Intention recognition method, device and equipment of dialogue information, and storage medium

Abstract

The application discloses a dialogue information intention recognition method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: obtaining dialogue text to be recognized; performing feature extraction processing on the dialogue text to obtain feature information of the dialogue text; performing multi-dimensional classification coding on the feature information, converting a plurality of discrete feature variables in the feature information into continuous feature variables to obtain a multi-dimensional feature vector of the feature information; performing classification processing on the multi-dimensional feature vector respectively to obtain multi-dimensional classification results of the feature information; and performing fusion recognition processing on the multi-dimensional classification results to determine an intention recognition result of the dialogue text. In the application, the intention recognition result is determined by analyzing and processing from multiple dimensions, which can improve the accuracy of the intention recognition result. Converting a plurality of discrete feature variables into multi-dimensional continuous feature variables can effectively reduce information loss in the subsequent feature processing process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a dialogue information intent recognition method and device, equipment and a storage medium. BACKGROUND

[0002] Task-oriented dialogue refers to an automatic communication dialogue between an artificial intelligence and a user. In the task-oriented dialogue, intent recognition needs to be performed on dialogue text input by the user to feed back appropriate response text to the user.

[0003] In related technologies, in the application process of the task-oriented dialogue, an automatic response system determines fixed response text in advance, and then, after obtaining dialogue text input by the user, determines a keyword in the dialogue text according to a pre-set keyword extraction rule, determines a next sentence response text corresponding to the keyword based on the keyword in the dialogue text, and then shows the next sentence response text to the user, so as to realize automatic dialogue.

[0004] However, in the above related technologies, the keyword cannot completely and accurately reflect semantic information of the dialogue text input by the user, thereby leading to inaccurate intent recognition of the dialogue text input by the user. SUMMARY

[0005] Embodiments of the present application provide a dialogue information intent recognition method, device, equipment and storage medium, which can improve the accuracy of the intent recognition result. The technical solution is as follows:

[0006] According to an aspect of an embodiment of the present application, a dialogue information intent recognition method is provided, and the method comprises:

[0007] Obtaining dialogue text to be recognized;

[0008] Performing feature extraction processing on the dialogue text to obtain feature information of the dialogue text; wherein the feature information comprises a plurality of discrete feature variables;

[0009] Performing multi-dimensional classification coding on the feature information to convert the plurality of discrete feature variables in the feature information into continuous feature variables to obtain a multi-dimensional feature vector of the feature information; wherein the multi-dimensional feature vector comprises a plurality of continuous feature variables of different dimensions;

[0010] Performing classification processing on the multi-dimensional feature vector respectively to obtain multi-dimensional classification results of the feature information, and the multi-dimensional classification results comprise a plurality of classification results of different dimensions;

[0011] The multi-dimension classification results are fused and recognized to determine the intention recognition result of the dialogue text.

[0012] According to an aspect of the embodiments of the present application, a training method of an intention recognition model is provided, and the method comprises:

[0013] At least one training sample of the intention recognition model is acquired, and the training sample comprises sample dialogue text and label information corresponding to the sample dialogue text.

[0014] Feature extraction processing is performed on the sample dialogue text to obtain feature information of the sample dialogue text, wherein the feature information comprises multiple discrete feature variables.

[0015] The feature information is subjected to multi-dimension classification coding, and the multiple discrete feature variables in the feature information are converted into continuous feature variables to obtain a multi-dimension feature vector of the feature information, wherein the multi-dimension feature vector comprises multiple feature vectors of different dimensions, and the feature vectors are continuous feature variables.

[0016] The multi-dimension feature vector is subjected to classification processing respectively to obtain multi-dimension classification results of the feature information, and the multi-dimension classification results comprise multiple classification results of different dimensions.

[0017] The multi-dimension classification results are subjected to fusion and recognition processing to determine the intention recognition result of the dialogue text.

[0018] Based on the multi-dimension classification results, the intention recognition result, and the label information corresponding to the sample dialogue text, a loss function value of the intention recognition model is calculated.

[0019] Based on the loss function value, parameters of the intention recognition model of the dialogue information are adjusted.

[0020] According to an aspect of the embodiments of the present application, a device for intention recognition of dialogue information is provided, and the device comprises:

[0021] A text acquisition module is configured to acquire dialogue text to be recognized.

[0022] A feature extraction module is configured to perform feature extraction processing on the dialogue text to obtain feature information of the dialogue text, wherein the feature information comprises multiple discrete feature variables.

[0023] A feature coding module is configured to perform multi-dimension classification coding on the feature information, convert the multiple discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimension feature vector of the feature information, wherein the multi-dimension feature vector comprises multiple continuous feature variables of different dimensions.

[0024] a feature classification module, configured to perform classification processing on the multi-dimensional feature vectors respectively to obtain multi-dimensional classification results of the feature information, the multi-dimensional classification results including classification results of the multiple different dimensions;

[0025] a result fusion module, configured to perform fusion recognition processing on the multi-dimensional classification results to determine an intent recognition result of the dialogue text.

[0026] According to an aspect of an embodiment of the present application, a training device of an intent recognition model is provided, and the device includes:

[0027] a sample acquisition module, configured to acquire at least one training sample of an intent recognition model, the training sample including a sample dialogue text and label information corresponding to the sample dialogue text;

[0028] a feature acquisition module, configured to perform feature extraction processing on the sample dialogue text to obtain feature information of the sample dialogue text; wherein the feature information includes multiple discrete feature variables;

[0029] a vector acquisition module, configured to perform multi-dimensional classification coding on the feature information, convert the multiple discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimensional feature vector of the feature information; wherein the multi-dimensional feature vector includes multiple feature vectors of different dimensions, and the feature vectors are continuous feature variables;

[0030] a vector classification module, configured to perform classification processing on the multi-dimensional feature vectors respectively to obtain multi-dimensional classification results of the feature information, the multi-dimensional classification results including classification results of multiple different dimensions;

[0031] a result acquisition module, configured to perform fusion recognition processing on the multi-dimensional classification results to determine an intent recognition result of the dialogue text.

[0032] a function value determination module, configured to calculate a loss function value of the intent recognition model based on the multi-dimensional classification results, the intent recognition result, and the label information corresponding to the sample dialogue text;

[0033] a parameter adjustment module, configured to adjust parameters of the intent recognition model of the dialogue information based on the loss function value.

[0034] According to an aspect of some embodiments of the present application, a computer device is provided, which comprises a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the above-mentioned intent recognition method for dialogue information or the above-mentioned training method for intent recognition model.

[0035] According to an aspect of some embodiments of the present application, a computer readable storage medium is provided, which stores at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by a processor to implement the above-mentioned intent recognition method for dialogue information or the above-mentioned training method for intent recognition model.

[0036] According to an aspect of some embodiments of the present application, a computer program product or a computer program is provided, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the above-mentioned intent recognition method for dialogue information or the above-mentioned training method for intent recognition model.

[0037] The technical solutions provided by the embodiments of the present application can bring the following beneficial effects:

[0038] By multi-dimensional classification coding and multi-dimensional classification processing of dialogue text, the multi-dimensional classification results of dialogue text are determined, and then the multi-dimensional classification results are fused and recognized to determine the intent recognition result of dialogue text. The intent recognition result is determined by analyzing and processing from multiple dimensions, which can improve the accuracy of the intent recognition result. Moreover, when the classification coding is performed, the multiple discrete feature variables are converted into multi-dimensional continuous feature variables, which can effectively reduce the information loss in the subsequent feature processing process, and can improve the accuracy and reliability of the intent recognition result. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 is a flowchart of the intent recognition method for dialogue information provided by an embodiment of the present application;

[0041] Figure 2 is a flowchart of a method for intent recognition of dialogue information provided by another embodiment of the present application;

[0042] Figure 3 is a flowchart of a method for training an intent recognition model provided by an embodiment of the present application;

[0043] Figure 4 an exemplary schematic diagram showing the structure of an intent recognition model is shown;

[0044] Figure 5 an exemplary schematic diagram showing the structure of a VAE model is shown;

[0045] Figure 6 an exemplary schematic diagram showing the application of an intent recognition model in a medical application scenario is shown;

[0046] Figure 7 is a block diagram of an apparatus for intent recognition of dialogue information provided by an embodiment of the present application;

[0047] Figure 8 is a block diagram of an apparatus for intent recognition of dialogue information provided by another embodiment of the present application;

[0048] Figure 9 is a block diagram of a training apparatus for an intent recognition model provided by an embodiment of the present application;

[0049] Figure 10 is a block diagram of a training apparatus for an intent recognition model provided by an embodiment of the present application;

[0050] Figure 11 is a structural block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0052] Artificial intelligence (AI) is the use of digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is the design principle and implementation method of various intelligent machines, so that the machine has the functions of perception, reasoning and decision making.

