Multi-agent reasoning system, data reasoning method, device, equipment and product

Abstract

The invention provides a multi-agent reasoning system, a data reasoning method, a data reasoning device, data reasoning equipment and a product, the method is applied to the multi-agent reasoning system comprising a plurality of agents, and the method comprises the following steps: a first agent performs key information extraction on reasoning reference data according to a reasoning problem to obtain reasoning key information; the second intelligent agent generates a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path comprises prompt information for reasoning based on the reasoning key information to solve the reasoning problem; and the third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, and the reasoning result comprises a reasoning process and / or a reasoning conclusion. According to the method, the data reasoning flexibility and reasoning efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a multi-agent reasoning system, data reasoning method, device, equipment and product. Background Art

[0002] Data reasoning refers to the process of reasoning specific data in a certain direction or under certain requirements to draw conclusions. Data reasoning is a common method of extracting or gaining insights from data and making decisions based on data. For example, in medical auxiliary diagnosis scenarios, it is often necessary to conduct comprehensive reasoning and analysis on the patient's medical records, examination reports, physical examination reports, etc. to determine the patient's disease and the corresponding diagnosis and treatment plan.

[0003] Traditional data reasoning is usually done by professionals who read, summarize, and understand the reference data, and then use experience to reason to get the reasoning results. This data reasoning method is inefficient, greatly affected by the professional level of personnel, and has poor reasoning flexibility. Summary of the invention

[0004] Based on the above-mentioned technical status, this application proposes a multi-agent reasoning system, data reasoning method, device, equipment and product, which can improve the flexibility and reasoning efficiency of data reasoning.

[0005] In a first aspect, the present application proposes a data reasoning method, which is applied to a multi-agent reasoning system including multiple agents. The method comprises:

[0006] The first agent extracts key information from the reasoning reference data according to the reasoning problem to obtain the key information of reasoning;

[0007] The second agent generates a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem;

[0008] The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0009] In some implementations, the method further includes:

[0010] The third agent evaluates the reasoning result to obtain an evaluation conclusion;

[0011] When the evaluation conclusion indicates that the reasoning result is wrong, the third agent calls the second agent to regenerate the reasoning path, and the third agent performs integrated reasoning based on the reasoning problem and the reasoning path regenerated by the second agent to regenerate the reasoning result corresponding to the reasoning problem.

[0012] In some implementations, the second agent, in the process of generating the reasoning path according to the reasoning problem and the reasoning key information, further generates a retrieval problem, wherein the retrieval problem is used to retrieve the knowledge required to execute the reasoning path from the knowledge base;

[0013] The method further comprises:

[0014] The fourth agent retrieves a search result corresponding to the search question from a knowledge base according to the search question; the search result includes relevant knowledge corresponding to the search question, and the knowledge base is constructed by the fifth agent;

[0015] The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, including:

[0016] The third agent performs integrated reasoning according to the reasoning problem, the retrieval results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

[0017] In some implementations, the fifth agent constructs the knowledge base by performing the following processing:

[0018] Perform knowledge segmentation on the knowledge source file to obtain knowledge labels;

[0019] Generating a knowledge triple based on the knowledge tag, wherein the knowledge triple includes a knowledge entity, attribute information of the knowledge entity, and the knowledge entity and a source of the attribute information of the knowledge entity;

[0020] Each knowledge triple is integrated with knowledge entities as the center to obtain the knowledge graph.

[0021] In some implementations, the attribute information of the knowledge entity includes the knowledge entity and an encoding of the chapter content where the attribute information of the knowledge entity is located in the knowledge entity and the source file of the knowledge entity.

[0022] In some implementations, the search results also include evaluation information on the relevant knowledge.

[0023] In some implementations, the fourth agent further determines whether the search question needs to be split;

[0024] In the case where it is determined that the retrieval problem needs to be split, the fourth agent calls the second agent to split the retrieval problem into a plurality of retrieval sub-problems;

[0025] Furthermore, the fourth agent retrieves from the knowledge base according to the multiple retrieval sub-questions and obtains retrieval results corresponding to each retrieval sub-question.

[0026] In some implementations, the reasoning problem includes a diagnosis and treatment reasoning problem, and the reasoning reference data includes a patient's medical documents.

[0027] In some implementations, the patient's medical documents include multiple medical documents of the patient at different periods;

[0028] The first agent extracts key information from the reasoning reference data according to the reasoning problem to obtain key reasoning information, including:

[0029] The first intelligent agent extracts patient portrait information from the multiple medical documents based on the reasoning problem, and generates a basic information portrait and a time-series medical portrait corresponding to the patient.

[0030] A second aspect of the present application proposes a multi-agent reasoning system, comprising multiple agents, wherein the multiple agents in the multi-agent system are configured to implement the data reasoning method as described in the first aspect or any implementation method of the first aspect.

