LLM model training method, device, equipment and storage medium thereof

Abstract

The embodiments of the present application belong to the field of digital medical technology and are applied to digital medical doctor-patient dialogue consultation training scenarios, and relate to an LLM model training method, apparatus, equipment, and storage medium thereof, including obtaining a natural language corpus to be trained on the LLM model; sequentially numbering each dialogue sentence in the text to be classified according to the dialogue time information; organizing a text dialogue sequence diagram according to the dialogue subject of the dialogue text and the sequence of the dialogue, and then screening each business node according to the part of speech of the words in the dialogue text, and training an LLM model based on the text dialogue sequence diagram and each business node, thereby ensuring that the LLM model training is performed according to the globality, intermediate expressiveness, and multi-dimensional interactivity of the dialogue, obtaining a dialogue model that is more in line with the medical scenario, facilitating the construction of the digital medical platform and the consultation dialogue business processing, and improving the semantic recognition accuracy of the digital medical industry dialogue system.

Description

Technical Field

[0001] The present application relates to the field of digital medical technology and is applied to digital medical doctor-patient dialogue consultation training scenarios, and in particular to an LLM model training method, apparatus, device and storage medium thereof. Background Art

[0002] With the development of the computer industry and artificial intelligence, and the advent of the big data era, traditional medical treatments are gradually shifting towards digital medicine. With the advancement of deep learning, language models (LMs) have made tremendous progress in the field of natural language processing (NLP), demonstrating impressive performance across a wide range of tasks. By performing unsupervised or self-supervised learning on massive amounts of text data, they acquire rich linguistic knowledge and general capabilities.

[0003] Large Language Models (LLMs) typically use an autoregressive or autoencoding approach. Given an input sequence, they generate the output words one by one, either from left to right or from the center outward. While simple and efficient, this approach also suffers from several drawbacks: a lack of globality, intermediate expressiveness, and multi-dimensional interactivity. Consequently, digital healthcare conversational systems still rely on these autoregressive or autoencoding approaches for language model training. However, these lacks globality, intermediate expressiveness, and multi-dimensional interactivity, making them ineffective for semantic analysis based on the overall conversational content. Consequently, these systems' semantic recognition results are inaccurate. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to propose an LLM model training method, apparatus, device and storage medium thereof to solve the problem that the prior art still lacks an LLM language model with globality, intermediate expressiveness and multi-dimensional interactivity, and cannot well combine the global conversation content for semantic expression analysis, resulting in inaccurate semantic recognition results of the digital medical industry conversation system.

[0005] In order to solve the above technical problems, the embodiment of the present application provides an LLM model training method, which adopts the following technical solutions:

[0006] An LLM model training method includes the following steps:

[0007] Obtaining a natural language corpus for LLM model training, wherein the natural language corpus contains conversation texts between doctors and corresponding consultants pre-organized from a target medical conversation platform, and the conversation texts carry conversation time information;

[0008] According to the preset cleaning rules, the conversation text is cleaned for non-focus data to obtain the cleaned text to be classified;

[0009] chronologically numbering each dialogue sentence in the text to be classified according to the dialogue time information;

[0010] Classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set composed of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the inquirer, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent;

[0011] According to the order of the time sequence numbers, alternately selecting target time sequence numbers from the first ordered classification set and the second ordered classification set to form a time sequence number sequence;

[0012] Selecting corresponding dialogue sentences according to the time sequence number sequence, using the dialogue sentence corresponding to the inquirer as the first dialogue party sentence, and using the dialogue sentence corresponding to the respondent as the second dialogue party sentence;

[0013] Based on the first and second dialogue party statements, the sequence of timing numbers, and the timing numbers corresponding to each dialogue statement, a directed acyclic graph is generated to obtain a dialogue text timing graph;

[0014] The conversation text sequence diagram is input into a pre-built LLM model, and each business node is filtered out according to the part of speech of the words in the conversation text. The pre-built LLM model is trained according to the conversation text sequence diagram and the each business node to obtain a trained LLM model.

[0015] Furthermore, the step of performing non-focus data cleaning on the conversation text according to a preset cleaning rule to obtain the cleaned text to be classified specifically includes:

[0016] Get the preset non-focus data form;

[0017] Using the conversation text as a search field, sequentially obtaining each non-focused data in the non-focused data form as a search term;

[0018] Filtering out non-focus data from the search domain according to the search term, and marking all filtered out non-focus data;

[0019] According to the marking processing results, all non-focus data are deleted to obtain the cleaned text to be classified.

[0020] Furthermore, the step of sequentially numbering each dialogue sentence in the text to be classified according to the dialogue time information specifically includes:

[0021] Arrange the order of each dialogue sentence in the dialogue text according to the dialogue time information corresponding to each dialogue sentence in the dialogue text;

[0022] According to the order of each dialogue sentence in the dialogue text, the dialogue sentences in the text to be classified are numbered using Arabic numerals from small to large to obtain a result of the numbering process;

[0023] Obtaining distinguishing identification information pre-set for different conversation subjects;

[0024] According to the processing results of the dialogue text and the distinction numbers, filtering out the dialogue sentences in the text to be classified corresponding to different dialogue subjects and the distinction numbers corresponding to the dialogue sentences in the text to be classified;

[0025] The distinguishing identification information and distinguishing numbers corresponding to different conversation subjects are spliced ​​together to obtain the time sequence number.