[0053] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes, such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning and other several major directions.

[0054] Natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies the theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language used in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies.

[0055] Machine learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It is dedicated to studying how computers can simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0056] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0057] The scheme provided by the embodiments of the present application relates to machine learning and other technologies of artificial intelligence, and the training sample is used to train an intent recognition model. The training sample includes sample dialogue text and label information corresponding to the sample dialogue text. After the training of the intent recognition model is completed, the dialogue text is input into the intent recognition model, the intent recognition result of the dialogue text is determined by the intent recognition model, and then the computer device selects the response text corresponding to the dialogue text from the dialogue text library according to the intent recognition result, so as to realize the automatic dialogue between the user and the intelligent customer service.

[0058] It should be noted that the intent recognition model provided by the present application can be widely applied to various application scenarios. Specifically as follows:

[0059] (1) In the medical application scenario, the user can understand the relevant regulations of the hospital through the automatic answering system of the hospital. The user inputs the content that the user wants to understand into the automatic answering system of the hospital, and then the automatic answering system generates dialogue text according to the content input by the user, processes the dialogue text by the intent recognition model, determines the intent recognition result of the dialogue text, and then selects the response text associated with the intent recognition result from the dialogue text library based on the intent recognition result, and displays the response text to the user, so that the user can communicate with the automatic answering system through the response text. Moreover, due to the particularity of the medical application scenario, the emergence of the automatic answering system can effectively provide some basic medical services to the user, and indirectly ensure the reasonable arrangement of medical resources. In the case of an emergency, sufficient medical resources can be arranged in time for the emergency.

[0060] (2) In the traffic application scenario, the user plans the route through the automatic answering system of the vehicle terminal. The user inputs the destination into the automatic answering system of the vehicle terminal, and then the automatic answering system generates dialogue text to be recognized according to the destination input by the user, combines the traffic travel field to which the vehicle terminal belongs, processes the dialogue text by the intent recognition system, determines the intent recognition result of the dialogue text, and then selects the response text associated with the intent recognition result from the dialogue text library based on the intent recognition result, and displays the response text to the user, so that the user can reasonably plan the route according to the information provided by the response text.

[0061] (3) In the shopping application scenario, the user purchases the items that meet the user's needs through the automatic answering system of the shopping application program. The user inputs the service function currently needed by the user into the automatic answering system of the shopping application program, and then the automatic answering system generates a dialogue text according to the service function currently needed by the user. The dialogue text is processed by the intent recognition model to determine the intent recognition result of the dialogue text. Then, based on the intent recognition result, the response text associated with the intent recognition result is selected from the dialogue text library, and the response text is displayed to the user. At this time, the user can determine the items that meet the user's needs through the response text for purchase. In this case, the response text can include a purchase link or an order phone number of the items. It should be noted that the shopping in the shopping application scenario refers to any consumption of actual items or virtual items.

[0062] Of course, the intent recognition model of the dialogue information in this application can also be used in other various fields, which are not exemplified here.

[0063] For ease of illustration, in the following method embodiments, only the computer device is taken as an example to introduce and illustrate the execution subject of each step. The computer device can be any electronic device with computing and storage capabilities. For example, the computer device can be a server, which can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (Content Delivery Network), and big data and artificial intelligence platform. For another example, the computer device can also be a terminal, which can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in this application. It should be noted that in the embodiments of this application, the execution subject of each step can be the same computer device, or can be executed by multiple different computer devices in cooperation, which is not limited herein. It should be further noted that in the embodiments of this application, the execution subject of the dialogue information intent recognition method described below and the execution subject of the training method of the intent recognition model can be the same computer device, or can be different computer devices, which is not limited in the embodiments of this application.

[0064] In the following, the technical solutions of the present application will be described in detail in conjunction with several embodiments.

[0065] Please refer to Figure 1Fig. 1 shows a flowchart of a method for identifying an intent of dialogue information according to an embodiment of the present application. The method can include the following steps (101-105):

[0066] In step 101, dialogue text to be identified is obtained.

[0067] The dialogue text refers to text obtained from content input by a user. Optionally, the content input by the user includes text information, voice information, dynamic image information, static image information, etc., which are not limited in the embodiments of the present application.

[0068] In a possible implementation, the computer device directly takes the content input by the user as the dialogue text to be identified.

[0069] In another possible implementation, after obtaining the content input by the user, the computer device performs optimization processing on the content, and then takes the content after the optimization processing as the dialogue text to be identified. The optimization processing includes but is not limited to at least one of the following: extracting text information from image information, converting voice information into text information, removing invalid content (such as punctuation marks, spaces, meaningless auxiliary words, etc.) in text information, etc.

[0070] In the embodiments of the present application, after detecting the content input by the user, the computer device obtains the dialogue text to be identified, and then determines the user intent according to the intent recognition result of the dialogue text. The user intent refers to the meaning expected to be expressed by the user through the dialogue text.

[0071] Optionally, the computer device can obtain the dialogue text to be identified in real time, or can obtain the dialogue text to be identified according to a time interval. In actual application, the computer device determines the obtaining manner of the dialogue text to be identified according to the actual application scenario.

[0072] In a possible implementation, the actual application scenario is an instant dialogue scenario, and the computer device obtains the dialogue text to be identified in real time. The instant dialogue scenario refers to a scenario in which the content input by the user needs to be responded in time, such as registration, ordering, real-time retrieval, etc. Optionally, in this case, when it is detected that the user provides new input content, the computer device obtains the content input by the user from itself or other devices in real time, and generates the dialogue text to be identified according to the content.

[0073] In another possible implementation, the actual application scenario is a non-immediate conversation scenario, and the computer device acquires the above-mentioned to-be-recognized conversation text at a certain time interval. The non-immediate conversation scenario refers to a scenario that does not require an immediate response to the content input by the user, such as disease query, path planning, and the like. Optionally, in this case, the computer acquires the target content input by the user from itself or another device at a certain time interval, and generates the above-mentioned to-be-recognized conversation text according to the target content. The target content can include multiple contents input by a single user, and different contents input by multiple users, which are not limited in the embodiments of the present application. It should be noted that the time interval can be any value, such as 1s, 1min, 1h, 1 day, or 1 month, and the computer device can flexibly set and adjust the time interval according to the actual situation, which is not limited in the embodiments of the present application.

[0074] Of course, in actual application, for a conversation of a user, the computer device can use different to-be-recognized conversation text acquisition manners according to different situations. Exemplarily, in the medical field, in the disease query stage, the demand of the user is to determine the corresponding disease according to the disease description information provided by the user, at this time, the immediacy requirement of the conversation scenario is not high, after the disease description information provided by the user is acquired, the computer device acquires the disease description information at a certain time interval, and determines the to-be-recognized conversation text according to the disease description information, and then the computer device determines the user intention, and determines the disease recognition result by the disease query unit through the disease description information. After the computer device provides the disease description result to the user, the conversation scenario changes to an immediate response scenario, and the intelligent customer service immediately responds to the follow-up question information of the user, at this time, the computer device acquires the content input by the user in real time, and determines the to-be-recognized conversation text according to the content.

[0075] In step 102, feature extraction processing is performed on the conversation text to obtain feature information of the conversation text.

[0076] In the embodiments of the present application, after the computer device acquires the above-mentioned to-be-recognized conversation text, the computer device performs feature extraction processing on the conversation text to obtain feature information of the conversation text. The feature information includes multiple discrete feature variables.

[0077] Optionally, after the computer device acquires the conversation text, the computer device performs feature extraction processing on the conversation text to acquire the features contained in the conversation text, and generates feature information composed of the features. At this time, the features are independent and discrete in the feature information, that is, the features are discrete feature variables.

[0078] In a possible implementation, the computer device obtains the feature information through text preprocessing. Optionally, after obtaining the dialogue text, the computer device performs word segmentation processing on the dialogue text, obtains each entity in the dialogue text, and then selects important entities as keywords according to the importance of each entity, and further determines the features contained in the dialogue text according to the keywords. The importance of the entity is obtained according to the actual meaning of the entity and the use of the entity in big data.

[0079] In another possible implementation, the computer device obtains the feature information through a feature extraction network. Optionally, after obtaining the dialogue text, the computer device inputs the dialogue text into the feature extraction network, and then obtains the feature information output by the feature extraction network.

[0080] In step 103, the feature information is classified and encoded in multiple dimensions, and multiple discrete feature variables in the feature information are converted into continuous feature variables to obtain a multi-dimensional feature vector of the feature information.

[0081] In the embodiment of the application, after obtaining the feature information, the computer device classifies and encodes the feature information in multiple dimensions, converts multiple discrete feature variables in the feature information into continuous feature variables, and obtains a multi-dimensional feature vector of the feature information.

