Information query method and device based on large language model

Abstract

The invention provides an information query method and device based on a large language model, and the method comprises the steps: obtaining first voice data inputted by a user, and converting the first voice data into voice character key contents through a preset large language model; then intention understanding analysis is carried out on the voice and character key content by using a large language model in combination with a pre-constructed intention knowledge base, suggested query information is generated based on an intention understanding analysis result by using the large language model, and then a query instruction is generated based on the suggested query information; and executing the query instruction to call the target service interface and / or query the target database, then generating natural language text content based on a query result by using a large language model, and displaying the natural language text content through a user interface. By adopting the method and the device, the simplicity, the response speed, the accuracy and the intuition of information query are improved, and the requirements of high efficiency and high accuracy of querying required information by a user can be met.

Description

Technical Field

[0001] The present invention relates to the field of computer information processing technology, and in particular to an information query method and device based on a large language model. Background Art

[0002] With the advent of the big data era, the amount of information has increased dramatically, and users' demand for high efficiency and high accuracy in information query is increasing. Traditional information query systems usually implement information query by extracting keywords contained in user input information, but this information query method has low accuracy in processing complex information query and natural language understanding, and usually requires users to use SQL statements to query to improve the accuracy of information query, which requires high programming ability of users. Therefore, traditional information query systems are difficult to meet users' needs for high efficiency and high accuracy in querying the required information. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide an information query method and device based on a large language model to alleviate the shortcomings of existing information query methods in processing complex information queries and natural language understanding.

[0004] In a first aspect, an embodiment of the present invention provides an information query method based on a large language model, comprising: obtaining first voice data input by a user, and converting the first voice data into voice-text key content using a preset large language model; using the large language model in combination with a pre-built intent knowledge base to perform intent understanding analysis on the voice-text key content, and using the large language model to generate suggested query information based on the intent understanding analysis result; wherein the intent understanding analysis result includes a target intent and its corresponding query keywords and query topics, and the suggested query information represents a target service interface and / or target database corresponding to the target intent; generating a query instruction based on the suggested query information, and executing the query instruction to call the target service interface and / or query the target database; using the large language model to generate natural language text content based on the query result, and displaying the natural language text content through a user interface; wherein the query result is structured information.

[0005] In a second aspect, an embodiment of the present invention further provides an information query device based on a large language model, comprising: a conversion module, used to obtain first voice data input by a user, and use a preset large language model to convert the first voice data into voice and text key content; an understanding module, used to use the large language model in combination with a pre-built intention knowledge base to perform intention understanding analysis on the voice and text key content, and use the large language model to generate suggested query information based on the intention understanding analysis result; wherein the intention understanding analysis result includes a target intention and its corresponding query keywords and query topics, and the suggested query information represents a target service interface and / or target database corresponding to the target intention; a query module, used to generate a query instruction based on the suggested query information, and execute the query instruction to call the target service interface and / or query the target database; a display generation module, used to use the large language model to generate natural language text content based on the query result, and display the natural language text content through a user interface; wherein the query result is structured information.

[0006] The embodiment of the present invention provides an information query method and device based on a large language model, which obtains first voice data input by a user, and uses a preset large language model to convert the first voice data into voice and text key content, then uses the large language model in combination with a pre-built intention knowledge base to perform intent understanding analysis on the voice and text key content, and uses the large language model to generate suggested query information based on the intention understanding analysis results, then generates a query instruction based on the suggested query information, and executes the query instruction to call the target service interface and / or query the target database, and then uses the large language model to generate natural language text content based on the query results, and displays the natural language text content through a user interface. Using the above technology, the large language model can be used to understand the voice and text key content converted from the voice data input by the user, and then the query result is obtained according to the understanding result of the large language model, so that the large language model is used to provide the user with easy-to-understand natural language text content based on the query result and intuitively display it through the user interface, thereby improving the simplicity, response speed, accuracy and intuitiveness of information query, and meeting the user's requirements for high efficiency and high accuracy in querying required information.