[0026] Furthermore, before executing the steps of inputting the conversation text sequence diagram into the pre-built LLM model, filtering out each service node according to the part of speech of words in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and each service node, the method further includes:

[0027] Obtaining node screening keywords pre-deployed in the LLM model, wherein the node screening keywords are pre-organized keywords that can screen out all business nodes, and the all business nodes include input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes;

[0028] Starting a node naming component pre-deployed in the LLM model, wherein the node naming component is capable of distinguishing names for different nodes;

[0029] The step of inputting the conversation text sequence diagram into a pre-built LLM model, filtering out each business node according to the part of speech of words in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and each business node specifically includes:

[0030] Parsing the dialogue text sequence diagram to obtain ordered dialogue sentences in the dialogue text sequence diagram;

[0031] Filter keywords based on the nodes pre-deployed in the LLM model to filter out all nodes included in the ordered dialogue sentences;

[0032] Performing distinguishing naming processing on all nodes in the ordered dialogue sentence according to the node naming component pre-deployed in the LLM model, and obtaining a distinguishing naming processing result;

[0033] According to the distinguishing naming processing result, identifying input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes in all business nodes;

[0034] Constructing an execution workflow within the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes;

[0035] The pre-built LLM model is trained according to the execution workflow and the ordered dialogue sentences.

[0036] Furthermore, the node screening keywords include key nouns corresponding to variable nodes, and key verbs corresponding to input nodes, output nodes, return nodes, method call nodes, and operation nodes. The step of screening all nodes included in the ordered dialogue sentence based on the node screening keywords pre-deployed in the LLM model specifically includes:

[0037] Splitting the ordered dialogue sentences according to parts of speech to obtain all nouns and all verbs contained in the ordered dialogue sentences;

[0038] Filtering keywords according to the node to select all key nouns from all the nouns, and filtering all key verbs from all the verbs;

[0039] Filter out corresponding variable nodes according to the key nouns corresponding to the variable nodes;

[0040] According to the key verbs corresponding to the input nodes, output nodes, return nodes, method call nodes and operation nodes, the corresponding input nodes, output nodes, return nodes, method call nodes and operation nodes are respectively screened out.

[0041] Furthermore, after executing the step of identifying input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes in all business nodes based on the distinguishing naming processing results, the method further includes:

[0042] According to the timing sequence in the dialog text timing diagram, setting execution priorities for all identified input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes, and setting data flow logic lines;

[0043] The step of constructing an execution workflow in the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes specifically includes:

[0044] The execution workflow is constructed based on the execution priorities corresponding to all input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes, as well as the data flow logic lines.

[0045] Furthermore, the step of training the pre-built LLM model according to the execution workflow and the ordered dialogue statements specifically includes:

[0046] Filtering keywords input into an input node from the ordered dialogue sentences;

[0047] Inputting the keywords input into the input node into the corresponding input node through the execution workflow;

[0048] According to the execution result of the execution workflow, all return results and output results in the execution process of the execution workflow are obtained;

[0049] Identify the execution accuracy rates corresponding to the return result and the output result, respectively, according to a preset accuracy rate algorithm formula and a preset reference result, wherein the preset reference result includes all correct return results and all correct output results;

[0050] If the execution accuracy rates corresponding to the returned result and the output result respectively meet the corresponding accuracy thresholds, the LLM model training is completed.

[0051] In order to solve the above technical problems, the embodiment of the present application also provides an LLM model training device, which adopts the following technical solutions:

[0052] An LLM model training device, comprising:

[0053] A training corpus acquisition module is used to acquire a natural language corpus set for LLM model training, wherein the natural language corpus set includes conversation texts between doctors and corresponding consultants pre-organized from the target medical conversation platform, and the conversation texts carry conversation time information;

[0054] A conversation text cleaning module is used to clean the non-focus data of the conversation text according to preset cleaning rules to obtain the cleaned text to be classified;

[0055] A dialogue sentence sequence numbering module, configured to sequentially number each dialogue sentence in the text to be classified according to the dialogue time information;

[0056] a dialogue sentence classification and arranging module, configured to classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set consisting of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the inquirer, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent;

[0057] a time sequence number sequence forming module, configured to alternately select target time sequence numbers from the first ordered classification set and the second ordered classification set according to the order of the time sequence numbers to form a time sequence number sequence;

[0058] a dialogue sentence classification and selection module, configured to select corresponding dialogue sentences according to the time sequence number sequence, and use the dialogue sentence corresponding to the inquirer as the first dialogue party sentence, and use the dialogue sentence corresponding to the respondent as the second dialogue party sentence;

[0059] a dialogue text timing diagram generating module, configured to generate a directed acyclic graph based on the first dialogue party's statement, the second dialogue party's statement, the timing number sequence, and the timing number corresponding to each dialogue statement, to obtain a dialogue text timing diagram;

[0060] The LLM model training module is used to input the conversation text timing diagram into a pre-built LLM model, filter out various business nodes according to the parts of speech of words in the conversation text, train the pre-built LLM model according to the conversation text timing diagram and the various business nodes, and obtain a trained LLM model.