[0031] The third aspect of the present application proposes another data reasoning method, including:

[0032] Extract key information from the reasoning reference data according to the reasoning problem to obtain the key information of reasoning;

[0033] Generate a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem;

[0034] Integrated reasoning is performed according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0035] In a fourth aspect, the present application provides a data reasoning device, which is applied to a multi-agent reasoning system including multiple agents, and the device includes:

[0036] An information integration unit, configured to extract key information from the reasoning reference data according to the reasoning problem through the first agent, so as to obtain the reasoning key information;

[0037] An analysis and reasoning unit, configured to generate a reasoning path through a second agent according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning and answering the reasoning problem based on the reasoning key information;

[0038] The integrated reasoning unit is used to perform integrated reasoning according to the reasoning problem and the reasoning path through a third agent to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0039] A fifth aspect of the present application provides a data reasoning device, comprising:

[0040] Input devices, output devices, and multi-agent reasoning systems;

[0041] The input device is connected to the processor and is used to input reasoning questions and reasoning reference data to the processor;

[0042] The output device is connected to the processor and is used to output the inference result generated by the processor;

[0043] The multi-agent reasoning system is configured to execute the data reasoning method as described in the first aspect or any implementation of the first aspect.

[0044] In a sixth aspect, the present application proposes a computer program product, comprising computer program instructions, which, when executed by multiple agents in a multi-agent reasoning system, enable the multi-agent reasoning system to execute the data reasoning method described in the first aspect or any implementation of the first aspect.

[0045] The data reasoning method proposed in this application completes the data reasoning process through the collaboration of different agents in a multi-agent reasoning system. In the multi-agent reasoning system, each agent automatically performs reasoning operations at different stages of data reasoning, thereby enabling fully automated data reasoning. Moreover, the multi-agent reasoning system automatically performs corresponding data reasoning operations based entirely on the input reasoning problems and reasoning reference data, without specifying reasoning rules, and its data reasoning is more flexible and can meet a variety of different data reasoning needs.

[0046] In addition, based on the intelligence capabilities of each agent, it can be ensured that each agent focuses on processing at different stages of reasoning, thereby ensuring the accuracy of reasoning in each reasoning link, as well as the accuracy of the entire reasoning process and the reasoning results. The reasoning process can also be included in the reasoning results, making the data reasoning conclusion more interpretable, enabling users to understand the origin of the reasoning conclusion more intuitively and efficiently, and improving the efficiency of users' information acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0048] Figure 1 A flowchart of a data reasoning method provided in an embodiment of the present application.

[0049] Figure 2-Figure 6 Flow charts of several other data inference methods provided in the embodiments of the present application.

[0050] Figure 7 A schematic diagram of the structure of a data inference device provided in an embodiment of the present application.

[0051] Figure 8 A schematic diagram of the structure of a data inference device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The technical solution of the embodiment of the present application is applicable to data reasoning application scenarios, specifically, to application scenarios in which given reasoning reference data is reasoned according to reasoning problems to answer reasoning problems. The technical solution of the embodiment of the present application can achieve efficient and accurate data reasoning, and make the reasoning conclusion more interpretable.

[0053] Traditional data reasoning is usually done by professionals who read, summarize, and understand the reference data, and then combine experience to reason and obtain the reasoning results. This data reasoning method is inefficient and greatly affected by the professional level of personnel, and the reasoning conclusions are less interpretable.

[0054] For example, in a medical diagnosis scenario, a large number of medical documents such as test reports, examination reports, and medical records will be generated during the patient's medical treatment process. These medical documents combined with professional medical knowledge can be used to judge the patient's disease and determine the diagnosis and treatment plan. At present, the above-mentioned disease diagnosis and treatment reasoning based on medical documents is mainly performed by professional doctors, that is, doctors determine the patient's disease and diagnosis and treatment plan by reading the patient's various medical documents. However, manual data reasoning will bring great work pressure to doctors, and with the enrichment and deepening of various examination items, more and more medical documents will be generated for doctors to consult, which will further reduce the efficiency of diagnosis and treatment. In addition, the above-mentioned diagnosis and treatment methods are greatly affected by the level of doctors, and for patients, they can only know the final diagnosis conclusion, and have no way of knowing the diagnosis process and basis.

[0055] In recent years, artificial intelligence technology has developed rapidly, bringing new opportunities for data reasoning. For example, in the medical field, more and more intelligent diagnosis solutions have emerged.

[0056] However, current intelligent reasoning solutions can often only perform reasoning under specific rules for a single data file. For example, in the field of medicine, intelligent diagnosis algorithms usually pre-design diagnosis or reasoning rules for a certain medical document. When the content of the medical document meets the corresponding diagnosis or reasoning rules, the diagnosis conclusion corresponding to the diagnosis or reasoning rules is directly output. For personalized reasoning requirements proposed by users, or for more diverse and larger data, current intelligent reasoning solutions are still difficult to achieve integrated reasoning, cannot effectively meet user needs, and can only be used as an auxiliary tool for manual reasoning.

[0057] In response to the above technical problems, this application proposes a new reasoning system, and based on the reasoning system, proposes a new data reasoning method. Based on the above reasoning system and reasoning method, more efficient and intelligent data reasoning can be achieved, and the reasoning conclusions can be made more interpretable.

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0059] The embodiment of the present application first proposes a multi-agent reasoning system, which includes multiple agents.

[0060] The above-mentioned intelligent agent refers to an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware or a system with autonomy, adaptability and interaction capabilities. The intelligent agent perceives changes in the environment (such as through sensors or data input), makes judgments and decisions based on the knowledge and algorithms it has learned, and then performs actions to affect the environment or achieve predetermined goals.

