A method and device for identifying business intent

Abstract

This application provides a method and apparatus for identifying business intent, comprising: acquiring a target identification statement in a current consultation dialogue; using a first natural language understanding model to identify the estimated intent indicated by the target identification statement; selecting a second target natural language understanding model from among various second natural language understanding models that corresponds to the business module indicated by the estimated intent; and using the second target natural language understanding model to identify the true intent indicated by the target identification statement from among the business intents in the second target natural language understanding model. This reduces the model size and improves model performance; simultaneously, the combined identification by the first and second natural language understanding models enhances the accuracy of intent recognition.

Description

Technical Field

[0001] This application relates to the field of intent recognition technology, and in particular to a method and apparatus for recognizing business intent. Background Technology

[0002] With the development of artificial intelligence-related technologies, the application of intelligent chatbots has become increasingly widespread, and people's expectations and requirements for artificial intelligence chatbots are also getting higher and higher.

[0003] Currently, intelligent chatbots often rely on Natural Language Understanding (NLU) models to identify the user intent indicated in the dialogue within each application scenario. However, as the business content in application scenarios continues to become richer and more extensive, the scale of NLU models is also constantly increasing. But currently, large-scale NLU models suffer from problems such as low recognition accuracy and poor recognition performance. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method and apparatus for identifying business intent. First, a first natural language understanding model is used to identify the estimated intent. Then, a second natural language understanding model corresponding to the business module is used to identify the specific true intent under the estimated intent. In this way, by splitting the business content into multiple business modules and training a second natural language understanding model for each business module, and then extracting data from the training data of each second natural language understanding model to train the first natural language understanding model, the model size can be reduced and the model performance improved. At the same time, the combined identification by the first and second natural language understanding models can further improve the accuracy of intent recognition.

[0005] This application provides a method for identifying business intent, the method comprising:

[0006] Retrieve the target-identified statements in the current consultation dialogue;

[0007] The first natural language understanding model is used to identify the estimated intent indicated by the target recognition statement;

[0008] A second target natural language understanding model corresponding to the business module indicated by the predicted intent is selected from each second natural language understanding model; wherein, the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to each business module is trained from the corpus data related to each business module; the first natural language understanding model is trained from the corpus data extracted from the corpus data related to each business module;

[0009] Using the second target natural language understanding model, the true intent indicated by the target recognition statement is identified from each business intent in the second target natural language understanding model.

[0010] Furthermore, each of the second natural language understanding models is trained through the following steps:

[0011] Based on the business content in the target business scenario, the business content is divided into multiple business modules;

[0012] For each business module, an initial second natural language understanding model corresponding to that business module is constructed;

[0013] Based on the business content in the business module, determine the intent that instructs the business module, the various business intents included under the intent of the business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can instruct each entity condition; wherein, for each business intent corresponding to multiple optional business response results, the entity conditions associated with the business intent are used to determine the actual business response result for the business intent from the multiple optional business response results;

[0014] The initial second natural language understanding model is trained by using the business intents included in the intent of using this business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can indicate each entity condition, to obtain the second natural language understanding model corresponding to the business module.

[0015] The first natural language understanding model was trained through the following steps:

[0016] Construct an initial first natural language understanding model corresponding to the target business scenario;

[0017] For each business module in the target business scenario, extract the expression statements from the expression statements in the second natural language understanding model corresponding to that business module, which is used to train the model.

[0018] The initial first natural language understanding model is trained using the intents of each business module, the extracted expressions, and the entity conditions associated with each business intent corresponding to multiple optional business response results, to obtain the first natural language understanding model.

[0019] Furthermore, after identifying the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model using the second target natural language understanding model, the recognition method further includes:

[0020] A pre-built knowledge graph is used to determine whether a business response result can be obtained for the true intent; wherein, the knowledge graph includes multiple business intents, at least one optional business response result corresponding to each business intent, entity conditions associated with each business intent corresponding to multiple optional business response results, and guiding words corresponding to each entity condition;

[0021] If possible, the service response result will be fed back to the user so that the user can handle the service according to the instructions of the service response result;

[0022] If not, the knowledge graph is used to determine the guiding scripts corresponding to each entity condition required to obtain the business response result;

[0023] The guiding script is provided to the user so that the user can supplement the required entity conditions based on the guiding script.

[0024] Furthermore, after using the first natural language understanding model to identify the estimated intent indicated by the target recognition statement, the recognition method further includes:

[0025] If the estimated intent is an intent to supplement entity conditions, then the second target natural language understanding model corresponding to the current consultation dialogue is used to identify the real entity conditions contained in the target identification statement; wherein, the second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the first real intent determined by other consultation statements before the target identification statement in the current consultation dialogue.

[0026] Based on the knowledge graph, determine whether there is a correlation between the real entity condition and the first real intent;

[0027] If there is a relationship, the knowledge graph is used to determine whether a business response result for the first true intent can be obtained based on the real entity conditions, and the business response result or guiding words for the first true intent are fed back to the user based on the determination result.

[0028] After the step of identifying the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model using the second target natural language understanding model, the recognition method further includes:

[0029] When the business content of any business module in the target business scenario changes, the training data of the second natural language understanding model corresponding to the business module is updated according to the changed business content, and the second natural language understanding model corresponding to the business module is retrained based on the updated training data.

[0030] Furthermore, the step of using a first natural language understanding model to identify the estimated intent indicated by the target recognition statement includes:

[0031] The first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent of each business module, and the first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent to supplement entity conditions.

[0032] The intention with the highest probability score is identified as the predicted intention.