[0061] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0062] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the LLM model training method described above when executing the computer-readable instructions.

[0063] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0064] A computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the LLM model training method as described above.

[0065] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0066] The LLM model training method described in the embodiment of the present application obtains a natural language corpus to be trained on the LLM model; sequentially numbers each dialogue sentence in the text to be classified according to the dialogue time information; classifies the sequential numbers corresponding to each dialogue sentence in the text to be classified according to the dialogue subject of the dialogue text, and obtains a first ordered classification set and a second ordered classification set composed of sequential numbers; forms a sequential number sequence according to the order of the sequential numbers; selects the corresponding dialogue sentence according to the sequential number sequence, takes the dialogue sentence corresponding to the consultant as the first dialogue party sentence, and takes the dialogue sentence corresponding to the respondent as the second dialogue party sentence. By sorting out a text dialogue sequential diagram according to different dialogue subjects and the order of dialogue in the dialogue text, and then filtering out each business node according to the word part of speech in the dialogue text, and training an LLM model based on the text dialogue sequential diagram and each business node, it is ensured that the LLM model training is carried out according to the globality, intermediate expression and multi-dimensional interactivity of the dialogue, and a dialogue model that is more suitable for medical scenarios is obtained, which is convenient for assisting the construction of digital medical platforms and the processing of consultation dialogue services, and improving the semantic recognition accuracy of digital medical industry dialogue systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the solutions in this application, a brief introduction will be given below to the drawings required for use in the description of the embodiments of this application. Obviously, the drawings described below are some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0068] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0069] Figure 2 A flowchart of an embodiment of the LLM model training method according to the present application;

[0070] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 202 is shown;

[0071] Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 203 is shown;

[0072] Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 208 is shown;

[0073] Figure 6 yes Figure 5 A flowchart of a specific embodiment of step 502 is shown;

[0074] Figure 7 yes Figure 5 A flowchart of a specific embodiment of step 506 is shown;

[0075] Figure 8 A schematic structural diagram of an embodiment of an LLM model training device according to the present application;

[0076] Figure 9 A schematic structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0078] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0079] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0080] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. Network 104 is a medium for providing communication links between terminal devices 101, 102, 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0081] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0082] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0083] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0084] It should be noted that the LLM model training method provided in the embodiment of the present application is generally executed by a server, and accordingly, the LLM model training device is generally set in the server.

[0085] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0086] Continue to refer Figure 2 , shows a flow chart of an embodiment of the LLM model training method according to the present application. The LLM model training method comprises the following steps:

[0087] Step 201: Obtain a natural language corpus for LLM model training, wherein the natural language corpus contains conversation texts between doctors and corresponding consultants pre-organized from a target medical conversation platform, and the conversation texts carry conversation time information.

[0088] In this embodiment, the conversation text between the doctor and the corresponding consultant, which is pre-organized from the target medical conversation platform, includes the disease symptom data content provided by the consultant, such as: the location of pain, insomnia symptoms, etc., personal information content, such as age, gender, etc., as well as the inquiry guidance sentences provided by the doctor, the test suggestions provided by the doctor, or the diagnosis results given by the doctor.

[0089] The LLM ( Large Language Model , LLM) model, that is, a large-scale language processing model, is often used to recognize large amounts of text and correct text errors under large amounts of data.

[0090] Step 202 , according to a preset cleaning rule, the dialogue text is cleaned for non-focus data to obtain the cleaned text to be classified, wherein the non-focus data refers to modal particles and title pronouns that frequently appear in dialogue scenarios.

[0091] Continue to refer Figure 3 , Figure 3 yes Figure 2 The flowchart of a specific embodiment of step 202 shown includes:

[0092] Step 301: Obtain a preset non-focus data form;

[0093] Specifically, the preset non-focus data form includes modal particles that frequently appear in dialogue scenarios, such as "hmm," "okay," "know," etc., or appellative pronouns, such as "I," "she," "father," etc.

[0094] Step 302: Using the conversation text as a search field, sequentially obtain each non-focused data in the non-focused data form as a search term;

[0095] Step 303: filtering out non-focus data from the search domain according to the search term, and marking all the filtered out non-focus data;

[0096] Step 304: Delete all non-focus data according to the marking processing result to obtain the cleaned text to be classified.

[0097] In this embodiment, non-focus data cleaning is performed on the conversation text, the main purpose of which is to clean out redundant words in the conversation text that are useless for model training.

[0098] Step 203: chronologically number each dialogue sentence in the text to be classified according to the dialogue time information.

[0099] Continue to refer Figure 4 , Figure 4 yes Figure 2 The flowchart of a specific embodiment of step 203 shown includes:

[0100] Step 401, sorting out the sequence of each dialogue sentence in the dialogue text based on the dialogue time information corresponding to each dialogue sentence in the dialogue text;

[0101] Step 402: Differentiate and number each dialogue sentence in the text to be classified using Arabic numerals from smallest to largest according to the order of each dialogue sentence in the dialogue text, and obtain a differential numbering result;

[0102] Step 403: Obtain distinguishing identification information pre-set for different conversation subjects;

[0103] In this embodiment, according to different dialogue subjects, such as doctor and patient, the distinguishing identification information is set as doctor and sufferer respectively.

