A noise recognition method and device, electronic equipment and readable storage medium

Abstract

The application provides a noise identification method and device, electronic equipment and a readable storage medium. In the process of broadcasting the first reply content of the first consultation dialogue proposed by a user, the first consultation dialogue is identified as noise according to a noise identification model. In the case that the first consultation dialogue is not noise, in response to a second consultation dialogue proposed by the user, a natural language understanding model is used to determine the dialogue intention of the second consultation dialogue, the possibility score of the second consultation dialogue having the dialogue intention and / or the second entity information carried by the second consultation dialogue. Whether the second consultation dialogue is a noise dialogue is determined according to the possibility score, the intention type of the dialogue intention and / or the second entity information. If yes, the second consultation dialogue is ignored, and the first reply content is continued to be broadcast. In this way, the voice interaction process between the user and the artificial intelligence dialogue robot can be avoided from being interrupted by the noise dialogue, the business handling efficiency is improved, and the business handling time is reduced.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a noise recognition method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] With the continuous development of artificial intelligence technology, the application of AI chatbots has become increasingly widespread and is penetrating into all aspects of life. As a result, people's expectations and requirements for AI chatbots are also increasing. People hope that AI chatbots can answer a wider range of questions and handle more complex business.

[0003] Although the technology of AI chatbots has been continuously improved and some simple background noise can be filtered out during the conversation, the filtering effect is not ideal. The voices of people and broadcasts around the user can still be received by the AI ​​chatbot, interrupting the content being broadcast by the current chatbot. This forces the user to ask questions again, increases the frequency of interaction between the user and the AI ​​chatbot, and delays the user's business processing time. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a noise recognition method, device, electronic device and readable storage medium, which can determine whether the received second consultation dialogue is a noise dialogue during the process of replying to users, so as to avoid the interference of noise dialogue on the voice interaction process, improve business processing efficiency and reduce business processing time.

[0005] This application provides a noise recognition method, the method comprising:

[0006] During the process of broadcasting the first response to the first inquiry dialogue raised by the user, noise recognition is performed on the first inquiry dialogue based on the noise recognition model;

[0007] If the first consultation dialogue is identified as non-noise by the noise recognition model, in response to the second consultation dialogue proposed by the user, the natural language understanding model is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue.

[0008] Based on the probability score, the intent type of the dialogue intent, and / or the second entity information, determine whether the second consultation dialogue is a noisy dialogue;

[0009] If so, ignore the second consultation dialogue and continue broadcasting the first response.

[0010] In one possible implementation, determining whether the second consultation dialogue is a noisy dialogue based on the probability score, the intent type of the dialogue intent, and / or the second entity information includes:

[0011] If the probability score of the second consultation dialogue having the dialogue intention is less than a preset score threshold, or if the probability score of the second consultation dialogue having the dialogue intention is less than the preset score threshold and the difference between it and the second high probability score is less than a preset difference threshold, then the second consultation dialogue is determined to be a noisy dialogue.

[0012] If the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is greater than or equal to a preset score threshold, or if the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then the second consultation dialogue is determined to be a non-noise dialogue.

[0013] If the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is greater than or equal to a preset score threshold, or if the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then based on the second entity information, it is determined whether the second consultation dialogue is a noise dialogue.

[0014] In one possible implementation, determining whether the second consultation dialogue is a noisy dialogue based on the second entity information includes:

[0015] Determine whether the second entity information carries the conditions for extracting the new entity from the second consultation dialogue;

[0016] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0017] If so, determine whether the second consultation dialogue is a noisy dialogue based on the newly added entity conditions.

[0018] In one possible implementation, determining whether the second consultation dialogue is a noisy dialogue based on the newly added entity condition includes:

[0019] Determine whether the newly added entity conditions are related to the business intent of the first consultation dialogue;

[0020] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0021] If so, using the newly added entity conditions, by searching the business knowledge graph, determine the content of the second response to the second consultation dialogue, and determine whether the content of the second response is consistent with the content of the first response;

[0022] If they match, the second consultation dialogue is determined to be a noisy dialogue;

[0023] If there is no discrepancy, the second consultation dialogue is determined to be a non-noise dialogue.

[0024] In one possible implementation, determining the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue using a natural language understanding model includes:

[0025] The probability score between the second consultation dialogue and each candidate intent is determined using a natural language understanding model, and the second entity information carried by the second consultation dialogue is extracted using the natural language understanding model.

[0026] The candidate intent with the highest probability score is determined as the dialogue intent of the second consultation dialogue.

[0027] This application embodiment also provides a noise recognition device, the device comprising:

[0028] The first noise detection module is used to identify noise in the first consultation dialogue based on the noise recognition model during the process of broadcasting the first reply content of the first consultation dialogue raised by the user.

[0029] The intent determination module is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue in response to the second consultation dialogue proposed by the user, when the first consultation dialogue is identified as non-noise by the noise recognition model.