[0061] In the embodiment of the present application, multiple agents are arranged in the reasoning system, each of which is used to perform different tasks, so that the entire reasoning system can achieve more flexible, efficient and accurate data reasoning.

[0062] In some embodiments, the above-mentioned multiple agents can be independent agents, for example, each agent can be run in an independent hardware environment, and the agents can communicate with each other, so as to realize the calling and collaborative interaction between different agents. For example, the agents are respectively set in different execution environments such as servers, computers, cloud servers, etc., and the agents are connected through wireless networks or wired networks to form a multi-agent reasoning system.

[0063] In other embodiments, the above-mentioned multiple intelligent agents can also be integrated in the same hardware environment. For example, multiple intelligent agents can be integrated in the same cloud server. In this case, each intelligent agent can serve as a different functional module in the cloud server, and each intelligent agent can share the hardware resources of the cloud server and can achieve collaborative interaction with each other.

[0064] In an embodiment of the present application, each agent in the above-mentioned multi-agent reasoning system is configured to perform different sub-tasks in data reasoning, so that multiple agents cooperate with each other to realize the full-process data reasoning process.

[0065] For any agent in the multi-agent reasoning system, it can be composed of a single deep learning model or multiple deep learning models, and the above-mentioned deep learning models include but are not limited to generative large language models. The deep learning models set in the same agent are aggregated at once through MOE (Mixture of Experts, mixed expert model, ensemble learning, etc.), so as to realize the functions of the agent.

[0066] The structures and training methods of the various agents in the multi-agent reasoning system may be the same or different, and are not limited in the embodiments of the present application.

[0067] In addition, the embodiments of the present application do not limit the type, quantity, and training method of the deep learning models contained in each agent. In the subsequent embodiments, the functions of each agent and the interactive cooperation method of each agent in the data reasoning process are only introduced by way of example through the embodiments, so as to realize the exemplary introduction of the functions of the multi-agent reasoning system proposed in the present application and the processing process of the data reasoning method. The realization of the functions of each agent and the interactive cooperation between each agent are mainly realized by training each agent. For the specific agent training method, please refer to the conventional neural network model training method, which will not be described in detail in the embodiments of the present application.

[0068] Based on the above-mentioned multi-agent reasoning system, another embodiment of the present application also proposes a data reasoning method implemented in the multi-agent reasoning system. The method achieves more efficient, flexible and accurate data reasoning through the interaction and collaboration between the agents in the multi-agent reasoning system.

[0069] Below, the structure and function of the multi-agent reasoning system proposed in this application, as well as the processing of the data reasoning method proposed in this application, will be introduced exemplarily through a series of embodiments in combination with the application scenarios of medical diagnosis reasoning. Through these introductions, those skilled in the art can further clarify the structure and function of the multi-agent reasoning system proposed in this application, and at the same time, clarify the specific processing of the data reasoning method proposed in this application applied to the above-mentioned multi-agent reasoning system.

[0070] In some embodiments, the multi-agent reasoning system includes a first agent, a second agent, and a third agent. The first agent is used to perform data integration and information extraction of reasoning reference data, the second agent is used to generate a reasoning path for solving reasoning problems based on the information extracted by the first agent, and the third agent is used to perform final integrated reasoning based on the reasoning path generated by the second agent and generate a reasoning result.

[0071] Based on this multi-agent reasoning system, see Figure 1 As shown, the data reasoning method proposed in the embodiment of the present application includes the following steps:

[0072] S101. The first intelligent agent extracts key information from the reasoning reference data according to the reasoning problem to obtain the reasoning key information.

[0073] The above-mentioned reasoning reference data refers to data objects that need to be reasoned and analyzed to mine deep knowledge and information. For example, in a medical diagnosis scenario, it is necessary to combine the patient's various medical documents to judge the patient's disease and treatment plan. At this time, the patient's various medical documents are the reasoning reference data. By performing reasoning analysis on these reasoning reference data, the patient's disease and treatment plan can be determined.

[0074] The above-mentioned inference reference data can be of any data format or any data content, for example, it can be data in electronic format stored in a memory, such as text data, image data, audio and video data, such as books, articles, reports, etc.

[0075] The above-mentioned inference reference data may be directly generated and stored in a computer system, or may be obtained by identifying physical information carriers such as books, articles, and reports.

[0076] The above-mentioned reasoning problem refers to the reasoning requirement or reasoning purpose of the above-mentioned reasoning reference data. The reasoning problem represents the reasoning direction, requirement, reasoning restriction condition, etc. that the user wants to reason about the above-mentioned reasoning reference data.

[0077] When a user calls the multi-agent reasoning system of an embodiment of the present application to perform data reasoning, the reasoning problem and reasoning reference data are first input into the multi-agent reasoning system. After the first agent in the system receives the reasoning problem and the reasoning reference data, it extracts key information from the reasoning reference data based on the reasoning problem to obtain reasoning key information.

[0078] Among them, the first intelligent agent extracts key information from the reasoning reference data, mainly by deduplicating, integrating, extracting, and summarizing the reasoning reference data, and extracting information from the reasoning reference data that is useful for answering the above-mentioned reasoning problems, thereby obtaining the key information for reasoning.

