) in the forward and backward directions. The LSTM model in the bidirectional long short-term memory network module (BLSTM) is used to solve the long-term dependency problem existing in the traditional RNN (Recurrent Neural Network). The forward LSTM processes each element in sequence from the beginning to the end of the sequence, while the backward LSTM processes each element from the end to the beginning of the sequence. The bidirectional long short-term memory network module (BLSTM) can use the past and future contextual information at the same time to better capture the overall characteristics of the sequence.
[0093] For example, in the sentence "I want to give up third-party insurance, which is not only expensive but also impractical", when processing the word "third-party insurance", the bidirectional long short-term memory network module (BLSTM) can simultaneously consider the preceding "I want to give up" and the following "which is not only expensive but also impractical", thereby gaining a more comprehensive understanding of the role of "third-party insurance" in this sentence.
[0094] By setting the self-attention module and the bidirectional long short-term memory network module in the second intent recognition model, the model can not only adaptively focus on different parts of the input second feature sequence, thereby better capturing the internal structure and relationship of the second feature sequence, but also utilize past and future contextual information to better capture the overall characteristics of the second feature sequence.
[0095] For further reference, Fig. 9 , as a response to the above Figure 1 The present application provides an embodiment of an intention recognition device based on artificial intelligence, and the device embodiment is similar to Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.
[0096] like Fig. 9 , the intention recognition device 400 based on artificial intelligence described in this embodiment includes:
[0097] A first acquisition module 401 is used to cascade the first convolutional neural network with a bidirectional gated recurrent unit to acquire a first intent recognition model, set a self-attention module on the encoder module of the second convolutional neural network, and cascade the output end of the second convolutional neural network and the output end of the self-attention module with a bidirectional long short-term memory network module to acquire a second intent recognition model;
[0098] A second acquisition module 402 is used to splice the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model;
[0099] The recognition module 403 is used to obtain user voice, convert the user voice into user text using natural speech processing technology, perform intent recognition on the user text using the joint intent recognition model, and generate an intent recognition result corresponding to the user voice.
[0100] By constructing a joint intent recognition model to perform secondary classification recognition on the classification recognition results, the final classification recognition results are obtained, which enhances the recognition effect of multi-scenario intentions, solves the problem that the current intelligent customer service robot has limited understanding ability and cannot obtain the user's optimal intentions, and improves the robustness of the model. In order to solve the above technical problems, the embodiment of the present application also proposes a computer device, which adopts the following technical solution: including a processor, a network module and a memory, and the processor and the memory are interconnected through a network module.
[0101] The computer device may be a computer, server, workstation or other device, or a mobile phone, tablet, vehicle-mounted mobile terminal or other device with program execution capability. The internal structure diagram of the computer device may be as follows: Fig.10 As shown, Fig.10 : This is a structural diagram of an embodiment of a computer device according to the present application. The computer device includes a processor, a memory and a network module. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, instructions or codes. The internal memory provides an environment for the operation of the operating system and instructions or codes in the non-volatile storage medium. When the instructions or codes are executed by the processor, a function or step of the above-mentioned intention recognition based on artificial intelligence is implemented. The network module of the computer device may include a network interface and / or a wireless network module, and the computer device may communicate with other devices or service platforms through the network module. In addition, the computer device may also include a display screen and an input device, etc.
[0102] Among them, the memory is used to store a computer program, the computer program includes program instructions, the processor is configured to call the program instructions, and when the processor executes the instructions or codes, the steps of the intention recognition method based on artificial intelligence as described above are implemented.
[0103] In order to solve the above problems, the embodiment of the present application also proposes a computer-readable storage medium, which adopts the following technical solution: the computer-readable storage medium stores a computer program, which includes program instructions, and the program instructions are executed by the processor to implement Figures 1 to 7 For the artificial intelligence-based intention recognition method provided in each step, please refer to the implementation method provided in the above steps, which will not be repeated here.
[0104] The computer-readable storage medium may be a device for positioning a multi-scale framework of a video moment provided in any of the above embodiments or an internal storage unit of the terminal device, such as a hard disk or memory of a computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the computer device.
[0105] Furthermore, the computer-readable storage medium may also include both an internal storage unit of the computer device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.
[0106] However, it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform 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 (ASIC), field-programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.
[0107] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.
[0108] Compared with the prior art, the embodiments of the present application mainly have the following technical effects: a first intention recognition model is constructed by cascading a first convolutional neural network with a bidirectional gated recurrent unit, a second intention recognition model is constructed by cascading a second convolutional neural network and a self-attention module with a bidirectional long short-term memory network module, and a joint intention recognition model is constructed to perform secondary classification and recognition on the classification recognition results to obtain the final classification and recognition results, thereby enhancing the recognition effect of multi-scenario intentions, solving the problem that the current intelligent customer service robot has limited understanding ability and cannot obtain the user's optimal intentions, and improving the robustness of the model.