[0082] The multi-dimensional feature vector includes multiple continuous feature variables of different dimensions. Optionally, in the embodiment of the application, the multi-dimensional feature vector includes a domain feature vector, an action feature vector, and an entity noun feature vector.

[0083] Optionally, in the embodiment of the application, after obtaining the feature information, the computer device classifies and encodes the feature information in the domain dimension by using a domain encoder, and then obtains a domain feature vector output by the domain encoder. The domain feature vector is used to indicate the domain to which the dialogue text belongs. For example, the domain can be a large domain such as medical treatment, transportation, and education, or the domain can be a small domain such as a cold, a fever, bus travel, walking, an examination, and borrowing a book, and the application is not limited in this regard.

[0084] Optionally, in the embodiment of the present application, after obtaining the feature information, the computer device classifies and encodes the feature information from the action dimension by using an action encoding classifier, and then obtains an action feature vector output by the action encoder. The action feature vector is used to indicate an action to be performed by an inputter (i.e., the user) of the dialogue text. For example, the action can be an action directly contained in the dialogue text, such as "read" in the dialogue text "read a book"; or the action can also be an action indirectly contained in the dialogue text, such as "read, borrow, return, etc." that can be contained in the dialogue text "book". Of course, in this case, the subsequent obtained intention recognition result for the dialogue text can include multiple intention recognition results, and the multiple intention recognition results are displayed to the user, and the user selects the real intention according to the actual situation.

[0085] Optionally, in the embodiment of the present application, after obtaining the feature information, the computer device classifies and encodes the feature information from the entity noun dimension by using an entity noun encoding classifier, and then obtains an entity noun feature vector output by the entity noun encoder. The entity noun feature vector is used to indicate an entity noun associated with the domain and the action. The number of the entity noun can be one or more, which is not limited in the embodiment of the present application. For example, the entity noun can be an entity noun directly contained in the dialogue text, such as "read" in the dialogue text "read a book"; or the entity noun can also be an entity noun indirectly contained in the dialogue text, such as "book, TV series, video, etc." that can be contained in the dialogue text "read". Of course, in this case, the subsequent obtained intention recognition result for the dialogue text can include multiple intention recognition results, and the multiple intention recognition results are displayed to the user, and the user selects the real intention according to the actual situation.

[0086] In step 104, the multi-dimensional feature vectors are classified respectively to obtain multi-dimensional classification results of the feature information.

[0087] In the embodiment of the present application, after obtaining the multi-dimensional feature vectors, the computer device classifies the multi-dimensional feature vectors respectively to obtain multi-dimensional classification results of the feature information. The multi-dimensional classification results include classification results of multiple different dimensions.

[0088] Optionally, corresponding to the multi-dimensional feature vectors in the above, the multi-dimensional classification results include domain classification results, action classification results, and entity noun classification results.

[0089] In step 105, the multi-dimension classification result is fused and recognized to determine the intention recognition result of the dialogue text.

[0090] In the embodiment of the present application, after the computer device obtains the multi-dimension classification result, the computer device fuses and recognizes the multi-dimension classification result to determine the intention recognition result of the dialogue text. The intention recognition result is used to indicate the dialogue intention of the user for the dialogue text.

[0091] Optionally, in the case that the multi-dimension classification result includes the domain classification result, the action classification result and the entity noun classification result, the computer device superimposes and recognizes the domain classification result, the action classification result and the entity noun classification result, and then determines the intention recognition result of the dialogue text according to the superimposed recognition result.

[0092] In summary, in the technical scheme provided by the embodiment of the present application, the multi-dimension classification result of the dialogue text is determined through multi-dimension classification coding and processing of the dialogue text, and then the multi-dimension classification result is fused and recognized to determine the intention recognition result of the dialogue text. The intention recognition result is determined through analysis and processing from multiple dimensions, which can improve the accuracy of the intention recognition result. Moreover, when the classification coding is performed, the multiple discrete feature variables are converted into multi-dimension continuous feature variables, which can effectively reduce the information loss in the subsequent feature processing process, and can improve the accuracy and reliability of the intention recognition result.

[0093] Next, the acquisition method of the target dimension classification result is introduced from the perspective of the target dimension. The target dimension can be any dimension, such as the domain dimension, the action dimension, the entity noun dimension, etc. The embodiment of the present application does not limit this.

[0094] In the exemplary embodiment, the step 104 includes the following steps:

[0095] 1. For the target dimension, a mean value calculation network is used to process the target dimension feature vector to obtain a mean vector of the target dimension feature vector;

[0096] 2. A variance calculation network is used to process the target dimension feature vector to obtain a variance vector of the target dimension feature vector;

[0097] 3. The variance vector is subjected to noise superposition processing by a target noise to obtain a processed variance vector;

[0098] 4. Based on the mean vector and the processed variance vector, a classification feature vector corresponding to the target dimension feature vector is determined;

[0099] 5. classifying the to-be-classified feature vector to obtain a target dimension classification result of the feature information.

[0100] The mean vector is used to indicate the distribution mean of the target dimension feature vector in the space. The variance vector is used to indicate the distribution variance of the target dimension feature vector in the space. In the angle of the target dimension, in the embodiment of the present application, after obtaining the target dimension feature vector, the computer device processes the target dimension feature vector by using a mean computer network, obtains the mean vector of the target dimension feature vector, and processes the target dimension feature vector by using a variance computer network, obtains the variance vector of the target dimension feature vector.

[0101] Then, the computer device performs noise superposition processing on the variance vector by using a target noise to obtain a processed variance vector, further determines the to-be-classified feature vector corresponding to the target dimension vector based on the mean vector and the processed variance vector, and classifies the to-be-classified feature vector to obtain a target dimension classification result of the feature information. The target noise can be any noise satisfying a normal distribution, such as Gaussian noise.

[0102] Optionally, the computer device classifies the to-be-classified feature vector by using different classifiers of different dimensions for the to-be-classified feature vectors of different dimensions. For example, for the domain dimension, the computer device classifies the to-be-classified feature vector of the domain dimension by using a domain classifier to obtain a domain classification result of the feature vector; for the action dimension, the computer device classifies the to-be-classified feature vector of the action dimension by using an action classifier to obtain an action classification result of the feature vector; and for the entity noun dimension, the computer device classifies the to-be-classified feature vector of the entity noun dimension by using an entity noun classifier to obtain an entity noun classification result of the feature vector.

[0103] Please refer to Figure 2 which shows a flowchart of an intent recognition method of dialogue information provided by another embodiment of the present application. The method can include the following steps (201-208):

[0104] Step 201, obtaining dialogue text to be recognized.

[0105] Step 202, performing feature extraction processing on the dialogue text to obtain feature information of the dialogue text.

[0106] Step 203, performing multi-dimensional classification coding on the feature information to convert multiple discrete feature variables in the feature information into continuous feature variables to obtain a multi-dimensional feature vector of the feature information.

[0107] In step 204, the multi-dimensional feature vectors are classified respectively to obtain multi-dimensional classification results of the feature information.

[0108] In step 205, the multi-dimensional classification results are fused and recognized to determine an intention recognition result of the dialogue text.

[0109] The steps 201-205 are the same as the steps 101-105 in the embodiment, and details are described in the steps 101-105 in the embodiment, which are not described here. Figure 1 The steps 201-205 are the same as the steps 101-105 in the embodiment, and details are described in the steps 101-105 in the embodiment, which are not described here. Figure 1 The steps 201-205 are the same as the steps 101-105 in the embodiment, and details are described in the steps 101-105 in the embodiment, which are not described here.

[0110] In step 206, according to the intention recognition result, a reply text associated with the intention recognition result is obtained from a dialogue text library as a candidate reply text.

[0111] In the embodiment of the present application, after the computer device obtains the above-mentioned intention recognition result, according to the intention recognition result, a reply text associated with the intention recognition result is obtained from a dialogue text library as a candidate reply text. The reply text is a reply text for the above-mentioned dialogue text. Optionally, after obtaining the reply text, the computer device can display the reply text to the user, and then the user inputs a new dialogue text according to the reply text in combination with the actual situation.

[0112] Optionally, the dialogue text library stores reply texts corresponding to each intention recognition result. In the embodiment of the present application, since the intention recognition result is determined according to the superposition of the domain classification result, the action classification result and the entity noun classification result, correspondingly, the reply texts of the intention recognition result are stored in the dialogue text library in the same dimension.