[0079] In some embodiments, it is disclosed that in a medical diagnosis scenario, a user can input diagnosis and treatment reasoning problems and a patient's medical documents into the above-mentioned multi-agent reasoning system to perform reasoning to determine the patient's disease.

[0080] In the above scenario, the diagnosis and treatment reasoning problem is the reasoning problem that is input into the reasoning system, such as "see what disease I have". The patient's medical documents are the reasoning reference data for the input reasoning system. After obtaining the above diagnosis and treatment reasoning problem and the patient's medical documents, the first agent extracts key information from the patient's medical documents based on the diagnosis and treatment reasoning problem, mainly to extract information useful for answering the above diagnosis and treatment reasoning problem.

[0081] The patient's medical documents may include medical records, physical examination reports, test results, patient self-reports, etc. Moreover, the patient's medical documents may be one medical document or multiple medical documents, for example, multiple medical documents generated by the patient at different stages of the medical treatment process.

[0082] When the patient's medical documents include multiple medical documents of the patient at different periods, when the first intelligent agent extracts key information from the patient's multiple medical documents based on the diagnosis and treatment reasoning problem, it first divides and classifies each medical document according to the evolution of the patient's condition, and then integrates, extracts, and summarizes the divided and classified medical documents to extract patient portrait information as key reasoning information to participate in the subsequent reasoning process.

[0083] Among them, the first intelligent agent extracts the basic information of the patient, such as age, gender, family medical history, etc., which have not changed during the patient's medical treatment, to form a basic information portrait of the patient.

[0084] The first intelligent agent integrates and summarizes the patient's clinical manifestations, test results, surgeries, medications, diseases, medical histories, etc. at different stages into a series of time-series medical portraits with timestamps.

[0085] For example, the first agent can obtain the following patient portrait information by extracting key information from multiple medical documents of the patient:

[0086] Chief complaint: acid reflux and heartburn for half a month.

[0087] Current medical history: The patient has had acid reflux and heartburn for nearly half a month, which is more obvious after meals, accompanied by pain behind the sternum, no vomiting, no nausea, and poor appetite. He came to the hospital for treatment today.

[0088] Past history: The patient was in good health.

[0089] Physical examination: clear consciousness, clear breath sounds in both lungs, tenderness (+) under the xiphoid process, no rebound tenderness, no muscle guarding, and shifting dullness.

[0090] (-), bowel sounds normal.

[0091] Inspection and testing:

[0092] 2020-03-08 Gastroscopy: mild damage to the esophageal mucosa, follow-up.

[0093] Age: 57

[0094] Gender:Male

[0095] The above-mentioned basic information portrait and time-series medical portrait together constitute the patient portrait information. Among them, the basic information portrait can represent the patient's basic information. This information can be used to judge the patient's physical fitness, basic physical condition, and diagnose certain diseases whose incidence rate changes with age; the time-series medical portrait reflects the patient's condition characteristics, disease evolution process, development trend, etc. Based on this information, we can look at the patient's entire course of illness and systematically grasp the patient's condition, judge the patient's disease and the severity of the disease.

[0096] It can be understood that the first intelligent agent extracts key information from the reasoning reference data according to the reasoning problem, and can eliminate information that is useless for solving the reasoning problem from the reasoning reference data. It can also integrate and summarize the reasoning reference data to extract information that is useful for solving the reasoning problem. Moreover, by organizing these useful information, the obtained key reasoning information can be made easier to reason, which is conducive to solving the reasoning problem and providing a data basis for subsequent data reasoning processes.

[0097] S102: The second agent generates a reasoning path according to the reasoning problem and the reasoning key information.

[0098] Specifically, after the first agent extracts the key reasoning information from the reasoning reference data, the key reasoning information is sent to the second agent together with the reasoning question, which triggers the second agent to execute the subsequent reasoning process.

[0099] After receiving the reasoning problem and the key reasoning information sent by the first agent, the second agent further analyzes and infers the reasoning problem and the key reasoning information to generate a reasoning path. The reasoning path generated by the second agent may include one reasoning path or multiple reasoning paths, which can be used to obtain the reasoning result corresponding to the reasoning problem.

[0100] Among them, the above-mentioned reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem, that is, based on which information in the reasoning key information can determine the reasoning conclusion corresponding to the reasoning problem, or determine the prompt information of the reasoning conclusion corresponding to a sub-problem in the reasoning problem.

[0101] For example, for the patient portrait information provided by the first agent, "Chief complaint: acid reflux and heartburn for half a month. Current medical history: The patient has had acid reflux and heartburn for nearly half a month, which is more obvious after meals, accompanied by pain behind the sternum, no vomiting, no nausea, and poor appetite. He came to the hospital for treatment today. Past history: The patient was previously healthy. Physical examination: clear consciousness, clear breath sounds in both lungs, tenderness under the xiphoid process (+), no rebound pain, no muscle guarding, mobile dullness (-), and normal bowel sounds. Tests and examinations: 2020-03-08 Gastroscopy: Mild damage to the esophageal mucosa, follow-up. Age: 57 years old Gender: Male", and the diagnosis and treatment reasoning question sent by the first agent, "See what disease I have", the second agent reasons about the above patient portrait information based on the diagnosis and treatment reasoning problem and generates a reasoning path.