[0030] The second noise detection module is used to determine whether the second consultation dialogue is a noise dialogue based on the probability score, the intent type of the dialogue intent, and / or the second entity information.

[0031] The continuous broadcast module is used to ignore the second consultation dialogue and continue broadcasting the first reply content if the answer is yes.

[0032] In one possible implementation, when the second noise detection module determines whether the second consultation dialogue is a noisy dialogue based on the probability score, the intent type of the dialogue intent, and / or the second entity information, the second noise detection module is used to:

[0033] If the probability score of the second consultation dialogue having the dialogue intention is less than a preset score threshold, or if the probability score of the second consultation dialogue having the dialogue intention is less than the preset score threshold and the difference between it and the second high probability score is less than a preset difference threshold, then the second consultation dialogue is determined to be a noisy dialogue.

[0034] If the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is greater than or equal to a preset score threshold, or if the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then the second consultation dialogue is determined to be a non-noise dialogue.

[0035] If the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is greater than or equal to a preset score threshold, or if the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then based on the second entity information, it is determined whether the second consultation dialogue is a noise dialogue.

[0036] In one possible implementation, when the second noise detection module is used to determine whether the second consultation dialogue is a noise dialogue based on the second entity information, the second noise detection module is used to:

[0037] Determine whether the second entity information carries the conditions for extracting the new entity from the second consultation dialogue;

[0038] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0039] If so, determine whether the second consultation dialogue is a noisy dialogue based on the newly added entity conditions.

[0040] In one possible implementation, when the second noise detection module is used to determine whether the second consultation dialogue is a noisy dialogue based on the newly added entity conditions, the second noise detection module is used to:

[0041] Determine whether the newly added entity conditions are related to the business intent of the first consultation dialogue;

[0042] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0043] If so, using the newly added entity conditions, by searching the business knowledge graph, determine the content of the second response to the second consultation dialogue, and determine whether the content of the second response is consistent with the content of the first response;

[0044] If they match, the second consultation dialogue is determined to be a noisy dialogue;

[0045] If there is no discrepancy, the second consultation dialogue is determined to be a non-noise dialogue.

[0046] In one possible implementation, when the intent determination module is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue using a natural language understanding model, the intent determination module is configured to:

[0047] The probability score between the second consultation dialogue and each candidate intent is determined using a natural language understanding model, and the second entity information carried by the second consultation dialogue is extracted using the natural language understanding model.

[0048] The candidate intent with the highest probability score is determined as the dialogue intent of the second consultation dialogue.

[0049] 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 noise recognition method described above are performed.

[0050] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the noise recognition method described above.

[0051] The noise recognition method, apparatus, electronic device, and readable storage medium provided in this application, during the process of broadcasting the first response content of a first consultation dialogue proposed by a user, perform noise recognition on the first consultation dialogue based on a noise recognition model; if the first consultation dialogue is identified as non-noise by the noise recognition model, in response to a second consultation dialogue proposed by the user, a natural language understanding model is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having a dialogue intent, and / or the second entity information carried by the second consultation dialogue; based on the probability score, the intent type of the dialogue intent, and / or the second entity information, it is determined whether the second consultation dialogue is a noisy dialogue; if so, the second consultation dialogue is ignored, and the first response content is continued to be broadcast. In this way, it is possible to avoid noise dialogue in the scenario interrupting the voice interaction process between the user and the artificial intelligence dialogue robot, improve business processing efficiency, and reduce business processing time.

[0052] 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

[0053] 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.

[0054] Figure 1 A flowchart illustrating a noise recognition method provided in an embodiment of this application;

[0055] Figure 2 This is a schematic diagram of a noise dialogue recognition process provided in an embodiment of this application;

[0056] Figure 3 This is a schematic diagram of a dialogue interaction process provided in an embodiment of this application;

[0057] Figure 4 This is a schematic diagram of the structure of a noise recognition device provided in an embodiment of this application;

[0058] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0059] 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.

[0060] Research has found that although the technology of AI chatbots can filter out some simple background noise during conversations, the filtering effect is not ideal. Voices from people around the user and broadcasts can still be heard by the AI ​​chatbot, interrupting the content being broadcast. This forces the user to ask questions again, increasing the frequency of interaction between the user and the AI ​​chatbot and delaying the user's business processing time.

[0061] Based on this, this application provides a noise recognition method to avoid interruptions in the voice interaction process with the AI ​​chatbot by noise, thereby improving the interaction efficiency between the user and the AI ​​chatbot and ultimately improving business processing efficiency.

[0062] Please see Figure 1 , Figure 1 This is a flowchart illustrating a noise recognition method provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the noise recognition method includes:

[0063] S101. During the process of broadcasting the first response to the first consultation dialogue raised by the user, noise recognition is performed on the first consultation dialogue based on the noise recognition model.

[0064] S102. If the first consultation dialogue is identified as non-noise by the noise recognition model, in response to the second consultation dialogue proposed by the user, the natural language understanding model is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue.