[0104] Step 404: Filter out, based on the dialogue text and the difference number processing results, the dialogue sentences in the text to be classified corresponding to different dialogue subjects and the difference numbers corresponding to the dialogue sentences in the text to be classified;

[0105] Step 405: Concatenate the distinguishing identification information and distinguishing numbers corresponding to different conversation entities to obtain the time sequence number.

[0106] In this embodiment, assuming that the dialogue sentences corresponding to patients are numbered 1, 2, 4, and 6, and the dialogue sentences corresponding to doctors are numbered 3, 5, 7, and 8, the sequential numbers include sufferer1, sufferer2, doctor3, sufferer4, doctor5, sufferer6, doctor7, and doctor8.

[0107] Step 204: classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set composed of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the consultant, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent.

[0108] In this embodiment, in order to avoid classifying and constructing sets of large amounts of data, an ordered set is directly constructed using the time sequence number of each dialogue statement.

[0109] Step 205 : Select target time sequence numbers from the first ordered classification set and the second ordered classification set alternately according to the order of the time sequence numbers to form a time sequence number sequence.

[0110] In this embodiment, a time sequence number sequence is constructed for the dialogue sentences between the doctor and the patient / consultant according to the order of the time sequence numbers. Therefore, the time sequence number sequence corresponds to the dialogue sentence sequence between the doctor and the patient / consultant.

[0111] Step 206: Select corresponding dialogue sentences according to the time sequence number sequence, use the dialogue sentence corresponding to the consultant as the first dialogue party sentence, and use the dialogue sentence corresponding to the respondent as the second dialogue party sentence.

[0112] Step 207 : Generate a directed acyclic graph based on the first dialogue party's statement, the second dialogue party's statement, the sequence of timing numbers, and the timing numbers corresponding to each dialogue statement to obtain a dialogue text timing graph.

[0113] Step 208: input the dialogue text sequence diagram into a pre-built LLM model, train the pre-built LLM model according to the dialogue text sequence diagram, and obtain a trained LLM model.

[0114] In this embodiment, the pre-built LLM model is a Chinese-Vicuna Chinese dialogue model, wherein the Chinese-Vicuna Chinese dialogue model is a Chinese dialogue model fine-tuned based on the Vicuna model using Chinese data + LORA solution.

[0115] In this embodiment, before executing the steps of inputting the conversation text sequence diagram into the pre-built LLM model, filtering out each business node according to the part of speech of the words in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and the each business node, the method further includes: obtaining node screening keywords pre-deployed in the LLM model, wherein the node screening keywords are pre-organized keywords that can filter out all business nodes, and the all business nodes include input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes; starting a node naming component pre-deployed in the LLM model, wherein the node naming component can distinguish different nodes by naming them.

[0116] Continue to refer Figure 5 , Figure 5 yes Figure 2 The flowchart of a specific embodiment of step 208 shown includes:

[0117] Step 501: parsing the dialogue text sequence diagram to obtain ordered dialogue sentences in the dialogue text sequence diagram;

[0118] Step 502: filtering keywords based on the nodes pre-deployed in the LLM model to filter out all nodes included in the ordered dialogue sentence;

[0119] In this embodiment, the node screening keywords include key nouns corresponding to variable nodes, and also include key verbs corresponding to input nodes, output nodes, return nodes, method call nodes, and operation nodes.

[0120] Continue to refer Figure 6 , Figure 6 yes Figure 5The flowchart of a specific embodiment of step 502 shown includes:

[0121] Step 601: Split the ordered dialogue sentence according to parts of speech to obtain all nouns and all verbs contained in the ordered dialogue sentence;

[0122] In this embodiment, the ordered dialogue sentences are split according to parts of speech. Specifically, a BERT-based NLP natural language processing model can be used for splitting, wherein the NLP natural language processing model is capable of recognizing parts of speech of different words, such as names, verbs, adjectives, modal particles, etc.

[0123] Step 602: filtering keywords from the nodes to select all key nouns from the nouns, and to select all key verbs from the verbs;

[0124] Step 603: Filter out corresponding variable nodes according to the key nouns corresponding to the variable nodes;

[0125] In this embodiment, the variable node, ie, the object generation node, is often named as a noun because the variable is often related to an entity class. Therefore, a key noun is pre-set as a keyword for screening the variable node.

[0126] Step 604 , according to the key verbs corresponding to the input nodes, output nodes, return nodes, method call nodes and operation nodes, the corresponding input nodes, output nodes, return nodes, method call nodes and operation nodes are filtered out respectively.

[0127] In this embodiment, input and output are often represented by input and out in the execution program, and input and output are also verbs in the dialogue text; return and return are also verbs, method calls often correspond to verbs, and calculations often express accumulation, modulo, division, etc., which are also verbs. Therefore, key verbs are pre-set as keywords for screening input nodes, output nodes, return nodes, method call nodes and calculation nodes.