[0102] For example, the reasoning path generated by the second agent is "First, according to the patient's chief complaint and current medical history, the patient has acid reflux and heartburn, especially after meals. And there is pain behind the sternum. This suggests that we should consider upper gastrointestinal disease." Based on this reasoning path, a preliminary reasoning conclusion about the patient's disease diagnosis can be drawn.

[0103] S103. The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0104] Specifically, after generating the reasoning path, the second agent sends the generated reasoning path and the reasoning problem to the third agent, triggering the third agent to execute the subsequent reasoning process.

[0105] The third intelligent agent performs integrated reasoning based on the received reasoning problem and reasoning path. For example, it integrates the reasoning basis, prompt information, etc. in the reasoning path to form a reasoning process and obtain the corresponding reasoning conclusion. Alternatively, when there are multiple reasoning paths, the third intelligent agent integrates the reasoning process and reasoning conclusion of each reasoning path respectively, and finally obtains the reasoning result corresponding to the above reasoning problem. The form of the reasoning result conforms to the form of the answer to the reasoning problem, and the reasoning result includes the reasoning process and / or reasoning conclusion.

[0106] Including the reasoning process in the reasoning result can make the reasoning result more interpretable, so that users can more intuitively understand the origin of the reasoning conclusion through the reasoning result, thereby enhancing the readability of the reasoning result.

[0107] From the above introduction, it can be seen that the data reasoning method proposed in the embodiment of the present application completes the data reasoning process through the collaboration of different agents in the multi-agent reasoning system. In the multi-agent reasoning system, each agent automatically performs reasoning operations at different stages of data reasoning, thereby enabling fully automated data reasoning. Moreover, the multi-agent reasoning system automatically performs corresponding data reasoning operations based entirely on the input reasoning problems and reasoning reference data, without specifying reasoning rules, and its data reasoning is more flexible and can meet a variety of different data reasoning needs.

[0108] In addition, based on the intelligence capabilities of each agent, it can be ensured that each agent focuses on processing at different stages of reasoning, thereby ensuring the accuracy of reasoning in each reasoning link, as well as the accuracy of the entire reasoning process and the reasoning results. The reasoning process can also be included in the reasoning results, making the data reasoning conclusion more interpretable, enabling users to understand the origin of the reasoning conclusion more intuitively and efficiently, and improving the efficiency of users' information acquisition.

[0109] In another embodiment, see Figure 2 As shown, after generating the reasoning result, the third agent further evaluates the reasoning result to obtain an evaluation conclusion.

[0110] Specifically, the third agent evaluates the reasoning result, specifically by combining the reasoning problem, the reasoning path, and the key information of the reasoning, and further analyzing whether the reasoning result obtained by the integrated reasoning based on the reasoning path sent by the second agent is correct. The evaluation conclusion includes judgment information on whether the reasoning result is correct or not, and may also include explanation information on the error of the reasoning result.

[0111] Then, the third agent further performs subsequent operations according to the evaluation conclusion.

[0112] If the evaluation conclusion indicates that the obtained inference result is correct, the third agent outputs the inference result.

[0113] If the evaluation conclusion indicates that the obtained reasoning result is wrong, the third agent sends the evaluation conclusion to the second agent and calls the second agent to regenerate the reasoning path. Correspondingly, the second agent re-reasons according to the evaluation conclusion sent by the third agent, the reasoning problem and the key reasoning information, regenerates the reasoning path, and sends it to the third agent.

[0114] The third agent performs integrated reasoning based on the reasoning problem and the reasoning path regenerated by the second agent, and regenerates the reasoning result corresponding to the reasoning problem.

[0115] Then, the third agent evaluates the regenerated reasoning results again to obtain an evaluation conclusion.

[0116] The above process is repeated until the third agent determines that the generated inference result is correct and outputs the inference result, or when the process is repeated a set number of times, the third agent outputs an inference abnormality prompt message.

[0117] Through the above-mentioned evaluation operation, the third agent further verifies the correctness of the reasoning conclusion, thereby ensuring the generation of correct reasoning results and improving the accuracy of data reasoning.

[0118] In some other embodiments, the multi-agent reasoning system further includes a fourth agent and a fifth agent. The fourth agent is used to retrieve the knowledge required for data reasoning from the knowledge base during the data reasoning process, and the fifth agent is used to build and maintain a retrieval-friendly knowledge base for the fourth agent to perform knowledge retrieval.

[0119] The fifth agent constructs the knowledge base by segmenting and integrating the knowledge source files, thereby constructing a search-friendly knowledge base. The above-mentioned knowledge source files can be any field and any type of knowledge carrier, including but not limited to books, articles, news manuscripts, network resource files, etc. Specifically, the fifth agent can select the knowledge source files in the corresponding field to construct the knowledge base according to the constructed knowledge base field or according to the application field of the multi-agent reasoning system.

[0120] In some embodiments, it is disclosed that the fifth agent constructs a knowledge base by executing the following steps A1-A3:

[0121] A1. Perform knowledge segmentation on the knowledge source file to obtain knowledge labels.

[0122] Exemplarily, the fifth agent may divide the content in the knowledge source file into knowledge segments according to the chapters, titles, etc. of the content in the file, and perform information summary or information generalization for each divided knowledge segment to obtain at least one knowledge tag.

