A multi-agent biochemical information extraction model and training method based on a large language model

Abstract

The present invention relates to a multi-agent biochemical information extraction model and training method based on a large language model, comprising the following specific steps: constructing a multi-agent biochemical information extraction model based on the large language model; obtaining a text training set containing biological or chemical information from a target database; inputting a first prompt word and a training set text to train a task agent; inputting a second prompt word and a training set text to train a data generation agent; inputting a third prompt word and a training set text to train an information mining agent; inputting prompt words to the trained multi-agent biochemical information extraction model, wherein the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results in a target database containing biological or chemical information according to the prompt words.

Description

Technical Field

[0001] The present invention relates to the field of large language models and biochemical information, and more specifically, to a multi-agent biochemical information extraction model and a training method based on a large language model. Background Art

[0002] The information extraction method based on the large language model has the advantages of high versatility, high accuracy and low labor cost when extracting general information. However, when applied to extracting key information in the fields of biology and chemistry, such as experimental conditions and data, the information extraction method based on the general large language model has limited accuracy in extracting vocabulary in its professional field. It is usually necessary to manually limit, build prompt words and cases to improve the accuracy of information extraction. When directly using a general large language model for experimental information extraction, hallucinations will occur, and wrong information and non-standard formats will be obtained. The accuracy rate is usually only 40%-60%. Therefore, when using a large model for information mining, it is still necessary to build prompt words and cases for model fine-tuning. For disciplines such as biology and chemistry with many professional vocabulary, huge and complex data, complex knowledge systems, and ever-changing content, it is particularly important to be able to extract information effectively and in a targeted manner for understanding the industry, creating databases, training models, and predicting trends. However, due to the huge and complex data and knowledge, it is technically difficult to select keywords in the fields of biology and chemistry and generate appropriate prompt words and cases based on them, and it usually takes professionals a lot of time to complete the training and preparation of the model. It can be seen that when extracting effective key information in the fields of biology and chemistry, it is impossible to use information extraction methods based on general large language models, but experts from multiple fields are required to manually supplement, adjust, screen and build models. This not only increases time and economic costs, but the volume and category of information obtained may change due to changes in subjective keyword selection.

[0003] Existing technologies such as the Chinese patent with patent number "CN118210914A" disclose an information mining and analysis method based on the LLM large model. The method includes the following steps: the large language model can retrieve relevant documents from the library knowledge base according to the researcher's keywords, topics or questions, including books, journal articles, conference papers, etc., and can also evaluate the documents according to the abstract or full text content, screen out the most relevant resources, including the latest research papers, papers included in top conferences, etc. and submit them to the researchers. The interaction is simple and easy to use, and the prompt engineering is universal. The general large language model can be applied to text analysis in various fields, including interview texts in the field of education. Its promotion also means that it can be widely used in different educational research and practice scenarios; at the same time, the interactive mode of natural language instructions makes the analysis process more intuitive and user-friendly. Everyone can refer to this set of prompt engineering frameworks to guide the large language model to perform private text analysis tasks in a personalized manner.

[0004] The problems existing in the above prior art are:

[0005] 1. The keywords and topics entered by researchers may not be precise or comprehensive enough, resulting in the omission of important relevant literature.

[0006] 2. Although large models can assess the relevance of literature, they may not be able to deeply understand the complex concepts and theories in the literature, thus affecting the quality of screening.

[0007] 3. General-purpose large language models cannot perform accurate information mining tasks at the data level, and it is difficult to complete tasks such as building knowledge graphs, databases, training data sets, etc. in specific fields.

[0008] 4. The single general-purpose large model is prone to errors and hallucinations when completing information extraction in the field of biochemistry. Compared with the multi-agent system, the accuracy and generation quality are lower in complex biochemical information extraction task scenarios, and it is unable to complete complex and systematic information extraction tasks. Summary of the invention

[0009] In order to solve the above problems existing in the prior art, the present invention proposes a multi-agent biochemical information extraction model and training method based on a large language model.