[0065] S103. Based on the probability score, the intent type of the dialogue intent, and / or the second entity information, determine whether the second consultation dialogue is a noisy dialogue.

[0066] S104. If so, ignore the second consultation dialogue and continue to broadcast the first reply.

[0067] The noise recognition method proposed in this application involves the following steps during the process of an AI-powered dialogue broadcasting the first response to a user's first inquiry. First, a noise recognition model is used to identify noise in the first inquiry. If the first inquiry is found to be noise-free, and the user initiates a second inquiry, the method determines the intention behind the second inquiry, the probability score of that intention, and / or the second entity information carried by the second inquiry. To prevent noise from interfering with the user's voice interaction with the AI ​​dialogue robot, the method determines whether the second inquiry is noise based on the probability score of the user's intention, the type of the intention, and / or the second entity information. If the second inquiry is determined to be noise, it is ignored, and the first response continues to be broadcast. This avoids noise interrupting the voice interaction between the user and the AI ​​dialogue robot, improving efficiency and reducing processing time.

[0068] In step S101, during the process of broadcasting the first response to the first consultation dialogue raised by the user, the system identifies whether the first consultation dialogue is noise based on the trained noise recognition model.

[0069] The noise recognition model is pre-trained using multiple first sample consultation dialogues and first sample labels indicating whether each first sample consultation dialogue is noise. By training the pre-built noise recognition model, a noise recognition model capable of identifying noise in consultation dialogues is obtained.

[0070] Here, when the AI ​​robot receives the first consultation dialogue, it will respond according to the settings regardless of whether the first consultation dialogue is background noise. If the first consultation dialogue is background noise, and the user asks for a second consultation dialogue, it will inevitably interrupt the first response. If the first consultation dialogue is not background noise, and the user asks for a second consultation dialogue, it is necessary to determine whether the second consultation dialogue is background noise in order to avoid unnecessary interruptions.

[0071] In step S102, if the first consultation dialogue is identified as non-noise by the noise recognition model, in response to the second consultation dialogue proposed by the user, the natural language understanding model is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue.

[0072] The dialogue intent includes business intent and entity intent in a specific scenario. When the dialogue intent of the second consultation dialogue is an entity intent, it indicates that the second consultation dialogue represents a certain entity condition. In this case, it can also be said that the user wants to supplement the entity condition carried by the first consultation dialogue through the second consultation dialogue.

[0073] Specific scenarios may include one or more of the following: banking services, community services, and hospital visits.

[0074] When the specific scenario is a banking business transaction scenario, the dialogue intent can represent the services provided by the bank. Specifically, the dialogue intent may include one or more of the following: deposit_processing_processing procedure, deposit_agency_processing materials, withdrawal_processing_processing procedure, withdrawal_agency_processing materials, loan_processing_processing procedure, and loan_agency_processing materials.

[0075] When the specific scenario is a community business processing scenario, the dialogue intent can represent the services provided by the community. Specifically, the dialogue intent may include one or more of the following: unemployment registration - processing - processing procedure, unemployment registration - agency - processing materials, residence permit - processing - processing procedure, residence permit agency - processing materials, employment and entrepreneurship certificate - application - application procedure, and employment and entrepreneurship certificate - agency - processing materials.

[0076] Natural Language Processing (NLU) technology refers to the technology of communicating between natural language and computers, also known as computational linguistics. On one hand, it is a branch of language information processing; on the other hand, it is one of the core topics of artificial intelligence (AI).

[0077] A Natural Language Understanding (NLU) model is a trained model that can recognize the intent indicated by natural language. It is mainly used to identify the intent corresponding to natural language. For example, the BERT model can be used. Here, multiple second-sample consultation dialogues and the second intent label corresponding to each second-sample consultation dialogue can be used to train the BERT model to obtain the NLU model.

[0078] NLU models can be built for specific business scenarios, such as banking, community services, and hospital visits. A pre-trained NLU model can identify entity information (i.e., entity words) carried in the consultation dialogue (the expressions used by users when inquiring about relevant business issues) to determine various business intentions related to the business scenario. For example, in a community service scenario, business intentions include: "unemployment registration process," "residence permit application process," and "employment and entrepreneurship certificate application materials." Furthermore, the pre-trained NLU model can extract entity information carried in the consultation dialogue during the identification process.

[0079] Here, the first response includes the result of the first consultation dialogue or the guiding dialogue generated based on the first consultation dialogue.

[0080] In one implementation, step S102 includes: using a natural language understanding model to determine the probability score between the second consultation dialogue and each candidate intent, and using the natural language understanding model to extract the second entity information carried by the second consultation dialogue; and determining the candidate intent with the highest probability score as the dialogue intent of the second consultation dialogue.