[0033] Furthermore, the step of using the second target natural language understanding model to identify the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model includes:

[0034] The second target natural language understanding model is used to determine the probability score of each business intent indicated by the target recognition statement;

[0035] If the probability score of the business intent with the highest probability score is less than or equal to a preset score threshold, and the difference in probability scores between the business intent with the highest probability score and the business intent with the second highest probability score is greater than a preset difference threshold, then the business intent with the highest probability score is determined as the true intent.

[0036] This application embodiment also provides a business intent identification device, the identification device comprising:

[0037] The acquisition module is used to acquire the target recognition statements in the current consultation dialogue;

[0038] The first recognition module is used to identify the estimated intent indicated by the target recognition statement using a first natural language understanding model;

[0039] A filtering module is used to filter out the second target natural language understanding model corresponding to the business module indicated by the estimated intent from each second natural language understanding model; wherein, the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to the intent of each business module is trained by the corpus data related to each business module; the first natural language understanding model is trained by the corpus data extracted from the corpus data related to each business module;

[0040] The second identification module is used to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model using the second target natural language understanding model.

[0041] Furthermore, the recognition device also includes a first training module; the first training module is used to train each of the second natural language understanding models through the following steps:

[0042] Based on the business content in the target business scenario, the business content is divided into multiple business modules;

[0043] For each business module, an initial second natural language understanding model corresponding to that business module is constructed;

[0044] Based on the business content in the business module, determine the intent that instructs the business module, the various business intents included under the intent of the business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can instruct each entity condition; wherein, for each business intent corresponding to multiple optional business response results, the entity conditions associated with the business intent are used to determine the actual business response result for the business intent from the multiple optional business response results;

[0045] The initial second natural language understanding model is trained by using the business intents included in the intent of using this business module, the expression statements that can reflect each business intent, and the entity conditions associated with each business intent corresponding to multiple optional business response results, to obtain the second natural language understanding model corresponding to the business module.

[0046] Furthermore, the recognition device also includes a second training module; the second training module is used to train the first natural language understanding model through the following steps:

[0047] Construct an initial first natural language understanding model corresponding to the target business scenario;

[0048] For each business module in the target business scenario, extract the expression statements from the expression statements in the second natural language understanding model corresponding to that business module, which is used to train the model.

[0049] The initial first natural language understanding model is trained using the intents of each business module, the extracted expressions, and the entity conditions associated with each business intent corresponding to multiple optional business response results, to obtain the first natural language understanding model.

[0050] Furthermore, the identification device also includes a feedback module; the feedback module is used for:

[0051] A pre-built knowledge graph is used to determine whether a business response result can be obtained for the true intent; wherein, the knowledge graph includes multiple business intents, at least one optional business response result corresponding to each business intent, entity conditions associated with each business intent corresponding to multiple optional business response results, and guiding words corresponding to each entity condition;

[0052] If possible, the service response result will be fed back to the user so that the user can handle the service according to the instructions of the service response result;

[0053] If not, the knowledge graph is used to determine the guiding scripts corresponding to each entity condition required to obtain the business response result;

[0054] The guiding script is provided to the user so that the user can supplement the required entity conditions based on the guiding script.

[0055] Furthermore, the identification device also includes a third identification module; the third identification module is used for:

[0056] If the estimated intent is an intent to supplement entity conditions, then the second target natural language understanding model corresponding to the current consultation dialogue is used to identify the real entity conditions contained in the target identification statement; wherein, the second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the first real intent determined by other consultation statements before the target identification statement in the current consultation dialogue.

[0057] Based on the knowledge graph, determine whether there is a correlation between the real entity condition and the first real intent;

[0058] If there is a relationship, the knowledge graph is used to determine whether a business response result for the first true intent can be obtained based on the real entity conditions, and the business response result or guiding words for the first true intent are fed back to the user based on the determination result.

[0059] Furthermore, the identification device also includes an update module; the update module is used for:

[0060] When the business content of any business module in the target business scenario changes, the training data of the second natural language understanding model corresponding to the business module is updated according to the changed business content, and the second natural language understanding model corresponding to the business module is retrained based on the updated training data.

[0061] Furthermore, when the first recognition module is used to identify the estimated intent indicated by the target recognition statement using the first natural language understanding model, the first recognition module is used to:

[0062] The first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent of each business module, and the first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent to supplement entity conditions.

[0063] The intention with the highest probability score is identified as the predicted intention.

[0064] Furthermore, when the second identification module is used to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model using the second target natural language understanding model, the second identification module is used to:

[0065] The second target natural language understanding model is used to determine the probability score of each business intent indicated by the target recognition statement;

[0066] If the probability score of the business intent with the highest probability score is less than or equal to a preset score threshold, and the difference in probability scores between the business intent with the highest probability score and the business intent with the second highest probability score is greater than a preset difference threshold, then the business intent with the highest probability score is determined as the true intent.

[0067] This application embodiment also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the business intent identification method described above are performed.

[0068] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the business intent identification method described above.

[0069] This application provides a method and apparatus for identifying business intent, comprising: acquiring a target identification statement in a current consultation dialogue; using a first natural language understanding model to identify the estimated intent indicated by the target identification statement; selecting a second target natural language understanding model from each second natural language understanding model that corresponds to the business module indicated by the estimated intent; wherein the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to each business module is trained from the corpus data related to each business module; the first natural language understanding model is trained from the corpus data extracted from the corpus data related to each business module; and using the second target natural language understanding model to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model.