[0010] The technical solution of the present invention is as follows:

[0011] On the one hand, the present invention proposes a multi-agent biochemical information extraction model training method based on a large language model, comprising the following specific steps:

[0012] Step S1: deploy a large language model for text generation tasks as the core input and output module of the agent; construct a multi-agent biochemical information extraction model including a task agent, a data generation agent, and an information mining agent;

[0013] Step S2, obtaining multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry;

[0014] Step S3, using the first prompt word and a training set of texts containing biological or chemical information as inputs to the task agent; training the task agent to extract important keywords related to the biological or chemical field from the training set; verifying the extraction results through experts, and feeding back the verification results to the task agent for reinforcement learning;

[0015] Step S4, using the constructed second prompt word and the randomly extracted text containing biological or chemical information as inputs of the data generation agent, the second prompt word containing the keywords extracted by the task agent; training the data generation agent to extract the key words extracted by the task agent as the extraction target, extract the keyword corresponding content from the text content of several biological or chemical information, and generate key examples for the information mining agent to learn; verifying the generated results through experts, and feeding back the verification results to the data generation agent for reinforcement learning;

[0016] Step S5: using the constructed third prompt word and several texts containing biological or chemical information as inputs to the information mining agent, wherein the third prompt word contains the keywords extracted by the task agent and the key examples generated by the data generation agent; training the information mining agent to mine and extract relevant biological or chemical information from the training set text, and output the information mining results in the format of key examples; verifying the extraction results through experts, and feeding back the verification results to the generation information mining agent for reinforcement learning;

[0017] Step S6, inputting prompt words into the trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from the target database containing biological or chemical information according to the prompt words.

[0018] As a preferred implementation, the first prompt word and a training set of texts containing biological or chemical information are used as inputs of the task agent, and the number of important keywords outputted can be adjusted by adjusting the first prompt word limit according to the complexity of the extraction target.

[0019] As a preferred implementation, in the step of using the constructed third prompt word and a number of texts containing biological or chemical information as inputs to the information mining agent:

[0020] The information mining agent divides the input content into multiple sub-contents and then performs input operations according to the token restrictions of the large language model input and output;

[0021] When outputting the information mining results, the information mining agent merges the output information mining results corresponding to multiple sub-contents.

[0022] As a preferred embodiment, the second prompt word also includes a limitation on the output text format, and the data generation agent generates corresponding key examples based on the limitation on the output text format by extracting the corresponding content of the keywords.

[0023] On the other hand, the present invention also provides a multi-agent biochemical information extraction model based on a large language model, comprising:

[0024] Model deployment module, deploying a large language model for text generation tasks as the core input and output module of the agent; building a multi-agent biochemical information extraction model including task agents, data generation agents and information mining agents;

[0025] A training set construction module obtains multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry;

[0026] The keyword generation module takes the first prompt word and a training set of texts containing biological or chemical information as inputs to the task agent; trains the task agent to extract important keywords related to the biological or chemical field from the training set; verifies the extraction results through experts, and feeds the verification results back to the task agent for reinforcement learning;

[0027] The key sample generation module uses the constructed second prompt word and the randomly extracted text containing biological or chemical information as the input of the data generation agent, and the second prompt word contains the keywords extracted by the task agent; the training data generation agent takes the important keywords extracted by the task agent as the extraction target, extracts the corresponding content of the keywords from the text content of several biological or chemical information, and generates key samples for the information mining agent to learn; the generated results are verified by experts, and the verification results are fed back to the data generation agent for reinforcement learning;

[0028] The information mining module uses the constructed third prompt words and several texts containing biological or chemical information as inputs to the information mining agent, where the third prompt words contain keywords extracted by the task agent and key examples generated by the data generation agent; the information mining agent is trained to mine and extract relevant biological or chemical information from the training set texts, and output the information mining results in the format of key examples; the extracted results are verified by experts, and the verification results are fed back to the generated information mining agent for reinforcement learning;

[0029] The model uses a module to input prompt words into a trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from a target database containing biological or chemical information according to the prompt words.

[0030] As a preferred implementation, the keyword generation module uses the constructed first prompt word and a training set of several texts containing biological or chemical information as input of the task agent, and can adjust the number of important keywords output by adjusting the first prompt word limit according to the complexity of the extraction target.

[0031] As a preferred implementation, the information mining module, in the step of using the constructed third prompt word and a number of texts containing biological or chemical information as information mining agent input:

[0032] The information mining agent divides the input content into multiple sub-contents and then performs input operations according to the token restrictions of the large language model input and output;

[0033] When outputting the information mining results, the information mining agent merges the output information mining results corresponding to multiple sub-contents.

[0034] As a preferred implementation, the key example generation module, the second prompt word also includes a limitation on the output text format, and the data generation agent generates corresponding key examples based on the output text format limitation for the extracted keyword corresponding content.

