Semantic analysis method, device and equipment based on field large model and storage medium

Abstract

The invention discloses a semantic analysis method and device based on a field large model, computer equipment and a storage medium. The method comprises the steps of obtaining voice data to be subjected to quality inspection; transmitting the voice data to a domain large model of multiple Agent nodes; wherein the domain large model is configured with a plurality of Agent nodes, and each Agent node represents a semantic quality inspection model; the field large model is associated with a field knowledge base of a preset field; calling at least one semantic Agent node of the domain large model to perform semantic analysis on the voice data to obtain a preliminary semantic quality inspection result; calling at least two commenting Agent nodes of the domain large model to comment the preliminary semantic quality inspection result, and obtaining a semantic commenting result of each commenting Agent node; and generating a semantic analysis result of the voice data according to the semantic comment result. Through cooperation of semantic quality inspection models and comment models in multiple professional fields, the accuracy and reliability of semantic analysis are improved.

Description

Technical Field

[0001] The present application relates to the field of semantic parsing, and in particular to a semantic parsing method, apparatus, device and storage medium based on a domain big model. Background Art

[0002] The agent call industry plays a vital role in finance, e-commerce, insurance, education, government affairs and other fields, and is an important bridge for communication between enterprises and customers. Most of the existing customer service or marketing recording quality inspection systems are based on simple regular expressions or traditional semantic similarity comparison algorithms, which are not flexible enough and have low accuracy. The existing quality inspection systems cannot truly understand customer semantics and are prone to misjudging users' real needs in some scenarios, resulting in poor user perception. How to improve the accuracy of semantic analysis in agent scenarios has become an urgent problem to be solved. Summary of the invention

[0003] The purpose of this application is to provide a semantic parsing method, device, equipment and storage medium based on a domain big model to improve the accuracy and reliability of semantic understanding.

[0004] In order to solve the above technical problems, the present application provides a semantic parsing method based on a large domain model, including: Obtain voice data to be tested; The voice data is transmitted to a domain big model of multiple Agent nodes; wherein the domain big model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; Based on a preconfigured scheduler, calling at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain preliminary semantic quality inspection results; Based on the preconfigured scheduler, at least two review Agent nodes of the domain big model are called to review the preliminary semantic quality inspection results to obtain the semantic review results of each review Agent node; The semantic analysis result of the speech data is generated according to the semantic review result of each review Agent node.

[0005] Optionally, the calling of at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain a preliminary semantic quality inspection result includes: Obtaining text input of the voice data; Perform matching according to the text input to obtain a corresponding domain knowledge base; Calling the Agent node corresponding to the domain knowledge base to perform quality inspection and annotation on the text input; wherein the quality inspection and annotation are based on the relevant information in the domain knowledge base, Generate preliminary semantic quality inspection results based on the quality inspection and annotation results.

[0006] Optionally, generating a preliminary semantic quality inspection result according to the quality inspection and annotation results includes: Obtaining standard text information according to the quality inspection and annotation results; Calculating a text feature distance between the text input and the standard text information; When the text feature distance is greater than a preset threshold, a preset dynamic node is activated according to the text feature distance; wherein a derived knowledge base is provided in the dynamic node, and the feature distance between the derived knowledge base and the domain knowledge base is equal to the text feature distance; The text input is corrected and recognized according to the moving node, and the quality inspection and annotation results are corrected according to the correction recognition result to generate the preliminary semantic quality inspection result.

[0007] Optionally, after acquiring the voice data to be quality-checked, the method further includes: Converting the voice data into text information; The text information is preprocessed and formatted so that the text information meets the data rules of the preset Agent node, and the input data of the domain large model is obtained.

[0008] Optionally, the preconfigured scheduler is further configured as: Using the preset distributed algorithm and parallel processing rules, the tasks of semantic quality inspection and semantic commenting are assigned to multiple Agent nodes for parallel processing; and, Using the preset dynamic scheduling strategy, the semantic quality inspection and semantic review tasks are dynamically allocated to different Agent nodes according to the current system load and task requirements.

[0009] Optionally, the semantic Agent node performs semantic analysis on the speech data to obtain a preliminary semantic quality inspection result, further comprising: Processing the text of the voice data based on a preset natural language processing algorithm to obtain a first result; Performing sentiment analysis on the speech data based on a preset sentiment analysis algorithm to obtain a second result; Extracting context information of the speech data; Based on the correlation between the context information, the first result, and the second result, semantic recognition is performed on the speech data to obtain a preliminary semantic quality inspection result.

[0010] Optionally, generating a semantic analysis result of the speech data according to the semantic review result of each review Agent node includes: Obtaining a diversity evaluation index, evaluating and optimizing the diversity of the semantic review results of each review Agent node, and obtaining an optimized review result; Obtain the trust and historical performance of each review agent node; According to the trust and historical performance of the review agent node, the weight of each review agent node is configured; Performing weighted processing on the optimized review results of each review Agent node; The semantic analysis result of the speech data is generated according to the weighted optimization review result.

[0011] In order to solve the above technical problems, the present application also provides a semantic parsing device based on a large domain model, comprising: A data acquisition module, used to acquire voice data to be inspected; A data input module, used to transmit the voice data to a domain model with multiple Agent nodes; wherein the domain model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain model is associated with a domain knowledge base of a preset domain; A semantic quality inspection module, configured to perform semantic analysis on the speech data based on a preconfigured scheduler by calling at least one semantic Agent node of the domain big model to obtain a preliminary semantic quality inspection result; A quality inspection and review module, which is used to call at least two review Agent nodes of the domain big model to review the preliminary semantic quality inspection results based on a preconfigured scheduler, and obtain the semantic review results of each review Agent node; The result feedback module is used to generate the semantic analysis result of the speech data according to the semantic review result of each review Agent node.

[0012] In order to solve the above technical problems, the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned semantic parsing method based on the domain big model.

[0013] In order to solve the above technical problems, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the above-mentioned semantic parsing method based on the domain big model.

