Voice interaction system optimization method based on data backflow and related equipment

Abstract

The invention discloses a voice interaction system optimization method based on data backflow and related equipment, and relates to the technical field of artificial intelligence and vehicle-mounted man-machine interaction, and the method comprises the steps: obtaining a user verbal skill, a first semantic slot position and a first falling domain intention, and determining a standard verbal skill in a preset standard statement library, and a second semantic slot position and a second falling domain intention corresponding to the standard verbal skill, and setting a label corresponding to the verbal skill of the user based on a preset label setting rule, the similarity, the first semantic slot position, the first falling domain intention, the second semantic intention and the second falling domain intention so as to optimize the voice interaction system. According to the method, the user verbal skill is acquired from the vehicle-mounted voice cloud log, the user verbal skill and the standard verbal skill in the standard statement library are subjected to multi-dimensional matching, the real label is set for the user verbal skill according to the matching results of different dimensions and the preset label setting rule, the real label and the user verbal skill are used as training samples for data backflow, and the user verbal skill is obtained. Therefore, the voice interaction system is optimized.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and in-vehicle human-machine interaction technology, and in particular to a method and related equipment for optimizing voice interaction systems based on data feedback. Background Technology

[0002] With the continuous development of artificial intelligence technology, in-vehicle human-machine interaction systems have gradually become an important technical means to improve driving experience and road safety. Voice interaction, as one of the core functions of in-vehicle human-machine interaction systems, enables drivers to conveniently operate various vehicle functions, such as navigation, music playback, and phone calls, through voice commands while driving, thereby reducing the driver's reliance on physical buttons and lowering driving risks.

[0003] However, existing in-vehicle voice interaction systems still have some problems in practical applications. During the optimization process, these systems lack an effective data feedback mechanism, failing to fully utilize actual user interaction data to optimize system performance. This results in the system being unable to dynamically adjust based on user habits and preferences, and thus unable to meet the personalized needs of different users.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is related technology. Summary of the Invention

[0005] The main purpose of this application is to provide a method and related equipment for optimizing a voice interaction system based on data feedback, aiming to solve the technical problem of how to optimize a voice interaction system through a data feedback mechanism.

[0006] To achieve the above objectives, this application proposes a voice interaction system optimization method based on data feedback, which includes: In response to the voice interaction system optimization command, user speech A is obtained from the vehicle voice cloud log, and the first semantic slot and the first domain intent corresponding to the user speech A are obtained. Determine the standard statement D in the preset standard statement library that is most similar to the user statement A, and determine the second semantic slot and the second domain intent corresponding to the standard statement D. The similarity value between the user statement A and the standard statement D is the similarity value B. Based on the preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent, the tag corresponding to the user's utterance A is set; Based on the user's dialogue A and the tags, the voice interaction system is optimized to obtain an optimized voice interaction system.

[0007] In one embodiment, the step of setting the tag corresponding to the user utterance A based on the preset tag setting rules, the similarity B, the first semantic slot, the first placement intent, the second semantic intent, and the second placement intent further includes: Based on the first semantic slot and the first domain intention, the domain status corresponding to the user utterance A is determined, and the domain status includes not domained and domained; Determine whether the similarity B is greater than or equal to a preset similarity threshold C to obtain a first determination result; Determine whether the first semantic slot and the second semantic slot are the same to obtain a second determination result; determine whether the first domain intent and the second domain intent are the same to obtain a third determination result. Based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result, and the third judgment result, the tag corresponding to the user's dialogue A is set.

[0008] In one embodiment, the step of setting the tag corresponding to the user's dialogue A based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result, and the third judgment result further includes: If the domain status is domain-defined, the first judgment result is greater than or equal to, and the second judgment result and the third judgment result are both the same, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as domain-defined correctly. If the domain status is domain-incorrect, the first judgment result is greater than or equal to, and at least one of the second judgment result and the third judgment result is different, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as domain-incorrect. If the domain landing status is not landed and the first judgment result is greater than or equal to, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as incomplete domain landing; If the domain status is domain and the first judgment result is less than, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as an incorrect domain. If the domain status is not domained and the first judgment result is less than, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as correct rejection.

[0009] In one embodiment, the step of determining the standard statement D in the preset standard statement library that is most similar to the user's statement A further includes: Based on a preset deep learning model, the user's statement A and multiple standard statements in the preset standard statement library are converted into high-dimensional semantic vectors, respectively, to obtain the first high-dimensional semantic vector corresponding to the user's statement A and multiple sets of second high-dimensional semantic vectors corresponding to the preset standard statement library. The deep learning model is constructed based on word vectors and sentence vectors. Based on any one of the multiple sets of second high-dimensional semantic vectors, a similarity calculation is performed with the first high-dimensional semantic vector to obtain multiple sets of similarity values. The maximum value among the multiple sets of similarity values ​​is the similarity value B. Based on the multiple sets of similarity values, the standard statement D that is most similar to the user's statement A in the preset standard statement library is determined.

[0010] In one embodiment, before the step of converting the user's utterance A and multiple standard statements in the preset standard statement library into high-dimensional semantic vectors based on a preset deep learning model, the method further includes: Obtain sample data, wherein the true high-dimensional semantic vector corresponding to the sample data is the third high-dimensional semantic vector; The sample data is processed using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data; Determine whether the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector; If they are inconsistent, adjust the parameters of the current deep learning model. Based on the current deep learning model with the adjusted parameters, return to the step of using the current deep learning model to process the sample data and obtain the fourth high-dimensional semantic vector corresponding to the sample data, until the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector, and obtain the preset deep learning model.

[0011] In one embodiment, the step of optimizing the voice interaction system based on user utterance A and the tag to obtain an optimized voice interaction system further includes: Based on user utterance A and the aforementioned tags, the natural language understanding model in the voice interaction system is retrained to optimize the intent recognition and slot filling functions of the natural language understanding model. Based on the optimized natural language understanding model, an optimized voice interaction system is obtained.