[0081] In this step, after obtaining the second consultation dialogue proposed by the user, the second consultation dialogue is input into a pre-trained natural language understanding model. The natural language understanding model is used to determine the probability score of the second consultation dialogue having any candidate intent. Here, the probability score represents the likelihood that the second consultation dialogue is any candidate intent; the higher the probability score, the higher the probability that the second consultation dialogue is that candidate intent. Therefore, the candidate intent with the highest probability score among multiple candidate intents can be determined as the dialogue intent possessed by the second consultation dialogue. The range of the probability score is (0, 1).

[0082] Simultaneously, a natural language understanding model is used to extract second entity information from the second consultation dialogue; here, the second entity information includes condition labels that can characterize whether the second consultation dialogue carries entity conditions and / or entity conditions carried in the second consultation dialogue.

[0083] In step S103, the possibility of the second consultation dialogue having a dialogue intention score, the intention type of the dialogue intention of the second consultation dialogue and / or the second entity information can be used to further determine whether the second consultation dialogue is a noisy dialogue.

[0084] Here, the types of intent in a dialogue include business intent and entity intent.

[0085] In one embodiment, step S103 includes: step S1031: if the probability score of the second consultation dialogue having the dialogue intention is less than a preset score threshold, or the probability score of the second consultation dialogue having the dialogue intention is less than the preset score threshold and the difference between it and the second high probability score is less than a preset difference threshold, then the second consultation dialogue is determined to be a noisy dialogue.

[0086] In this step, the probability score of the second consultation dialogue having a dialogue intention is compared with a preset score threshold. If the probability score of having a dialogue intention is less than the preset score threshold, it can be said that the second consultation dialogue does not actually have any candidate intention. At this time, the second consultation dialogue can be determined to be a noisy dialogue.

[0087] Alternatively, if the second consultation dialogue has a business intent, but the probability score of the second consultation dialogue having a business intent is less than a preset score threshold, the difference between the probability score of the second consultation dialogue having a business intent and the second highest probability score is compared with a preset difference threshold. If the difference between the two is still less than the preset difference threshold, it can also be said that the second consultation dialogue does not actually have any candidate intent. At this time, the second consultation dialogue can be determined to be a noisy dialogue.

[0088] Among them, the dialogue intent of the second consultation dialogue is the candidate dialogue intent with the highest probability score among the two consultation dialogues; the intent types of the candidate dialogue intent include business intent and entity intent in a specific scenario; when the dialogue intent of the second consultation dialogue is an entity intent, it indicates that the second consultation dialogue represents a certain entity condition, which can also indicate that the user wants to supplement the entity condition carried by the first consultation dialogue through the second consultation dialogue.

[0089] 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. The sources of the preset score threshold and preset difference threshold are as follows: After training the NLU intent model, it is tested using a large number of test sets (the test sets contain test consultation dialogues and test intent labels corresponding to each test consultation dialogue). The test results mainly include the scores corresponding to the highest-scoring business intent and the second-highest-scoring business intent matched by the test consultation dialogue in the NLU model, and the difference between the two. Where the highest-scoring dialogue intent is the same as the intent indicated by the test intent label, it is considered a correct recognition; otherwise, it is considered an incorrect recognition.

[0090] Statistical results show that when the highest dialogue intent score is greater than or equal to 0.7, the credibility of the intent result is relatively high and can be considered a credible result.

[0091] When the highest-scoring dialogue intent score is less than 0.7, further calculation is performed on the difference between the highest-scoring and second-highest-scoring dialogue intents in each training sample group. The calculated differences for each group are statistically analyzed and categorized to determine an effective difference threshold. There are two categorization methods: one is based on total values, such as [0,1), [0.1,1), [0.2,1); the other is based on intervals, such as [0,0.1), [0.1,0.2), [0.2,0.3).

[0092] Statistical analysis revealed that when the difference between the highest-scoring and second-highest-scoring dialogue intent scores is between 0.2 and 0.3, the number of misidentifications of unreliable results as reliable results in a test sample is controlled to within 3, 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 are kept within an acceptable range. Therefore, when the highest-scoring intent score is less than 0.7, or when the difference between the highest-scoring and second-highest-scoring dialogue intent scores is greater than or equal to 0.25, the highest-scoring dialogue intent is considered a reliable result.

[0093] Step S1032: If the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is greater than or equal to a preset score threshold, or if the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then the second consultation dialogue is determined to be a non-noise dialogue.

[0094] In this step, when the intent of the second consultation dialogue is a business intent, if the probability score of the second consultation dialogue having a business intent is greater than or equal to a preset score threshold, or if the probability score of the second consultation dialogue having a business intent is less than the preset score threshold and the difference between it and the second high probability score is greater than or equal to a preset difference threshold, it can be said that the second consultation dialogue has an actual business intent. At this time, it can be considered that the user wants to consult relevant information about the second business intent through the second consultation dialogue. Therefore, the second consultation dialogue is determined to be a non-noise dialogue, and the artificial intelligence dialogue robot needs to respond to the second consultation dialogue.