[0070] In this way, by breaking down business content into multiple business modules and training a second natural language understanding model for each business module, and then extracting data from the training data of each second natural language understanding model to train the first natural language understanding model, the size of the model can be reduced and the performance of the model can be improved. At the same time, the combined recognition of the first and second natural language understanding models can further improve the accuracy of intent recognition.

[0071] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0072] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 A flowchart illustrating a method for identifying business intent provided in an embodiment of this application is shown;

[0074] Figure 2 This illustration shows one of the structural schematic diagrams of a business intent identification device provided in an embodiment of this application;

[0075] Figure 3 This is a second schematic diagram of the structure of a business intent identification device provided in an embodiment of this application;

[0076] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0078] Research has shown that with the development of artificial intelligence-related technologies, the application of intelligent chatbots has become increasingly widespread, and people's expectations and requirements for AI chatbots are also rising.

[0079] Currently, intelligent chatbots often rely on Natural Language Understanding (NLU) models to identify the user intent indicated in the dialogue within each application scenario. However, as the business content in application scenarios continues to become richer and more extensive, the scale of NLU models is also constantly increasing. But currently, large-scale NLU models suffer from problems such as low recognition accuracy and poor recognition performance.

[0080] Natural Language Understanding (NLU) is a general term for methods, models, or tasks that support machines in understanding text content. NLU research uses computers to simulate the human language communication process, enabling computers to understand and use natural languages ​​such as Chinese and English, thus achieving natural language communication between humans and machines. A Natural Language Understanding model (NLU model) is a trained model capable of recognizing the intent indicated by natural language. Its main purpose is to identify the intent corresponding to natural language; examples include the BERT model.

[0081] Based on this, embodiments of this application provide a method and apparatus for identifying business intent, so as to reduce the size of the natural language understanding model, improve the performance of the natural language understanding model, and improve the accuracy of intent recognition.

[0082] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for identifying business intent provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the identification method includes:

[0083] S101. Obtain the target recognition statement in the current consultation dialogue.

[0084] In one possible implementation, a dialogue window can be provided in the electronic device as a channel for human-computer interaction. For example, a user can have a consultation dialogue with an intelligent chatbot through the dialogue window. The user can enter different consultation statements in the dialogue window to inquire about the business they want to handle. Then, the intelligent chatbot can obtain the target recognition statement in the current consultation dialogue entered by the user through the dialogue window.

[0085] S102. Use a first natural language understanding model to identify the estimated intent indicated by the target recognition statement.

[0086] It should be noted that in practical implementation, natural language understanding models are often built for specific application scenarios, such as government service scenarios. A well-trained natural language understanding model can identify multiple intents related to a specific application scenario, which can indicate the business that the user wants to handle. For example, in a government service scenario, the services that users can handle include "unemployment registration," "residence permit," and "minimum living allowance for urban and rural residents." Accordingly, the business intents that can be identified from the user's inquiry should include "unemployment registration process," "residence permit application process," and "minimum living allowance application process for urban and rural residents," etc.

[0087] In some cases, after a natural language understanding model identifies a specific business intent, it cannot determine the precise final response to the user because this intent corresponds to multiple possible business responses. Therefore, it needs to provide the user with guiding statements to help them supplement the necessary entity conditions and ultimately determine the actual response based on these conditions. For example, in the case of "residence permit application," after identifying the user's intent as "residence permit application process," the model needs to provide the guiding statement, "Are you applying for a residence permit for yourself or for someone else?" After receiving this guidance, the user can enter "I want to apply for a residence permit for someone else" in the current dialogue. In this way, the natural language understanding model can extract the entity condition "Application recipient: someone else" from the user's re-entered statement. In this way, after one or more rounds of dialogue, the user has completed all the entity conditions required to obtain the business response result corresponding to the business intent according to the guiding script. At this time, there is no need to provide the guiding script to the user again, and the business response result that should be provided to the user can be determined, such as "The application process for residence permit agency is: xxx; the required application materials include: xxx", etc.

[0088] Clearly, as the business content in application scenarios continues to be enriched and expanded, the amount of business-related data continues to increase, and thus the scale of the trained natural language understanding model also continues to grow.

[0089] Therefore, the identification method provided in this application can pre-divide the business content of the target business scenario into multiple business modules, train a second natural language understanding model corresponding to each business module using the corpus data related to each business module, and then extract corpus data from the corpus data related to each business module to train a first natural language understanding model. For example, the business content in the target business scenario "government service scenario" can be pre-divided into business modules such as "civil affairs," "urban and rural residents' social security," and "grain subsidies," and a second natural language understanding model corresponding to each business module can be trained using the corpus data related to each business module, namely, the second natural language understanding model corresponding to the "civil affairs" business module, the second natural language understanding model corresponding to the "urban and rural residents' social security" business module, and the second natural language understanding model corresponding to the "grain subsidies" business module, etc.; and then corpus data can be extracted from the corpus data related to each business module to train a first natural language understanding model corresponding to the "government service scenario."

[0090] Specifically, in one possible implementation, each of the second natural language understanding models can be trained through the following steps:

[0091] Step 1: Based on the business content in the target business scenario, divide the business content into multiple business modules.

[0092] Step 2: For each business module, construct an initial second natural language understanding model corresponding to that business module.

[0093] Step 3: Based on the business content in the business module, determine the intent that indicates the business module, the various business intents included under the intent of the business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can indicate each entity condition.

[0094] Specifically, for each business intent corresponding to multiple optional business response results, the entity conditions associated with that business intent are used to determine the actual business response result for that business intent from the multiple optional business response results.