[0007] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0008] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0010] Figure 1 A schematic diagram of a flow chart of an information query method based on a large language model in an embodiment of the present invention;

[0011] Figure 2 This is an example diagram of an information query method based on a large language model in an embodiment of the present invention;

[0012] Figure 3 It is a structural schematic diagram of an information query device based on a large language model in an embodiment of the present invention;

[0013] Figure 4 Schematic diagram of the structure of another information query device based on a large language model in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] At present, traditional information query systems have low accuracy in processing complex information queries and natural language understanding, and usually require users to use SQL statements to query in order to improve the accuracy of information queries, which requires high programming skills from users. Therefore, traditional information query systems are difficult to meet users' needs for high efficiency and high accuracy in querying the information they need.

[0016] Based on this, the present invention provides an information query method and device based on a large language model, which can alleviate the shortcomings of existing information query methods in processing complex information queries and natural language understanding.

[0017] To facilitate understanding of this embodiment, firstly, a method for information query based on a large language model disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method may include the following steps:

[0018] Step S102, obtaining first voice data input by a user, and using a preset large language model to convert the first voice data into voice text key content.

[0019] Step S104, use the large language model in combination with a pre-built intent knowledge base to perform intent understanding analysis on the key contents of the speech and text, and use the large language model to generate suggested query information based on the intent understanding analysis results.

[0020] Among them, the intent understanding analysis results may include the target intent and its corresponding query keywords and query topics, and the suggested query information may represent the target service interface and / or target database corresponding to the target intent.

[0021] Step S106: generating a query instruction based on the suggested query information, and executing the query instruction to call a target service interface and / or query a target database.

[0022] Step S108: Generate natural language text content based on the query result using the large language model, and display the natural language text content through a user interface.

[0023] The query result may be structured information.

[0024] An information query method based on a large language model provided in an embodiment of the present invention obtains first voice data input by a user, and uses a preset large language model to convert the first voice data into voice-text key content, then uses the large language model in combination with a pre-built intention knowledge base to perform intent understanding analysis on the voice-text key content, and uses the large language model to generate suggested query information based on the intention understanding analysis result, then generates a query instruction based on the suggested query information, and executes the query instruction to call a target service interface and / or query a target database, and then uses the large language model to generate natural language text content based on the query result, and displays the natural language text content through a user interface. By adopting the above technology, a large language model can be used to understand the voice-text key content converted from the voice data input by the user, and then the query result is obtained according to the understanding result of the large language model, so that the large language model is used to provide the user with easy-to-understand natural language text content based on the query result and intuitively display it through the user interface, thereby improving the simplicity, response speed, accuracy and intuitiveness of information query, and meeting the user's requirements for high efficiency and high accuracy in querying the required information.

[0025] As a possible implementation, before performing the above step S104 of using a large language model in combination with a pre-built intent knowledge base to perform intent understanding analysis on key contents of speech and text, it may also include: constructing an intent knowledge base based on RAG (Retrieval-Augmented Generation) technology.

[0026] The intent knowledge base may include multiple intent vector sets pointing to intents, and the intent vectors in each intent vector set may be word vectors, sentence vectors, vectors that appear in groups in the form of dialogues, etc. Based on this, the intent understanding analysis of the key contents of speech and text using the large language model combined with the pre-built intent knowledge base in the above step S104 may include:

[0027] Step 1: Use a large language model to generate a query vector set corresponding to the key content of the voice and text, and use the large language model to determine a target intent vector set in the intent knowledge base that matches the query vector set.

[0028] Exemplarily, a large language model can be used to perform word segmentation, remove modal particles, and other processing on key content of spoken text, and then vectorize the processed text data, and form a query vector set with the query vectors obtained after vectorization; after obtaining the query vector set, the large language model can be used to calculate the matching similarity (such as cosine similarity, Euclidean distance, etc.) between each query vector in the query vector set and each intent vector in each intent vector set, and the large language model can be used to determine the intent vector set containing the target intent vector whose matching similarity is greater than a preset similarity threshold as the target intent vector set.