[0012] Furthermore, to achieve the above objectives, this application also proposes a voice interaction system optimization device based on data feedback, the voice interaction system optimization device based on data feedback comprising: The acquisition module is used to respond to the voice interaction system optimization command, acquire user speech A from the vehicle voice cloud log, and acquire the first semantic slot and the first domain intent corresponding to the user speech A. The determining module is used to determine the standard statement D that is most similar to the user statement A in the preset standard statement library, and to determine the second semantic slot and the second domain intent corresponding to the standard statement D. The similarity value between the user statement A and the standard statement D is the similarity value B. The setting module is used to set the tag corresponding to the user's utterance A based on the preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent; An optimization module is used to optimize the voice interaction system based on user speech A and the tag, so as to obtain an optimized voice interaction system.

[0013] In one embodiment, the setting module further includes: The first determining unit is configured to determine the domain status corresponding to the user utterance A based on the first semantic slot and the first domain intention, wherein the domain status includes not domained and domained; The first judgment unit is used to determine whether the similarity B is greater than or equal to a preset similarity threshold C, and to obtain a first judgment result; The second judgment unit is used to determine whether the first semantic slot and the second semantic slot are the same, and to obtain a second judgment result; and to determine whether the first domain intention and the second domain intention are the same, and to obtain a third judgment result. The first setting unit is used to set the tag corresponding to the user's speech A based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result and the third judgment result.

[0014] In one embodiment, the setting module further includes: The second setting unit is used to set the tag corresponding to the user's speech A as "correct domain" based on the preset tag setting rules if the domain status is "domain", the first judgment result is greater than or equal to, and the second judgment result and the third judgment result are the same. The third setting unit is used to set the tag corresponding to the user's speech A as a domain error based on a preset tag setting rule if the domain status is a domain, the first judgment result is greater than or equal to, and at least one of the second judgment result and the third judgment result is different. The fourth setting unit is used to set the tag corresponding to the user's speech A as incomplete domain coverage if the domain coverage status is not completed and the first judgment result is greater than or equal to the domain coverage status, based on the preset tag setting rules. The fifth setting unit is used to set the tag corresponding to the user's speech A as an incorrect domain if the domain status is a domain and the first judgment result is less than. The sixth setting unit is used to set the tag corresponding to the user's speech A as correct rejection based on the preset tag setting rules if the domain status is not domained and the first judgment result is less than.

[0015] In one embodiment, the determining module further includes: The conversion unit is used to convert the user's utterance A and multiple standard statements in the preset standard statement library into high-dimensional semantic vectors based on a preset deep learning model, respectively, to obtain a first high-dimensional semantic vector corresponding to the user's utterance A and multiple sets of second high-dimensional semantic vectors corresponding to the preset standard statement library. The deep learning model is constructed based on word vectors and sentence vectors. The calculation unit is used to perform similarity calculations with the first high-dimensional semantic vector based on any one of the multiple sets of second high-dimensional semantic vectors, and obtain multiple sets of similarity values, wherein the maximum value among the multiple sets of similarity values ​​is the similarity value B; The second determining unit is used to determine, based on the multiple sets of similarity values, the standard statement D in the preset standard statement library that is most similar to the user's statement A.

[0016] In one embodiment, the determining module further includes: The acquisition unit is used to acquire sample data, wherein the real high-dimensional semantic vector corresponding to the sample data is the third high-dimensional semantic vector; The data processing unit is used to process the sample data using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data. The third judgment unit is used to determine whether the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector; The iterative training unit is used to adjust the parameters of the current deep learning model if there is a discrepancy. Based on the current deep learning model with adjusted parameters, it returns to the step of processing the sample data using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data, until the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector, thus obtaining the preset deep learning model.

[0017] In one embodiment, the optimization module further includes: An optimization unit is used to retrain the natural language understanding model in the voice interaction system based on user utterance A and the label, so as to optimize the intent recognition function and slot filling function of the natural language understanding model. The acquisition unit is used to obtain the optimized voice interaction system based on the optimized natural language understanding model.

[0018] Furthermore, to achieve the above objectives, this application also proposes a voice interaction system optimization device based on data feedback, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the voice interaction system optimization method based on data feedback as described above.

[0019] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the voice interaction system optimization method based on data feedback as described above.

[0020] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the data feedback-based voice interaction system optimization method described above.

[0021] One or more technical solutions proposed in this application have at least the following technical effects: This application proposes a voice interaction system optimization method and related equipment based on data feedback, relating to the fields of artificial intelligence and in-vehicle human-machine interaction technology. Compared to existing technologies, existing voice interaction systems lack an effective data feedback mechanism during optimization, failing to fully utilize actual user interaction data to optimize system performance. This results in the system being unable to dynamically adjust according to user habits and preferences, and thus unable to meet the personalized needs of different users. In this application, firstly, in response to a voice interaction system optimization command, user speech A is obtained from the in-vehicle voice cloud log, and the first semantic slot and the second semantic slot corresponding to user speech A are obtained. First, a landing intent is determined. Then, the standard speech D most similar to the user speech A in the preset standard speech library is determined, and the second semantic slot and the second landing intent corresponding to the standard speech D are determined. The similarity value between the user speech A and the standard speech D is similarity value B. Further, based on the preset tag setting rules, the similarity value B, the first semantic slot, the first landing intent, the second semantic intent and the second landing intent, the tag corresponding to the user speech A is set. Finally, based on the user speech A and the tag, the voice interaction system is optimized to obtain the optimized voice interaction system.

[0022] Understandably, this application obtains user dialogue from in-vehicle voice cloud logs, matches the user dialogue with standard dialogue in the standard dialogue library in multiple dimensions (semantic slots, similarity, and contextual intent), sets real labels for user dialogue based on the matching results of different dimensions and preset label setting rules, and uses real labels and user dialogue as training samples for data feedback to optimize the voice interaction system. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating an embodiment of the voice interaction system optimization method based on data feedback in this application. Figure 2 A simplified flowchart illustrating the data feedback-based voice interaction system optimization method provided in Embodiment 1 of this application; Figure 3 This is a flowchart illustrating Embodiment 2 of the voice interaction system optimization method based on data feedback in this application. Figure 4 This is a schematic diagram of the module structure of the voice interaction system optimization device based on data feedback according to an embodiment of this application; Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the data feedback-based voice interaction system optimization method in the embodiments of this application.