[0014] The beneficial effects of the embodiments created by the present application are: by acquiring voice data to be inspected; transmitting the voice data to a domain big model of multiple Agent nodes; wherein the domain big model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; based on a preconfigured scheduler, calling at least one semantic Agent node of the domain big model to perform semantic analysis on the voice data to obtain preliminary semantic quality inspection results; based on a preconfigured scheduler, calling at least two review Agent nodes of the domain big model to review the preliminary semantic quality inspection results to obtain the semantic review results of each review Agent node; generating the semantic analysis results of the voice data according to the semantic review results of each review Agent node; utilizing the collaboration of semantic quality inspection models and review models in multiple professional fields to effectively process the voice data to be inspected, and provide accurate and comprehensive semantic analysis results, thereby improving the accuracy and reliability of semantic analysis, thereby meeting the needs of voice data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A basic flow chart of a semantic parsing method based on a domain big model according to a specific embodiment of the present application; Figure 2 A schematic diagram of the basic structure of a semantic parsing device based on a domain big model according to a specific embodiment of the present application; Figure 3 This is a basic structural block diagram of a computer device according to a specific embodiment of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0017] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0018] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0019] It will be understood by those skilled in the art that the term "terminal" as used herein includes both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capability, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of performing two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices, which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which may combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with a radio frequency receiver. The "terminal" used here can be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable and / or configured to operate locally, and / or in a distributed form, at any other location on the earth and / or in space. The "terminal" used here can also be a communication terminal, an Internet terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device) and / or a mobile phone with music / video playback function, or a smart TV, a set-top box and other devices.

[0020] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with capabilities equivalent to those of a personal computer. It is a hardware device that has the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0021] It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.

[0022] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.

[0023] Unless explicitly specified, the AI ​​models referenced or may be referenced in this application can be deployed on a remote server and remotely called on the client, or can be deployed and directly called by a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0024] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.

[0025] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.

[0026] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0027] See also Figure 1 , Figure 1 The figure is a basic flow chart of the semantic parsing method based on the domain big model in this embodiment.

[0028] like Figure 1 As shown, including: S1100, obtaining voice data to be inspected; This embodiment is applied to voice quality inspection scenarios in multiple fields such as finance, e-commerce, insurance, and education. In this embodiment, a voice data quality inspection system is configured to monitor / detect the voice call data of users and agents for quality inspection. The voice data quality inspection system can be deployed on the server for implementation and accessed by the client remotely through the online service interface provided by the server, or it can be directly deployed and run on the client for access. Among them, the voice data quality inspection system is used to obtain voice data to be inspected. Voice data can be obtained in a variety of ways, such as from voice files uploaded by users, real-time voice input, voice call records, etc. After the system receives the voice data to be inspected, it prepares to transmit it to the domain big model for semantic analysis.

[0029] It should be pointed out that the voice data quality inspection system of this embodiment can be integrated into the call recording system: the voice quality inspection system is integrated into the call recording system. The call recording system usually records relevant information of the call, including call time, participants, call content, etc. After the voice quality inspection system is integrated with the call recording system, the calls that need to be quality inspected can be marked in the call records, and then the call data can be extracted for use by the quality inspection system. In one embodiment, the integration is implemented through an API, and the call recording system can send the call data that needs to be quality inspected to the API interface of the voice quality inspection system, and then receive the quality inspection results. This method can realize an automated quality inspection process and seamlessly integrate with the existing call recording system, which is convenient for use and management.

[0030] It should be noted that, in one implementation, the user can manually upload voice files to the voice quality inspection system. The user saves the voice call data to be inspected as audio files (such as WAV or MP3 format), and then uploads these files through the user interface or API provided by the voice quality inspection system.

[0031] It should be noted that, in one embodiment, the voice quality inspection system supports real-time streaming transmission, and can directly obtain data from the ongoing voice call for quality inspection, transmit the audio stream to the voice quality inspection system during the call, and perform quality inspection and analysis in real time. Real-time streaming transmission is integrated with the voice call system to transmit the audio stream to the quality inspection system.

[0032] S1200, transmitting the voice data to a domain big model with multiple Agent nodes; wherein the domain big model is configured with multiple Agent nodes, and each Agent node represents a semantic quality inspection model; After obtaining the voice data to be inspected, the voice data is transmitted to the domain model of multiple agent nodes, wherein the domain model is configured with multiple agent nodes. The domain model can be regarded as a system containing multiple agent nodes, each agent node represents a semantic quality inspection model, focusing on a specific field or task, and each agent node has independent semantic analysis capabilities, and can independently perform semantic quality inspection and comments. In addition, data transmission between each agent node can be achieved through network communication, ensuring that the voice data can be quickly and reliably transmitted to each agent node of the domain model.

[0033] It should be noted that this embodiment can be a domain model containing multiple Agent nodes, each Agent node represents a semantic quality inspection model, responsible for semantic analysis tasks in a specific domain. By transmitting voice data to the domain model, the collaborative work of multiple Agent nodes can be used to perform more accurate and comprehensive semantic analysis.

[0034] It should be noted that the domain big model of this embodiment can be associated with the domain knowledge base of the preset domain, and the domain knowledge base can contain domain-related semantic information, vocabulary, rules and patterns, etc. By associating with the domain knowledge base, the domain big model can better understand and interpret the meaning of the speech data and improve the accuracy and efficiency of semantic quality inspection.

[0035] S1300, based on a preconfigured scheduler, calling at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain a preliminary semantic quality inspection result; After the voice data is transmitted to the domain big model of multiple agent nodes, based on the preconfigured scheduler, at least one semantic agent node of the domain big model is called to perform semantic analysis on the voice data to obtain preliminary semantic quality inspection results. The preconfigured scheduler in the domain big model selects at least one semantic agent node to perform semantic analysis on the voice data according to the task requirements. The selected semantic agent node uses the semantic quality inspection model of its professional field to analyze the voice data input to obtain preliminary semantic quality inspection results. Specifically, the semantic agent node uses an end-to-end AI model to perform part-of-speech tagging, entity recognition, sentiment analysis, and keyword extraction on the voice data, and then extracts the features of the voice data, extracts features related to sound, tone, speaking speed, etc. from the voice data, and performs semantic analysis on the voice data based on the association between the voice data and semantics to obtain preliminary semantic quality inspection results. The preliminary semantic quality inspection results include a preliminary understanding of the semantic information such as intent, entity, and keyword contained in the voice data.