[0095] Step S1033: If the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is greater than or equal to a preset score threshold, or if the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is less than the preset score threshold and the difference between the second and the second high probability score is greater than or equal to a preset difference threshold, then determine whether the second consultation dialogue is a noise dialogue based on the second entity information.

[0096] In this step, when the dialogue intent of the second consultation dialogue is an entity intent, if the probability score of the second consultation dialogue having the entity intent is greater than or equal to a preset score threshold, or if the probability score of the second consultation dialogue having the entity intent is less than the preset score threshold and the difference between it and the second high probability score is greater than or equal to a preset difference threshold, it indicates that the second consultation dialogue can supplement the entity conditions carried by the first consultation dialogue.

[0097] However, in both of the above situations, the user may provide incorrect entity conditions, that is, the entity conditions extracted from the second consultation dialogue cannot supplement the entity conditions of the first consultation dialogue; or in other words, no entity conditions are actually extracted from the second consultation dialogue, so they cannot supplement the entity conditions of the first consultation dialogue. Therefore, even if the second consultation dialogue has entity intent, it is still necessary to combine the second entity information of the second consultation dialogue to determine whether the second consultation dialogue is a noisy dialogue.

[0098] In one implementation, determining whether the second consultation dialogue is a noisy dialogue based on the second entity information includes: determining whether the second entity information carries a newly added entity condition extracted from the second consultation dialogue; if not, determining that the second consultation dialogue is a noisy dialogue; if so, determining whether the second consultation dialogue is a noisy dialogue based on the newly added entity condition.

[0099] In this step, if there are new entity conditions in the second consultation dialogue, the second entity information carried by the second consultation dialogue should include not only the conditional label, but also the entity conditions carried in the second consultation dialogue; determine whether the second entity information of the second consultation dialogue carries the new entity conditions extracted from the second consultation dialogue; if the second entity information does not carry the new entity conditions, then the second consultation dialogue is determined to be a noisy dialogue.

[0100] If the second entity information carries new entity conditions, there are two possibilities: first, the new entity conditions carried by the second entity information can be used to supplement the entity conditions carried by the first consultation dialogue; second, the new entity conditions carried by the second entity information cannot supplement the entity conditions carried by the first consultation dialogue. Therefore, even if the second entity information carries new entity conditions, it is still necessary to determine whether the second consultation dialogue is a noisy dialogue based on the new entity conditions.

[0101] In one implementation, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating a noise dialogue recognition process provided in an embodiment of this application. Figure 2 As shown, determining whether the second consultation dialogue is a noisy dialogue based on the newly added entity conditions includes:

[0102] Step S201: Determine whether the newly added entity condition is related to the business intent of the first consultation dialogue.

[0103] In this step, it is determined whether the newly added entity condition is related to the business intent of the first consultation dialogue, that is, whether the newly added entity condition can serve as an entity condition that triggers the first business intent.

[0104] Step S202: If not, determine that the second consultation dialogue is a noisy dialogue.

[0105] In this step, if the newly added entity conditions are not related to the business intent of the first consultation dialogue, that is, the newly added entity conditions carried by the second consultation dialogue cannot further supplement the entity conditions required by the first consultation dialogue, then the second consultation dialogue is identified as a noise dialogue.

[0106] Step S203: If yes, using the newly added entity conditions, determine the second response content of the second consultation dialogue by searching the business knowledge graph, and determine whether the second response content is consistent with the first response content.

[0107] In this step, if the newly added entity conditions are related to the business intent of the first consultation dialogue, that is, the newly added entity conditions carried by the second consultation dialogue can further supplement the entity conditions required by the first consultation dialogue to obtain a more accurate response result; at this time, the newly added entity conditions can be used to determine the second response content for answering the second consultation dialogue by searching the pre-generated business knowledge graph, and then it can be judged again whether the second response content is consistent with the first response content.

[0108] Step S204: If they match, determine that the second consultation dialogue is a noisy dialogue.

[0109] Step S205: If there is no consistency, determine that the second consultation dialogue is a non-noise dialogue.

[0110] If there is a discrepancy in this step, it means that the user has raised a new question. In this case, the second consultation dialogue can be considered a valid consultation dialogue, and a response should be given to the second consultation dialogue.

[0111] For example, a user's first inquiry to the AI ​​chatbot is, "I would like to inquire about the ID card application process." To clarify the user's intended business, the AI ​​chatbot guides the user to supplement the required physical conditions for applying for an ID card through feedback prompts. At this point, the AI ​​chatbot provides the user with the prompts, "Which of the following services do you wish to apply for: 1. New ID card application; 2. ID card replacement?" Based on the user's response to the prompts, the AI ​​chatbot clarifies that the user intends to "replace an ID card" and provides the user with the first response, "ID card replacement process XXX."