[0026] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0028] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0029] The main solution in this application embodiment is: In this embodiment, for ease of description, the following description will focus on the voice interaction system optimization device based on data feedback.

[0030] Due to the lack of an effective data feedback mechanism in the optimization process of existing voice interaction systems, they cannot fully utilize the actual user interaction data to optimize system performance. As a result, the system cannot dynamically adjust according to the user's usage habits and preferences, and cannot meet the personalized needs of different users.

[0031] This application provides a solution in which: first, in response to a voice interaction system optimization command, user speech A is obtained from the vehicle voice cloud log, and the first semantic slot and first domain intent corresponding to user speech A are obtained; then, a standard speech D most similar to user speech A in a preset standard speech library is determined, and the second semantic slot and second domain intent corresponding to standard speech D are determined, wherein the similarity value between user speech A and standard speech D is a similarity value B; further, based on preset tag setting rules, the similarity value B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent, a tag corresponding to user speech A is set; finally, based on user speech A and the tag, the voice interaction system is optimized to obtain an optimized voice interaction system.

[0032] Understandably, this application obtains user dialogue from in-vehicle voice cloud logs, matches the user dialogue with standard dialogue in the standard dialogue library in multiple dimensions (semantic slots, similarity, and contextual intent), sets real labels for user dialogue based on the matching results of different dimensions and preset label setting rules, and uses real labels and user dialogue as training samples for data feedback to optimize the voice interaction system.

[0033] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as a voice interaction system optimization device based on data feedback. The following description uses a voice interaction system optimization device based on data feedback as an example to illustrate this embodiment and the subsequent embodiments.

[0034] Based on this, embodiments of this application provide an optimization method for a voice interaction system based on data feedback, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the voice interaction system optimization method based on data feedback of this application.

[0035] In this embodiment, the voice interaction system optimization method based on data feedback includes steps S100~S400: Step S100: In response to the voice interaction system optimization command, obtain user speech A from the vehicle voice cloud log, and obtain the first semantic slot and the first domain intent corresponding to the user speech A. It should be noted that responding to a voice interaction system optimization instruction means that the system receives an instruction whose goal is to optimize the performance or functionality of the voice interaction system.

[0036] It should be noted that voice interaction systems typically require regular optimization to improve user experience, increase recognition accuracy, or streamline conversation flow. Optimization commands may be initiated by the system administrator or triggered automatically by the system based on its own operational status.

[0037] Retrieving user speech A from in-vehicle voice cloud logs refers to extracting specific user voice input (speech A) from in-vehicle voice interaction logs stored in the cloud.

[0038] It should be noted that the in-vehicle voice cloud log records the history of dialogue between the user and the in-vehicle voice interaction system, including the user's voice commands and system responses.

[0039] It should be noted that user dialogue A refers to the voice command or statement uttered by the user in a specific scenario. For example, the user might say, "Navigate to the nearest gas station."

[0040] Obtaining the first semantic slot corresponding to the user's utterance A means extracting key information (semantic slots) from the user's utterance A. This information is an important component of understanding the user's intent.

[0041] It should be noted that semantic slots refer to specific units of information extracted from a user's speech in natural language processing. For example, in the sentence "navigate to the nearest gas station," "gas station" is a semantic slot, indicating the user's desired destination.

[0042] It should be noted that the first semantic slot refers to the key information identified in user utterance A. For example, if the user says, "Set the temperature to 25 degrees," then "25 degrees" is the extracted semantic slot.

[0043] Obtaining the first domain intent corresponding to the user's utterance A means determining the intent category (domain intent) to which the user's utterance A belongs, that is, the specific task or goal that the user wants to accomplish through this sentence.

[0044] The domain intent refers to the domain or category to which the user's utterance expresses intent. For example, the domain intent of the phrase "navigate to the nearest gas station" might be "navigation".

[0045] The primary intent of a user's statement A is the main intent identified. For example, if a user says, "Play a Jay Chou song," the primary intent of this statement might be "play music."

[0046] Understandably, in this embodiment, after receiving the optimization instruction, the system extracts a sentence the user previously spoke from the cloud logs (user utterance A) and analyzes the key information (first semantic slot) and the task the user wants to complete (first domain intent) within that sentence. This process aims to optimize the performance of the voice interaction system, such as improving the accuracy of speech recognition, optimizing the dialogue flow, or enhancing the user experience.

[0047] Step S200: Determine the standard statement D in the preset standard statement library that is most similar to the user statement A, and determine the second semantic slot and the second domain intent corresponding to the standard statement D. The similarity value between the user statement A and the standard statement D is similarity value B. It should be noted that determining the standard statement D that is most similar to the user's statement A in the preset standard statement library means finding the standard statement D that is most similar to the user's statement A in the preset standard statement library.

[0048] The preset standard phrase library is a database containing a variety of standard phrases. These standard phrases are predefined by the system and are used to match and compare with the user's actual phrases.

[0049] Standard phrase D refers to the phrase in the standard phrase library that is most similar to the user's phrase A. For example, if the user's phrase A is "set the air conditioner temperature to 25 degrees", the standard phrase D might be "set the air conditioner temperature to 25 degrees".

[0050] It should be noted that determining the second semantic slot corresponding to the standard utterance D means extracting the key information (second semantic slot) from the standard utterance D.

[0051] The second semantic slot refers to the key information identified in the standard utterance D. For example, if the standard utterance D is "set the air conditioner temperature to 25 degrees", then "25 degrees" is the second semantic slot.

[0052] Determining the second domain intent corresponding to the standard phrase D means determining the intent category (second domain intent) to which the standard phrase D belongs.