[0123] Alternatively, the fifth intelligent agent may also perform semantic recognition and classification on the content in the knowledge source file, extract different semantic fragments therefrom, and perform information summary or information generalization on each semantic fragment to obtain at least one knowledge label.

[0124] For example, assuming that the fifth agent builds a medical diagnosis knowledge base, the knowledge base can be built based on medical textbooks. The fifth agent divides the textbook into chapters and paragraphs, and extracts knowledge summaries from each paragraph to obtain corresponding knowledge labels.

[0125] For example, for the section "Coronary Atherosclerotic Heart Disease" in textbook A, the fifth intelligent agent divides it into three sections and generates three knowledge labels for each section: "Pathophysiology of Coronary Atherosclerotic Heart Disease", "Clinical Manifestations of Coronary Atherosclerotic Heart Disease", and "Treatment of Coronary Atherosclerotic Heart Disease".

[0126] The above-mentioned chapter segmentation process is performed on textbook B, and a knowledge label of “prevention of coronary atherosclerotic heart disease” can be generated for the segmentation result.

[0127] Similarly, for other medical textbooks and other types of medical knowledge source files, such as medical papers, medical cases, etc., knowledge tags can be generated by referring to the above method.

[0128] A2. Generate knowledge triples based on the knowledge tags.

[0129] The knowledge triplet includes a knowledge entity, attribute information of the knowledge entity, and the knowledge entity and the source of the attribute information of the knowledge entity.

[0130] Specifically, after the fifth intelligent agent obtains the knowledge label through the processing of step A1, it organizes the knowledge for each knowledge label separately, determines the knowledge entity in the knowledge label, the attribute information of the knowledge entity, and the source of the knowledge entity and the attribute information of the knowledge entity, and generates knowledge triples based on the above-mentioned knowledge entities, attribute information and sources.

[0131] For example, for the four knowledge labels of “Pathophysiology of coronary atherosclerotic heart disease”, “Clinical manifestations of coronary atherosclerotic heart disease”, “Treatment of coronary atherosclerotic heart disease”, and “Prevention of coronary atherosclerotic heart disease”, the fifth agent generates corresponding knowledge triples: <Coronary atherosclerotic heart disease, pathophysiology, textbook A>, <Coronary atherosclerotic heart disease, clinical manifestations, textbook A>, <Coronary atherosclerotic heart disease, treatment, textbook A>, and <Coronary atherosclerotic heart disease, prevention, textbook B>.

[0132] In other embodiments, the attribute information of the above-mentioned knowledge entity includes the encoding of the knowledge entity and the chapter content of the knowledge entity and the source file of the knowledge entity, that is, the chapter content of the source file containing the knowledge entity and the attribute information of the knowledge entity is encoded, such as generating its md5 encoding, and linking it to the knowledge entity as part of the attribute information of the knowledge entity. Through this encoding, when the attribute information of any knowledge entity is queried, the complete knowledge content containing the knowledge entity and its attribute information can be obtained through the corresponding content encoding.

[0133] A3. Perform knowledge entity-centered integration processing on each knowledge triple to obtain a knowledge graph.

[0134] Specifically, after the fifth intelligent agent obtains a series of knowledge triples through the processing of step A2, it integrates each knowledge triple with a knowledge entity as the center to obtain a knowledge graph, so that the attribute information and source information of the same knowledge entity are aggregated together, so that fast and comprehensive knowledge retrieval can be performed based on the constructed knowledge graph.

[0135] For example, for the knowledge entity "coronary atherosclerotic heart disease", the knowledge triples of <coronary atherosclerotic heart disease, pathophysiology, textbook A>, <coronary atherosclerotic heart disease, clinical manifestations, textbook A>, <coronary atherosclerotic heart disease, treatment, textbook A>, and <coronary atherosclerotic heart disease, prevention, textbook B> are integrated with the knowledge entity "coronary atherosclerotic heart disease" as the center, thereby obtaining a knowledge graph with the knowledge entity "coronary atherosclerotic heart disease" as the center and the attribute information and sources of the knowledge entity as nodes.

[0136] After constructing the knowledge base, the fifth agent continues to maintain the knowledge base, such as updating and supplementing the knowledge base regularly or in real time, so that the knowledge of the knowledge base becomes increasingly rich, thereby providing an increasingly solid knowledge foundation for data reasoning.

[0137] Based on the above-mentioned knowledge base and the above-mentioned multi-agent reasoning system architecture, the second agent can also generate a retrieval question in the process of generating a reasoning path according to the reasoning problem and the reasoning key information. The retrieval question is used to retrieve the knowledge required to generate the reasoning path and / or execute the reasoning path from the knowledge base.

[0138] Specifically, in the process of parsing the key reasoning information to determine the reasoning path based on the reasoning problem, the second agent further parses the knowledge required to generate the reasoning path or execute the reasoning path, and generates a retrieval question for retrieving the knowledge from the knowledge base.

[0139] In some embodiments, see Figure 3 As shown, the second agent parses the key information of reasoning based on the reasoning question to determine the reasoning path. At the same time, if the execution of the reasoning path requires reference to additional knowledge, the second agent generates a retrieval question for retrieving the knowledge while generating the reasoning path, and sends the generated reasoning path to the third agent, and sends the generated retrieval question to the fourth agent.