[0095] Step 4: Using the intents included in the business module, the expressions that reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expressions that indicate each entity condition, train the initial second natural language understanding model to obtain the second natural language understanding model corresponding to the business module.

[0096] In this way, the second natural language understanding model corresponding to each trained business module can identify the specific business intent indicated by the target recognition statement and the real entity conditions contained in the target recognition statement based on the target recognition statement.

[0097] Furthermore, in one possible implementation, the first natural language understanding model can be trained through the following steps:

[0098] Step 1: Construct the initial first natural language understanding model corresponding to the target business scenario.

[0099] Step 2: For each business module in the target business scenario, extract the expression statements from the expression statements in the second natural language understanding model corresponding to that business module, which is used to train the model.

[0100] Step 3: Use the intent of each business module, the extracted expressions, and the entity conditions associated with each business intent corresponding to multiple optional business response results to train the initial first natural language understanding model to obtain the first natural language understanding model.

[0101] In this way, the trained first natural language understanding model can identify the predicted intent based on the target recognition statement as either an intent to instruct any business module or an intent to supplement entity conditions.

[0102] In one possible implementation, in step S102, a pre-trained first natural language understanding model can be used to perform preliminary identification of the target recognition statement and determine the estimated intent indicated by the target recognition statement.

[0103] In specific implementation, step S102 may include the following steps:

[0104] S1021. Using a first natural language understanding model, determine the probability scores of the intent indicated by the target recognition statement as the intent of each business module, and use the first natural language understanding model to determine the probability scores of the intent indicated by the target recognition statement as the intent to supplement entity conditions.

[0105] S1022. The intention with the highest probability score is determined as the estimated intention.

[0106] Here, as mentioned earlier, to determine the business feedback result for the business intent, it is necessary to guide the user to supplement entity conditions. Therefore, the intent that the first natural language understanding model can identify should include the intent to instruct each business module and the intent to supplement entity conditions. For example, the intent label to instruct each business module can be defined as the name of the corresponding business module, such as "Civil Affairs", "Urban and Rural Residents' Basic Medical Insurance", and "Grain Subsidy", and the intent label to supplement entity conditions can be defined as "supplement entity".

[0107] In practical implementation, the first natural language understanding model can be used in any way available in the prior art to determine the probability scores of the intent indicated by the target recognition statement as the intent of each business module, and the probability scores of the intent indicated by the target recognition statement as the intent used to supplement entity conditions. The probability scores characterize the likelihood that the schematic diagram pointed to by the target recognition statement corresponds to each intent label. That is, the higher the probability score, the greater the likelihood that the schematic diagram pointed to by the target recognition statement corresponds to the intent label. Therefore, the intent with the highest probability score can be determined as the estimated intent.

[0108] The intent identified by the first natural language understanding model is only a rough estimate. It is used to distribute the identified sentences to subsequent second natural language understanding models for further refined identification, thereby determining a more detailed specific business intent under the estimated intent.

[0109] S103. Determine that the estimated intent is an intent to instruct any business module, or that the estimated intent is an intent to supplement entity conditions.

[0110] In this step, it is determined whether the estimated intent is an intent to instruct any business module or an intent to supplement entity conditions; if the estimated intent is determined to be an intent to instruct any business module, then step S104 is executed:

[0111] Step S104: Select the second target natural language understanding model from each second natural language understanding model that corresponds to the business module indicated by the estimated intent.

[0112] As mentioned earlier, the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to each business module is trained using the corpus data related to each business module; the first natural language understanding model is trained using corpus data extracted from the corpus data related to each business module. Therefore, after the first natural language understanding model identifies the estimated intent, the second natural language understanding model corresponding to the business module indicated by the estimated intent can be selected from among the various second natural language understanding models. This second natural language understanding model is then used as the second target natural language understanding model, and further refined intent recognition is performed by the second target natural language understanding model.

[0113] S105. Using the second target natural language understanding model, identify the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model.

[0114] Here, each business intent in each second natural language understanding model refers to the business intent that the second natural language understanding model can identify, that is, the business intent related to the business module corresponding to the second natural language understanding model. For example, the business intents that the second natural language understanding model corresponding to the "Urban and Rural Residents' Basic Social Security" business module can identify include: "Urban and Rural Residents' Basic Social Security Registration Process" and "Urban and Rural Residents' Minimum Living Allowance Application Process," etc. In this step, a second target natural language understanding model can be used to identify the true intent indicated by the target recognition statement from the business intents in the second target natural language understanding model. The true intent can indicate the specific business that the user wants to handle.

[0115] In one possible implementation, step S105 may include the following steps:

[0116] S1051. Use the second target natural language understanding model to determine the probability score of each business intent indicated by the target recognition statement.

[0117] Similarly, in practical implementation, the likelihood score of the intent indicated by the consultation statement as each business intent can be determined using any method in the existing technology based on the second target natural language understanding model. It is worth noting that although the likelihood score characterizes the probability that the diagram pointed to by the target identification statement represents each business intent—that is, the higher the likelihood score, the greater the probability that the diagram pointed to by the target identification statement represents the business intent corresponding to that intent label—even the business intent with the highest likelihood score does not necessarily accurately represent the true intent of the specific business indicated by the consultation statement. Therefore, to further increase the accuracy of intent recognition, it is necessary to further determine whether the business intent with the highest likelihood score is credible.