[0053] The second contextual intent refers to the primary intent identified in the standard statement D. For example, if the standard statement D is "set the air conditioner temperature to 25 degrees", its contextual intent is "air conditioner adjustment".

[0054] It should be noted that the similarity value B between the user's dialogue A and the standard dialogue D refers to calculating the similarity between the user's dialogue A and the standard dialogue D, and recording this similarity value as the similarity value B.

[0055] The similarity score is a quantitative metric used to measure the degree of similarity between user utterance A and standard utterance D. It is usually calculated using natural language processing techniques (such as cosine similarity, edit distance, etc.).

[0056] The similarity value B is a specific numerical value that represents the similarity between user's statement A and standard statement D. For example, the similarity value B might be 0.9, indicating that the two are very similar.

[0057] Understandably, in this embodiment, the system finds the standard statement D most similar to user statement A in a preset standard statement library, and extracts key information (second semantic slot) and intent category (second domain intent) from standard statement D. Simultaneously, the system also calculates a similarity value B between user statement A and standard statement D to quantify their degree of similarity.

[0058] Specifically, the step of determining the standard statement D in the preset standard statement library that is most similar to the user's statement A further includes steps S210 to S230: Step S210: Based on a preset deep learning model, the user's statement A and multiple standard statements in the preset standard statement library are converted into high-dimensional semantic vectors respectively, to obtain the first high-dimensional semantic vector corresponding to the user's statement A and multiple sets of second high-dimensional semantic vectors corresponding to the preset standard statement library. The deep learning model is constructed based on word vectors and sentence vectors. In this embodiment, a preset deep learning model is used to convert user utterance A and multiple standard statements from a preset standard statement library into high-dimensional semantic vectors. The purpose is to convert text data into a numerical form that the model can process for subsequent similarity calculations.

[0059] A default deep learning model is a model built on word and sentence vectors, used to convert text into high-dimensional semantic vectors. Common models include BERT, Word2Vec, and Sentence-BERT.

[0060] High-dimensional semantic vectors refer to the transformation of text data into vectors in a high-dimensional space. The dimension of each vector is usually between several hundred and several thousand, and they can capture the semantic information of the text.

[0061] Step S220: Based on any one of the multiple sets of second high-dimensional semantic vectors, similarity calculation is performed with the first high-dimensional semantic vector to obtain multiple sets of similarity values. The maximum value among the multiple sets of similarity values ​​is the similarity value B. In this embodiment, similarity is calculated between the first high-dimensional semantic vector of user statement A and the second high-dimensional semantic vector of each standard statement. The purpose is to quantify the degree of similarity between user statement A and each standard statement.

[0062] Understandably, the similarity value between each standard statement and user statement A forms a group, ultimately resulting in multiple groups of similarity values.

[0063] Step S230: Based on the multiple sets of similarity values, determine the standard statement D in the preset standard statement library that is most similar to the user statement A.

[0064] Understandably, in this embodiment, a deep learning model is used to convert user utterance A and multiple standard statements from a preset standard statement library into high-dimensional semantic vectors. Then, the similarity value between user utterance A and each standard statement is calculated, ultimately determining the standard statement D most similar to user utterance A. This process leverages the semantic understanding capabilities of deep learning models, achieving efficient matching between user utterances and standard statements through vectorization and similarity calculation.

[0065] Step S300: Based on the preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent, set the tag corresponding to the user's utterance A; Preset tagging rules are a set of predefined rules used by the system to assign tags based on different input conditions (such as similarity, semantic slots, intent, etc.). These rules are typically designed to optimize system performance and user experience.

[0066] Setting the label corresponding to the user's dialogue A means assigning a suitable label to the user's dialogue A based on the above factors.

[0067] A tag is an identifier used to represent the category, intent, or other attributes of a user's speech. Tags help systems better understand and process user input, for example, in subsequent conversation management, intent recognition, or function execution.

[0068] Understandably, in this embodiment, the system, based on preset tag setting rules, combines the similarity value B between user utterance A and standard utterance D, the first semantic slot and the first domain intent in user utterance A, and the second semantic slot and the second domain intent in standard utterance D, and comprehensively analyzes this information to assign a suitable tag to user utterance A.

[0069] Specifically, the step of setting the tag corresponding to the user utterance A based on the preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent further includes steps S310~S340: Step S310: Based on the first semantic slot and the first domain intention, determine the domain status corresponding to the user utterance A, wherein the domain status includes not domained and domained; In this embodiment, based on the first semantic slot and the first domain intent in user utterance A, it is determined whether user utterance A can be clearly classified into a specific intent domain (domain).

[0070] In this embodiment, the domain-based status includes domain-based and domain-unbased. Domain-unbased means that user utterance A cannot be clearly classified into any known intent domain, which may be due to ambiguity, incomplete information, or being beyond the system's understanding. Domain-based means that user utterance A can be clearly classified into a specific intent domain, and the system can understand and process the utterance.

[0071] Step S320: Determine whether the similarity B is greater than or equal to a preset similarity threshold C, and obtain a first determination result; In this embodiment, the similarity value B between user speech A and standard speech D is compared to see if it reaches a preset threshold C.

[0072] If the similarity B is greater than or equal to the threshold C, it means that user's wording A and standard wording D are sufficiently similar, and they can be considered semantically matched. Conversely, if the similarity B is less than the threshold C, it means that user's wording A and standard wording D have significant semantic differences, and further analysis or adjustment may be needed.

[0073] Step S330: Determine whether the first semantic slot and the second semantic slot are the same to obtain a second determination result; determine whether the first domain placement intention and the second domain placement intention are the same to obtain a third determination result. In this embodiment, semantic slot determination and domain intent determination are performed separately. Semantic slot determination is used to verify whether the key information in user statement A and standard statement D is consistent. For example, if the user says "set the temperature to 25 degrees", the standard statement is "set the temperature to 25 degrees", and the semantic slot "25 degrees" is consistent in both.

[0074] The intent determination function is used to verify whether the intents of user statement A and standard statement D are consistent. For example, if the intent of user statement A is "temperature adjustment" and the intent of standard statement D is also "temperature adjustment," then the intents of both are consistent. Figure 1 To.