[0036] It should be noted that this embodiment configures and manages a preset scheduler for calling each semantic agent node of the domain model to perform semantic analysis tasks. The scheduler can dynamically select and allocate appropriate semantic agent nodes to process voice data according to pre-set strategies and algorithms. Through reasonable scheduling and allocation, the efficiency and quality of semantic analysis can be improved.

[0037] It should be noted that in this embodiment, multiple semantic Agent nodes can work together at the same time to process voice data in parallel. Each node can be responsible for different semantic analysis tasks, such as lexical analysis, syntactic analysis, semantic role labeling, etc. Through the parallel processing of multiple Agent nodes, the speed of semantic analysis can be accelerated, and the performance and throughput of the overall system can be improved.

[0038] S1400, based on a preconfigured scheduler, calling at least two review Agent nodes of the domain big model to review the preliminary semantic quality inspection result, and obtaining the semantic review result of each review Agent node; Based on the preconfigured scheduler, at least one semantic agent node of the domain big model is called to perform semantic analysis on the voice data to obtain preliminary semantic quality inspection results. Based on the preconfigured scheduler, at least two review agent nodes of the domain big model are called to review the preliminary semantic quality inspection results to obtain the semantic review results of each review agent node. The preconfigured scheduler of the domain big model further calls at least two review agent nodes to review the preliminary semantic quality inspection results, wherein each review agent node uses a specific review model and algorithm to review and analyze the preliminary semantic quality inspection results from different angles and dimensions, and the semantic review results of the review agent node include scoring, suggestions, corrections or further supplementary information for the preliminary semantic quality inspection results. In one embodiment, based on the review of rules and rule engines, at least one set of rules and rule engines are configured to review the preliminary semantic quality inspection results; the rules define various semantic quality indicators and specifications, such as correctness, consistency, logical relationships, etc., and the rule engine can evaluate and review the semantic quality inspection results according to these rules. The review implementation method based on rules and rule engines has the following beneficial effects: Flexibility: rules can be defined and adjusted according to specific needs, and can quickly adapt to different semantic quality assessment requirements. Customizability: suitable review rules can be formulated according to specific fields and application scenarios to meet specific semantic review needs. Scalability: the rule set can be continuously expanded and updated as needs change and develop, thereby improving the accuracy and comprehensiveness of reviews. In another embodiment, the review model is trained using machine learning technology, and the preliminary semantic quality inspection results are used as input to generate corresponding semantic review results. The review implementation method based on machine learning has the following beneficial effects: Automation and rapidity: through the machine learning model, automatic review of the semantic quality inspection results can be achieved, improving the efficiency and speed of the review; learning ability: the machine learning model can continuously learn and optimize according to the training data, gradually improving the accuracy and quality of the review; robustness: the machine learning model can handle a large number of semantic quality inspection results, and has good adaptability and generalization capabilities.

[0039] It should be noted that each review agent node in this embodiment can be trained based on different evaluation criteria, models and algorithms. Through the collaborative work of multiple review agent nodes, more comprehensive and accurate semantic review results can be obtained. This embodiment provides each review agent node with knowledge and professional background in a specific field. The review agent node is associated with the domain knowledge base to understand the semantic requirements and specifications of the specific field. Through the review agent node with domain expertise, more targeted semantic review results can be provided. S1500: Generate a semantic analysis result of the speech data according to the semantic review result of each review Agent node.

[0040] After obtaining the preliminary semantic quality inspection results and the semantic review results of each review agent node, the semantic analysis results of the voice data are generated according to the semantic review results of each review agent node. According to the semantic review results of each review agent node, the system comprehensively considers the weights and credibility of different review results to generate the final semantic analysis results of the voice data. The semantic analysis results may include detailed explanations of the voice data, intent classification, key information extraction, semantic tags, etc., to provide a comprehensive understanding and analysis of the meaning of the voice data. Furthermore, the semantic analysis results are converted into natural language expressions to make them more in line with human language habits and understanding methods. By optimizing the expression form of the analysis results, the user's acceptance and use effect of the semantic analysis results can be improved.

[0041] In the above implementation, the voice data to be inspected is obtained; the voice data is transmitted to a domain model of multiple Agent nodes; wherein the domain model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; based on a preconfigured scheduler, at least one semantic Agent node of the domain model is called to perform semantic analysis on the voice data to obtain a preliminary semantic quality inspection result; based on a preconfigured scheduler, at least two review Agent nodes of the domain model are called to review the preliminary semantic quality inspection result to obtain a semantic review result of each review Agent node; a semantic analysis result of the voice data is generated according to the semantic review result of each review Agent node; through the above steps, the semantic analysis method based on the domain model can effectively process the voice data to be inspected and provide accurate and comprehensive semantic analysis results. Voice data can be semantically analyzed by multiple semantic Agent nodes at the same time, greatly shortening the processing time. At the same time, the introduction of the review agent node and the integration of the semantic review results can accurately evaluate and correct the preliminary semantic quality inspection results, and improve the accuracy and reliability of the quality inspection; by associating the domain knowledge base of the preset domain, the domain big model can better understand and interpret the meaning of the speech data, and the semantic parsing results will be more targeted and reliable, and can better meet the semantic requirements of specific fields; this method utilizes the collaboration of semantic quality inspection models and review models in multiple professional fields to improve the accuracy and reliability of semantic analysis, thereby meeting the needs of speech data processing.