[0035] On the other hand, the present invention also provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for training a multi-agent biochemical information extraction model based on a large language model as described in any embodiment of the present invention is implemented.

[0036] On the other hand, the present invention also provides a computer-readable medium for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement a multi-agent biochemical information extraction model training method based on a large language model as described in any embodiment of the present invention.

[0037] The present invention has the following beneficial effects:

[0038] Compared with existing data mining methods that require thousands or tens of thousands of labeled data, the present invention can complete complex biological and chemical data mining tasks by verifying less than a hundred data through a large multi-agent language model, and significantly reduces the time, labor and knowledge costs required for data mining of related biological and chemical data, ultimately achieving the purpose of automatically selecting relevant keywords, automatically building prompt words, and automatically conducting information mining.

[0039] At the same time, compared with other extraction methods, this framework also has significant advantages in the accuracy of keyword extraction after a small amount of training. Compared with other information extraction methods of large language models, this framework has significant advantages in accuracy, precision, recall, and evaluation indicators such as BLEU and ROUGE in information extraction and professional vocabulary recognition in the field of biochemistry. Compared with the 40%-60% accuracy of directly using large models, the multi-agent system trained by this framework can achieve an accuracy of more than 90%. In addition, since the extracted keywords will iterate as the text is continuously updated, the information extracted using this framework will also be continuously updated. Therefore, this framework helps to extract information from large amounts of text more comprehensively and objectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0042] It should be understood that the step numbers used in this article are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0043] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0044] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0045] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0046] Embodiment 1:

[0047] In order to make the purpose, technical scheme and advantages of the present invention more clear, the following will be combined with the specific embodiments of the present application and refer to the attached Figure 1 , clearly and completely describe the technical solution of the present invention.

[0048] To solve the problems in the prior art, the present invention provides a multi-agent biochemical information extraction model training method based on a large language model, comprising the following steps:

[0049] Step S1: deploy a large language model for text generation tasks as the core input and output module of the agent; construct a multi-agent biochemical information extraction model including a task agent, a data generation agent, and an information mining agent;

[0050] Step S101: Select a suitable language model architecture, such as Transformer, BERT, GPT, etc.

[0051] Step S102: configure the hardware and software environment. The hardware environment includes GPU / CPU resources, storage devices, etc., and the software environment includes a deep learning framework (TensorFlow, PyTorch, etc.).

[0052] Step S2, obtaining multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry;

[0053] Step S3, taking the first prompt word and a training set of several texts containing biological or chemical information as inputs of the task agent; training the task agent to extract important keywords related to the biological or chemical field from the training set; and verifying the extraction results through experts, and feeding back the verification results to the task agent for reinforcement learning; the keywords output by the task agent are used as tasks for data mining by the subsequent agent system; reducing the time and difficulty of expert task formulation; and improving the comprehensiveness of data mining tasks;

[0054] Step S301: construct the first prompt word for keyword extraction as the first part of the input of the task agent. The prompt word specifies and restricts the text analysis and output content of the keyword extraction agent.

[0055] When building the first prompt word of the task agent, by stipulating that important keywords should be general, representative, and comprehensive in the first prompt word to avoid their specificity and repetition, the large language model can perform more accurate and comprehensive mining of important keywords in the text, especially improving the accuracy of vocabulary extraction in professional fields. This restriction on the direction of information analysis and search by the large language model ensures that important keywords with high frequency can be obtained in the output. When faced with more complex text information, the accuracy of the output content can be improved by increasing the number of important keywords allowed to be output in the prompt word.

[0056] Step S302: input a number of text contents containing biological or chemical information randomly selected from the training set as the second part of the input of the task agent.

[0057] To ensure that the output series of keywords is comprehensive and covers all the key contents mentioned in the training set text, randomly selected training set text descriptions are used as the second part of the input to the task agent.

[0058] Step S303: Output a series of keywords and store them in Python list format or other common data structure formats.

[0059] Step S304: Have experts verify the output series of keyword results, and feed back the verification results to the task agent for reinforcement learning to complete the construction of the task agent.

[0060] Step S4, using the constructed second prompt word and the randomly extracted text containing biological or chemical information as inputs of the data generation agent, the second prompt word containing the keywords extracted by the task agent; training the data generation agent to extract the key words extracted by the task agent as the extraction target, extract the keyword corresponding content from the text content of several biological or chemical information, and generate key examples for the information mining agent to learn; verifying the generated results through experts, and feeding back the verification results to the data generation agent for reinforcement learning;

[0061] Step S401, combining the keywords acquired by the task agent and the text with a limited output format to build a second prompt word as the first part of the input of the data generation agent.