[0109] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention. The non-company software tools or components appearing in the embodiments of this application are only examples and do not represent actual use.
Claims
1. An intention recognition method based on artificial intelligence, characterized in that it includes: Cascading the first convolutional neural network with the bidirectional gated recurrent unit to obtain a first intent recognition model, setting a self-attention module on the encoder module of the second convolutional neural network, and cascading the output end of the second convolutional neural network and the output end of the self-attention module with the bidirectional long short-term memory network module to obtain a second intent recognition model; Concatenating the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model; Acquire user voice, convert the user voice into user text using natural speech processing technology, perform intent recognition on the user text using the joint intent recognition model, and generate an intent recognition result corresponding to the user voice.
2. The intention recognition method based on artificial intelligence according to claim 1, characterized in that: The step of acquiring user voice and converting the user voice into user text using natural speech processing technology includes: Acquiring user voice, preprocessing the user voice, and extracting acoustic features from the preprocessed user voice; The extracted acoustic features are recognized using natural speech processing technology to obtain the user text.
3. The intention recognition method based on artificial intelligence according to claim 2 is characterized in that: The acquiring of user voice, preprocessing of the user voice, and extraction of acoustic features of the preprocessed user voice include: Filtering the user's voice to remove noise from the user's voice; The acoustic features of the filtered user voice are extracted to obtain the frequency features, duration features and amplitude features corresponding to the user voice.
4. The intention recognition method based on artificial intelligence according to claim 3 is characterized in that: The step of using natural speech processing technology to recognize the extracted acoustic features to obtain the user text includes: The frequency feature, duration feature and amplitude feature are converted into factor representation feature and syllable representation feature by using the acoustic model in the natural speech processing technology; The factor representation features and syllable representation features are decoded using the speech model in the natural speech processing technology to obtain a preliminary user text, and the preliminary user text is corrected to obtain the user text.
5. The intention recognition method based on artificial intelligence according to claim 1, characterized in that: The step of using the joint intention recognition model to perform intention recognition on the user text and generating an intention recognition result corresponding to the user voice includes: Using the first intention recognition model to perform intent recognition on the user text to obtain a first recognition result; Using the second intent recognition model to perform intent recognition on the user text to obtain a second recognition result; Inputting the first recognition result and the second recognition result into a joint classifier, performing secondary classification and recognition on the first recognition result and the second recognition result using confidence, and obtaining corresponding first confidence and second confidence respectively; comparing the first confidence level with the second confidence level; The recognition result with a higher confidence level is used as the intention recognition result corresponding to the user voice.
6. The artificial intelligence-based intention recognition method according to claim 5, characterized in that: The step of using the first intention recognition model to perform intent recognition on the user text to obtain a first recognition result includes: Using the first convolutional neural network to perform feature extraction on the input user text to obtain a first feature sequence; Using the bidirectional gated recurrent unit to perform bidirectional context information extraction on the first feature sequence, synchronously updating the current state at each time step, and using the current hidden state feature as the first output feature sequence; The first output feature sequence is sequentially input into the first fully connected layer and the first softmax layer to obtain a first recognition result.
7. The artificial intelligence-based intention recognition method according to claim 5, characterized in that: The step of using the second intent recognition model to perform intent recognition on the user text to obtain a second recognition result includes: Using the second convolutional neural network to perform feature extraction on the input user text to obtain a second feature sequence; Calculate the attention weight of each element of the second feature sequence and other elements using the self-attention module; Using the attention weight, weighted sum is performed on each element of the second feature sequence to obtain a third feature sequence; Extracting the overall features of the third feature sequence using the bidirectional long short-term memory network module to obtain a second output feature sequence; The second output feature sequence is sequentially input into the second fully connected layer and the second softmax layer to obtain the second recognition result.
8. An optimal intention recognition device based on a joint model, using an intention recognition method based on artificial intelligence as described in any one of claims 1 to 7, characterized in that: include: A first acquisition module is used to cascade the first convolutional neural network with a bidirectional gated recurrent unit to obtain a first intent recognition model, set a self-attention module on the encoder module of the second convolutional neural network, and cascade the output end of the second convolutional neural network and the output end of the self-attention module with a bidirectional long short-term memory network module to obtain a second intent recognition model; A second acquisition module, used to splice the first intent recognition model and the second intent recognition model to obtain a joint intent recognition model; The recognition module is used to obtain user voice, convert the user voice into user text using natural speech processing technology, perform intent recognition on the user text using the joint intent recognition model, and generate an intent recognition result corresponding to the user voice.
9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based intention recognition method as described in any one of claims 1 to 7 are implemented.
10. 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 artificial intelligence-based intention recognition method as described in any one of claims 1 to 7.