[0118] S1052. If the probability score of the business intention with the highest probability score is less than or equal to a preset score threshold, and the difference in probability scores between the business intention with the highest probability score and the business intention with the second highest probability score is greater than a preset difference threshold, or if the probability score of the business intention with the highest probability score is greater than a preset score threshold, then the business intention with the highest probability score is determined as the true intention.

[0119] Specifically, if the probability score of the business intention with the highest probability score is greater than a preset score threshold, or if the probability score of the business intention with the highest probability score is less than or equal to the preset score threshold, but the difference in probability scores between the business intention with the highest probability score and the business intention with the second highest probability score is greater than a preset difference threshold, in both cases the business intention with the highest probability score is considered credible and is determined as the true intention.

[0120] The preset score threshold can be a fixed constant, such as 0.7; similarly, the preset difference threshold can also be a fixed constant, such as 0.25. For example, the sources of the preset score threshold and preset difference threshold are as follows: After training each second natural language understanding model, testing is conducted using a large number of test sets (the test sets contain test consultation statements and test intent labels corresponding to each test consultation statement). The test results mainly include the probability scores corresponding to the business intent with the highest probability score and the second highest probability score determined by each second natural language understanding model, as well as the difference between the two. The business intent with the highest probability score that matches the intent indicated by the test intent label is considered correctly identified; otherwise, it is considered incorrectly identified.

[0121] Statistical results show that when the highest probability score is greater than or equal to 0.7, the credibility of the intended result is relatively high and can be considered a credible result. When the highest probability score is less than 0.7, further calculation is performed on the difference between the highest probability score and the second highest probability score in each training sample group. The calculated differences for each group are statistically analyzed and categorized to calculate an effective difference threshold. There are two categorization methods: one is based on the total amount, such as [0,1), [0.1,1), [0.2,1); the other is based on the interval, such as [0,0.1), [0.1,0.2), [0.2,0.3).

[0122] Statistical analysis revealed that when the difference between the highest and second-highest probabilities is between 0.2 and 0.3, the number of misidentifications (where unreliable results are mistakenly identified as reliable results) can be controlled to within 3 in a test sample, while the number of correctly identified results is greater than or equal to 42. This demonstrates that the positive benefits far outweigh the negative impacts for a test sample, and the negative impacts can be kept within an acceptable range. Therefore, when the highest probability score is less than 0.7, and the difference between the highest and second-highest probability scores is greater than or equal to 0.25, the highest-scoring intent is considered a reliable result.

[0123] In one possible implementation, after step S105 uses the second target natural language understanding model to identify the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model, the identification method further includes:

[0124] S106. Use a pre-built knowledge graph to determine whether a business response result can be obtained for the true intent.

[0125] If possible, proceed to step S107: provide feedback on the business response result to the user so that the user can process the business according to the instructions of the business response result.

[0126] If not, proceed to step S108: use the knowledge graph to determine the guiding scripts corresponding to each entity condition required to obtain the business response result.

[0127] S109. Provide feedback to the user with the guiding script so that the user can supplement the required entity conditions according to the guiding script.

[0128] Here, the pre-built knowledge graph is constructed specifically for the target business scenario. It is a network structure that identifies the relationships between business intents and entity conditions based on the various business modules within the target business scenario. Specifically, the knowledge graph includes at least one business intent under the intent corresponding to each business module in the target business scenario; entity conditions associated with each business intent corresponding to multiple optional business response results (wherein, the entity conditions corresponding to each business intent corresponding to multiple optional business response results can be used to determine the actual business response result for that business intent from the multiple optional business response results); at least one optional business response result that can be fed back to the user for that business intent based on different entity conditions; and guiding messages that should be fed back to the user when any entity condition is missing (i.e., guiding messages corresponding to each entity condition).

[0129] Therefore, after identifying the true intent indicated by the target recognition statement using the second-target natural language understanding model, it can be determined whether a business response result corresponding to the true intent can be obtained by querying the knowledge graph. In one possible implementation, the true intent in the knowledge graph corresponds to only one possible business response result, that is, the business response result for the true intent can be determined without supplementing entity conditions. In this case, the business response result can be directly fed back to the user so that the user can handle the business according to the instructions of the business response result.

[0130] In another possible implementation, after determining the true intent, a knowledge graph is consulted to identify multiple possible business responses corresponding to that intent. At this point, it's impossible to determine which response should be given to the user. Guiding statements are used to encourage the user to supplement the required entity conditions. The final response is then determined based on these supplemented conditions. For example, after identifying the user's intent as "residence permit application process," the knowledge graph can determine the corresponding entity conditions, including "applicant," "administrative division," and "place of household registration." Guiding statements can then be gradually provided to the user to guide them in supplementing the "administrative division" and "place of household registration." After guiding the user to supplement these conditions, the knowledge graph and the corresponding entity conditions can be used to determine the appropriate response: "The residence permit application process is: xxx; the required materials include: xxx, etc."

[0131] In one possible implementation, the user's query may include entity conditions in addition to the estimated intent indicated by the business module. For example, the user's query might be "I want to apply for a residence permit on behalf of someone else." In this case, besides recognizing the estimated intent indicated by the target recognition statement, the entity condition "Application subject: applying on behalf of someone else" included in the query can also be identified, thereby reducing the number of rounds of dialogue with the user and helping the user to handle business more quickly.

[0132] Furthermore, if the estimated intent determined in step S103 is an intent to supplement entity conditions, then step S110 is executed:

[0133] S110. Use the second target natural language understanding model corresponding to the current consultation dialogue to identify the real entity conditions contained in the target recognition statement.