[0140] After receiving the search question, the fourth agent retrieves corresponding knowledge from the knowledge base constructed by the fifth agent according to the search question, and obtains the search results corresponding to the search question.

[0141] The above-mentioned search results include relevant knowledge corresponding to the search question. In other embodiments, after the fourth agent retrieves the corresponding knowledge from the knowledge base constructed by the fifth agent according to the received search question, the retrieved knowledge data is sorted, summarized and refined as the search result corresponding to the search question.

[0142] In another embodiment, after the fourth agent retrieves the relevant knowledge corresponding to the search question from the knowledge base according to the search question, it also evaluates the retrieved relevant knowledge, judges the correctness of the relevant knowledge, and obtains evaluation information of the relevant knowledge. Exemplarily, the above evaluation information can be, for example, "correct", "wrong", "ambiguous", etc. Moreover, the fourth agent sends the evaluation information as part of the search result together with the relevant knowledge in the search result to the third agent.

[0143] After receiving the reasoning question and reasoning path sent by the second agent, and the retrieval result sent by the fourth agent, the third agent performs integrated reasoning based on the reasoning question, the retrieval result and the reasoning path to generate a reasoning result corresponding to the reasoning question.

[0144] In the integrated reasoning process of the third agent, relevant knowledge is further combined to perform integrated reasoning on the reasoning path, and the evaluation information of relevant knowledge can also be referenced to determine the influence weight of relevant knowledge on the reasoning process, so that the third agent can perform more accurate and comprehensive integrated reasoning, and improve the accuracy of reasoning results and reasoning efficiency.

[0145] In other embodiments, the second agent may need to refer to some relevant knowledge to continue to determine the reasoning path in the process of parsing the reasoning key information based on the reasoning question to determine the reasoning path. In this case, the second agent determines the knowledge required to generate the reasoning path in the process of generating the reasoning path, and generates a retrieval question for retrieving the knowledge from the knowledge base.

[0146] After the second agent generates a retrieval question, it sends the retrieval question to the fourth agent. The fourth agent retrieves relevant knowledge corresponding to the retrieval question from the knowledge base constructed by the fifth agent and feeds back the retrieval result to the second agent.

[0147] Then, the second agent continues to determine the reasoning path based on the received retrieval results, combined with the reasoning problem, the previous reasoning conclusion, and the description of the reasoning path, and obtains the final reasoning path.

[0148] The above-mentioned search results include relevant knowledge corresponding to the search question sent by the second agent, and may also include evaluation information of the relevant knowledge.

[0149] The above-mentioned prior reasoning conclusion is specifically the reasoning conclusion that the second agent has inferred when generating the retrieval question. The reasoning conclusion may be an incomplete reasoning path. The second agent may continue to reason on the incomplete reasoning path in combination with the retrieval until the final complete reasoning path is obtained.

[0150] The description of the reasoning path is an explanation or description of the process of generating a reasoning path by reasoning about the key information based on the retrieval question. The second agent generates a description of the reasoning path based on its understanding of the reasoning question and the key information.

[0151] After receiving the retrieval results, the second agent continues to reason based on the retrieval results to generate a subsequent reasoning path. In the process of generating the subsequent reasoning path, if the second agent again determines that additional relevant knowledge is needed to derive the reasoning path, the second agent can also generate a retrieval question again and send the retrieval question to the fourth agent. After the fourth agent retrieves the relevant knowledge, it sends the retrieval results to the second agent. The second agent continues to reason based on the received retrieval results until the final reasoning path is generated.

[0152] For example, in the medical diagnosis reasoning scenario, for the patient portrait information provided by the first agent, "Main complaint: acid reflux and heartburn for half a month. Current medical history: The patient has had acid reflux and heartburn for nearly half a month, which is more obvious after meals, accompanied by pain behind the sternum, no vomiting, no nausea, and poor appetite. He came to see a doctor today. Past medical history: The patient was previously healthy. Physical examination: clear mind, clear breath sounds in both lungs, tenderness under the xiphoid process (+), no rebound pain, no muscle guarding, mobile dullness (-), and normal bowel sounds. Tests and examinations: 2020-03-08 Gastroscopy: mild damage to the esophageal mucosa, follow-up. Age: 5 7 years old, gender: male", and the diagnosis and treatment reasoning question sent by the first agent "See what disease I have", the second agent infers the patient portrait information based on the diagnosis and treatment reasoning question to generate the reasoning path. When it infers to the intermediate conclusion "First, according to the patient's chief complaint and current medical history, the patient has acid reflux and heartburn, especially after meals. And there is pain behind the sternum. This suggests that we should consider upper gastrointestinal diseases.", the second agent believes that it is necessary to further determine the upper gastrointestinal diseases that cause acid reflux and heartburn before continuing to reason. At this time, the second agent generates a retrieval question "Upper gastrointestinal diseases that can cause acid reflux and heartburn", and sends the retrieval question to the fourth agent. Based on the retrieval question, the fourth agent retrieves relevant knowledge from the medical knowledge base constructed by the fifth agent, "Acid reflux and heartburn may be symptoms of diseases such as gastroesophageal reflux disease, chronic gastritis, and gastric ulcers", and feeds the relevant knowledge back to the second agent. The second agent continues to reason based on this knowledge to obtain a more complete reasoning path "First, according to the patient's chief complaint and current medical history, the patient has acid reflux and heartburn, especially after meals. And there is pain behind the sternum. This suggests that we should consider upper gastrointestinal diseases. Especially the possibility of gastroesophageal reflux disease (GERD) and chronic gastritis. The patient's past medical history does not seem to have any other obvious medical history, and the gastroscopy mentioned in the examination showed mild damage to the esophageal mucosa."