[0128] Step 503: Perform distinguishing naming processing on all nodes in the ordered dialogue sentence according to the node naming component pre-deployed in the LLM model, and obtain a distinguishing naming processing result;

[0129] Step 504: Identify input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes in all business nodes based on the distinguishing naming processing result;

[0130] In this embodiment, after executing the step of identifying the input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes in all business nodes according to the distinguishing naming processing results, the method also includes: setting execution priorities for all identified input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes according to the timing sequence in the dialogue text timing diagram, and setting data flow logic lines.

[0131] Step 505: construct an execution workflow in the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes;

[0132] In this embodiment, the step of constructing an execution workflow in the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes specifically includes: constructing the execution workflow based on the execution priorities corresponding to all input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes, as well as the data flow logic lines.

[0133] Step 506: Training the pre-built LLM model according to the execution workflow and the ordered dialogue statements.

[0134] Continue to refer Figure 7 , Figure 7 yes Figure 5 The flowchart of a specific embodiment of step 506 shown includes:

[0135] Step 701, filtering out keywords input into an input node from the ordered dialogue sentences;

[0136] Step 702: input the keyword input into the input node into the corresponding input node through the execution workflow;

[0137] Step 703: Obtain all return results and output results during the execution of the workflow according to the execution result of the workflow;

[0138] Step 704: Identify the execution accuracy rates corresponding to the return results and the output results, respectively, based on a preset accuracy rate algorithm formula and a preset reference result, wherein the preset reference result includes all correct return results and all correct output results;

[0139] Step 705: If the execution accuracy rates corresponding to the returned result and the output result respectively meet corresponding accuracy thresholds, the LLM model training is completed.

[0140] In this embodiment, an LLM model is trained by pre-organizing the conversation texts between doctors and corresponding consultants from the target medical conversation platform, and a text conversation sequence diagram is organized according to the different conversation subjects and the conversation sequence in the conversation text. Then, each business node is screened out according to the part of speech of the words in the conversation text, and an LLM model is trained based on the text conversation sequence diagram and the each business node. This ensures that the LLM model is trained based on the globality, intermediate expressiveness and multi-dimensional interactivity of the conversation, and obtains a conversation model that is more in line with the medical scenario, which is convenient for assisting the construction and business processing of the digital medical platform and improving the semantic recognition accuracy of the digital medical industry conversation system.

[0141] The present application obtains a natural language corpus to be trained on an LLM model; sequentially numbers each dialogue sentence in the text to be classified according to the dialogue time information; classifies the sequential numbers corresponding to each dialogue sentence in the text to be classified according to the dialogue subject of the dialogue text, and obtains a first ordered classification set and a second ordered classification set composed of sequential numbers; forms a sequential number sequence according to the order of the sequential numbers; selects corresponding dialogue sentences according to the sequential number sequence, and uses the dialogue sentence corresponding to the consultant as the first dialogue party sentence, and the dialogue sentence corresponding to the respondent as the second dialogue party sentence. By sorting out a text dialogue sequential diagram according to the different dialogue subjects and the order of the dialogue in the dialogue text, and then filtering out each business node according to the word part of speech in the dialogue text, and training an LLM model based on the text dialogue sequential diagram and each business node, it is ensured that the LLM model is trained according to the globality, intermediate expression and multi-dimensional interactivity of the dialogue, and a dialogue model that is more suitable for medical scenarios is obtained, which is convenient for assisting the construction of digital medical platforms and the processing of consultation dialogue services, and improving the semantic recognition accuracy of digital medical industry dialogue systems.

[0142] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0143] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0144] In an embodiment of the present application, the present application obtains a natural language corpus to be trained on the LLM model; performs time sequence numbering on each dialogue sentence in the text to be classified according to the dialogue time information; organizes a text dialogue sequence diagram according to the dialogue subject of the dialogue text and the sequence of dialogues, and then filters out each business node according to the part of speech of the words in the dialogue text, and trains an LLM model based on the text dialogue sequence diagram and the each business node, thereby ensuring that the LLM model is trained according to the globality, intermediate expressiveness and multi-dimensional interactivity of the dialogue, obtaining a dialogue model that is more in line with the medical scenario, facilitating the construction of the digital medical platform and the processing of consulting dialogue business, and improving the semantic recognition accuracy of the dialogue system in the digital medical industry.

[0145] Further references Figure 8 , as a response to the above Figure 2 The present application provides an embodiment of an LLM model training device, which is similar to the embodiment of the present invention. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0146] like Figure 8 As shown, the LLM model training device 800 of this embodiment includes: a training corpus acquisition module 801, a dialogue text cleaning module 802, a dialogue sentence time sequence numbering module 803, a dialogue sentence classification and organization module 804, a time sequence number sequence formation module 805, a dialogue sentence classification and selection module 806, a dialogue text time sequence diagram generation module 807, and an LLM model training module 808. Among them:

[0147] The training corpus acquisition module 801 is used to acquire a natural language corpus set for LLM model training, wherein the natural language corpus set includes conversation texts between doctors and corresponding consultants pre-organized from the target medical conversation platform, and the conversation texts carry conversation time information;

[0148] The dialogue text cleaning module 802 is configured to clean the dialogue text for non-focus data according to preset cleaning rules, thereby obtaining a cleaned text to be classified, wherein the non-focus data refers to modal particles and title pronouns that frequently appear in dialogue scenarios;