[0113] Exemplarily, the dialogue text library stores the response text corresponding to the target domain, the response text corresponding to the target domain includes the response text directly aiming at the target domain (used when the intention recognition result cannot include the accurate action and entity noun), and in the response text corresponding to the target domain, the response text corresponding to each action is included (used when the intention recognition result accurately indicates the domain and the action but cannot accurately indicate the entity noun), in addition, for the response text corresponding to the target action, the response text corresponding to each entity noun is included (used when the intention recognition result accurately indicates the domain, the action and the entity noun); or the dialogue text library stores the response text corresponding to the target action, the response text corresponding to the target action includes the response text directly aiming at the target action (used when the intention recognition result cannot include the accurate domain and entity noun), and in the response text corresponding to the target action, the response text corresponding to each domain is included (used when the intention recognition result accurately indicates the domain and the action but cannot accurately indicate the entity noun), in addition, for the response text corresponding to the target domain, the response text corresponding to each entity noun is included (used when the intention recognition result accurately indicates the domain, the action and the entity noun); or the dialogue text library stores the response text corresponding to the target entity noun, the response text corresponding to the target entity noun includes the response text directly aiming at the target entity noun (used when the intention recognition result cannot include the accurate domain and action), and in the response text corresponding to the target entity noun, the response text corresponding to each domain is included (used when the intention recognition result accurately indicates the domain and the entity noun but cannot accurately indicate the action), in addition, for the response text corresponding to the target domain, the response text corresponding to each action is included (used when the intention recognition result accurately indicates the domain, the action and the entity noun). Optionally, the computer device can mark each response text in the form of "domain-action-entity noun" to realize the quick search of the response text, wherein if the response text does not exist corresponding domain, "empty" is used to represent the domain, and the action and the entity noun are similar to the domain.

[0114] In step 207, the quality evaluation index of each candidate response text is acquired when the number of candidate response texts is not unique.

[0115] In the embodiment of the present application, after the computer device obtains the candidate response text, the computer device determines the processing mode for the candidate response text based on the number of the candidate response text. If the number of the candidate response text is unique, the computer device directly takes the candidate response text as the response text corresponding to the dialogue text. If the number of the candidate response text is not unique, the computer device obtains the quality evaluation index of each candidate response text, and selects the response text corresponding to the dialogue text from the candidate response text based on the quality evaluation index.

[0116] The quality evaluation index is used to indicate the quality of the response text. Optionally, the quality evaluation index includes, but is not limited to, at least one of the following: the adoption rate of the response text, the amount of information contained in the response text, the number of uses of the response text, and the like. The adoption rate of the response text refers to the ratio between the number of accurate responses of the response text and the number of uses of the response text. The accurate response of the response text refers to the case that the user continues to successfully carry out the dialogue for the response text. For example, in the case that no negative keywords such as "wrong" and "incorrect" appear in the dialogue text of the user for the response text, it is determined that the user continues to successfully carry out the dialogue for the response text.

[0117] Step 208: From the candidate response text, the candidate response text whose quality evaluation index meets the condition is selected as the response text corresponding to the dialogue text.

[0118] In the embodiment of the present application, after the computer device obtains the quality evaluation index, the computer device selects the candidate response text whose quality evaluation index meets the condition from the candidate response text as the response text corresponding to the dialogue text. Optionally, the condition can be a condition for a single index in the quality evaluation index, or a condition for multiple indexes in the quality evaluation index. The computer device or the staff can flexibly set and adjust the condition according to the actual situation, and the embodiment of the present application does not limit this.

[0119] In summary, in the technical scheme provided by the embodiment of the present application, the dialogue text corresponding response text is obtained through the intent recognition result, which ensures the smooth progress of the automatic dialogue in the intelligent service, and the response text meeting the condition is selected according to the quality evaluation index of each candidate response text, which ensures the reliability of the response text.

[0120] For reference Figure 3 which shows a flowchart of a training method of an intent recognition model provided by an embodiment of the present application. The method can include the following steps (301-307):

[0121] Step 301: Obtain at least one training sample of the intent recognition model.

[0122] The intent recognition model refers to a deep learning model for determining an intent recognition result of a dialogue text based on feature information of the dialogue text. In the embodiment of the present application, the computer device obtains at least one training sample of the intent recognition model before training the intent recognition model. The training sample includes a sample dialogue text and label information corresponding to the sample dialogue text. Optionally, the computer device obtains the training sample based on historical data of the automatic answering system, or the computer device searches for and obtains the training sample from a network environment.

[0123] In step 302, feature extraction processing is performed on the sample dialogue text to obtain feature information of the sample dialogue text.

[0124] In the embodiment of the present application, after obtaining the sample dialogue text, the computer device performs feature extraction processing on the sample dialogue text to obtain feature information of the sample dialogue text. The feature information includes a plurality of discrete feature variables.

[0125] Optionally, the computer device performs feature extraction processing on the sample dialogue text through a feature extraction layer in the intent recognition model, and then obtains feature information output by the feature extraction network.

[0126] In step 303, multi-dimensional classification coding is performed on the feature information to convert the plurality of discrete feature variables in the feature information into continuous feature variables, and a multi-dimensional feature vector of the feature information is obtained.

[0127] In the embodiment of the present application, after obtaining the feature information, the computer device performs multi-dimensional classification coding on the feature information to convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtains a multi-dimensional feature vector of the feature information. The multi-dimensional feature vector includes a plurality of continuous feature variables of different dimensions.

[0128] Optionally, the computer device respectively performs classification coding on the feature information from different dimensions through different-dimensional encoders in the intent recognition model, and then respectively obtains a plurality of continuous feature variables of different dimensions output by the different-dimensional encoders.

[0129] In step 304, classification processing is performed on the multi-dimensional feature vector to obtain a multi-dimensional classification result of the feature information.

[0130] In the embodiment of the present application, after obtaining the multi-dimensional feature vector, the computer device performs classification processing on the multi-dimensional feature vector to obtain a multi-dimensional classification result of the feature information. The multi-dimensional classification result includes a plurality of classification results of different dimensions.

[0131] Optionally, when obtaining the multi-dimensional classification result, the computer device processes the plurality of dimensional feature vectors respectively through a mean value calculation network in the intent recognition model, obtains a plurality of mean value vectors respectively corresponding to the plurality of dimensional feature vectors, and processes the plurality of dimensional feature vectors respectively through a variance calculation network in the intent recognition model, to obtain a plurality of variance vectors respectively corresponding to the plurality of dimensional feature vectors. Then, the computer device superimposes target noise based on the plurality of mean value vectors and the plurality of variance vectors respectively corresponding to the plurality of dimensional feature vectors through a noise superimposition network in the intent recognition model, to determine a plurality of to-be-classified feature vectors respectively corresponding to the plurality of dimensional feature vectors, and then processes the to-be-classified feature vectors of different dimensions respectively through different-dimensional classifiers in the intent recognition model, to obtain the multi-dimensional classification result.

[0132] In step 305, the multi-dimensional classification result is fused and recognized to determine the intent recognition result of the sample dialogue text.

[0133] In the embodiment of the present application, after obtaining the multi-dimensional classification result, the computer device fuses and recognizes the multi-dimensional classification result to determine the intent recognition result of the sample dialogue text.

[0134] Optionally, the computer device superimposes and recognizes the multi-dimensional classification result through a result fusion layer in the intent recognition model to determine the intent recognition result of the sample dialogue text.

[0135] In step 306, a loss function value of the intent recognition model is calculated based on the multi-dimensional classification result, the intent recognition result, and label information corresponding to the sample dialogue text.

[0136] In the embodiment of the present application, the computer device calculates a loss function value of the intent recognition model based on the multi-dimensional classification result, the intent recognition result, and label information corresponding to the sample dialogue text. The loss function is used to measure the reliability of the intent recognition model.

[0137] In step 307, the parameters of the intent recognition model are adjusted based on the loss function value.

[0138] In the embodiment of the present application, after obtaining the loss function value, the computer device adjusts the parameters of the intent recognition model based on the loss function value, and continues to train the intent recognition model with the adjusted parameters using the training sample until the loss function value converges.

[0139] To sum up, in the technical scheme provided by the embodiments of the present application, the intent recognition model is trained by using training samples. In the training process, the multiple discrete feature variables are converted into continuous feature variables of multiple dimensions when classification coding is performed, which can effectively reduce the information loss of the intent recognition model in use and improve the accuracy of the intent recognition result.

[0140] Optionally, in the embodiments of the present application, the above-mentioned multi-dimensional classification result includes a domain classification result, an action classification result, and an entity noun classification result, and the above-mentioned label information includes a domain classification label, an action classification label, and an entity noun classification label corresponding to the sample dialogue text, and an intent recognition label corresponding to the sample dialogue text. The label is the accurate result corresponding to the intent recognition model. Optionally, the computer device splits the intent recognition label based on a preset rule to obtain the domain classification label, the action classification label, and the entity noun classification label. Next, the method of obtaining the loss function value is introduced.

[0141] In the example embodiments, the step 306 includes the following steps:

[0142] 1. Based on the domain classification result, the action classification result, and the entity noun classification result, and in combination with the domain classification label, the action classification label, and the entity noun classification label, a first loss function value corresponding to the intent recognition model is determined;

[0143] 2. Based on the intent recognition result and the intent recognition label, a second loss function value corresponding to the intent recognition model is determined;

[0144] 3. Based on the first loss function value and the second loss function value, a loss function value of the intent recognition model is determined.