[0075] Step S340: Based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result, and the third judgment result, set the tag corresponding to the user's speech A.

[0076] Understandably, in this embodiment, the system analyzes the similarity, semantic slots, and intent consistency between user utterance A and standard utterance D through a series of judgment steps, and finally assigns a suitable label to user utterance A according to preset rules. This label can help the system more accurately understand the user's intent and optimize subsequent processing logic and function implementation.

[0077] Specifically, the step of setting the tag corresponding to the user's dialogue A based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result, and the third judgment result further includes steps S341 to S345: Step S341: If the domain status is domain, the first judgment result is greater than or equal to, and the second judgment result and the third judgment result are the same, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as domain correct. For example, to aid in understanding this embodiment, please refer to Figure 2 , Figure 2 A simplified flowchart of a voice interaction system optimization method based on data feedback is provided.

[0078] It should be noted that in this embodiment, user utterance A not only clearly falls into a certain intent domain, but also perfectly matches standard utterance D in terms of semantic slots and domain intent, while also meeting the similarity requirements. Therefore, the system considers this to be a completely correct domain placement result.

[0079] Step S342: If the domain status is domain, the first judgment result is greater than or equal to, and at least one of the second judgment result and the third judgment result is different, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as domain error. It should be noted that in this embodiment, even if user utterance A can be classified into a certain intent domain and the similarity meets the requirements, the system considers it to be an incorrect domain placement result because at least one of the semantic slots or domain placement intents does not match the standard utterance D.

[0080] Step S343: If the domain landing status is not landed and the first judgment result is greater than or equal to, then based on the preset tag setting rule, the tag corresponding to the user's speech A is set as not landed. It should be noted that, in this embodiment, although the similarity between user utterance A and standard utterance D meets the requirements, the system considers it an incomplete domain placement result because it cannot be clearly classified into a certain intent domain.

[0081] Step S344: If the domain status is domain and the first judgment result is less than, then based on the preset tag setting rule, the tag corresponding to the user's speech A is set as an incorrect domain. It should be noted that in this embodiment, user utterance A can be classified into a certain intent domain, but because the similarity does not meet the requirements, the system considers it to be an incorrect domain classification result.

[0082] Step S345: If the domain status is not domained and the first judgment result is less than, then based on the preset tag setting rules, the tag corresponding to the user's speech A is set as correct rejection.

[0083] It should be noted that in this embodiment, user utterance A cannot be clearly categorized into a specific intent domain, and the similarity does not meet the requirements. In this case, the system correctly rejects the recognition of the utterance, avoiding erroneous processing.

[0084] Understandably, by comprehensively considering the consistency of contextualization status, similarity, semantic slots, and contextualization intent, the system can more accurately evaluate the processing results of user speech and assign appropriate labels. These labels can help the system better understand and optimize the voice interaction process, thereby improving the user experience.

[0085] Step S400: Based on the user's dialogue A and the tags, optimize the voice interaction system to obtain an optimized voice interaction system.

[0086] In this embodiment, the voice interaction system is adjusted and improved based on the user's speech A and its tags. After optimization and adjustment, a voice interaction system with improved performance is obtained.

[0087] It should be noted that the purpose of optimization is to improve system performance, including but not limited to improving the accuracy of speech recognition, improving the precision of intent recognition, optimizing dialogue management processes, and enhancing user experience.

[0088] A voice interaction system is a system that can receive user voice commands, understand their intentions, and respond accordingly. It is commonly used in smart devices, in-vehicle systems, smart homes, and other fields.

[0089] Understandably, in this embodiment, the system utilizes user utterance A and its corresponding tags to optimize and adjust the voice interaction system, thereby obtaining a performance-enhanced voice interaction system. User utterance A and its tags provide specific information about user needs, and the system adjusts its processing logic, optimizes algorithms, or improves the dialogue flow based on this information to better meet user needs.

[0090] Specifically, the step of optimizing the voice interaction system based on user speech A and the tag to obtain the optimized voice interaction system further includes steps S410~S420: Step S410: Based on user speech A and the tags, retrain the natural language understanding model in the voice interaction system to optimize the intent recognition function and slot filling function of the natural language understanding model. It should be noted that Natural Language Understanding (NLU) models are models responsible for understanding the natural language input from users and extracting intent and semantic slots.

[0091] In this embodiment, the specific process of optimizing the model is as follows: First, data preparation is performed (collecting user utterance A and its corresponding labels; this data serves as training data). Then, user utterance A and its labels are input into the natural language understanding model. The model learns from this data, adjusting its internal parameters to improve the accuracy of intent recognition and slot filling for similar utterances.

[0092] It should be noted that the optimization goals are: to enable the model to more accurately identify the intent of the user's speech; and to enable the model to more accurately extract key information (such as time, location, and values) from the user's speech.

[0093] Step S420: Based on the optimized natural language understanding model, the optimized voice interaction system is obtained.

[0094] It should be noted that a voice interaction system refers to a system that includes multiple modules such as speech recognition, natural language understanding, and dialogue management.

[0095] In this embodiment, the specific process of obtaining the optimized voice interaction system based on the optimized natural language understanding model is as follows: First, the optimized natural language understanding model is replaced or updated in the voice interaction system. Then, the optimized voice interaction system is tested to verify its performance improvement. Tests include the accuracy of intent recognition and slot filling accuracy. Finally, the optimized system is deployed to a real-world application environment for user use.

[0096] Understandably, in this embodiment, by utilizing user utterance A and its corresponding tags, the natural language understanding model in the voice interaction system is retrained, optimizing the model's intent recognition and slot filling functions. Then, the optimized model is updated in the voice interaction system, resulting in a performance-improved, optimized voice interaction system.