[0112] If, at this point, the user raises a second inquiry, "New ID Card Application," the AI ​​chatbot recognizes "New ID Card Application" as an entity intent, extracts the new entity condition "New ID Card Application," and then uses this new entity condition to determine the second response to "New ID Card Application" by searching the business knowledge graph. The second response is "The process for applying for a new ID card is YYY." Since the first response is inconsistent with the second response, the second inquiry is determined to be a non-noise dialogue, and the "The process for applying for a new ID card is YYY" is broadcast to the user.

[0113] If, at this point, the user raises a second inquiry, "ID card replacement," the AI ​​chatbot recognizes "ID card replacement" as the entity intent, extracts the new entity condition "ID card replacement," and then uses this new entity condition to determine the second response to "ID card replacement" by searching the business knowledge graph. The second response is "The process for applying for a new ID card is XXX." Since the first and second responses are identical, the second inquiry is considered background noise, and the AI ​​chatbot does not need to respond to it; that is, the second inquiry does not interrupt the AI ​​chatbot's recitation of the first response. Here, the business knowledge graph is pre-structured based on the business environment of the AI ​​chatbot. The business knowledge graph is a network structure that shows the relationship between business intents and entity conditions. It includes each entity condition, the corresponding guiding question for each entity condition (i.e., the guiding question the AI ​​chatbot should provide when the user's inquiry lacks certain entity conditions), and the response result to be given after the user completes all entity conditions.

[0114] In step S104, if it is determined that the second consultation dialogue is a noisy dialogue, then the second consultation dialogue is ignored and the first response content is continued to be broadcast.

[0115] Furthermore, the noise recognition method further includes: step S105 (as shown in the image). Figure 1 As shown), if the second consultation dialogue is determined to be a non-noise dialogue, it means that a response is needed for the second consultation dialogue raised by the user. At this time, based on the dialogue intent of the second consultation dialogue and / or the second entity information extracted from the second consultation dialogue, the second response content for the second consultation dialogue can be determined by searching the business knowledge graph, and the second response content can be broadcast to the user.

[0116] Here, if the second response content of the second consultation dialogue has already been obtained through the business knowledge graph during the process of determining whether the second consultation dialogue is a noisy dialogue, then there is no need to obtain it again; the already obtained second response content can be broadcast directly.

[0117] Please see Figure 3 , Figure 3 This is a schematic diagram of a dialogue interaction process provided in an embodiment of this application. Figure 3As shown, step 301: broadcast the first response of the first consultation dialogue to the user; step 302: determine the dialogue intent and / or second entity information of the second consultation dialogue received by the user through a Natural Language Understanding (NLU) model; step 303: determine whether the second consultation dialogue has a credible dialogue intent; if yes, proceed to step 304; if no, proceed to step 309; step 304: determine whether the dialogue intent of the second consultation dialogue is a business intent or an entity intent; if it is a business intent, proceed to step 310; if it is an entity intent, proceed to step 305; step 305: determine whether the second entity information carries any newly added entity conditions. If not, proceed to step 309; if yes, proceed to step 306: determine whether the newly added entity conditions are related to the business intent of the first consultation dialogue; if not, proceed to step 309; if yes, proceed to step 307; step 307: using the newly added entity conditions carried by the second entity information, determine the second response content of the second consultation dialogue by searching the business knowledge graph; step 308: determine whether the second response content is consistent with the first response content; if consistent, proceed to step 309; if inconsistent, proceed to step 310; step 309: determine that the second consultation dialogue is a noisy dialogue; step 310: determine that the second consultation dialogue is a non-noisy dialogue.

[0118] The noise recognition method provided in this application, during the process of broadcasting the first response to a user's first consultation dialogue, performs noise recognition on the first consultation dialogue based on a noise recognition model. If the first consultation dialogue is identified as non-noise by the noise recognition model, in response to a user's second consultation dialogue, a natural language understanding model is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having a dialogue intent, and / or the second entity information carried by the second consultation dialogue. Based on the probability score, the intent type of the dialogue intent, and / or the second entity information, it is determined whether the second consultation dialogue is a noisy dialogue. If so, the second consultation dialogue is ignored, and the first response content continues to be broadcast. This avoids noisy dialogues in the scenario from interrupting the voice interaction process between the user and the AI ​​chatbot, improving business processing efficiency and reducing business processing time.

[0119] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a noise recognition device provided in an embodiment of this application. Figure 4 As shown, the noise recognition device 400 includes:

[0120] The first noise detection module 410 is used to identify noise in the first consultation dialogue based on the noise recognition model during the process of broadcasting the first reply content of the first consultation dialogue raised by the user.

[0121] The intent determination module 420 is used to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue in response to the second consultation dialogue proposed by the user, when the first consultation dialogue is identified as non-noise by the noise recognition model.

[0122] The second noise judgment module 430 is used to determine whether the second consultation dialogue is a noise dialogue based on the probability score, the intent type of the dialogue intent and / or the second entity information.

[0123] The continuous broadcast module 440 is used to ignore the second consultation dialogue and continue broadcasting the first reply content if the condition is met.