[0029] Step 2: Use the large language model to generate query keywords, query topics, and intent understanding content based on the target intent corresponding to the target intent vector set, combined with the large language model analysis.

[0030] As a possible implementation manner, a first correspondence between intent and service interface and / or a second correspondence between intent and database may be pre-established on the large language model, and the suggested query information may include interface parameters of the target service interface (such as interface name, IP address, etc.) and / or database parameters of the target database (such as database name, database table name, database field, etc.) as well as query keywords and query subjects, and both the first correspondence and the second correspondence may adopt data mapping relationships; based on this, in the above step S104, using the large language model to generate suggested query information based on the intention understanding analysis results may include: using the large language model to determine the target service interface based on the first correspondence, and / or, using the large language model to determine the target database based on the second correspondence; using the large language model to combine the interface parameters and / or database parameters as well as query keywords and query subjects into suggested query information.

[0031] Exemplarily, when a mapping relationship between intent and service interface is pre-established on the large language model, the large language model only needs to obtain the specified intent to use the mapping relationship to determine the specified service interface corresponding to the specified intent, thereby facilitating subsequent calls to the query service provided by the specified service interface; when a mapping relationship between intent and database is pre-established on the large language model, the large language model only needs to obtain the specified intent to use the mapping relationship to determine the specified database corresponding to the specified intent, thereby facilitating subsequent queries to the specified database.

[0032] As a possible implementation, generating a query instruction based on the suggested query information in the above step S106 may include: if the suggested query information includes interface parameters, generating a first query instruction based on the interface parameters, query keywords and query subjects; if the suggested query information includes database parameters, generating a second query instruction based on the database parameters, query keywords and query subjects; wherein the first query instruction is used to call the query service provided by the target service interface, and the second query instruction is used to query the target database.

[0033] As a possible implementation, executing a query instruction to call the target service interface and / or query the target database in the above step S106 may include: executing a first query instruction to call the query service provided by the target service interface, and / or executing a second query instruction to query the target database.

[0034] As a possible implementation, after performing the above step S108 using the large language model to generate natural language text content based on the query results, it may also include: using the large language model to generate visualization information based on the query results, and displaying the visualization information through a user interface; wherein the visualization information may be presented in the form of dialogues, charts, sentences, paragraphs, etc., which are not limited.

[0035] As a possible implementation, after performing the above step S108 of using the large language model to generate natural language text content based on the query result, it may also include: using the large language model to convert the natural language text content into second voice data, and playing the second voice data.

[0036] Exemplarily, the large language model can support the TTS (Text To Speech) function. After the large language model generates natural language text content that is easy for humans to understand from the structured information obtained after executing the query instruction, the large language model can use the TTS function to convert the natural language text content into voice data, and then play the voice data through a voice output device.

[0037] For ease of understanding, the implementation of the above-mentioned information query method based on a large language model is described as follows by taking a specific application as an example.

[0038] See also Figure 2 As shown, the above-mentioned information query method based on the large language model can be mainly implemented by the large language model, the intention understanding agent (intelligent agent), the query generation agent, the execution agent, and the result generation agent. The knowledge base is pre-built on the large language model using the RAG technology; the specific implementation method of the above-mentioned information query method based on the large language model may include the following steps:

[0039] Step S1, the user inputs voice to express the content he wants to obtain, and the large language model recognizes the voice content of the user input and generates voice text content based on the recognition result.

[0040] In step S2, the intent understanding agent uses a large language model combined with a knowledge base to perform intent understanding analysis on the speech and text content to generate keywords, topics, and intent understanding content of the target query intent.