[0097] This application proposes a voice interaction system optimization method and related equipment based on data feedback, relating to the fields of artificial intelligence and in-vehicle human-machine interaction technology. Compared to existing technologies, existing voice interaction systems lack an effective data feedback mechanism during optimization, failing to fully utilize actual user interaction data to optimize system performance. This results in the system being unable to dynamically adjust according to user habits and preferences, and thus unable to meet the personalized needs of different users. In this application, firstly, in response to a voice interaction system optimization command, user speech A is obtained from the in-vehicle voice cloud log, and the first semantic slot and the second semantic slot corresponding to user speech A are obtained. First, a landing intent is determined. Then, the standard speech D most similar to the user speech A in the preset standard speech library is determined, and the second semantic slot and the second landing intent corresponding to the standard speech D are determined. The similarity value between the user speech A and the standard speech D is similarity value B. Further, based on the preset tag setting rules, the similarity value B, the first semantic slot, the first landing intent, the second semantic intent and the second landing intent, the tag corresponding to the user speech A is set. Finally, based on the user speech A and the tag, the voice interaction system is optimized to obtain the optimized voice interaction system.

[0098] Understandably, this application obtains user dialogue from in-vehicle voice cloud logs, matches the user dialogue with standard dialogue in the standard dialogue library in multiple dimensions (semantic slots, similarity, and contextual intent), sets real labels for user dialogue based on the matching results of different dimensions and preset label setting rules, and uses real labels and user dialogue as training samples for data feedback to optimize the voice interaction system.

[0099] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Before the step of converting the user's utterance A and multiple standard statements in the preset standard statement library into high-dimensional semantic vectors based on a preset deep learning model, the method further includes steps A1 to A4: Step A1: Obtain sample data, wherein the real high-dimensional semantic vector corresponding to the sample data is the third high-dimensional semantic vector; Sample data refers to a set of known text data that is used to adjust model parameters during training.

[0100] The third high-dimensional semantic vector refers to the true high-dimensional semantic vector corresponding to the sample data. It is usually calculated in advance by a reliable method and used as the target for model training.

[0101] It should be noted that these sample data have been pre-labeled with real high-dimensional semantic vectors for subsequent model evaluation and optimization.

[0102] Step A2: Process the sample data using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data; Current deep learning models refer to models that are being trained or optimized, such as BERT and Word2Vec.

[0103] The fourth high-dimensional semantic vector refers to the high-dimensional semantic vector generated by the model after processing the sample data.

[0104] Step A3: Determine whether the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector; In this embodiment, the accuracy of the model is evaluated by comparing whether the real high-dimensional semantic vector (third high-dimensional semantic vector) of the sample data and the high-dimensional semantic vector (fourth high-dimensional semantic vector) generated by the model are consistent.

[0105] It should be noted that in this embodiment, the similarity (such as cosine similarity) or distance (such as Euclidean distance) between two vectors is typically used to determine whether they are close enough. If two vectors are very close, they are considered to be consistent; otherwise, they are considered to be inconsistent.

[0106] Step A4: If there is no consistency, adjust the parameters of the current deep learning model. Based on the current deep learning model with adjusted parameters, return to the step of using the current deep learning model to process the sample data and obtain the fourth high-dimensional semantic vector corresponding to the sample data, until the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector, and obtain the preset deep learning model.

[0107] It should be noted that if the third high-dimensional semantic vector is inconsistent with the fourth high-dimensional semantic vector, the parameters of the current deep learning model are adjusted. Then, the model with adjusted parameters is used again to process the sample data until the generated high-dimensional semantic vector is consistent with the true high-dimensional semantic vector.

[0108] Parameter tuning refers to adjusting parameters such as weights and biases of a model through optimization algorithms (such as gradient descent) to improve the model's performance.

[0109] It should be noted that model training is an iterative process, which continues until the high-dimensional semantic vectors generated by the model are consistent with the real high-dimensional semantic vectors.

[0110] Understandably, in this embodiment, sample data and its actual high-dimensional semantic vector are acquired, and the sample data is processed using the current deep learning model to generate the model's output high-dimensional semantic vector. Then, the actual high-dimensional semantic vector and the model-generated high-dimensional semantic vector are compared for consistency. If they are inconsistent, the model parameters are adjusted, and the sample data is reprocessed until the generated high-dimensional semantic vector matches the actual high-dimensional semantic vector, thus obtaining the preset deep learning model.

[0111] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the voice interaction system optimization method based on data feedback in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0112] It should be noted that all user-related data involved in this application (such as user attribute data, user behavior data, and user geographical location, etc., the data types here should be modified according to the adaptability of the solution content) were obtained with the user's permission or consent; that is to say, when this application is applied to specific products or technologies, user permission is required to obtain and process the relevant data, and the processing of the relevant data must comply with the relevant laws, regulations and regulatory standards of the relevant countries and regions.

[0113] For example, when it is necessary to obtain a user's current geographical location, a location acquisition prompt can be displayed on the user's terminal. After receiving confirmation from the user regarding the location acquisition prompt, the terminal can obtain the user's current geographical location.

[0114] This application also provides a voice interaction system optimization device based on data feedback, please refer to... Figure 4 The voice interaction system optimization device based on data feedback includes: The acquisition module 10 is used to respond to the voice interaction system optimization command, acquire user speech A from the vehicle voice cloud log, and acquire the first semantic slot and the first domain intent corresponding to the user speech A. The determining module 20 is used to determine the standard statement D that is most similar to the user statement A in the preset standard statement library, and to determine the second semantic slot and the second domain intent corresponding to the standard statement D. The similarity value between the user statement A and the standard statement D is the similarity value B. Setting module 30, the setting module is used to set the tag corresponding to the user utterance A based on preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent and the second domain intent; The optimization module 40 is used to optimize the voice interaction system based on the user's utterance A and the tag, so as to obtain an optimized voice interaction system.

[0115] In one embodiment, the setting module further includes: The first determining unit is configured to determine the domain status corresponding to the user utterance A based on the first semantic slot and the first domain intention, wherein the domain status includes not domained and domained; The first judgment unit is used to determine whether the similarity B is greater than or equal to a preset similarity threshold C, and to obtain a first judgment result; The second judgment unit is used to determine whether the first semantic slot and the second semantic slot are the same, and to obtain a second judgment result; and to determine whether the first domain intention and the second domain intention are the same, and to obtain a third judgment result. The first setting unit is used to set the tag corresponding to the user's speech A based on the preset tag setting rules, the domain status, the first judgment result, the second judgment result and the third judgment result.