[0145] The first loss function is used to measure the accuracy of the multi-dimensional classification result of the intent recognition model. The second loss function is used to measure the accuracy of the intent recognition result of the intent recognition model. In the embodiments of the present application, in order to ensure the reliability of each classifier in the intent recognition model, the computer device determines a first loss function value corresponding to the intent recognition model based on the domain classification result, the action classification result, and the entity noun classification result, and in combination with the domain classification label, the action classification label, and the entity noun classification label. In order to ensure the accuracy of the intent recognition result of the intent recognition model, the computer device determines a second loss function value corresponding to the intent recognition model based on the intent recognition result and the intent recognition label. Then, the computer device determines a loss function value of the intent recognition model according to the first loss function value and the second loss function value.

[0146] Optionally, in the acquisition of the first loss function value, based on the domain classification result, the domain classification label, the action classification label and the entity noun classification label, a loss function value corresponding to a domain classifier in the intent recognition model is determined; based on the action classification result, the domain classification label, the action classification label and the entity noun classification label, a loss function value corresponding to an action classifier in the intent recognition model is determined; based on the entity noun classification result, the domain classification label, the action classification label and the entity noun classification label, a loss function value corresponding to an entity noun classifier in the intent recognition model is determined; and then the first loss function value is determined according to the loss function value corresponding to the domain classifier, the loss function value corresponding to the action classifier and the loss function value corresponding to the entity noun classifier. Wherein, the computer device can add the loss function value corresponding to the domain classifier, the loss function value corresponding to the action classifier and the loss function value corresponding to the entity noun classifier to obtain the first loss function value.

[0147] Exemplarily, assuming that s d is a feature vector to be classified in the domain dimension, s a is a feature vector to be classified in the action dimension, and s s is a feature vector to be classified in the entity noun dimension, the intent recognition result of the intent recognition model is:

[0148]

[0149] Wherein, Decoder represents a decoder, [s d , s a , s s ] = h + σ∈, ∈ ~ N(0, I), h represents a mean vector, σ represents a variance vector, and ∈ represents Gaussian noise conforming to a normal distribution;

[0150] The loss function L i enc of a certain classifier in the first loss function is:

[0151]

[0152] Wherein, W i represents parameters of the classifier, s i represents an intent recognition label, a i represents a classifier output result, A i represents a classification label of the classifier, d represents a domain, a represents an action, and s represents an entity noun.

[0153] The second loss function L dec is:

[0154] ​

[0155] wherein |S| represents the training sample, n represents the sample quantity of the training sample, S n represents the intent recognition label, represents the intent recognition result.

[0156] Next, the structure of the intent recognition model is introduced. Optionally, the intent recognition model comprises a feature extraction layer, a domain encoder, an action encoder, an entity noun encoder, a mean calculation network, a variance calculation network, a noise superposition network, a domain classifier, an action classifier, an entity noun classifier, and a result fusion layer. The feature extraction layer is configured to perform feature extraction processing on the dialogue text to obtain feature information corresponding to the dialogue text, wherein the feature information comprises a plurality of discrete feature variables. The domain encoder is configured to perform domain classification coding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a domain feature vector of the feature information, wherein the domain feature vector is a continuous feature variable. The action encoder is configured to perform action classification coding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain an action feature vector of the feature information, wherein the action feature vector is a continuous feature variable. The entity noun encoder is configured to perform entity noun classification coding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain an entity noun feature vector of the feature information, wherein the entity noun feature vector is a continuous feature variable. The mean calculation network is configured to process a plurality of dimension feature vectors respectively, and obtain a mean vector corresponding to each of the plurality of dimension feature vectors, wherein the mean vector is used to indicate the distribution mean of the dimension feature vector in space. The variance calculation network is configured to process a plurality of dimension feature vectors respectively, and obtain a variance vector corresponding to each of the plurality of dimension feature vectors, wherein the variance vector is used to indicate the distribution variance of the dimension feature vector in space. The noise superposition network is configured to superimpose a target noise based on the mean vector and the variance vector corresponding to each of the plurality of dimension feature vectors, and determine a to-be-classified feature vector corresponding to each of the plurality of dimension feature vectors. The domain classifier is configured to perform classification processing on the to-be-classified feature vector of the domain dimension, and obtain a domain dimension classification result of the feature vector. The action classifier is configured to perform classification processing on the to-be-classified feature vector of the action dimension, and obtain an action dimension classification result of the feature vector. The entity noun classifier is configured to perform classification processing on the to-be-classified feature vector of the entity noun dimension, and obtain an entity noun dimension classification result of the feature vector. The result fusion layer is configured to superimpose and recognize the domain classification result, the action classification result, and the entity noun classification result, and determine the intent recognition result of the dialogue text.

[0157] Exemplarily, in combination with reference to Figure 4An intent recognition method of an intent recognition model is introduced. A computer device inputs dialogue text to the intent recognition model, and a feature extraction layer in the intent recognition model performs feature extraction processing on the dialogue text to obtain feature information of the dialogue text. Further, a domain encoder performs domain classification coding on the feature information to obtain a domain feature vector h d ; an action encoder performs action classification coding on the feature information to obtain an action feature vector h a ; and an entity noun encoder performs entity noun classification coding on the feature information to obtain an entity noun feature vector h s . Then, a mean calculation network and a variance calculation network are used to obtain a mean vector and a variance vector corresponding to each dimension feature vector respectively, and a noise superposition network is used to obtain a domain dimension feature vector S d , an action dimension feature vector S a , and an entity noun dimension feature vector S s . Then, a domain classifier is used to classify the feature vector S d to determine a domain classification result, an action classifier is used to classify the feature vector S a to determine an action classification result, and an entity noun classifier is used to classify the feature vector S s to determine an entity noun classification result. Then, the domain classification result, the action classification result, and the entity noun classification result are superimposed and fused, and a decoder is used to decode the superimposed and fused result to obtain an intent recognition result of the dialogue text.

[0158] In addition, the improvement of the intent recognition model in the present application is introduced in combination with the VAE (Variational Autoencoder) model in the related art. As shown in Figure 5 , in the VAE model, the input has only six dimensions (X1, X2, X3, X4, X5, X6), each dimension is composed of 0 or 1, then two neural networks are used to calculate the mean and variance corresponding to each dimension respectively, and then a standard high noise is introduced to form new sampling variables (Z1, Z2, Z3, Z4, Z5, Z6), and finally a generator is used to transform the sampling variables into generated samples

[0159] However, the intent recognition model in the present application converts discrete feature variables into continuous feature variables for processing, which effectively reduces the loss of information in the VAE model.

[0160] Moreover, the following Table 1 and Table 2 are data obtained through experiments:

[0161] Table 1 Comparison of information loss between VAE model and intent recognition model in the present application

[0162] Model Information loss VAE 0.1 Intention recognition model of the present application 0.02

[0163] Table 2 Comparison of convergence number between VAE model and intent recognition model in the present application

[0164] Model Converged data VAE 100 Intention recognition model of the present application 20

[0165] From the above, the information loss of the intent recognition model in the present application is lower than that of the VAE model in the data processing process, which can effectively ensure the accuracy of the intent recognition result, and the intent recognition model in the present application requires fewer training steps, thereby reducing the cost required for model training to a certain extent.

[0166] Next, the intent recognition method in the present application will be introduced in combination with reference to Figure 6 a medical application scenario. Specifically, the following steps are included:

[0167] Step 601, obtaining a dialogue text to be recognized, which is input by a user through an automatic answering type medical consultation system.

[0168] Step 602, inputting the dialogue text into the intent recognition model to obtain an intent recognition result output by the intent recognition model.

[0169] Optionally, the intent recognition model includes a feature extraction layer, a domain encoder, an action encoder, an entity noun encoder, a mean calculation network, a variance calculation network, a noise superposition network, a domain classifier, an action classifier, an entity noun classifier, and a result fusion layer.

[0170] The feature extraction layer performs feature extraction processing on the dialogue text to obtain feature information corresponding to the dialogue text, and the feature information includes a plurality of discrete feature variables.

[0171] The domain encoder performs domain classification encoding on the feature information, converts the plurality of discrete feature variables in the feature information into continuous feature variables, obtains a domain feature vector of the feature information, and the domain feature vector is a continuous feature variable; the action encoder performs action classification encoding on the feature information, converts the plurality of discrete feature variables in the feature information into continuous feature variables, obtains an action feature vector of the feature information, and the action feature vector is a continuous feature variable; and the entity noun encoder is used for performing entity noun classification encoding on the feature information, converting the plurality of discrete feature variables in the feature information into continuous feature variables, obtaining an entity noun feature vector of the feature information, and the entity noun feature vector is a continuous feature variable.