[0062] Key examples are usually used as the third prompt word component of information mining. They can be used as cases to guide information mining agents, standardize the output content of information mining, and improve the accuracy of information mining agents. Since general examples need to be written manually, data generation agents can be trained to automatically generate key examples that can be used in information mining, thereby reducing the labor and time costs used in constructing key examples for information mining. Prompt words are also needed to guide data generation agents to generate key examples.

[0063] Step S402: input a number of text contents containing biological or chemical information randomly selected from the training set as the second part of the input of the data generation agent.

[0064] Step S403: output key examples and store them in Python dictionary format or other common data structure formats.

[0065] Step S404: The output key sample results are verified by experts, and the verification results are fed back to the sample generation agent for reinforcement learning to complete the construction of the sample generation agent.

[0066] Step S5: using the constructed third prompt word and several texts containing biological or chemical information as inputs to the information mining agent, wherein the third prompt word contains the keywords extracted by the task agent and the key examples generated by the data generation agent; training the information mining agent to mine and extract relevant biological or chemical information from the training set text, and output the information mining results in the format of key examples; verifying the extraction results through experts, and feeding back the verification results to the generation information mining agent for reinforcement learning;

[0067] Step S501, construct the third prompt word for information mining, including important keywords extracted from the task agent, the task of information mining and the format of data output, as the first part of the input of the information mining agent.

[0068] Step S502: construct the third prompt word of the information mining agent, including the key examples obtained from the data generation agent as the second part of the input of the information mining agent.

[0069] Step S503, considering the token limitation of the input and output of the large language model, optionally, the text is divided into several batches as the third part input of the information mining agent, the construction of the information mining agent is completed, and information mining is performed on the training set.

[0070] Step S504: Based on the expert feedback results, the information mining agent is subjected to a continuous reinforcement learning process to complete the training of the information mining agent and improve the accuracy of mining biological and chemical related information.

[0071] Step S505: The output results of each batch are in the format of a Python dictionary or other common data structure, and then converted into a Pandas data frame or other common data table format and stored.

[0072] Step S506: Merge the output results of each batch as extracted information, which is subsequently applied to the construction of knowledge graphs, databases, or training data sets.

[0073] Step S6, inputting prompt words into the trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from the target database containing biological or chemical information according to the prompt words.

[0074] In a specific example of finding pharmacokinetic experimental information of a substance, such as a water-soluble experiment, the first prompt word of the task agent built is as follows:

[0075] Model input:

[0076] Please summarize the key experimental conditions for the water solubility experiment of the substance. Analyze the file and output the five most common experimental condition types. These experimental conditions should belong to a certain category, such as pH, temperature and other related general conditions, and cannot be a specific value or condition. Output this information in the format of a Python list. Each output in the list must represent a unique experimental condition type, and repeating similar experimental condition types should be avoided.

[0077] A specific output example is:

[0078] '''Python

[0079] ['pH', 'temperature', 'time']

[0080] Make sure this list is comprehensive and covers all the major types of experimental conditions mentioned in the text.

[0081] Subsequently, 40 randomly selected experimental information descriptions and the prompt words of the constructed task agent were input into the task agent as model input. After analysis, the output of the model was:

[0082] ['pH', 'Temperature', 'Time', 'Experimental Method', 'Solvent Composition']

[0083] Experts in the field of pharmacokinetics then verified the above output results to see whether the obtained keywords were the key experimental conditions for the water-solubility experiment.

[0084] After finding the above keywords through the task agent, you can use the data generation agent to generate samples of water-soluble experiments. For example:

[0085] Model input:

[0086] Define the task of biomedical information mining; the output is in the format of a Python dictionary, with the key values ​​including 'original text', 'pH', 'temperature', 'time', 'experimental method', and 'solvent composition'. If none, enter 'none'. Please analyze all sentences in all texts.

[0087] Subsequently, the above prompt words and the original descriptions of 40 relevant experimental conditions were input into the data generation agent. After analysis, 40 samples were automatically generated, such as:

[0088] [{Original text: Water solubility of substances at pH 7.4

[0089] pH: 7.4

[0090] Temperature: None

[0091] Time: None

[0092] Experimental method: None

[0093] Solvent composition: water}, ...]

[0094] Experts in the field of pharmacokinetics then verified the above output results to see whether the key examples were correctly extracted from the original text.