[0134] The second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the estimated intent determined by other consultation statements before the target identification statement in the current consultation dialogue.

[0135] Here, if the estimated intent identified by the first natural language understanding model is used to supplement entity conditions, it actually indicates that the specific business intent has been determined based on other consultation statements preceding the target identification statement in the current consultation dialogue. In this case, the second target natural language understanding model that identified the specific business intent (that is, the second target natural language understanding model corresponding to the estimated intent determined based on other consultation statements preceding the target identification statement) can be directly invoked to continue identifying the target identification statement. Further, if the second target natural language understanding model cannot be determined from other consultation statements preceding the target identification statement in the current consultation dialogue (for example, if the target identification statement is the first statement in the current consultation dialogue, in which case the first statement is used to supplement entity conditions), then corresponding guiding dialogue is provided to the user to determine the estimated intent for the user to instruct any business module.

[0136] Specifically, the real entity conditions contained in the target recognition statement can be identified using the second target natural language understanding model corresponding to the current consultation dialogue.

[0137] S111. Based on the knowledge graph, determine whether there is a correlation between the real entity condition and the first real intent.

[0138] Here, the actual entity conditions supplemented by the user may not be related to the first true intent. For example, the entity conditions related to the first true intent may be determined by the knowledge graph as "applicant", "administrative division", and "place of household registration", but the actual entity condition supplemented by the user is "age". This actual entity condition does not belong to the entity conditions corresponding to the first true intent "residence permit application process", and this actual entity condition is not related to the first true intent. In other words, if the actual entity condition is not the entity condition required to obtain the business response result, the user can be prompted with corresponding guidance to indicate that the input is incorrect, or to guide the user to re-enter the entity condition, etc.

[0139] In specific implementation, if there is a correlation between the real entity conditions and the first real intent, then step S112 is executed: based on the real entity conditions, the knowledge graph is used to determine whether a business response result for the first real intent can be obtained, and based on the determination result, the business response result or guiding words for the first real intent are fed back to the user.

[0140] In one possible implementation, the second natural language understanding models are trained through the following steps:

[0141] Step 1: Based on the business content in the target business scenario, divide the business content into multiple business modules.

[0142] Step 2: For each business module, construct an initial second natural language understanding model corresponding to that business module.

[0143] Step 3: Based on the business content in the business module, determine the intent that indicates the business module, the various business intents included under the intent of the business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can indicate each entity condition.

[0144] Step 4: Using the intents included in the business module, the expressions that reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expressions that indicate each entity condition, train the initial second natural language understanding model to obtain the second natural language understanding model corresponding to the business module.

[0145] Furthermore, after using the second target natural language understanding model in step S105 to identify the true intent indicated by the target recognition statement from each business intent in the second target natural language understanding model, the recognition method further includes: when the business content of any business module in the target business scenario changes, updating the training data of the second natural language understanding model corresponding to the business module according to the changed business content, and retraining the second natural language understanding model corresponding to the business module based on the updated training data.

[0146] After updating the training data of the second natural language understanding model corresponding to the changed business content, the training data of the first natural language understanding model also needs to be updated accordingly to retrain the first natural language understanding model. For example, if a new business intent is added to the second natural language understanding model, the new business intent and related training data are added to the training data of the first natural language understanding model; if new corpus data is added to a certain business intent of the second natural language understanding model, a portion of the corpus data is extracted from the new corpus data and added to the training data of the first natural language understanding model; if the corpus data related to a certain business intent of the second natural language understanding model is modified or deleted, the corresponding corpus data in the training data of the first natural language understanding model must also be modified or deleted synchronously.

[0147] In this way, when the business content of any business module changes, only the training data of the second natural language understanding model corresponding to that business module needs to be updated, and the second natural language understanding model corresponding to that business module can be retrained based on the updated training data, without having to retrain other second natural language understanding models, thereby shortening the training time of the model and reducing the number of training sessions.

[0148] This application provides a method for identifying business intent, comprising: acquiring a target identification statement in a current consultation dialogue; using a first natural language understanding model to identify the estimated intent indicated by the target identification statement; selecting a second target natural language understanding model from each second natural language understanding model that corresponds to the business module indicated by the estimated intent; wherein the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to each business module is trained from the corpus data related to each business module; the first natural language understanding model is trained from the corpus data extracted from the corpus data related to each business module; and using the second target natural language understanding model to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model.

[0149] In this way, by breaking down the originally large-scale natural language understanding model into a first natural language understanding model and multiple second natural language understanding models, the size of a single model is greatly reduced, which is beneficial for model training and improving model performance. In practical applications, a combined recognition method in which the first natural language understanding model initially identifies the estimated intent, and then the second target natural language understanding model corresponding to the estimated intent performs further refined recognition, can further improve the accuracy of intent recognition.

[0150] Please see Figure 2 , Figure 3 , Figure 2 This is one of the structural schematic diagrams of a business intent recognition device provided in an embodiment of this application. Figure 3 This is a second schematic diagram of a business intent identification device provided in an embodiment of this application. Figure 2 As shown, the identification device 200 includes:

[0151] The acquisition module 210 is used to acquire the target recognition statement in the current consultation dialogue;

[0152] The first identification module 220 is used to identify the estimated intent indicated by the target identification statement using a first natural language understanding model;

[0153] The filtering module 230 is used to filter out the second target natural language understanding model corresponding to the business module indicated by the estimated intent from each second natural language understanding model; wherein, the business content of the target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to the intent of each business module is trained by the corpus data related to each business module; the first natural language understanding model is trained by the corpus data extracted from the corpus data related to each business module;

[0154] The second identification module 240 is used to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model using the second target natural language understanding model.