[0149] A dialogue sentence sequence numbering module 803 is configured to sequentially number each dialogue sentence in the text to be classified according to the dialogue time information;

[0150] The dialogue sentence classification and arrangement module 804 is configured to classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set consisting of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the inquirer, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent;

[0151] A time sequence number sequence forming module 805 is configured to alternately select target time sequence numbers from the first ordered classification set and the second ordered classification set according to the order of the time sequence numbers to form a time sequence number sequence;

[0152] A dialogue sentence classification and selection module 806 is configured to select corresponding dialogue sentences according to the time sequence number sequence, and use the dialogue sentence corresponding to the inquirer as the first dialogue party sentence and the dialogue sentence corresponding to the respondent as the second dialogue party sentence;

[0153] A dialogue text timing diagram generating module 807 is configured to generate a directed acyclic graph based on the first dialogue party's statement, the second dialogue party's statement, the timing number sequence, and the timing number corresponding to each dialogue statement, thereby obtaining a dialogue text timing diagram;

[0154] The LLM model training module 808 is used to input the conversation text timing diagram into the pre-built LLM model, filter out each business node according to the part of speech of the words in the conversation text, train the pre-built LLM model according to the conversation text timing diagram and the each business node, and obtain a trained LLM model.

[0155] The present application obtains a natural language corpus to be trained on an LLM model; sequentially numbers each dialogue sentence in the text to be classified according to the dialogue time information; classifies the sequential numbers corresponding to each dialogue sentence in the text to be classified according to the dialogue subject of the dialogue text, and obtains a first ordered classification set and a second ordered classification set composed of sequential numbers; forms a sequential number sequence according to the order of the sequential numbers; selects corresponding dialogue sentences according to the sequential number sequence, and uses the dialogue sentence corresponding to the consultant as the first dialogue party sentence, and the dialogue sentence corresponding to the respondent as the second dialogue party sentence. By sorting out a text dialogue sequential diagram according to the different dialogue subjects and the order of the dialogue in the dialogue text, and then filtering out each business node according to the word part of speech in the dialogue text, and training an LLM model based on the text dialogue sequential diagram and each business node, it is ensured that the LLM model is trained according to the globality, intermediate expression and multi-dimensional interactivity of the dialogue, and a dialogue model that is more suitable for medical scenarios is obtained, which is convenient for assisting the construction of digital medical platforms and the processing of consultation dialogue services, and improving the semantic recognition accuracy of digital medical industry dialogue systems.

[0156] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware using computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0157] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0158] To solve the above technical problems, the present application also provides a computer device. Figure 9 , Figure 9 This is a basic structural block diagram of the computer device in this embodiment.

[0159] The computer device 9 includes a memory 9a, a processor 9b, and a network interface 9c that are interconnected through a system bus. It should be noted that the figure only shows a computer device 9 with components 9a-9c, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0160] The computer device may be a desktop computer, notebook computer, PDA, cloud server, etc. The computer device may interact with the user via a keyboard, mouse, remote control, touchpad, or voice control device.

[0161] The memory 9a includes at least one type of readable storage medium, including flash memory, a hard disk, a multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, a magnetic disk, an optical disk, etc. In some embodiments, the memory 9a may be an internal storage unit of the computer device 9, such as the hard disk or internal memory of the computer device 9. In other embodiments, the memory 9a may also be an external storage device of the computer device 9, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash memory card, etc. Of course, the memory 9a may also include both the internal storage unit of the computer device 9 and its external storage device. In this embodiment, the memory 9a is generally used to store the operating system and various application software installed on the computer device 9, such as computer-readable instructions for an LLM model training method. In addition, the memory 9a can also be used to temporarily store various types of data that have been output or are to be output.

[0162] In some embodiments, the processor 9b can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 9b is generally used to control the overall operation of the computer device 9. In this embodiment, the processor 9b is used to execute computer-readable instructions stored in the memory 9a or process data, such as computer-readable instructions for executing the LLM model training method.

[0163] The network interface 9c may include a wireless network interface or a wired network interface. The network interface 9c is generally used to establish a communication connection between the computer device 9 and other electronic devices.

[0164] The computer device proposed in this embodiment belongs to the field of digital medical technology and is applied to the dialogue consultation training scenario between digital medical doctors and patients. This application obtains a natural language corpus to be trained on the LLM model; sequentially numbers each dialogue sentence in the text to be classified according to the dialogue time information; classifies the sequential numbers corresponding to each dialogue sentence in the text to be classified according to the dialogue subject of the dialogue text, and obtains a first ordered classification set and a second ordered classification set composed of sequential numbers; forms a sequential number sequence according to the order of the sequential numbers; selects the corresponding dialogue sentence according to the sequential number sequence, and uses the dialogue sentence corresponding to the consultant as the first dialogue party sentence, and uses the dialogue sentence corresponding to the respondent as the second dialogue party sentence. By organizing a text conversation sequence diagram based on the different conversation subjects and the sequence of conversations in the conversation text, and then filtering out each business node based on the part of speech of the words in the conversation text, an LLM model is trained based on the text conversation sequence diagram and the each business node. This ensures that the LLM model is trained based on the globality, intermediate expressiveness and multi-dimensional interactivity of the conversation, and obtains a conversation model that is more in line with the medical scenario, which is convenient for assisting the construction of the digital medical platform and the consulting conversation business processing, and improves the semantic recognition accuracy of the digital medical industry conversation system.