[0041] Specifically, the knowledge base includes multiple intent vectors pointed to by query intent (such as word vectors, sentence vectors, vectors that appear in groups in the form of dialogue, etc.). The intent understanding agent can first use the large language model to convert the voice and text content into one or more query vectors, and then use the large language model to determine the target intent vector in the knowledge base that matches each query vector by matching the query vector with the intent vector in the knowledge base, and then use the large language model to generate the keywords, topics and intent understanding content of the target query intent corresponding to the target intent vector. The specific operation method of matching the query vector with the intent vector in the knowledge base can be: for each query vector, calculate the cosine similarity between the query vector and each intent vector in the knowledge base, and use the intent vector in the knowledge base whose cosine similarity is greater than the preset cosine similarity threshold as the target intent vector that matches the query vector.

[0042] Step S3, the query generation agent uses the large language model to understand the content based on keywords, topics and intent to determine the target query service interface for the recommended call and the target database for the recommended query, and appends keywords and topics.

[0043] Specifically, a first mapping relationship between query intent and query service interface information and a second mapping relationship between query intent and database table name are pre-established on the large language model. After the query generation agent uses the large language model to obtain the keywords, topics and intent understanding content of the target query intent, it can use the large language model to determine the target query service interface information corresponding to the target query intent based on the first mapping relationship and the target database table name corresponding to the target query intent based on the second mapping relationship. The query service interface with the target query service interface information is the target query service interface recommended for call, and the database with the target database table name is the target database recommended for query; then the query generation agent binds the target query service interface information and the target database table name as well as the keywords, topics and intent understanding content of the target query intent.

[0044] Step S4, the execution agent uses the large language model to generate query instructions, execute query instructions and obtain query results based on the target query service interface and the target database as well as the keywords, topics and intent understanding content of the target query intent.

[0045] Specifically, the execution agent can first use the large language model to generate query instructions for calling the query service provided by the target query service interface and querying the target database according to the target query service interface information and the target database table name as well as the keywords, topics and intent understanding content of the target query intent, and then use the large language model to execute the query instructions to respectively call the query service provided by the target query service interface and query the target database and respectively obtain the results returned by the target query service interface and the query target database.

[0046] In step S5, the result generation agent uses the large language model to understand the results returned by the query and generates natural language text content that is easy for humans to understand based on the subject of the target query intent.

[0047] Step S6, the result generation agent uses the large language model to generate visual information based on the results returned by the query and displays it on the interface. At the same time, the TTS function of the large language model is used to convert the natural language text content into voice data and play it through an external playback device.

[0048] The visual information displayed on the interface can mainly take the following three forms: dialogue, chart, and paragraph.

[0049] In addition, the results returned by the query can be displayed directly on the interface.

[0050] The above information query method based on large language model brings the following beneficial effects:

[0051] By leveraging the reasoning power of the large model, the intent of the voice and text content generated based on the user's input voice is understood, the query service interface and database are analyzed to generate query instructions, the query instructions are executed and the query results are obtained, which improves the simplicity, response speed and accuracy of information query; the query results are converted into natural language text that is easy for humans to understand, and the intuitiveness of information query is improved by combining interface display and voice playback.

[0052] Based on the above-mentioned information query method based on a large language model, an embodiment of the present invention further provides an information query device based on a large language model, see Figure 3 As shown, the device may include the following modules:

[0053] The conversion module 302 is used to obtain first voice data input by a user and convert the first voice data into voice text key content using a preset large language model.

[0054] The understanding module 304 is used to use the large language model in combination with a pre-built intent knowledge base to perform intent understanding analysis on the key content of the speech and text, and use the large language model to generate suggested query information based on the intent understanding analysis results; wherein the intent understanding analysis results include the target intent and its corresponding query keywords and query topics, and the suggested query information represents the target service interface and / or target database corresponding to the target intent.

[0055] The query module 306 is configured to generate a query instruction based on the suggested query information, and execute the query instruction to call the target service interface and / or query the target database.

[0056] The generation and display module 308 is used to generate natural language text content based on the query result using the large language model, and display the natural language text content through a user interface; wherein the query result is structured information.