[0155] Furthermore, such as Figure 3 As shown, the recognition device further includes a first training module 250; the first training module 250 is used to train each of the second natural language understanding models through the following steps:

[0156] Based on the business content in the target business scenario, the business content is divided into multiple business modules;

[0157] For each business module, an initial second natural language understanding model corresponding to that business module is constructed;

[0158] Based on the business content in the business module, determine the intent that instructs the business module, the various business intents included under the intent of the business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can instruct each entity condition; wherein, for each business intent corresponding to multiple optional business response results, the entity conditions associated with the business intent are used to determine the actual business response result for the business intent from the multiple optional business response results;

[0159] The initial second natural language understanding model is trained by using the business intents included in the intent of using this business module, the expression statements that can reflect each business intent, the entity conditions associated with each business intent corresponding to multiple optional business response results, and the expression statements that can indicate each entity condition, to obtain the second natural language understanding model corresponding to the business module.

[0160] Furthermore, such as Figure 3 As shown, the recognition device further includes a second training module 260; the second training module 260 is used to train the first natural language understanding model through the following steps:

[0161] Construct an initial first natural language understanding model corresponding to the target business scenario;

[0162] For each business module in the target business scenario, extract the expression statements from the expression statements in the second natural language understanding model corresponding to that business module, which is used to train the model.

[0163] The initial first natural language understanding model is trained using the intent of each business module, the extracted expressions, and the entity conditions corresponding to each business intent, to obtain the first natural language understanding model.

[0164] Furthermore, such as Figure 3 As shown, the identification device further includes a feedback module 270; the feedback module 270 is used for:

[0165] A pre-built knowledge graph is used to determine whether a business response to the stated true intent can be obtained.

[0166] If possible, the service response result will be fed back to the user so that the user can handle the service according to the instructions of the service response result;

[0167] If not, the knowledge graph is used to determine the guiding scripts corresponding to each entity condition required to obtain the business response result;

[0168] The guiding script is provided to the user so that the user can supplement the required entity conditions based on the guiding script.

[0169] Furthermore, such as Figure 3 As shown, the identification device further includes a third identification module 280; the third identification module 280 is used for:

[0170] If the estimated intent is an intent to supplement entity conditions, then the second target natural language understanding model corresponding to the current consultation dialogue is used to identify the real entity conditions contained in the target identification statement; wherein, the second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the first real intent determined by other consultation statements before the target identification statement in the current consultation dialogue.

[0171] Based on the knowledge graph, determine whether there is a correlation between the real entity condition and the first real intent;

[0172] If there is a relationship, the knowledge graph is used to determine whether a business response result for the first true intent can be obtained based on the real entity conditions, and the business response result or guiding words for the first true intent are fed back to the user based on the determination result.

[0173] Furthermore, such as Figure 3As shown, the identification device further includes an update module 290; the update module 290 is used for:

[0174] When the business content of any business module in the target business scenario changes, the training data of the second natural language understanding model corresponding to the business module is updated according to the changed business content, and the second natural language understanding model corresponding to the business module is retrained based on the updated training data.

[0175] Furthermore, when the first recognition module 220 uses the first natural language understanding model to identify the estimated intent indicated by the target recognition statement, the first recognition module 220 is used to:

[0176] The first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent of each business module, and the first natural language understanding model is used to determine the probability scores of the intent indicated by the target recognition statement as the intent to supplement entity conditions.

[0177] The intention with the highest probability score is identified as the predicted intention.

[0178] Furthermore, when the second identification module 240 is used to identify the true intent indicated by the target identification statement from each business intent in the second target natural language understanding model using the second target natural language understanding model, the second identification module 240 is used to:

[0179] The second target natural language understanding model is used to determine the probability score of each business intent indicated by the target recognition statement;

[0180] If the probability score of the business intent with the highest probability score is less than or equal to a preset score threshold, and the difference in probability scores between the business intent with the highest probability score and the business intent with the second highest probability score is greater than a preset difference threshold, then the business intent with the highest probability score is determined as the true intent.

[0181] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0182] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1The steps of a business intent identification method in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0183] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of a business intent identification method in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0184] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0185] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0186] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0187] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0188] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0189] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying a service intent, characterized by, The identification method comprises: obtaining a target identification sentence in a current consultation dialogue; identifying an estimated intention indicated by the target identification sentence using a first natural language understanding model; selecting a second target natural language understanding model corresponding to a business module indicated by the estimated intention from each second natural language understanding model; wherein the business content of a target business scenario is pre-divided into multiple business modules; the second natural language understanding model corresponding to each business module is trained by corpus data related to each business module; and the first natural language understanding model is trained by corpus data extracted from corpus data related to each business module; identifying a real intention indicated by the target identification sentence from each business intention in the second target natural language understanding model using the second target natural language understanding model; After identifying the estimated intention indicated by the target identification sentence using the first natural language understanding model, the identification method further comprises: if the estimated intention is an intention for supplementing entity conditions, identifying real entity conditions contained in the target identification sentence using the second target natural language understanding model corresponding to the current consultation dialogue; wherein the second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the first real intention determined by other consultation sentences before the target identification sentence in the current consultation dialogue; determining whether there is an association relationship between the real entity conditions and the first real intention according to a knowledge graph; wherein the knowledge graph comprises multiple business intentions, at least one optional business reply result corresponding to each business intention, entity conditions associated with each business intention corresponding to multiple optional business reply results, and guide rhetoric corresponding to each entity condition; if there is an association relationship, determining whether a business reply result for the first real intention can be obtained using the knowledge graph according to the real entity conditions, and feeding back the business reply result or the guide rhetoric for the first real intention to the user according to the determination result.