[0116] In one embodiment, the setting module further includes: The second setting unit is used to set the tag corresponding to the user's speech A as "correct domain" based on the preset tag setting rules if the domain status is "domain", the first judgment result is greater than or equal to, and the second judgment result and the third judgment result are the same. The third setting unit is used to set the tag corresponding to the user's speech A as a domain error based on a preset tag setting rule if the domain status is a domain, the first judgment result is greater than or equal to, and at least one of the second judgment result and the third judgment result is different. The fourth setting unit is used to set the tag corresponding to the user's speech A as incomplete domain coverage if the domain coverage status is not completed and the first judgment result is greater than or equal to the domain coverage status, based on the preset tag setting rules. The fifth setting unit is used to set the tag corresponding to the user's speech A as an incorrect domain if the domain status is a domain and the first judgment result is less than. The sixth setting unit is used to set the tag corresponding to the user's speech A as correct rejection based on the preset tag setting rules if the domain status is not domained and the first judgment result is less than.

[0117] In one embodiment, the determining module further includes: The conversion unit is used to convert the user's utterance A and multiple standard statements in the preset standard statement library into high-dimensional semantic vectors based on a preset deep learning model, respectively, to obtain a first high-dimensional semantic vector corresponding to the user's utterance A and multiple sets of second high-dimensional semantic vectors corresponding to the preset standard statement library. The deep learning model is constructed based on word vectors and sentence vectors. The calculation unit is used to perform similarity calculations with the first high-dimensional semantic vector based on any one of the multiple sets of second high-dimensional semantic vectors, and obtain multiple sets of similarity values, wherein the maximum value among the multiple sets of similarity values ​​is the similarity value B; The second determining unit is used to determine, based on the multiple sets of similarity values, the standard statement D in the preset standard statement library that is most similar to the user's statement A.

[0118] In one embodiment, the determining module further includes: The acquisition unit is used to acquire sample data, wherein the real high-dimensional semantic vector corresponding to the sample data is the third high-dimensional semantic vector; The data processing unit is used to process the sample data using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data. The third judgment unit is used to determine whether the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector; The iterative training unit is used to adjust the parameters of the current deep learning model if there is a discrepancy. Based on the current deep learning model with adjusted parameters, it returns to the step of processing the sample data using the current deep learning model to obtain the fourth high-dimensional semantic vector corresponding to the sample data, until the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector, thus obtaining the preset deep learning model.

[0119] In one embodiment, the optimization module further includes: An optimization unit is used to retrain the natural language understanding model in the voice interaction system based on user utterance A and the label, so as to optimize the intent recognition function and slot filling function of the natural language understanding model. The acquisition unit is used to obtain the optimized voice interaction system based on the optimized natural language understanding model.

[0120] The voice interaction system optimization device based on data feedback provided in this application adopts the voice interaction system optimization method based on data feedback in the above embodiments, and can solve the technical problem of voice interaction system optimization based on data feedback. Compared with related technologies, the beneficial effects of the voice interaction system optimization device based on data feedback provided in this application are the same as the beneficial effects of the voice interaction system optimization method based on data feedback provided in the above embodiments, and other technical features in the voice interaction system optimization device based on data feedback are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0121] This application provides a voice interaction system optimization device based on data feedback. The voice interaction system optimization device based on data feedback includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the voice interaction system optimization method based on data feedback in the first embodiment described above.

[0122] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of a voice interaction system optimization device based on data feedback suitable for implementing embodiments of this application. The voice interaction system optimization device based on data feedback in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The voice interaction system optimization device based on data feedback shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0123] like Figure 5As shown, the voice interaction system optimization device based on data feedback may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the voice interaction system optimization device based on data feedback. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the data-backflow-based voice interaction system optimization device to exchange data with other devices wirelessly or via wired communication. Although a data-backflow-based voice interaction system optimization device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0124] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0125] The voice interaction system optimization device based on data feedback provided in this application, employing the voice interaction system optimization method based on data feedback in the above embodiments, can solve the technical problems. Compared with related technologies, the beneficial effects of the voice interaction system optimization device based on data feedback provided in this application are the same as those of the voice interaction system optimization method based on data feedback provided in the above embodiments, and other technical features in this voice interaction system optimization device based on data feedback are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0126] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included 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.

[0128] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the data backflow-based voice interaction system optimization method in the above embodiments.

[0129] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0130] The aforementioned computer-readable storage medium may be included in a voice interaction system optimization device based on data return flow; or it may exist independently and not assembled into a voice interaction system optimization device based on data return flow.

[0131] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the data-backflow-based voice interaction system optimization device, cause the data-backflow-based voice interaction system optimization device to: In response to the voice interaction system optimization command, user speech A is obtained from the vehicle voice cloud log, and the first semantic slot and the first domain intent corresponding to the user speech A are obtained. Determine the standard statement D in the preset standard statement library that is most similar to the user statement A, and determine the second semantic slot and the second domain intent corresponding to the standard statement D. The similarity value between the user statement A and the standard statement D is the similarity value B. Based on the preset tag setting rules, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent, the tag corresponding to the user's utterance A is set; Based on the user's dialogue A and the tags, the voice interaction system is optimized to obtain an optimized voice interaction system.

[0132] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0133] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0134] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0135] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described data-backflow-based voice interaction system optimization method, thereby solving the technical problem of optimizing a data-backflow-based voice interaction system. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the data-backflow-based voice interaction system optimization method provided in the above embodiments, and will not be repeated here.

[0136] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described data feedback-based voice interaction system optimization method.

[0137] The computer program product provided in this application can solve the technical problem of optimizing a voice interaction system based on data feedback. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the voice interaction system optimization method based on data feedback provided in the above embodiments, and will not be repeated here.