[0172] The mean value calculation network respectively processes the plurality of dimension feature vectors to obtain a plurality of mean value vectors respectively corresponding to the plurality of dimension feature vectors, and the mean value vectors are used to indicate distribution means of the dimension feature vectors in space.

[0173] The noise superposition network superposes target noise based on the mean value vectors and the variance vectors respectively corresponding to the plurality of dimension feature vectors to determine a plurality of to-be-classified feature vectors respectively corresponding to the plurality of dimension feature vectors.

[0174] The domain classifier classifies the to-be-classified feature vectors of the domain dimension to obtain a domain dimension classification result of the feature vectors, the action classifier classifies the to-be-classified feature vectors of the action dimension to obtain an action dimension classification result of the feature vectors, and the entity noun classifier classifies the to-be-classified feature vectors of the entity noun dimension to obtain an entity noun dimension classification result of the feature vectors.

[0175] The result fusion layer superimposes and identifies the domain classification result, the action classification result, and the entity noun classification result to determine an intent recognition result of the dialogue text.

[0176] In step 603, whether the dialogue intent of the user is a specific disease consultation is determined according to the intent recognition result. If the dialogue intent of the user is a specific disease consultation, step 604 is performed; if the dialogue intent of the user is not a specific disease consultation, steps 605-608 are performed.

[0177] In step 604, the user is allocated a manual consultation service.

[0178] In step 605, a response text associated with the intent recognition result is obtained from a dialogue text library as a candidate response text.

[0179] In step 606, in the case where the number of candidate response texts is not unique, a quality evaluation index of each candidate response text is obtained.

[0180] In step 607, a candidate response text whose quality evaluation index satisfies a condition is selected from the candidate response texts as a response text corresponding to the dialogue text.

[0181] In step 608, the user is shown the response text, and a new dialogue text to be recognized is obtained, and the steps are repeated from step 601 until the dialogue ends.

[0182] The following is an embodiment of the device of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0183] Reference is made to Figure 7 which shows a block diagram of an apparatus for intent recognition of dialogue information according to an embodiment of the present application. The apparatus has the functions of implementing the method for intent recognition of dialogue information described above, which can be implemented by hardware, or by executing corresponding software by hardware. The apparatus can be a computer device, or can be arranged in a computer device. The apparatus 700 can include a text acquisition module 710, a feature extraction module 720, a feature encoding module 730, a feature classification module 740, and a result fusion module 750.

[0184] The text acquisition module 710 is configured to acquire dialogue text to be recognized.

[0185] The feature extraction module 720 is configured to perform feature extraction processing on the dialogue text to obtain feature information of the dialogue text, wherein the feature information includes a plurality of discrete feature variables.

[0186] The feature encoding module 730 is configured to perform multi-dimensional classification coding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimensional feature vector of the feature information, wherein the multi-dimensional feature vector includes a plurality of continuous feature variables of different dimensions.

[0187] The feature classification module 740 is configured to perform classification processing on the multi-dimensional feature vector respectively, obtain multi-dimensional classification results of the feature information, and the multi-dimensional classification results include a plurality of classification results of different dimensions.

[0188] The result fusion module 750 is configured to perform fusion recognition processing on the multi-dimensional classification results, and determine an intent recognition result of the dialogue text.

[0189] In an exemplary embodiment, the multi-dimensional feature vector includes a domain feature vector, an action feature vector, and an entity noun feature vector; wherein the domain feature vector is used to indicate a domain to which the dialogue text belongs, the action feature vector is used to indicate an action to be performed by an inputter of the dialogue text, and the entity noun feature vector is used to indicate an entity noun associated with the domain and the action.

[0190] In an example embodiment, the feature classification module 740 is configured to, for a target dimension, process a target dimension feature vector by using a mean value calculation network to obtain a mean value vector of the target dimension feature vector, wherein the mean value vector is used to indicate a distribution mean value of the target dimension feature vector in a space; process the target dimension feature vector by using a variance calculation network to obtain a variance vector of the target dimension feature vector, wherein the variance vector is used to indicate a distribution variance of the target dimension feature vector in the space; perform noise superposition processing on the variance vector by using a target noise to obtain a processed variance vector; determine a to-be-classified feature vector corresponding to the target dimension feature vector based on the mean value vector and the processed variance vector; and perform classification processing on the to-be-classified feature vector to obtain a target dimension classification result of the feature information.

[0191] In an example embodiment, as shown in Figure 8 The device 700 further includes a candidate obtaining module 760, an index obtaining module 770, and a text selection module 780.

[0192] The candidate obtaining module 760 is configured to obtain, according to the intent recognition result, a reply text associated with the intent recognition result from a dialogue text library as a candidate reply text.

[0193] The index obtaining module 770 is configured to, in a case where the number of candidate reply texts is not unique, obtain a quality evaluation index of each candidate reply text.

[0194] The text selection module 780 is configured to select, from the candidate reply texts, a candidate reply text that satisfies a quality evaluation index condition as a reply text corresponding to the dialogue text.

[0195] In summary, in the technical scheme provided by the embodiments of the present application, the multi-dimensional classification coding and multi-maintenance classification processing of dialogue texts are performed to determine multi-dimensional classification results of the dialogue texts, and then the multi-dimensional classification results are fused and recognized to determine an intent recognition result of the dialogue text, so that the accuracy of the intent recognition result can be improved. Moreover, when the classification coding is performed, multiple discrete feature variables are converted into multi-dimensional continuous feature variables, which can effectively reduce the information loss in the subsequent feature processing process and improve the accuracy and reliability of the intent recognition result.

[0196] For reference Figure 9FIG. 9 is a block diagram illustrating a training device of an intent recognition model according to an embodiment of the present application. The device has the function of implementing the training method of the intent recognition model described above, which can be implemented by hardware or by corresponding software executed by hardware. The device can be a computer device or can be arranged in a computer device. The device 900 can include a sample obtaining module 910, a feature obtaining module 920, a vector obtaining module 930, a vector classification module 940, a result obtaining module 950, a function value determining module 960, and a parameter adjusting module 970.

[0197] The sample obtaining module 910 is configured to obtain at least one training sample of the intent recognition model, the training sample including sample dialogue text and label information corresponding to the sample dialogue text.

[0198] The feature obtaining module 920 is configured to perform feature extraction processing on the sample dialogue text to obtain feature information of the sample dialogue text, wherein the feature information includes a plurality of discrete feature variables.

[0199] The vector obtaining module 930 is configured to perform multi-dimensional classification coding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimensional feature vector of the feature information, wherein the multi-dimensional feature vector includes a plurality of feature vectors of different dimensions, and the feature vectors are continuous feature variables.

[0200] The vector classification module 940 is configured to perform classification processing on the multi-dimensional feature vector respectively to obtain multi-dimensional classification results of the feature information, the multi-dimensional classification results including a plurality of classification results of different dimensions.

[0201] The result obtaining module 950 is configured to perform fusion recognition processing on the multi-dimensional classification results to determine an intent recognition result of the dialogue text.

[0202] The function value determining module 960 is configured to calculate a loss function value of the intent recognition model based on the multi-dimensional classification results, the intent recognition result, and the label information corresponding to the sample dialogue text.

[0203] The parameter adjusting module 970 is configured to adjust parameters of the intent recognition model of the dialogue information based on the loss function value.

[0204] In an exemplary embodiment, the multi-dimensional classification results include a domain classification result, an action classification result, and an entity noun classification result; and the label information includes a domain classification label, an action classification label, and an entity noun classification label corresponding to the sample dialogue text, and an intent recognition label corresponding to the sample dialogue text.

[0205] In an example embodiment, the function value determination module 960 comprises a first obtaining unit, a second obtaining unit and a function obtaining unit.

[0206] The first obtaining unit is configured to determine a first loss function value corresponding to the intent recognition model based on the domain classification result, the action classification result and the entity noun classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label, wherein the first loss function is used to measure the accuracy of the multi-dimensional classification result of the intent recognition model.

[0207] The second obtaining unit is configured to determine a second loss function value corresponding to the intent recognition model based on the intent recognition result and the intent recognition label, wherein the second loss function is used to measure the accuracy of the intent recognition result of the intent recognition model.

[0208] The function obtaining unit is configured to determine the loss function value of the intent recognition model based on the first loss function value and the second loss function value.

[0209] In an example embodiment, the first obtaining unit is configured to determine a loss function value corresponding to a domain classifier in the intent recognition model based on the domain classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label; determine a loss function value corresponding to an action classifier in the intent recognition model based on the action classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label; determine a loss function value corresponding to an entity noun classifier in the intent recognition model based on the entity noun classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label; and determine the first loss function value according to the loss function value corresponding to the domain classifier, the loss function value corresponding to the action classifier and the loss function value corresponding to the entity noun classifier.