[0095] After generating the key experimental conditions and examples of the water solubility of the substance, these two parts will be input into the information mining agent as the command and case of the prompt word to mine the information about these experimental conditions in the input text, such as:

[0096] The output is in Python dictionary format, with keys 'index', 'original text', 'pH', 'temperature', 'time', 'method', and 'solvent composition'. If none, enter 'None'. Please analyze all sentences in all texts.

[0097] Case section:

[0098] [{Input: water solubility of the substance at pH 7.4,

[0099] Output example: {index: 1, original text: 'Water solubility of a substance at pH 7.4', pH: 7.4, temperature: none, time: none, experimental method: none, solvent composition: water}}, ..]

[0100] Subsequently, the original text descriptions of the above prompt words and experimental conditions were input into the information mining agent. Since the large language model used has an upper limit on the number of tokens processed, all texts were divided into groups of 20 and analyzed in turn. After the information mining of the text, the information about the water-soluble experimental conditions was output in the format of a Python dictionary. Subsequently, this information was converted from the Python dictionary into a Pandas data frame for storage and then merged together.

[0101] The output of the information mining agent is verified by experts and becomes a standard data set, which is fed back to the information mining agent in the form of prompt words or model fine-tuning, and a cyclic reinforcement learning process is carried out to improve the accuracy of biological and chemical information mining in professional literature and data. The more standard data sets there are, the better the performance of the information mining agent, but the corresponding expert verification consumes more time and manpower; the present invention proposes that only through a small amount of expert-assisted data cleaning, the multi-agent solution can effectively analyze key biological and chemical information from the literature and complete high-quality data information extraction.

[0102] Embodiment 2:

[0103] This embodiment provides a multi-agent biochemical information extraction model based on a large language model, including:

[0104] Model deployment module, deploying a large language model for text generation tasks as the core input and output module of the agent; building a multi-agent biochemical information extraction model including task agents, data generation agents and information mining agents;

[0105] A training set construction module obtains multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry;

[0106] The keyword generation module takes the first prompt word and a training set of texts containing biological or chemical information as inputs to the task agent; trains the task agent to extract important keywords related to the biological or chemical field from the training set; verifies the extraction results through experts, and feeds the verification results back to the task agent for reinforcement learning;

[0107] The key sample generation module uses the constructed second prompt word and the randomly extracted text containing biological or chemical information as the input of the data generation agent, and the second prompt word contains the keywords extracted by the task agent; the training data generation agent takes the important keywords extracted by the task agent as the extraction target, extracts the corresponding content of the keywords from the text content of several biological or chemical information, and generates key samples for the information mining agent to learn; the generated results are verified by experts, and the verification results are fed back to the data generation agent for reinforcement learning;

[0108] The information mining module uses the constructed third prompt words and several texts containing biological or chemical information as inputs to the information mining agent, where the third prompt words contain keywords extracted by the task agent and key examples generated by the data generation agent; the information mining agent is trained to mine and extract relevant biological or chemical information from the training set texts, and output the information mining results in the format of key examples; the extracted results are verified by experts, and the verification results are fed back to the generated information mining agent for reinforcement learning;

[0109] The model uses a module to input prompt words into a trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from a target database containing biological or chemical information according to the prompt words.

[0110] Embodiment three:

[0111] This embodiment provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, a multi-agent biochemical information extraction model training method based on a large language model as described in any embodiment of the present invention is implemented.

[0112] Embodiment 4:

[0113] This embodiment provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement a multi-agent biochemical information extraction model training method based on a large language model as described in any embodiment of the present invention.