2. The identification method according to claim 1, characterized in that, Each second natural language understanding model is trained by the following steps: According to the business content in the target business scenario, the business content is divided into multiple business modules; for each business module, an initial second natural language understanding model corresponding to the business module is constructed; According to the business content in the business module, the intention indicating the business module, each business intention included under the intention of the business module, expression sentences that can embody each business intention, each entity condition associated with each business intention corresponding to multiple optional business reply results, and expression sentences that can indicate each entity condition are determined; wherein for each business intention corresponding to multiple optional business reply results, each entity condition associated with the business intention is used to determine the actual business reply result for the business intention from multiple optional business reply results; The initial second natural language understanding model is trained using the intents included in the intents of each business module, the expression sentences that can embody the intents of each business module, the entity conditions associated with each business intent corresponding to a plurality of optional business reply results, and the expression sentences that can indicate the entity conditions, to obtain a second natural language understanding model corresponding to the business module.

3. The identification method according to claim 2, characterized in that, The first natural language understanding model is obtained by training through the following steps: An initial first natural language understanding model corresponding to the target business scenario is constructed; For each business module in the target business scenario, expression sentences are extracted from the expression sentences used to train the second natural language understanding model corresponding to the business module; The initial first natural language understanding model is trained using the intents of each business module, the extracted expression sentences, and the entity conditions associated with each business intent corresponding to a plurality of optional business reply results, to obtain the first natural language understanding model.

4. The identification method according to claim 1, characterized in that, After the second target natural language understanding model is used to identify the real intent indicated by the target recognition sentence from the business intents in the second target natural language understanding model, the identification method further includes: Using a pre-constructed knowledge graph to determine whether a business reply result for the real intent can be obtained; If so, the business reply result is fed back to the user, so that the user handles the business according to the indication of the business reply result; If not, the knowledge graph is used to determine the guiding language corresponding to each entity condition required to obtain the business reply result; The guiding language is fed back to the user, so that the user supplements each entity condition required according to the guiding language.

5. The identification method according to claim 1, characterized in that, After the step of using the second target natural language understanding model to identify the real intent indicated by the target recognition sentence from the business intents in the second target natural language understanding model, the identification method further includes: When the business content of any business module in the target business scenario changes, the training data of the second natural language understanding model corresponding to the business module is updated according to the changed business content, and the second natural language understanding model corresponding to the business module is retrained based on the updated training data.

6. The identification method according to claim 1, characterized in that, The step of using the first natural language understanding model to identify the estimated intent indicated by the target recognition sentence includes: Using the first natural language understanding model to determine the possibility score of the intent indicated by the target recognition sentence being the intent of each business module and the possibility score of the intent indicated by the target recognition sentence being the intent for supplementing entity conditions; The intent with the highest possibility score is determined as the estimated intent.

7. The identification method of claim 1, wherein, The step of using the second target natural language understanding model to identify the real intent indicated by the target recognition sentence from the business intents in the second target natural language understanding model includes: Using the second target natural language understanding model to determine the possibility score of the intent indicated by the target recognition sentence being each business intent; If the possibility score of the service intention with the highest possibility score is less than or equal to a preset score threshold, and a difference between the possibility score of the service intention with the highest possibility score and the possibility score of the service intention with the second highest possibility score is greater than a preset difference threshold, the service intention with the highest possibility score is determined as the real intention.

8. A device for identifying business intent, characterized in that, The recognition device comprises: An acquisition module configured to acquire a target recognition sentence in a current consultation dialogue; A first recognition module configured to recognize, using a first natural language understanding model, an estimated intention indicated by the target recognition sentence; A screening module configured to screen, from each second natural language understanding model, a second target natural language understanding model corresponding to a business module of the estimated intention indicated by the target recognition sentence; wherein the business content of a target business scenario is pre-divided into a plurality of business modules; the second natural language understanding model corresponding to the intention of each business module is trained by corpus data related to each business module; and the first natural language understanding model is trained by corpus data extracted from the corpus data related to each business module; A second recognition module configured to recognize, using the second target natural language understanding model, a real intention indicated by the target recognition sentence from each business intention in the second target natural language understanding model; The recognition device further comprises a third recognition module; the third recognition module is configured to: If the estimated intention is an intention for supplementing an entity condition, recognize, using the second target natural language understanding model corresponding to the current consultation dialogue, a real entity condition contained in the target recognition sentence; wherein the second target natural language understanding model corresponding to the current consultation dialogue refers to the second target natural language understanding model corresponding to the first real intention determined by other consultation sentences before the target recognition sentence in the current consultation dialogue; Determine, according to a knowledge graph, whether there is an association relationship between the real entity condition and the first real intention; wherein the knowledge graph comprises a plurality of business intentions, at least one optional business reply result corresponding to each business intention, an entity condition associated with each business intention corresponding to a plurality of optional business reply results, and a guide script corresponding to each entity condition; If there is an association relationship, determine, according to the real entity condition, whether a business reply result for the first real intention can be obtained using the knowledge graph, and feed back the business reply result for the first real intention or the guide script to the user according to the determination result.

9. An electronic device, comprising: The processor, the memory and the bus, the memory stores machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, the machine readable instructions are executed by the processor to execute the steps of the business intention recognition method in any one of claims 1 to 7. ​

Images