[0042] In some implementations, S1300 calls at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain preliminary semantic quality inspection results, including: S1311, obtaining text input of the voice data; In order to perform semantic analysis in this embodiment, it is first necessary to convert the voice data into text form. Specifically, it is achieved through voice recognition technology to convert the voice signal into corresponding text input. The voice recognition system can use sound features and voice models to transcribe the voice data into text strings. The transcribed text will serve as input for subsequent steps.

[0043] S1312, matching is performed according to the text input to obtain a corresponding domain knowledge base; After obtaining the text input of the voice data, matching is performed based on the text input to obtain the corresponding domain knowledge base, and the text input is used to match the corresponding domain knowledge base. The domain knowledge base is a pre-built database that contains knowledge, rules and patterns in a specific field. In this embodiment, text matching algorithms are used, such as keyword matching, semantic similarity matching, etc., to match the text input with the content in the domain knowledge base to find the most relevant domain knowledge base. The matching process can use natural language processing technology and semantic models to improve accuracy and matching.

[0044] S1313, calling the Agent node corresponding to the domain knowledge base to perform quality inspection and annotation on the text input; wherein the quality inspection and annotation are based on the relevant information in the domain knowledge base; After matching the text input and obtaining the corresponding domain knowledge base, the corresponding Agent node is called to perform quality inspection and annotation. Each domain knowledge base corresponds to an Agent node, which has the professional ability to analyze the semantics of the domain. The Agent node can use the relevant information, rules and patterns in the domain knowledge base to perform quality inspection and annotation on the text input. This includes checking the grammatical correctness and semantic consistency of the text, as well as annotating important information such as entities and relationships in the text.

[0045] S1314. Generate preliminary semantic quality inspection results based on the quality inspection and annotation results.

[0046] After calling the Agent node corresponding to the domain knowledge base to perform quality inspection and annotation on the text input, a preliminary semantic quality inspection result is generated based on the quality inspection and annotation results, and the quality inspected and annotated text input is combined with relevant information in the domain knowledge base to generate a preliminary understanding and quality inspection of semantics, including understanding and classification of the intent, emotion, theme, etc. of the text. In one embodiment, the preliminary semantic quality inspection result can be in a structured form, such as a label, classification or other semantic representation form, for subsequent application and further processing.

[0047] This implementation method can ensure that semantic analysis and quality inspection are based on professional knowledge and information in a specific field by matching the domain knowledge base and calling the corresponding Agent node, thereby improving the accuracy and effectiveness of the analysis; by calling multiple Agent nodes, different semantic analysis technologies and algorithms can be used to comprehensively consider semantic information from multiple aspects to provide more comprehensive and accurate quality inspection results; Customization: According to specific domain requirements and application scenarios, you can choose to call the appropriate Agent node, and expand and customize the corresponding knowledge base as needed to meet specific semantic analysis requirements; Automated processing: By automatically calling Agent nodes and domain knowledge bases, you can achieve fast and efficient semantic analysis and quality inspection of voice data, thereby improving processing speed and efficiency.

[0048] In some implementations, generating preliminary semantic quality inspection results based on the quality inspection and annotation results includes: S1314a, obtaining standard text information according to the quality inspection and marking results; When the Agent node generates the quality inspection and annotation results, it extracts the standard text information corresponding to the quality inspection and annotation results from the preset database according to the content of the quality inspection and annotation results. The standard text information records the standard text information with the same field, scene, score, and label as the quality inspection and annotation results.

[0049] S1314b, calculating the text feature distance between the text input and the standard text information; Calculate the text feature distance between the text input and the standard text information. The text feature distance can be: Hamming distance, Mahalanobis distance or Euclidean distance. The text feature distance between the text input and the standard text information represents the similarity between the two texts. The closer the feature distance between the two, the more similar the two texts are; otherwise, the less similar they are.

[0050] S1314c, when the text feature distance is greater than a preset threshold, activating a preset dynamic node according to the text feature distance; wherein a derived knowledge base is provided in the dynamic node, and a feature distance between the derived knowledge base and the domain knowledge base is equal to the text feature distance; When the text feature distance is greater than a preset threshold, it indicates that the similarity between the two texts is small and the preliminary semantic quality inspection result needs to be corrected.

[0051] The correction method is: activate the dynamic node corresponding to the Agent node. The dynamic node in this embodiment is a node pre-configured with the same application and function as the Agent node. The dynamic node is free between different Agent nodes, and has different text feature distances with different Agent nodes. Before being awakened, the dynamic node is in standby state. The Agent node and the dynamic node work asynchronously. When the dynamic node is awakened, the Agent node can continue to receive new tasks.

[0052] The derived knowledge base is generated based on the content of the domain knowledge base. Synonyms, near-senses or antonyms are replaced for some of the content in the domain knowledge base to generate the corresponding derived knowledge base.

[0053] For the Agent node, each dynamic node has a different feature distance with it. After the text feature distance is calculated, the dynamic node of the corresponding value or the corresponding value range is awakened according to the value represented by the text feature distance.

[0054] S1314d. Correct and recognize the text input according to the dynamic node, and correct the quality inspection and annotation results according to the correction recognition result to generate the preliminary semantic quality inspection result.

[0055] After the corresponding dynamic node is awakened, it will re-recognize, quality check and annotate the text input based on its internal derived knowledge base. And according to the results of the re-quality check and annotation, the quality check and annotation results output by the Agent node are replaced to generate preliminary semantic quality check results.

[0056] In this embodiment, the quality inspection and annotation results output by the Agent node are used to call the corresponding standard text information. Then the text feature distance between the standard text information and the text input is calculated. The text feature distance is used to wake up the corresponding dynamic node. The derived knowledge base in the dynamic node is equal to the text feature distance with the Agent node, that is, the information in the derived knowledge base is more accurate when used to evaluate the text input. The text input is corrected and recognized according to the dynamic node. Since the knowledge base used has a higher degree of fit with the text input, the recognition result obtained is also more accurate.