[0165] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by a processor to enable the processor to perform the steps of the LLM model training method as described above.

[0166] The computer-readable storage medium proposed in this embodiment belongs to the field of digital medical technology and is applied to the dialogue consultation training scenario between digital medical doctors and patients. This application obtains a natural language corpus to be trained on the LLM model; according to the dialogue time information, each dialogue sentence in the text to be classified is sequentially numbered; according to the dialogue subject of the dialogue text, the sequential numbers corresponding to each dialogue sentence in the text to be classified are classified to obtain a first ordered classification set and a second ordered classification set composed of sequential numbers; according to the order of the sequential numbers, a sequential number sequence is formed; according to the sequential number sequence, the corresponding dialogue sentence is selected, and the dialogue sentence corresponding to the consultant is used as the first dialogue party sentence, and the dialogue sentence corresponding to the respondent is used as the second dialogue party sentence. By organizing a text conversation sequence diagram based on the different conversation subjects and the sequence of conversations in the conversation text, and then filtering out each business node based on the part of speech of the words in the conversation text, an LLM model is trained based on the text conversation sequence diagram and the each business node. This ensures that the LLM model is trained based on the globality, intermediate expressiveness and multi-dimensional interactivity of the conversation, and obtains a conversation model that is more in line with the medical scenario, which is convenient for assisting the construction of the digital medical platform and the consulting conversation business processing, and improves the semantic recognition accuracy of the digital medical industry conversation system.

[0167] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of this application.

[0168] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. A LLM model training method, characterized in that: The steps include: Obtaining a natural language corpus for LLM model training, wherein the natural language corpus contains conversation texts between doctors and corresponding consultants pre-organized from a target medical conversation platform, and the conversation texts carry conversation time information; According to the preset cleaning rules, the conversation text is cleaned for non-focus data to obtain the cleaned text to be classified; chronologically numbering each dialogue sentence in the text to be classified according to the dialogue time information; Classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set composed of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the inquirer, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent; According to the order of the time sequence numbers, alternately selecting target time sequence numbers from the first ordered classification set and the second ordered classification set to form a time sequence number sequence; Selecting corresponding dialogue sentences according to the time sequence number sequence, using the dialogue sentence corresponding to the inquirer as the first dialogue party sentence, and using the dialogue sentence corresponding to the respondent as the second dialogue party sentence; Based on the first and second dialogue party statements, the sequence of timing numbers, and the timing numbers corresponding to each dialogue statement, a directed acyclic graph is generated to obtain a dialogue text timing graph; Inputting the conversation text sequence diagram into a pre-built LLM model, filtering out each service node according to the word part of speech in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and each service node to obtain a trained LLM model, wherein, before executing the steps of inputting the conversation text sequence diagram into the pre-built LLM model, filtering out each service node according to the word part of speech in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and each service node, the method further includes: Obtaining node screening keywords pre-deployed in the LLM model, wherein the node screening keywords are pre-organized keywords that can screen out all business nodes, and the all business nodes include input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes; A node naming component pre-deployed in the LLM model is started, wherein the node naming component can distinguish names for different nodes.

2. The LLM model training method according to claim 1, characterized in that The step of performing non-focus data cleaning on the conversation text according to the preset cleaning rules to obtain the cleaned text to be classified specifically includes: Get the preset non-focus data form; Using the conversation text as a search field, sequentially obtaining each non-focused data in the non-focused data form as a search term; Filtering out non-focus data from the search domain according to the search term, and marking all filtered out non-focus data; According to the marking processing results, all non-focus data are deleted to obtain the cleaned text to be classified.

3. The LLM model training method according to claim 1, characterized in that The step of sequentially numbering each dialogue sentence in the text to be classified according to the dialogue time information specifically includes: Arrange the order of each dialogue sentence in the dialogue text according to the dialogue time information corresponding to each dialogue sentence in the dialogue text; According to the order of each dialogue sentence in the dialogue text, the dialogue sentences in the text to be classified are numbered using Arabic numerals from small to large to obtain a result of the numbering process; Obtaining distinguishing identification information pre-set for different conversation subjects; According to the processing results of the dialogue text and the distinction numbers, filtering out the dialogue sentences in the text to be classified corresponding to different dialogue subjects and the distinction numbers corresponding to the dialogue sentences in the text to be classified; The distinguishing identification information and distinguishing numbers corresponding to different conversation subjects are spliced ​​together to obtain the time sequence number.