[0057] By adopting the above-mentioned information query device based on the large language model, the large language model can be used to understand the key content of the voice and text converted from the voice data input by the user, and then the query results are obtained according to the understanding results of the large language model, so as to use the large language model to provide users with easy-to-understand natural language text content based on the query results and intuitively display it through the user interface, thereby improving the simplicity, response speed, accuracy and intuitiveness of information query, and can meet the user's needs for high efficiency and high accuracy in querying the required information.

[0058] A first correspondence between intent and service interface and / or a second correspondence between intent and database may be pre-established on the large language model, and the suggested query information may include interface parameters of the target service interface and / or database parameters of the target database as well as the query keywords and the query subject; based on this, the understanding module 304 may also be used to: use the large language model to determine the target service interface based on the first correspondence, and / or, use the large language model to determine the target database based on the second correspondence; use the large language model to combine the interface parameters and / or the database parameters as well as the query keywords and the query subject into the suggested query information.

[0059] The above-mentioned query module 306 can also be used for: if the suggested query information includes the interface parameters, then generating a first query instruction based on the interface parameters, the query keywords and the query subject; if the suggested query information includes the database parameters, then generating a second query instruction based on the database parameters, the query keywords and the query subject.

[0060] The query module 306 may also be used to: execute the first query instruction to call the query service provided by the target service interface, and / or execute the second query instruction to query the target database.

[0061] The generation and display module 308 may also be used to: after using the large language model to generate natural language text content based on the query results, use the large language model to generate visualization information based on the query results, and display the visualization information through a user interface.

[0062] The generation and display module 308 may also be used to: after using the large language model to generate natural language text content based on the query result, use the large language model to convert the natural language text content into second voice data, and play the second voice data.

[0063] In the above Figure 3 Based on the information query device based on the large language model shown in FIG. , the embodiment of the present invention also provides another information query device based on the large language model, see Figure 4 As shown, the device may also include:

[0064] The construction module 310 is used to construct the intention knowledge base based on the RAG technology.

[0065] The above-mentioned intention knowledge base may include a plurality of intention-pointing intention vector sets; based on this, the above-mentioned understanding module 304 may also be used to: use the large language model to generate a query vector set corresponding to the key content of the voice and text, and use the large language model to determine the target intention vector set in the intention knowledge base that matches the query vector set; use the large language model to generate the query keywords, the query topics and the intention understanding content based on the target intention corresponding to the target intention vector set, combined with the large language model analysis.

[0066] The above-mentioned understanding module 304 can also be used to: use the large language model to calculate the matching similarity between each query vector in the query vector set and each intention vector in each intention vector set, and use the large language model to determine the intention vector set where the target intention vector whose matching similarity is greater than a preset similarity threshold is located as the target intention vector set.

[0067] The information query device based on a large language model provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned information query method embodiment based on a large language model. For the sake of brief description, for matters not mentioned in the embodiment of the information query device based on a large language model, reference may be made to the corresponding contents in the aforementioned information query method embodiment based on a large language model.

[0068] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0069] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An information query method based on a large language model, characterized in that: include: Acquire first voice data input by a user, and convert the first voice data into voice text key content using a preset large language model; Using the large language model in combination with a pre-built intent knowledge base to perform intent understanding analysis on the key contents of the speech text, and using the large language model to generate suggested query information based on the intent understanding analysis results; wherein the intent understanding analysis results include a target intent and its corresponding query keywords and query topics, and the suggested query information represents a target service interface and / or a target database corresponding to the target intent; Generate a query instruction based on the suggested query information, and execute the query instruction to call the target service interface and / or query the target database; The large language model is used to generate natural language text content based on the query result, and the natural language text content is displayed through a user interface; wherein the query result is structured information.