[0057] In some embodiments, the domain knowledge base includes at least one of the following: Domain-specific vocabularies; Domain-related entity and relationship information; Domain-specific semantic rules and constraints; In this embodiment, the domain knowledge base includes a domain-related vocabulary, which records terms, professional terms, etc. related to the field, and these words help improve the accuracy of semantic analysis and understanding. Using the domain-related vocabulary, vocabulary matching and synonym replacement can be performed. In the semantic analysis process, the vocabulary in the matching text input is compared with the vocabulary in the vocabulary to identify keywords and terms. In addition, using the synonym information in the vocabulary, some words in the text are replaced with more commonly used or more accurate terms related to it to improve the accuracy of semantic understanding. Further, based on the domain-related vocabulary, weights or importance can be given to different words. These weights can be used for keyword extraction to help the system quickly identify important information and key content in the text. Keyword extraction can be used in application scenarios such as semantic analysis, summary generation, and subject classification, which helps to understand and process the text more deeply.

[0058] The domain knowledge base contains domain-related entities and relationships, which are used to identify and understand entities in the text and the relationships between them. Through the entity information recorded in the domain knowledge base, entity extraction and named entity recognition can be implemented. Entity extraction refers to the identification of specific entities from the text, such as names of people, places, and organizations. Named entity recognition is the classification of extracted entities, such as identifying names as character entities and place names as geographic entities. These entity recognition results can be used for subsequent semantic relationship analysis and semantic understanding. Based on domain-related entity and relationship information, relationship extraction and knowledge graph construction can be performed. Relationship extraction refers to extracting relationships between entities from the text, such as relationships between people, relationships between geographic entities, etc. The results of these relationship extractions can be used to construct domain-specific knowledge graphs to help the system understand and represent the information in the text more comprehensively. The domain knowledge base contains domain-related semantic rules and constraints, which are used to constrain and regulate the process of semantic analysis. Using domain-related semantic rules, grammar and semantic rule verification can be performed; grammar rule verification is used to check the grammatical correctness of the text to ensure that the text meets grammatical requirements, and semantic rule verification is used to check the semantic consistency of the text to ensure that the meaning of the text meets the domain-specific requirements. This can improve the accuracy and reliability of semantic analysis. Based on domain-related semantic rules and constraints, logical reasoning and rationality checks can be performed. Logical reasoning is to reason and infer the statements in the text by applying the logical rules and reasoning mechanisms in the domain knowledge base to draw new conclusions or expand existing information. Logical reasoning can help the system obtain a deeper and more comprehensive semantic understanding from the text. At the same time, rationality checks are used to verify whether the views, statements or arguments expressed in the text meet the rules and constraints set in the domain knowledge base. Through rationality checks, unreasonable or contradictory statements can be excluded, and the accuracy of understanding and analysis of the text can be improved.

[0059] The specific content of the domain knowledge base in this embodiment, including the domain-related vocabulary, entity and relationship information, and semantic rules and constraints, brings the following benefits and effects to semantic analysis and understanding: Improving the accuracy of semantic understanding: The vocabulary, entity and relationship information, and semantic rules and constraints in the domain knowledge base enable the system to more accurately identify, understand and represent the meaning in the text during the semantic analysis process; By using domain-specific knowledge, the system can better grasp the context, identify professional terms and entities, and accurately parse the relationship between them, thereby improving the accuracy and reliability of semantic understanding; Enhance information extraction and summary generation: The specific content in the domain knowledge base can be used for tasks such as keyword extraction, entity extraction, and relationship extraction. These information extraction technologies can help the system extract important information related to the domain from a large amount of text, generate concise and accurate summaries, or extract key information for further analysis and application. By utilizing the information extraction technology in the domain knowledge base, the efficiency and accuracy of information processing can be greatly improved. The domain knowledge base can be constructed and customized according to the needs of a specific field. By collecting and organizing domain-related vocabulary, entities, and relationship information, as well as formulating domain-specific semantic rules and constraints, the system can have a higher level of specialization and adaptability in a specific domain, and provide customized solutions that meet domain requirements. This enables the system to better understand and process domain-specific text information, and provide users with more accurate and intelligent semantic analysis.

[0060] In some implementations, the semantic parsing method based on the domain big model is characterized in that after obtaining the speech data to be quality inspected, it also includes: Converting the voice data into text information; The text information is preprocessed and formatted so that the text information meets the data rules of the preset Agent node, and the input data of the domain large model is obtained to convert the voice data into text information.

[0061] In this embodiment, in the semantic parsing method based on the domain big model, after obtaining the voice data to be inspected, it also includes converting the voice data into text information. Specifically, the voice data is converted into text information using automatic speech recognition technology. The conversion into text information can facilitate subsequent quality inspection and analysis. Further, the text information is preprocessed and formatted so that the text information meets the data rules of the preset Agent node. Specifically, the text is cleaned and normalized, including removing special characters, punctuation marks and extra spaces, unifying uppercase and lowercase, etc., to ensure the consistency and standardization of the input data; cleaning and normalizing the text can improve the accurate understanding of the user's intention and ensure the consistency and correctness of subsequent processing.

[0062] This implementation method converts the voice data into text information; pre-processes and formats the text information so that the text information meets the data rules of the preset Agent node. The input data of the domain big model is obtained to convert the voice data into text information, which helps the subsequent semantic analysis and model application, and improves the performance and accuracy of the system.

[0063] In some implementations, the preconfigured scheduler is further configured to: Using the preset distributed algorithm and parallel processing rules, the tasks of semantic quality inspection and semantic review are assigned to multiple Agent nodes for parallel processing; In this embodiment, the preconfigured scheduler is also configured to perform task division and distribution. Specifically, the semantic quality inspection and semantic comment tasks to be processed are divided into multiple small tasks, and these tasks are assigned to different Agent nodes using a distributed algorithm. For example, task division and distribution are performed according to the type of task, priority or other specific rules, so as to achieve parallel processing of tasks. In addition, the preconfigured message passing and communication mechanism is used to enable each Agent node to cooperate and exchange information with each other. Specifically, this is achieved through message queues, distributed data transmission or other communication mechanisms to ensure the coordination of tasks and the integration of results. In large-scale semantic quality inspection tasks, by assigning tasks to multiple Agent nodes for parallel processing, the processing efficiency and speed of the tasks can be significantly improved. For real-time semantic comment tasks, assigning tasks to multiple Agent nodes through parallel processing rules can achieve rapid response and real-time performance.