4. The LLM model training method according to claim 1, characterized in that The step of inputting the conversation text sequence diagram into a pre-built LLM model, filtering out each business node according to the part of speech of words in the conversation text, and training the pre-built LLM model according to the conversation text sequence diagram and each business node specifically includes: Parsing the dialogue text sequence diagram to obtain ordered dialogue sentences in the dialogue text sequence diagram; Filter keywords based on the nodes pre-deployed in the LLM model to filter out all nodes included in the ordered dialogue sentences; Performing distinguishing naming processing on all nodes in the ordered dialogue sentence according to the node naming component pre-deployed in the LLM model, and obtaining a distinguishing naming processing result; According to the distinguishing naming processing result, identifying input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes in all business nodes; Constructing an execution workflow within the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes; The pre-built LLM model is trained according to the execution workflow and the ordered dialogue sentences.

5. The LLM model training method according to claim 4, characterized in that: The node screening keywords include key nouns corresponding to variable nodes, and key verbs corresponding to input nodes, output nodes, return nodes, method call nodes, and operation nodes. The step of screening all nodes included in the ordered dialogue sentence based on the node screening keywords pre-deployed in the LLM model specifically includes: Splitting the ordered dialogue sentences according to parts of speech to obtain all nouns and all verbs contained in the ordered dialogue sentences; Filtering keywords according to the node to select all key nouns from all the nouns, and filtering all key verbs from all the verbs; Filter out corresponding variable nodes according to the key nouns corresponding to the variable nodes; According to the key verbs corresponding to the input nodes, output nodes, return nodes, method call nodes and operation nodes, the corresponding input nodes, output nodes, return nodes, method call nodes and operation nodes are respectively screened out.

6. The LLM model training method according to claim 4, characterized in that: After executing the step of identifying input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes in all business nodes according to the distinguishing naming processing results, the method further includes: According to the timing sequence in the dialog text timing diagram, setting execution priorities for all identified input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes, and setting data flow logic lines; The step of constructing an execution workflow in the LLM model based on the input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes specifically includes: The execution workflow is constructed based on the execution priorities corresponding to all input nodes, output nodes, method call nodes, variable nodes, return nodes and operation nodes, as well as the data flow logic lines.

7. The LLM model training method according to claim 4, characterized in that: The step of training the pre-built LLM model according to the execution workflow and the ordered dialogue statements specifically includes: Filtering keywords input into an input node from the ordered dialogue sentences; Inputting the keywords input into the input node into the corresponding input node through the execution workflow; According to the execution result of the execution workflow, all return results and output results in the execution process of the execution workflow are obtained; Identify the execution accuracy rates corresponding to the return result and the output result, respectively, according to a preset accuracy rate algorithm formula and a preset reference result, wherein the preset reference result includes all correct return results and all correct output results; If the execution accuracy rates corresponding to the returned result and the output result respectively meet the corresponding accuracy thresholds, the LLM model training is completed.

8. An LLM model training device, characterized in that: The LLM model training device is used to implement the steps of the LLM model training method according to any one of claims 1 to 7, and the LLM model training device includes: A training corpus acquisition module is used to acquire a natural language corpus set for LLM model training, wherein the natural language corpus set includes conversation texts between doctors and corresponding consultants pre-organized from the target medical conversation platform, and the conversation texts carry conversation time information; A conversation text cleaning module is used to clean the non-focus data of the conversation text according to preset cleaning rules to obtain the cleaned text to be classified; A dialogue sentence sequence numbering module, configured to sequentially number each dialogue sentence in the text to be classified according to the dialogue time information; a dialogue sentence classification and arranging module, configured to classify the time sequence numbers corresponding to the dialogue sentences in the text to be classified according to the dialogue subject of the dialogue text, and obtain a first ordered classification set and a second ordered classification set consisting of the time sequence numbers, wherein the time sequence numbers in the first ordered classification set correspond to the dialogue sentences of the inquirer, and the time sequence numbers in the second ordered classification set correspond to the dialogue sentences of the respondent; a time sequence number sequence forming module, configured to alternately select target time sequence numbers from the first ordered classification set and the second ordered classification set according to the order of the time sequence numbers to form a time sequence number sequence; a dialogue sentence classification and selection module, configured to select corresponding dialogue sentences according to the time sequence number sequence, and use the dialogue sentence corresponding to the inquirer as the first dialogue party sentence, and use the dialogue sentence corresponding to the respondent as the second dialogue party sentence; a dialogue text timing diagram generating module, configured to generate a directed acyclic graph based on the first dialogue party's statement, the second dialogue party's statement, the timing number sequence, and the timing number corresponding to each dialogue statement, to obtain a dialogue text timing diagram; The LLM model training module is configured to input the conversation text sequence diagram into a pre-constructed LLM model, filter out each service node based on the word part of speech in the conversation text, and train the pre-constructed LLM model based on the conversation text sequence diagram and each service node to obtain a trained LLM model. Prior to executing the steps of inputting the conversation text sequence diagram into the pre-constructed LLM model, filtering out each service node based on the word part of speech in the conversation text, and training the pre-constructed LLM model based on the conversation text sequence diagram and each service node, the method further comprises: Obtaining node screening keywords pre-deployed in the LLM model, wherein the node screening keywords are pre-organized keywords that can screen out all business nodes, and the all business nodes include input nodes, output nodes, method call nodes, variable nodes, return nodes, and operation nodes; A node naming component pre-deployed in the LLM model is started, wherein the node naming component can distinguish names for different nodes.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the LLM model training method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the LLM model training method according to any one of claims 1 to 7.

Images