[0114] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0115] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0117] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, 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, which is stored in a storage medium and includes 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 method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0118] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-agent biochemical information extraction model training method based on a large language model, characterized in that: The specific steps include: Step S1: deploy a large language model for text generation tasks as the core input and output module of the agent; construct a multi-agent biochemical information extraction model including a task agent, a data generation agent, and an information mining agent; Step S2, obtaining multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry; Step S3, using the constructed first prompt word and a training set of several texts containing biological or chemical information as inputs of the task agent; The training task agent extracts important keywords related to the biological or chemical fields from the training set; the extracted results are verified by experts and the verification results are fed back to the task agent for reinforcement learning; Step S4, using the constructed second prompt word and the randomly extracted text containing biological or chemical information as inputs of the data generation agent, and the second prompt word contains the keywords extracted by the task agent as the main task of case generation; The training data generation agent uses the important keywords extracted by the task agent as the extraction target, extracts the corresponding content of the keywords from the text content of several biological or chemical information, and generates key examples for the information mining agent to learn; the generated results are verified by experts, and the verification results are fed back to the data generation agent for reinforcement learning; The second prompt word also includes a limitation on the output text format; the data generation agent generates corresponding key examples according to the limitation on the output text format by extracting the corresponding content of the keyword; Step S5: using the constructed third prompt word and several texts containing biological or chemical information as inputs to the information mining agent, wherein the third prompt word contains the keywords extracted by the task agent and the key examples generated by the data generation agent; training the information mining agent to mine and extract relevant biological or chemical information from the training set text, and output the information mining results in the format of key examples; verifying the extraction results through experts, and feeding back the verification results to the generation information mining agent for reinforcement learning; Step S6, inputting prompt words into the trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from the target database containing biological or chemical information according to the prompt words.

2. The multi-agent biochemical information extraction model training method based on a large language model according to claim 1, characterized in that: The first prompt word constructed and a training set of several texts containing biological or chemical information are used as inputs of the task agent, and the number of important keywords outputted can be adjusted by adjusting the first prompt word limit according to the complexity of the extraction target.

3. The multi-agent biochemical information extraction model training method based on a large language model according to claim 1, characterized in that: In the step of using the constructed third prompt word and a number of texts containing biological or chemical information as inputs to the information mining agent: The information mining agent divides the input content into multiple sub-contents and then performs input operations according to the token restrictions of the large language model input and output; When outputting the information mining results, the information mining agent merges the output information mining results corresponding to multiple sub-contents.

4. A multi-agent biochemical information extraction model based on a large language model, characterized in that: include: Model deployment module, deploying a large language model for text generation tasks as the core input and output module of the agent; building a multi-agent biochemical information extraction model including task agents, data generation agents and information mining agents; A training set construction module obtains multiple text training sets containing biological or chemical information from a target database; the target database is journal articles, patents and databases in the field of biology or chemistry; The keyword generation module takes the first prompt word and a training set of texts containing biological or chemical information as inputs to the task agent; trains the task agent to extract important keywords related to the biological or chemical field from the training set; verifies the extraction results through experts, and feeds the verification results back to the task agent for reinforcement learning; The key example generation module uses the constructed second prompt word and the randomly extracted text containing biological or chemical information as the input of the data generation agent. The second prompt word contains the keywords extracted by the task agent as the main task of case generation; The training data generation agent uses the important keywords extracted by the task agent as the extraction target, extracts the corresponding content of the keywords from the text content of several biological or chemical information, and generates key examples for the information mining agent to learn; the generated results are verified by experts, and the verification results are fed back to the data generation agent for reinforcement learning; The second prompt word also includes a limitation on the output text format; the data generation agent generates corresponding key examples according to the limitation on the output text format by extracting the corresponding content of the keyword; The information mining module uses the constructed third prompt words and several texts containing biological or chemical information as inputs to the information mining agent, where the third prompt words contain keywords extracted by the task agent and key examples generated by the data generation agent; the information mining agent is trained to mine and extract relevant biological or chemical information from the training set texts, and output the information mining results in the format of key examples; the extracted results are verified by experts, and the verification results are fed back to the generated information mining agent for reinforcement learning; The model uses a module to input prompt words into a trained multi-agent biochemical information extraction model; the multi-agent biochemical information extraction model extracts and outputs corresponding information mining results from a target database containing biological or chemical information according to the prompt words.

5. The multi-agent biochemical information extraction model based on a large language model according to claim 4, characterized in that: The keyword generation module uses the constructed first prompt word and a training set of several texts containing biological or chemical information as inputs of the task agent, and can adjust the number of important keywords output by adjusting the first prompt word limit according to the complexity of the extraction target.

6. The multi-agent biochemical information extraction model based on a large language model according to claim 4, characterized in that: The information mining module, in the step of using the constructed third prompt word and a number of texts containing biological or chemical information as information mining agent input: The information mining agent divides the input content into multiple sub-contents and then performs input operations according to the token restrictions of the large language model input and output; When outputting the information mining results, the information mining agent merges the output information mining results corresponding to multiple sub-contents.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for training a multi-agent biochemical information extraction model based on a large language model as described in any one of claims 1 to 3 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a multi-agent biochemical information extraction model training method based on a large language model as described in any one of claims 1 to 3.

Images