[0064] In this embodiment, the preconfigured scheduler is also configured to use a preset dynamic scheduling strategy to dynamically allocate semantic quality inspection and semantic comment tasks to different Agent nodes according to the current system load and task requirements; by real-time monitoring of the system load, including the number of current tasks, processing speed and resource utilization, etc. According to the preset scheduling strategy, the tasks of semantic quality inspection and semantic comment are dynamically allocated to idle or low-load Agent nodes to achieve load balancing and optimal utilization of system resources. In one embodiment, according to the priority and importance of the task, combined with the preset scheduling algorithm, the tasks of semantic quality inspection and semantic comment are dynamically scheduled, and the scheduling order and allocation strategy of the task can be determined by priority queues, dynamic weight allocation or other scheduling algorithms. For example, in a high-concurrency scenario, the tasks of semantic quality inspection and semantic comment are allocated to different Agent nodes through a dynamic scheduling strategy, which can achieve load balancing and avoid overloading of some nodes and affecting the overall system performance. For example, in an online speech recognition platform, tasks can be dynamically scheduled according to system load and task requirements to achieve efficient real-time speech recognition services. Alternatively, for different types of semantic quality inspection and semantic commentary tasks, dynamic scheduling can be performed based on the priority and importance of the tasks; for example, in a voice advertisement review system, review tasks for sensitive content can be given priority to ensure timely and accurate review results.

[0065] This implementation can flexibly manage the distributed processing of semantic quality inspection and semantic comment tasks through a preconfigured scheduler, dynamically adjust task priorities and scheduling rules according to real-time conditions to improve system efficiency, performance and reliability. Such scheduling strategies and mechanisms are very beneficial for scenarios such as large-scale data processing, real-time voice applications and distributed AI applications.

[0066] In some implementations, the semantic Agent node performs semantic analysis on the speech data to obtain a preliminary semantic quality inspection result, including: S1321. Processing the text of the voice data based on a preset natural language processing algorithm to obtain a first result; Specifically, speech recognition technology is used to convert speech data into text form: the converted text is cleaned and preprocessed, including noise removal, spelling error correction, stemming, stop word removal, etc. This can improve the effect of subsequent natural language processing algorithms, and keyword extraction algorithms (such as TF-IDF, TextRank, etc.) are used to extract keywords from the text. Keywords can be used for feature representation and semantic analysis of subsequent tasks, and natural language processing technology is used for part-of-speech tagging and entity recognition to obtain the part-of-speech of words in sentences and identify named entities. Based on the above processing, the first result is obtained.

[0067] S1322, performing sentiment analysis on the voice data based on a preset sentiment analysis algorithm to obtain a second result; Specifically, a pre-built sentiment dictionary is used for analysis. The sentiment dictionary contains the correspondence between words and sentiment polarity. The sentiment dictionary can include positive, negative, and neutral words. The sentiment dictionary and other feature extraction techniques (such as n-gram, part-of-speech tagging, etc.) are used to extract sentiment features from the text of the speech data. These features can include sentiment word frequency, sentiment intensity, sentiment sentence pattern, etc. The trained sentiment classification model is used to perform sentiment classification prediction on the speech data to obtain sentiment analysis results, such as positive, negative, or neutral.

[0068] S1323: Extract context information of the voice data.

[0069] Specifically, the size of the context window is predefined, which can be a number of sentences before and after or a fixed time window. The size of the context window depends on the task requirements and the characteristics of the speech data. Context text is extracted from the speech data, including context sentences or text within a specific time window. The context information is extracted by matching the timestamp of the speech data with the corresponding text. The context text is converted into a vector form that can be processed by a computer using text representation technology (such as word embedding, text vectorization, etc.). In this way, the context information can be input as a feature into the subsequent semantic analysis model.

[0070] S1324. Perform semantic recognition on the speech data based on the correlation between the context information, the first result, and the second result to obtain a preliminary semantic quality inspection result.

[0071] Specifically, the first result (text processing result) and the second result (sentiment analysis result) are fused with context information. Different features can be combined by vector concatenation, weighted average, etc. to form a comprehensive feature representation, and the trained semantic analysis model is used to predict the semantic classification of the comprehensive features of the speech data to obtain preliminary semantic quality inspection results.

[0072] The semantic Agent node of this embodiment can perform multi-dimensional semantic analysis on voice data, including text processing, sentiment analysis, context information extraction and comprehensive semantic analysis, so as to obtain preliminary semantic quality inspection results, provide valuable information for subsequent task processing and decision-making, and improve the accuracy and comprehensiveness of semantic analysis.

[0073] In some implementations, generating the semantic analysis result of the speech data according to the semantic review result of each review Agent node includes: S1511, obtaining a diversity evaluation index, evaluating and optimizing the diversity of the semantic review results of each review Agent node, and obtaining an optimized review result; Specifically, a diversity evaluation index is obtained. The diversity evaluation index is used to evaluate the diversity of semantic review results, such as vocabulary coverage, sentence diversity, semantic category distribution, etc. These indicators can reflect the richness and diversity of semantic review results. According to the defined diversity index, the diversity score of the semantic review results of each review Agent node is calculated. In one embodiment, the calculation is performed using statistical methods or information entropy and other techniques. According to the diversity evaluation results, the semantic review results are optimized. The diversity of the results can be improved by introducing a diversity-enhanced generation model, adjusting model parameters, or using data enhancement methods to obtain optimized review results.

[0074] S1512, obtaining the trustworthiness and historical performance of each review Agent node; Specifically, the reliability and performance of each review agent node are evaluated based on its trust index and historical performance. For example, the level of trust is determined by using weighted average of evaluation indicators, threshold determination, and other methods.