[0210] In an example embodiment, as shown in Figure 10 The apparatus 900 further comprises a label obtaining module 980.

[0211] The label obtaining module 980 is configured to perform splitting processing on the intent recognition label based on a preset rule, and obtain the domain classification label, the action classification label and the entity noun classification label.

[0212] In an example embodiment, the intention recognition model comprises: a feature extraction layer, a domain encoder, an action encoder, an entity noun encoder, a mean value calculation network, a variance calculation network, a noise superposition network, a domain classifier, an action classifier, an entity noun classifier, and a result fusion layer.

[0213] The feature extraction layer is configured to perform feature extraction processing on the dialogue text to obtain feature information corresponding to the dialogue text.

[0214] The domain encoder is configured to perform domain classification encoding on the feature information, convert the multiple discrete feature variables in the feature information into continuous feature variables, and obtain a domain feature vector of the feature information; wherein the domain feature vector is a continuous feature variable.

[0215] The action encoder is configured to perform action classification encoding on the feature information, convert the multiple discrete feature variables in the feature information into continuous feature variables, and obtain an action feature vector of the feature information; wherein the action feature vector is a continuous feature variable.

[0216] The entity noun encoder is configured to perform entity noun classification encoding on the feature information, convert the multiple discrete feature variables in the feature information into continuous feature variables, and obtain an entity noun feature vector of the feature information; wherein the entity noun feature vector is a continuous feature variable.

[0217] The mean value calculation network is configured to process multiple dimension feature vectors respectively to obtain a mean value vector corresponding to each of the multiple dimension feature vectors; wherein the mean value vector is used to indicate the distribution mean of the dimension feature vector in space.

[0218] The variance calculation network is configured to process multiple dimension feature vectors respectively to obtain a variance vector corresponding to each of the multiple dimension feature vectors; wherein the variance vector is used to indicate the distribution variance of the dimension feature vector in space.

[0219] The noise superposition network is configured to superimpose target noise based on the mean value vector and the variance vector corresponding to each of the multiple dimension feature vectors, to determine a to-be-classified feature vector corresponding to each of the multiple feature dimension vectors.

[0220] The domain classifier is configured to perform classification processing on the to-be-classified feature vector of the domain dimension to obtain a domain dimension classification result of the feature vector.

[0221] The action classifier is configured to perform classification processing on the to-be-classified feature vector of the action dimension to obtain an action dimension classification result of the feature vector.

[0222] The entity noun classifier is configured to perform classification processing on a feature vector of an entity noun dimension to be classified to obtain an entity noun dimension classification result of the feature vector.

[0223] The result fusion layer is configured to superimpose the domain classification result, the action classification result and the entity noun classification result to determine an intent recognition result of the dialogue text.

[0224] To sum up, in the technical scheme provided by the embodiments of the present application, the intent recognition model is trained by using training samples. In the training process, multiple discrete feature variables are converted into multi-dimensional continuous feature variables during classification coding, which can effectively reduce the information loss of the intent recognition model in the use process and improve the accuracy of the intent recognition result.

[0225] It should be noted that the device provided in the above embodiments, in realizing its functions, is only exemplified by the division of the above functional modules. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is described in detail in the method embodiments, which will not be repeated here.

[0226] Please refer to Figure 11 which shows a structural block diagram of a computer device provided by an embodiment of the present application. The computer device can be used to implement the functions of the dialogue information intent recognition method or the training method of the intent recognition model described above. Specifically:

[0227] The computer device 1100 includes a central processing unit (CPU) 1101, a system memory 1104 including a random access memory (RAM) 1102 and a read-only memory (ROM) 1103, and a system bus 1105 connecting the system memory 1104 and the central processing unit 1101. The computer device 1100 also includes a basic input / output system (Input / Output, I / O system) 1106 to help transfer information between various devices in the computer, and a mass storage device 1107 for storing an operating system 1113, application programs 1114 and other program modules 1115.

[0228] The basic input / output system 1106 includes the various components needed to display information and input information from a user, such as a display 1108 and input devices 1109, such as a mouse, keyboard, or electronic stylus. The display 1108 and input devices 1109 are connected to the central processing unit 1101 through an input / output controller 1110 that is connected to the system bus 1105. The basic input / output system 1106 can also include the input / output controller 1110 for receiving and processing input from a number of other devices, including a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1110 provides output to the display screen, a printer, or other type of output device.

[0229] The mass storage device 1107 is connected to the central processing unit 1101 through a mass storage controller (not shown) that is connected to the system bus 1105. The mass storage device 1107 and its associated computer-readable media provide nonvolatile storage for the computer device 1100. That is, the mass storage device 1107 can include a computer- readable medium (not shown) such as a hard disk or a CD-ROM drive.

[0230] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid state memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Computer storage media would not, however, include communication media including wired or wireless signaling media that communicate program code in a modulated data signal. The system memory 1104 and mass storage device 1107 described above can be embodied as a memory.

[0231] According to various embodiments of the present application, the computer device 1100 can also operate in connection with a remote computer through a network, such as the Internet. That is, the computer device 1100 can connect to the network 1112 through a network interface unit 1111 connected to the system bus 1105, or can connect to other types of networks or remote computer systems (not shown) using the network interface unit 1111.

[0232] The memory further includes a computer program stored in the memory and configured to be executed by one or more processors to implement the above-described intent recognition method for dialogue information or to implement the above-described training method for the intent recognition model.

[0233] In exemplary embodiments, a computer-readable storage medium is also provided, in which at least one instruction, at least one program, a code set, or an instruction set is stored, and the at least one instruction, the at least one program, the code set, or the instruction set, when executed by a processor, implements the above-described intent recognition method for dialogue information or the above-described training method for the intent recognition model.

[0234] Optionally, the computer-readable storage medium can include a ROM (Read Only Memory), a RAM (Random Access Memory), a SSD (Solid State Drives), or an optical disc, etc. Among them, the random access memory can include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0235] In exemplary embodiments, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-described intent recognition method for dialogue information or executes the above-described training method for the intent recognition model.

[0236] It should be understood that "multiple" mentioned herein refers to two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A existing alone, A and B existing together, and B existing alone. The character " / " generally represents that the associated objects before and after it are in an "or" relationship. In addition, the step numbers described herein only exemplarily show a possible execution order between steps. In some other embodiments, the above steps can also be executed in a sequence different from the number order, such as two steps with different numbers being executed at the same time, or two steps with different numbers being executed in an order opposite to the illustration, which is not limited in the embodiments of the present application.

[0237] The above only describes exemplary embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying an intent of a dialogue information, characterized by, The method comprises: acquiring dialogue text to be identified; performing feature extraction processing on the dialogue text to obtain feature information of the dialogue text; wherein the feature information includes multiple discrete feature variables; performing multi-dimensional classification coding on the feature information to convert the multiple discrete feature variables in the feature information into continuous feature variables to obtain a multi-dimensional feature vector of the feature information; wherein the multi-dimensional feature vector includes a domain feature vector, an action feature vector, and an entity noun feature vector; wherein the domain feature vector is used to indicate a domain to which the dialogue text belongs, the action feature vector is used to indicate an action to be performed by an inputter of the dialogue text, and the entity noun feature vector is used to indicate an entity noun associated with the domain and the action; performing classification processing on the multi-dimensional feature vector respectively to obtain multi-dimensional classification results of the feature information, the multi-dimensional classification results including classification results of the multiple different dimensions; performing fusion recognition processing on the multi-dimensional classification results to determine an intent recognition result of the dialogue text.

2. The method of claim 1, wherein, The classification processing on the multi-dimensional feature vector respectively to obtain multi-dimensional classification results of the feature information comprises: for a target dimension, performing processing on a target dimension feature vector by using a mean value calculation network to acquire a mean value vector of the target dimension feature vector; wherein the mean value vector is used to indicate a distribution mean value of the target dimension feature vector in space; performing processing on the target dimension feature vector by using a variance calculation network to acquire a variance vector of the target dimension feature vector; wherein the variance vector is used to indicate a distribution variance of the target dimension feature vector in space; performing noise superposition processing on the variance vector by using a target noise to obtain a processed variance vector; based on the mean value vector and the processed variance vector, determining a to-be-classified feature vector corresponding to the target dimension feature vector; performing classification processing on the to-be-classified feature vector to obtain a target dimension classification result of the feature information.