2. The information query method based on a large language model according to claim 1, characterized in that: Before using the large language model to perform intent understanding analysis on the key contents of the speech text based on the pre-built intent knowledge base, the method further includes: Building the intention knowledge base based on RAG technology; The intention knowledge base includes a plurality of intention vector sets pointing to intentions; using the large language model in combination with the pre-built intention knowledge base to perform intention understanding analysis on the key contents of the speech and text, including: Using the large language model to generate a query vector set corresponding to the key content of the speech text, and using the large language model to determine a target intention vector set in the intention knowledge base that matches the query vector set; The large language model is used to generate the query keywords, the query topics and the intent understanding content based on the target intent corresponding to the target intent vector set in combination with the large language model analysis.

3. The information query method based on a large language model according to claim 2, characterized in that: Using the large language model to determine a target intent vector set in the intent knowledge base that corresponds to and matches the query vector set includes: The large language model is used to calculate the matching similarity between each query vector in the query vector set and each intent vector in each intent vector set, and the large language model is used to determine the intent vector set where the target intent vector whose matching similarity is greater than a preset similarity threshold is located as the target intent vector set.

4. The information query method based on a large language model according to claim 1, characterized in that: A first correspondence between an intent and a service interface and / or a second correspondence between an intent and a database is pre-established on the large language model, and the suggested query information includes interface parameters of the target service interface and / or database parameters of the target database as well as the query keywords and the query subject; Generate suggested query information based on the intention understanding analysis result using the large language model, including: Determine the target service interface based on the first corresponding relationship using the large language model, and / or determine the target database based on the second corresponding relationship using the large language model; The interface parameters and / or the database parameters, the query keywords and the query subject are combined into the suggested query information using the large language model.

5. The information query method based on a large language model according to claim 4, characterized in that: Generating a query instruction based on the suggested query information includes: If the suggested query information includes the interface parameter, generating a first query instruction based on the interface parameter, the query keyword and the query subject; If the suggested query information includes the database parameter, a second query instruction is generated based on the database parameter, the query keyword and the query subject.

6. The information query method based on a large language model according to claim 5, characterized in that: Executing the query instruction to call the target service interface and / or query the target database includes: The first query instruction is executed to call the query service provided by the target service interface, and / or the second query instruction is executed to query the target database.

7. The information query method based on a large language model according to claim 1, characterized in that: After using the large language model to generate natural language text content based on the query result, the method further includes: The large language model is used to generate visualization information based on the query result, and the visualization information is displayed through a user interface.

8. The information query method based on a large language model according to claim 1, characterized in that: After using the large language model to generate natural language text content based on the query result, the method further includes: The natural language text content is converted into second voice data using the large language model, and the second voice data is played.

9. An information query device based on a large language model, characterized in that: include: A conversion module, used to obtain first voice data input by a user, and convert the first voice data into voice text key content using a preset large language model; An understanding module, used to use the large language model in combination with a pre-built intention knowledge base to perform intention understanding analysis on the key contents of the speech text, and use the large language model to generate suggested query information based on the intention understanding analysis result; wherein the intention understanding analysis result includes a target intention and its corresponding query keywords and query topics, and the suggested query information represents a target service interface and / or a target database corresponding to the target intention; A query module, configured to generate a query instruction based on the suggested query information, and execute the query instruction to call the target service interface and / or query the target database; A generation and display module is used to generate natural language text content based on the query result using the large language model, and display the natural language text content through a user interface; wherein the query result is structured information.

10. The information query device based on a large language model according to claim 9, characterized in that: Also includes: A construction module, used for constructing the intention knowledge base based on RAG technology; The intention knowledge base includes a plurality of intention vector sets pointed to by intentions; the understanding module is further used for: Using the large language model to generate a query vector set corresponding to the key content of the speech text, and using the large language model to determine a target intention vector set in the intention knowledge base that matches the query vector set; The large language model is used to generate the query keywords, the query topics and the intent understanding content based on the target intent corresponding to the target intent vector set in combination with the large language model analysis.

Images