[0075] S1513, according to the trust and historical performance of the review Agent node, configure the weight of each review Agent node; Specifically, according to the trust evaluation results, a corresponding weight is assigned to each review agent node. Agent nodes with higher reliability can be given higher weights to make them more influential in generating semantic analysis results.

[0076] S1514: weighting the optimized review results of each review Agent node.

[0077] Specifically, the optimized review results of each review Agent node are weighted, and weighted average, weighted summation or other weighted strategies may be used to merge the results according to the weights of the nodes to obtain a weighted comprehensive review result.

[0078] S1515: Generate semantic analysis results of the speech data according to the weighted optimization review results.

[0079] Based on the weighted optimized review results, the semantic analysis results of the voice data are generated. This can be a piece of text that describes the semantic information, emotional tendency, etc. in the voice data. The generated semantic analysis results are output in a readable or displayable format, such as text, so that users can understand and use them. This can be used as the final output of the semantic agent node to provide semantic interpretation and analysis of the voice data.

[0080] Through the above expansion steps, this implementation method can evaluate and optimize the diversity of the semantic review results of each review Agent node, consider the trust and historical performance of the node, configure the node weight, and finally generate the semantic analysis results of the voice data based on the weighted results. This can improve the accuracy and diversity of semantic analysis, while taking into account the reliability and performance of different nodes, thereby providing a more comprehensive and reliable semantic analysis service.

[0081] Please refer to Figure 2 , Figure 2 Schematic diagram of the basic structure of the semantic parsing device based on the domain big model of this embodiment.

[0082] like Figure 2 As shown, a semantic parsing device based on a domain big model includes: a data acquisition module 1100, used to acquire voice data to be quality inspected; a data input module 1200, used to transmit the voice data to the domain big model of multiple Agent nodes; wherein the domain big model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; a semantic quality inspection module 1300, used to call at least one semantic Agent node of the domain big model to perform semantic analysis on the voice data based on a preconfigured scheduler to obtain a preliminary semantic quality inspection result; a quality inspection review module 1400, used to call at least two review Agent nodes of the domain big model to review the preliminary semantic quality inspection result based on a preconfigured scheduler to obtain a semantic review result of each review Agent node; a result feedback module 1500, used to generate a semantic parsing result of the voice data according to the semantic review result of each review Agent node.

[0083] The above-mentioned semantic parsing device based on the domain big model obtains the voice data to be inspected; transmits the voice data to the domain big model of multiple agent nodes; wherein the domain big model is configured with multiple agent nodes, each agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; based on a preconfigured scheduler, calls at least one semantic agent node of the domain big model to perform semantic analysis on the voice data to obtain a preliminary semantic quality inspection result; based on the preconfigured scheduler, calls at least two review agent nodes of the domain big model to review the preliminary semantic quality inspection result to obtain the semantic review result of each review agent node; generates the semantic parsing result of the voice data according to the semantic review result of each review agent node; utilizes the collaboration of semantic quality inspection models and review models in multiple professional fields to effectively process the voice data to be inspected, and provide accurate and comprehensive semantic parsing results, thereby improving the accuracy and reliability of semantic analysis, thereby meeting the needs of voice data processing.

[0084] Optionally, the semantic quality inspection module is further used to: Obtaining text input of the voice data; Perform matching according to the text input to obtain a corresponding domain knowledge base; Calling the Agent node corresponding to the domain knowledge base to perform quality inspection and annotation on the text input; wherein the quality inspection and annotation are based on the relevant information in the domain knowledge base, Generate preliminary semantic quality inspection results based on the quality inspection and annotation results.

[0085] Optionally, the semantic quality inspection module is further used to: Obtaining standard text information according to the quality inspection and annotation results; Calculating a text feature distance between the text input and the standard text information; When the text feature distance is greater than a preset threshold, a preset dynamic node is activated according to the text feature distance; wherein a derived knowledge base is provided in the dynamic node, and the feature distance between the derived knowledge base and the domain knowledge base is equal to the text feature distance; The text input is corrected and recognized according to the moving node, and the quality inspection and annotation results are corrected according to the correction recognition result to generate the preliminary semantic quality inspection result.

[0086] Optionally, the domain knowledge base includes at least one of the following: Domain-specific vocabularies; Domain-related entity and relationship information; Domain-specific semantic rules and constraints.

[0087] Optionally, the data input module is further used for: Converting the voice data into text information; The text information is preprocessed and formatted so that the text information meets the data rules of the preset Agent node, and the input data of the domain large model is obtained.

[0088] Optionally, the preconfigured scheduler is further configured as: Use the preset distributed algorithm and parallel processing rules to assign semantic quality inspection and semantic comment tasks to multiple Agent nodes for parallel processing; and Using the preset dynamic scheduling strategy, the semantic quality inspection and semantic review tasks are dynamically allocated to different Agent nodes according to the current system load and task requirements.

[0089] Optionally, the semantic quality inspection module is further used to: Processing the text of the voice data based on a preset natural language processing algorithm to obtain a first result; Performing sentiment analysis on the speech data based on a preset sentiment analysis algorithm to obtain a second result; Extracting context information of the speech data; Based on the correlation between the context information, the first result, and the second result, semantic recognition is performed on the speech data to obtain a preliminary semantic quality inspection result.

[0090] Optionally, the result feedback module is further used for: Obtaining a diversity evaluation index, evaluating and optimizing the diversity of the semantic review results of each review Agent node, and obtaining an optimized review result; Obtain the trust and historical performance of each review agent node; According to the trust and historical performance of the review agent node, the weight of each review agent node is configured; Performing weighted processing on the optimized review results of each review Agent node; The semantic analysis result of the speech data is generated according to the weighted optimization review result.

[0091] To solve the above technical problems, the present application also provides a computer device. Figure 3 , Figure 3 This is a basic structural block diagram of the computer device in this embodiment.