3. The method according to claim 1 or 2, characterized in that, After the fusion recognition processing on the multi-dimensional classification results to determine the intent recognition result of the dialogue text, the method further comprises: according to the intent recognition result, acquiring a response text associated with the intent recognition result from a dialogue text library as a candidate response text; in a case where the number of the candidate response texts is not unique, acquiring a quality evaluation index of each of the candidate response texts; from the candidate response texts, selecting a candidate response text whose quality evaluation index meets a condition as a response text corresponding to the dialogue text. 4.A method for training an intent recognition model, the method comprising: The method comprises: acquiring at least one training sample of an intent recognition model, the training sample including a sample dialogue text and label information corresponding to the sample dialogue text; performing feature extraction processing on the sample dialogue text to obtain feature information of the sample dialogue text; wherein the feature information includes multiple discrete feature variables; The feature information is multi-dimensionally classified and coded, and a plurality of discrete feature variables in the feature information are converted into continuous feature variables to obtain a multi-dimension feature vector of the feature information; wherein the multi-dimension feature vector includes a domain feature vector, an action feature vector and an entity noun feature vector; wherein the domain feature vector is used to indicate a domain to which the dialogue text belongs, the action feature vector is used to indicate an action to be performed by an inputter of the dialogue text, and the entity noun feature vector is used to indicate an entity noun associated with the domain and the action; The multi-dimension feature vector is classified respectively to obtain a multi-dimension classification result of the feature information, and the multi-dimension classification result includes classification results of the plurality of different dimensions; The multi-dimension classification result is fused and recognized to determine an intent recognition result of the sample dialogue text; Based on the multi-dimension classification result, the intent recognition result and label information corresponding to the sample dialogue text, a loss function value of the intent recognition model is calculated; Parameters of the intent recognition model are adjusted based on the loss function value.

5. The method of claim 4, wherein The multi-dimension classification result includes a domain classification result, an action classification result and an entity noun classification result; The label information includes a domain classification label, an action classification label and an entity noun classification label corresponding to the sample dialogue text, and an intent recognition label corresponding to the sample dialogue text.

6. The method of claim 5, wherein, The calculation of the loss function value of the intent recognition model based on the multi-dimension classification result, the intent recognition result and the label information corresponding to the sample dialogue text includes: Based on the domain classification result, the action classification result, the entity noun classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label, a first loss function value corresponding to the intent recognition model is determined; wherein the first loss function is used to measure the accuracy of the multi-dimension classification result of the intent recognition model; Based on the intent recognition result and the intent recognition label, a second loss function value corresponding to the intent recognition model is determined; wherein the second loss function is used to measure the accuracy of the intent recognition result of the intent recognition model; Based on the first loss function value and the second loss function value, the loss function value of the intent recognition model is determined.

7. The method of claim 6, wherein, The determination of the first loss function value corresponding to the intent recognition model based on the domain classification result, the action classification result, the entity noun classification result, and in combination with the domain classification label, the action classification label and the entity noun classification label includes: Based on the domain classification result, in combination with the domain classification label, the action classification label and the entity noun classification label, a loss function value corresponding to a domain classifier in the intent recognition model is determined; Based on the action classification result, in combination with the domain classification label, the action classification label and the entity noun classification label, a loss function value corresponding to an action classifier in the intent recognition model is determined; Determine, based on the entity noun classification result, a loss function value corresponding to an entity noun classifier in the intent recognition model in combination with the domain classification label, the action classification label, and the entity noun classification label; Determine the first loss function value according to the loss function value corresponding to the domain classifier, the loss function value corresponding to the action classifier, and the loss function value corresponding to the entity noun classifier.

8. The method of claim 5, wherein, The method further comprises: Split the intent recognition label based on a preset rule to obtain the domain classification label, the action classification label, and the entity noun classification label.

9. The method according to any one of claims 5 to 7, characterized in that, The intent recognition model comprises a feature extraction layer, a domain encoder, an action encoder, an entity noun encoder, a mean value calculation network, a variance calculation network, a noise superposition network, a domain classifier, an action classifier, an entity noun classifier, and a result fusion layer. The feature extraction layer is configured to perform feature extraction processing on the dialogue text to obtain feature information corresponding to the dialogue text; the feature information comprises a plurality of discrete feature variables; The domain encoder is configured to perform domain classification encoding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a domain feature vector of the feature information; the domain feature vector is a continuous feature variable; The action encoder is configured to perform action classification encoding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain an action feature vector of the feature information; the action feature vector is a continuous feature variable; The entity noun encoder is configured to perform entity noun classification encoding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain an entity noun feature vector of the feature information; the entity noun feature vector is a continuous feature variable; The mean value calculation network is configured to process a plurality of dimension feature vectors respectively to obtain a mean value vector corresponding to each of the plurality of dimension feature vectors; the mean value vector is used to indicate a distribution mean value of the dimension feature vector in space; The variance calculation network is configured to process a plurality of dimension feature vectors respectively to obtain a variance vector corresponding to each of the plurality of dimension feature vectors; the variance vector is used to indicate a distribution variance of the dimension feature vector in space; The noise superposition network is configured to superimpose target noise based on the mean value vector and the variance vector corresponding to each of the plurality of dimension feature vectors to determine a to-be-classified feature vector corresponding to each of the plurality of dimension feature vectors; The domain classifier is configured to perform classification processing on the to-be-classified feature vector of the domain dimension to obtain a domain dimension classification result of the feature vector; The action classifier is configured to perform classification processing on the to-be-classified feature vector of the action dimension to obtain an action dimension classification result of the feature vector; The entity noun classifier is configured to perform classification processing on the to-be-classified feature vector of the entity noun dimension to obtain an entity noun dimension classification result of the feature vector; The result fusion layer is configured to superimpose the domain classification result, the action classification result and the entity noun classification result to determine an intent recognition result of the dialogue text.

10. An apparatus for identifying an intent of a dialogue, characterized by comprising: The device comprises: A text acquisition module is configured to acquire dialogue text to be recognized. A feature extraction module is configured to perform feature extraction processing on the dialogue text to obtain feature information of the dialogue text, wherein the feature information comprises a plurality of discrete feature variables. A feature encoding module is configured to perform multi-dimensional classification encoding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimensional feature vector of the feature information, wherein the multi-dimensional feature vector comprises a domain feature vector, an action feature vector and an entity noun feature vector, wherein the domain feature vector is configured to indicate a domain to which the dialogue text belongs, the action feature vector is configured to indicate an action to be performed by an inputter of the dialogue text, and the entity noun feature vector is configured to indicate an entity noun associated with the domain and the action. A feature classification module is configured to perform classification processing on the multi-dimensional feature vector respectively to obtain a multi-dimensional classification result of the feature information, wherein the multi-dimensional classification result comprises classification results of a plurality of different dimensions. A result fusion module is configured to perform fusion recognition processing on the multi-dimensional classification result to determine an intent recognition result of the dialogue text. 11.A device for training an intention recognition model, comprising: The device comprises: A sample acquisition module is configured to acquire at least one training sample of an intent recognition model, wherein the training sample comprises a sample dialogue text and label information corresponding to the sample dialogue text. A feature acquisition module is configured to perform feature extraction processing on the sample dialogue text to obtain feature information of the sample dialogue text, wherein the feature information comprises a plurality of discrete feature variables. A vector acquisition module is configured to perform multi-dimensional classification encoding on the feature information, convert the plurality of discrete feature variables in the feature information into continuous feature variables, and obtain a multi-dimensional feature vector of the feature information, wherein the multi-dimensional feature vector comprises a domain feature vector, an action feature vector and an entity noun feature vector, wherein the domain feature vector is configured to indicate a domain to which the dialogue text belongs, the action feature vector is configured to indicate an action to be performed by an inputter of the dialogue text, and the entity noun feature vector is configured to indicate an entity noun associated with the domain and the action. A vector classification module is configured to perform classification processing on the multi-dimensional feature vector respectively to obtain a multi-dimensional classification result of the feature information, wherein the multi-dimensional classification result comprises classification results of a plurality of different dimensions. A result acquisition module is configured to perform fusion recognition processing on the multi-dimensional classification result to determine an intent recognition result of the dialogue text. A function value determination module is configured to calculate a loss function value of the intent recognition model based on the multi-dimensional classification result, the intent recognition result and the label information corresponding to the sample dialogue text. A parameter adjustment module is configured to adjust parameters of the intent recognition model of the dialogue information based on the loss function value.

12. A computer device, comprising: The computer device comprises a processor and a memory, and at least one program is stored in the memory, and the at least one program is loaded and executed by the processor to implement the intent recognition method of the dialogue information according to any one of claims 1 to 3, or implement the training method of the intent recognition model according to any one of claims 4 to 9.

13. A computer-readable storage medium, characterized in that, The storage medium stores at least one program, and the at least one program is loaded and executed by the processor to implement the intent recognition method of the dialogue information according to any one of claims 1 to 3, or implement the training method of the intent recognition model according to any one of claims 4 to 9.

14. A computer program product, characterised in that, The computer program product comprises computer instructions stored in a computer readable storage medium, and the processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to implement the intent recognition method of the dialogue information according to any one of claims 1 to 3, or implement the training method of the intent recognition model according to any one of claims 4 to 9.

Images