[0138] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for optimizing a voice interaction system based on data reflux, characterized in that, The voice interaction system optimization method based on data backflow comprises: In response to a voice interaction system optimization instruction, a user utterance A is obtained from a vehicle-mounted voice cloud log, and a first semantic slot and a first domain intention corresponding to the user utterance A are obtained; A standard utterance D most similar to the user utterance A in a preset standard statement library is determined, and a second semantic slot and a second domain intention corresponding to the standard utterance D are determined, and a similarity value between the user utterance A and the standard utterance D is a similarity value B; Based on a preset label setting rule, the similarity B, the first semantic slot, the first domain intention, the second semantic intention, and the second domain intention, the label corresponding to the user utterance A is set; Based on the user utterance A and the label, the voice interaction system is optimized to obtain an optimized voice interaction system.

2. The method of claim 1, wherein the data flow based voice interaction system optimization method is characterized by, The step of setting the label corresponding to the user utterance A based on the preset label setting rule, the similarity B, the first semantic slot, the first domain intention, the second semantic intention, and the second domain intention further comprises: Based on the first semantic slot and the first domain intention, a domain state corresponding to the user utterance A is determined, and the domain state comprises non-domain and domain; It is judged whether the similarity B is greater than or equal to a preset similarity threshold C to obtain a first judgment result; It is judged whether the first semantic slot is the same as the second semantic slot to obtain a second judgment result, and it is judged whether the first domain intention is the same as the second domain intention to obtain a third judgment result; Based on the preset label setting rule, the domain state, the first judgment result, the second judgment result, and the third judgment result, the label corresponding to the user utterance A is set.

3. The method of optimizing voice interaction systems based on data recirculation according to claim 2, characterized in that, The step of setting the label corresponding to the user utterance A based on the preset label setting rule, the domain state, the first judgment result, the second judgment result, and the third judgment result further comprises: If the domain state is domain, the first judgment result is greater than or equal to, and the second judgment result and the third judgment result are both the same, then based on the preset label setting rule, the label corresponding to the user utterance A is set as domain correct; If the domain state is domain, the first judgment result is greater than or equal to, and at least one of the second judgment result and the third judgment result is not the same, then based on the preset label setting rule, the label corresponding to the user utterance A is set as domain error; If the domain state is non-domain, the first judgment result is greater than or equal to, then based on the preset label setting rule, the label corresponding to the user utterance A is set as non-completed domain; If the domain state is domain, the first judgment result is less than, then based on the preset label setting rule, the label corresponding to the user utterance A is set as error domain; If the domain state is non-domain, the first judgment result is less than, then based on the preset label setting rule, the label corresponding to the user utterance A is set as correct rejection.

4. The method of claim 1, wherein the data flow based voice interaction system optimization method is characterized by, The step of determining the standard utterance D most similar to the user utterance A in the preset standard utterance library further comprises: Based on a preset deep learning model, the user utterance A and the plurality of standard utterances in the preset standard utterance library are respectively converted into high-dimensional semantic vectors, to obtain a first high-dimensional semantic vector corresponding to the user utterance A and a plurality of second high-dimensional semantic vectors corresponding to the preset standard utterance library, and the deep learning model is constructed based on word vectors and sentence vectors; Based on any one of the plurality of second high-dimensional semantic vectors, the first high-dimensional semantic vector is subjected to similarity calculation, to obtain a plurality of similarity values, and the maximum value in the plurality of similarity values is a similarity value B; Based on the plurality of similarity values, the standard utterance D most similar to the user utterance A in the preset standard utterance library is determined.

5. The method of optimizing voice interaction systems based on data recirculation according to claim 4, characterized in that, Before the step of converting the user utterance A and the plurality of standard utterances in the preset standard utterance library into high-dimensional semantic vectors based on the preset deep learning model, the following steps are further included: Sample data is acquired, and a true high-dimensional semantic vector corresponding to the sample data is a third high-dimensional semantic vector; The sample data is processed using a current deep learning model, to obtain a fourth high-dimensional semantic vector corresponding to the sample data; It is determined whether the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector; If not, the parameters of the current deep learning model are adjusted, and the step of processing the sample data using the current deep learning model based on the current deep learning model with adjusted parameters to obtain the fourth high-dimensional semantic vector corresponding to the sample data is returned until the third high-dimensional semantic vector is consistent with the fourth high-dimensional semantic vector, to obtain the preset deep learning model.

6. The method of optimizing voice interaction systems based on data reflux as claimed in claim 1, wherein, The step of optimizing the voice interaction system based on the user utterance A and the label to obtain an optimized voice interaction system further comprises: Based on the user utterance A and the label, the natural language understanding model in the voice interaction system is retrained to optimize the intent recognition function and the slot filling function of the natural language understanding model; Based on the optimized natural language understanding model, an optimized voice interaction system is obtained.

7. A voice interaction system optimization apparatus based on data backflow, characterized by, The voice interaction system optimization device based on data backflow comprises: An acquisition module is configured to acquire a user utterance A from a vehicle-mounted voice cloud log and acquire a first semantic slot and a first domain intent corresponding to the user utterance A in response to a voice interaction system optimization instruction; A determination module is configured to determine a standard utterance D most similar to the user utterance A in a preset standard utterance library and determine a second semantic slot and a second domain intent corresponding to the standard utterance D, and a similarity value between the user utterance A and the standard utterance D is a similarity value B; A setting module is configured to set a label corresponding to the user utterance A based on a preset label setting rule, the similarity B, the first semantic slot, the first domain intent, the second semantic intent, and the second domain intent. An optimization module is configured to optimize the voice interaction system based on the user dialogue A and the label, and obtain an optimized voice interaction system.

8. A data flow based voice interaction system optimization device, characterized by, The device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the voice interaction system optimization method based on data backflow according to any one of claims 1 to 6.

9. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the voice interaction system optimization method based on data backflow according to any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by a processor to implement the steps of the voice interaction system optimization method based on data backflow according to any one of claims 1 to 6.

Images