[0124] Furthermore, when the second noise judgment module 430 determines whether the second consultation dialogue is a noisy dialogue based on the probability score, the intent type of the dialogue intent, and / or the second entity information, the second noise judgment module 430 is used to:

[0125] If the probability score of the second consultation dialogue having the dialogue intention is less than a preset score threshold, or if the probability score of the second consultation dialogue having the dialogue intention is less than the preset score threshold and the difference between it and the second high probability score is less than a preset difference threshold, then the second consultation dialogue is determined to be a noisy dialogue.

[0126] If the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is greater than or equal to a preset score threshold, or if the second consultation dialogue has a business intention and the probability score of the second consultation dialogue having a business intention is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then the second consultation dialogue is determined to be a non-noise dialogue.

[0127] If the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is greater than or equal to a preset score threshold, or if the dialogue intent of the second consultation dialogue is an entity intent and the probability score of the second consultation dialogue having an entity intent is less than the preset score threshold and the difference between the second consultation dialogue and the second high probability score is greater than or equal to a preset difference threshold, then based on the second entity information, it is determined whether the second consultation dialogue is a noise dialogue.

[0128] Furthermore, when the second noise detection module 430 determines whether the second consultation dialogue is a noise dialogue based on the second entity information, the second noise detection module 430 is used to:

[0129] Determine whether the second entity information carries the conditions for extracting the new entity from the second consultation dialogue;

[0130] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0131] If so, determine whether the second consultation dialogue is a noisy dialogue based on the newly added entity conditions.

[0132] Furthermore, when the second noise detection module 430 determines whether the second consultation dialogue is a noise dialogue based on the newly added entity conditions, the second noise detection module 430 is used to:

[0133] Determine whether the newly added entity conditions are related to the business intent of the first consultation dialogue;

[0134] If not, the second consultation dialogue is determined to be a noisy dialogue;

[0135] If so, using the newly added entity conditions, by searching the business knowledge graph, determine the content of the second response to the second consultation dialogue, and determine whether the content of the second response is consistent with the content of the first response;

[0136] If they match, the second consultation dialogue is determined to be a noisy dialogue;

[0137] If there is no discrepancy, the second consultation dialogue is determined to be a non-noise dialogue.

[0138] Furthermore, when the intent determination module 420 uses a natural language understanding model to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue, the intent determination module 420 is used to:

[0139] The probability score between the second consultation dialogue and each candidate intent is determined using a natural language understanding model, and the second entity information carried by the second consultation dialogue is extracted using the natural language understanding model.

[0140] The candidate intent with the highest probability score is determined as the dialogue intent of the second consultation dialogue.

[0141] The noise recognition device provided in this application embodiment, during the process of broadcasting the first response content of a first consultation dialogue proposed by a user, performs noise recognition on the first consultation dialogue based on a noise recognition model; if the first consultation dialogue is identified as non-noise by the noise recognition model, in response to a second consultation dialogue proposed by the user, it uses a natural language understanding model to determine the dialogue intent of the second consultation dialogue, the probability score of the second consultation dialogue having a dialogue intent, and / or the second entity information carried by the second consultation dialogue; based on the probability score, the intent type of the dialogue intent, and / or the second entity information, it determines whether the second consultation dialogue is a noisy dialogue; if so, the second consultation dialogue is ignored, and the first response content is continued to be broadcast. In this way, it can avoid noise dialogue in the scenario from interrupting the voice interaction process between the user and the artificial intelligence dialogue robot, improve business processing efficiency, and reduce business processing time.

[0142] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device 500 includes a processor 510, a memory 520, and a bus 530.

[0143] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 and the memory 520 communicate via the bus 530. When the machine-readable instructions are executed by the processor 510, they can perform the operations described above. Figure 1 The steps of the noise recognition method in the illustrated method embodiment can be found in the method embodiment for specific implementation methods, which will not be repeated here.

[0144] 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 the noise recognition method in the illustrated method embodiment can be found in the method embodiment for specific implementation, and will not be repeated here. Those skilled in the art will clearly 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.

[0145] 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.

[0146] 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.

[0147] 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.

[0148] 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.

[0149] 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.