[0092] like Figure 3As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected via a system bus. Among them, the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence, and the computer-readable instructions are executed by the processor. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor can execute a semantic parsing method based on a large domain model. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0093] In this embodiment, the processor is used to execute Figure 2 The memory stores the program codes and various data required to execute the above modules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program codes and data required to execute all sub-modules in the semantic parsing device based on the domain big model, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

[0094] The computer device acquires voice data to be inspected; transmits the voice data to a domain big model of multiple agent nodes; wherein the domain big model is configured with multiple agent nodes, each agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; based on a preconfigured scheduler, at least one semantic agent node of the domain big model is called to perform semantic analysis on the voice data to obtain a preliminary semantic quality inspection result; based on the preconfigured scheduler, at least two review agent nodes of the domain big model are called to review the preliminary semantic quality inspection result to obtain a semantic review result of each review agent node; a semantic analysis result of the voice data is generated according to the semantic review result of each review agent node; the collaboration of semantic quality inspection models and review models in multiple professional fields is utilized to effectively process the voice data to be inspected, and provide accurate and comprehensive semantic analysis results, thereby improving the accuracy and reliability of semantic analysis, thereby meeting the needs of voice data processing.

[0095] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the semantic parsing method based on the domain big model described in any of the above embodiments.

[0096] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0097] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0098] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A semantic parsing method based on a large domain model, characterized in that: include: Obtain voice data to be tested; The voice data is transmitted to a domain big model of multiple Agent nodes; wherein the domain big model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain big model is associated with a domain knowledge base of a preset domain; Based on a preconfigured scheduler, calling at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain preliminary semantic quality inspection results; Based on the preconfigured scheduler, at least two review Agent nodes of the domain big model are called to review the preliminary semantic quality inspection results to obtain the semantic review results of each review Agent node; The semantic analysis result of the speech data is generated according to the semantic review result of each review Agent node.

2. The semantic parsing method based on a domain big model according to claim 1 is characterized in that: The calling of at least one semantic Agent node of the domain big model to perform semantic analysis on the speech data to obtain a preliminary semantic quality inspection result includes: Obtaining text input of the voice data; Perform matching according to the text input to obtain a corresponding domain knowledge base; Calling the Agent node corresponding to the domain knowledge base to perform quality inspection and annotation on the text input, wherein the quality inspection and annotation are based on relevant information in the domain knowledge base; Generate preliminary semantic quality inspection results based on the quality inspection and annotation results.

3. The semantic parsing method based on a domain big model according to claim 2 is characterized in that: Generating preliminary semantic quality inspection results according to the quality inspection and annotation results includes: Obtaining standard text information according to the quality inspection and annotation results; Calculating a text feature distance between the text input and the standard text information; When the text feature distance is greater than a preset threshold, a preset dynamic node is activated according to the text feature distance; wherein a derived knowledge base is provided in the dynamic node, and the feature distance between the derived knowledge base and the domain knowledge base is equal to the text feature distance; The text input is corrected and recognized according to the moving node, and the quality inspection and annotation results are corrected according to the correction recognition result to generate the preliminary semantic quality inspection result.

4. The semantic parsing method based on a domain big model according to claim 1 is characterized in that: After obtaining the voice data to be quality-checked, the method further includes: Converting the voice data into text information; The text information is preprocessed and formatted so that the text information meets the data rules of the preset Agent node, and the input data of the domain large model is obtained.

5. The semantic parsing method based on a domain big model according to claim 1 is characterized in that: The preconfigured scheduler is also configured to: Using the preset distributed algorithm and parallel processing rules, the tasks of semantic quality inspection and semantic commenting are assigned to multiple Agent nodes for parallel processing; and, Using the preset dynamic scheduling strategy, the semantic quality inspection and semantic review tasks are dynamically allocated to different Agent nodes according to the current system load and task requirements.

6. The semantic parsing method based on a domain big model according to claim 1 is characterized in that: The semantic Agent node performs semantic analysis on the speech data to obtain a preliminary semantic quality inspection result, and further includes: Processing the text of the voice data based on a preset natural language processing algorithm to obtain a first result; Performing sentiment analysis on the speech data based on a preset sentiment analysis algorithm to obtain a second result; Extracting context information of the speech data; Based on the correlation between the context information, the first result, and the second result, semantic recognition is performed on the speech data to obtain a preliminary semantic quality inspection result.

7. The semantic parsing method based on a domain big model according to claim 1 is characterized in that: The generating of the semantic analysis result of the speech data according to the semantic review result of each review Agent node includes: Obtaining a diversity evaluation index, evaluating and optimizing the diversity of the semantic review results of each review Agent node, and obtaining an optimized review result; Obtain the trust and historical performance of each review agent node; According to the trust and historical performance of the review agent node, the weight of each review agent node is configured; Performing weighted processing on the optimized review results of each review Agent node; The semantic analysis result of the speech data is generated according to the weighted optimization review result.

8. A semantic parsing device based on a domain big model, characterized in that: include: A data acquisition module, used to acquire voice data to be inspected; A data input module, used to transmit the voice data to a domain model with multiple Agent nodes; wherein the domain model is configured with multiple Agent nodes, each Agent node represents a semantic quality inspection model; the domain model is associated with a domain knowledge base of a preset domain; A semantic quality inspection module, configured to perform semantic analysis on the speech data based on a preconfigured scheduler by calling at least one semantic Agent node of the domain big model to obtain a preliminary semantic quality inspection result; A quality inspection and review module, which is used to call at least two review Agent nodes of the domain big model to review the preliminary semantic quality inspection results based on a preconfigured scheduler, and obtain the semantic review results of each review Agent node; The result feedback module is used to generate the semantic analysis result of the speech data according to the semantic review result of each review Agent node.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the semantic parsing method based on the domain big model as claimed in any one of claims 1 to 7.

10. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to perform the steps of the semantic parsing method based on a large domain model as claimed in any one of claims 1 to 7.

Images