[0150] 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 of noise identification, characterized by, The method comprises: In the process of broadcasting the first reply content of the first consultation dialogue proposed by the user, the first consultation dialogue is identified as noise according to a noise identification model; the first reply content comprises a response result of the first consultation dialogue or a guided dialogue generated according to the first consultation dialogue; In the case that the first consultation dialogue is identified as non-noise by the noise identification model, in response to a second consultation dialogue proposed by the user, a natural language understanding model is used to determine a dialogue intent possessed by the second consultation dialogue, a possibility score of the second consultation dialogue possessing the dialogue intent and / or second entity information carried by the second consultation dialogue; According to the possibility score, the intent type of the dialogue intent and / or the second entity information, it is determined whether the second consultation dialogue is a noise dialogue; the determination whether the second consultation dialogue is a noise dialogue according to the possibility score, the intent type of the dialogue intent and / or the second entity information comprises: if the possibility score of the second consultation dialogue possessing the dialogue intent is less than a preset score threshold, it is determined that the second consultation dialogue is a noise dialogue; if the dialogue intent possessed by the second consultation dialogue is a business intent and the possibility score of the second consultation dialogue possessing the business intent is greater than or equal to the preset score threshold, or the dialogue intent possessed by the second consultation dialogue is a business intent and the possibility score of the second consultation dialogue possessing the business intent is less than the preset score threshold and the difference between the possibility score and a second high possibility score is greater than or equal to a preset difference threshold, it is determined that the second consultation dialogue is a non-noise dialogue; if the dialogue intent possessed by the second consultation dialogue is an entity intent and the possibility score of the second consultation dialogue possessing the entity intent is greater than or equal to the preset score threshold, or the dialogue intent possessed by the second consultation dialogue is an entity intent and the possibility score of the second consultation dialogue possessing the entity intent is less than the preset score threshold and the difference between the possibility score and a second high possibility score is greater than or equal to a preset difference threshold, it is determined whether the second consultation dialogue is a noise dialogue according to the second entity information; If the second consultation dialogue is a noise dialogue, the second consultation dialogue is ignored and the first reply content is continued to be broadcast; wherein, in the case that the first consultation dialogue is identified as noise by the noise identification model, in response to a second consultation dialogue proposed by the user, the broadcasting of the first reply content is stopped.

2. The noise recognition method according to claim 1, characterized by, The determination whether the second consultation dialogue is a noise dialogue according to the second entity information comprises: It is determined whether the second entity information carries a newly added entity condition extracted from the second consultation dialogue; If not, it is determined that the second consultation dialogue is a noise dialogue; If yes, it is determined whether the second consultation dialogue is a noise dialogue according to the newly added entity condition.

3. The method of claim 2, wherein, The determination whether the second consultation dialogue is a noise dialogue according to the newly added entity condition comprises: It is determined whether the newly added entity condition has an association relationship with a business intent possessed by the first consultation dialogue; If not, it is determined that the second consultation dialogue is a noise dialogue; If yes, the second reply content of the second consultation dialogue is determined by searching the business knowledge graph using the added entity condition, and whether the second reply content is consistent with the first reply content is determined. If yes, the second consultation dialogue is determined as a noise dialogue. If no, the second consultation dialogue is determined as a non-noise dialogue.

4. The method of claim 1, wherein, The determination of the dialogue intent possessed by the second consultation dialogue, the possibility score of the second consultation dialogue having the dialogue intent, and / or the second entity information carried by the second consultation dialogue using the natural language understanding model comprises: The possibility score between the second consultation dialogue and each candidate intent is determined using the natural language understanding model, and the second entity information carried by the second consultation dialogue is extracted using the natural language understanding model. The candidate intent with the highest possibility score is determined as the dialogue intent possessed by the second consultation dialogue.

5. A noise identification device, characterized by, The device comprises: A first noise judgment module configured to perform noise identification on a first consultation dialogue according to a noise identification model during broadcasting of first reply content of the first consultation dialogue proposed by a user; the first reply content comprises a response result of the first consultation dialogue or a guide dialogue generated according to the first consultation dialogue; An intent determination module configured to, in a case where the first consultation dialogue is identified as a non-noise by the noise identification model, determine, in response to a second consultation dialogue proposed by the user, a dialogue intent possessed by the second consultation dialogue, a possibility score of the second consultation dialogue having the dialogue intent, and / or second entity information carried by the second consultation dialogue using a natural language understanding model. The second noise judgment module is configured to determine whether the second consultation dialogue is a noise dialogue according to the possibility score, the intent type of the dialogue intent, and / or the second entity information. The second noise judgment module is configured to: determine whether the second consultation dialogue carries a newly added entity condition extracted from the second consultation dialogue; 6. The device of claim 5, wherein if not, determine that the second consultation dialogue is a noise dialogue; if yes, determine whether the second consultation dialogue is a noise dialogue according to the newly added entity condition. The noise recognition device is further configured to, in a case where the first consultation dialogue is identified as noise by the noise recognition model, stop broadcasting the first reply content in response to a second consultation dialogue raised by the user. The second noise judgment module is configured to:

7. An electronic device, comprising: determine whether the second entity information carries a newly added entity condition extracted from the second consultation dialogue; if not, determine that the second consultation dialogue is a noise dialogue; 8. A computer-readable storage medium, characterized in that, if yes, determine whether the second consultation dialogue is a noise dialogue according to the newly added entity condition. The noise recognition device comprises: 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 and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to perform the steps of the noise recognition method according to any one of claims 1 to 4. The computer-readable storage medium stores a computer program, and the computer program is executed by the processor to perform the steps of the noise recognition method according to any one of claims 1 to 4.

Images