[0153] At this point, the second agent believes that it needs to further determine the upper gastrointestinal diseases that can cause mild damage to the esophageal mucosa before it can continue reasoning. The second agent generates a retrieval question "Upper gastrointestinal diseases that can cause mild damage to the esophageal mucosa" and sends the retrieval question to the fourth agent. Based on the retrieval question, the fourth agent retrieves relevant knowledge from the medical knowledge base constructed by the fifth agent: "Esophageal mucosal damage may be caused by diseases such as reflux esophagitis and esophageal ulcers", and feeds the relevant knowledge back to the second agent. The second agent continues to reason based on this knowledge and obtains a more complete reasoning path: "First, according to the patient's chief complaint and current medical history, the patient has acid reflux and heartburn, which are especially obvious after meals. And there is pain behind the sternum. This This suggests that we should consider upper gastrointestinal diseases. Especially the possibility of gastroesophageal reflux disease (GERD) and chronic gastritis. The patient's past medical history does not seem to have any other obvious medical history, so we can temporarily rule out chronic complications or recurrences related to previous diseases. Physical examination revealed tenderness under the xiphoid process, which may indicate gastric inflammation, esophageal or gastric problems. There was no rebound tenderness or muscle guarding, so acute abdomen related to membrane irritation signs can be ruled out first. The gastroscopy mentioned in the test showed mild damage to the esophageal mucosa, which suggests possible gastroesophageal reflux leading to esophagitis. The damage to the esophageal mucosa is consistent with the patient's symptoms of heartburn."

[0154] In another embodiment, see Figure 4 As shown, when the fourth agent retrieves relevant knowledge from the knowledge base based on the retrieval question, it can further determine whether the retrieval question needs to be split.

[0155] For example, when receiving a search question, the fourth agent determines whether the search question needs to be split into multiple search sub-questions based on the complexity of the search question, and then searches for corresponding knowledge based on the search sub-questions. For example, if the search question is long or there is a lot of information to be retrieved, the fourth agent determines that the search question needs to be split; if the search question is short or only one piece of information needs to be retrieved, the fourth agent determines that the search question does not need to be split.

[0156] Alternatively, after receiving the search question, the fourth agent first conducts a trial search based on the search question, retrieves relevant knowledge from the knowledge base, and then evaluates the retrieved relevant knowledge to determine evaluation information. Based on the evaluation information, the fourth agent can determine whether it is necessary to split the search question. If the evaluation information indicates that the relevant knowledge obtained by the trial search is wrong or ambiguous, the fourth agent determines that it is necessary to split the search question so that more correct relevant knowledge can be retrieved from the knowledge base through the search sub-questions obtained by the split. If the evaluation information indicates that the relevant knowledge obtained by the trial search is correct, the fourth agent can determine that it is not necessary to split the search question.

[0157] When it is determined that the retrieval problem needs to be split, the fourth agent sends a split request to the second agent, so as to call the second agent to split the retrieval problem into multiple retrieval sub-problems.

[0158] After receiving the split request, the second agent splits the search question into sub-questions to obtain multiple sub-questions, and then sends the split sub-questions to the fourth agent. Exemplarily, the second agent can parse the search intent in the search question through semantic recognition and split each search intent into a search sub-question.

[0159] The fourth agent retrieves the retrieval results corresponding to each sub-question from the knowledge base according to the received multiple sub-questions, and feeds them back to the second agent.

[0160] Furthermore, the fourth agent can also integrate and summarize the search results corresponding to each sub-question, obtain a comprehensive search result and feed it back to the second agent.

[0161] For example, in a medical diagnosis reasoning scenario, the retrieval question that the second agent sends to the fourth agent is "After pacemaker implantation: 1. Segmental movement disorder of the left ventricular wall myocardium 2. Enlarged left heart, decreased left ventricular systolic function 3. Clinical significance of aortic valve regurgitation (mild) and mitral valve regurgitation (mild to moderate)". After receiving the retrieval question, the fourth agent determines that it needs to be split, and sends a split request to the second agent. After receiving the split request, the second agent splits the above retrieval question into three retrieval sub-questions, namely: "Clinical significance of segmental movement disorder of the left ventricular wall myocardium", "Clinical significance of enlarged left heart, decreased left ventricular systolic function", and "Clinical significance of aortic valve regurgitation (mild) and mitral valve regurgitation (mild to moderate)". Then, the second intelligent agent sends the above three retrieval sub-questions to the fourth intelligent agent. The fourth intelligent agent retrieves the relevant knowledge corresponding to each retrieval sub-question from the knowledge base. On this basis, the fourth intelligent agent integrates the relevant knowledge retrieved for each retrieval sub-question to obtain the final retrieval result "After pacemaker implantation: 1. Segmental movement disorder of the left ventricular wall myocardium 2. Enlarged left heart and decreased left ventricular systolic function 3. Aortic valve regurgitation (mild), the clinical significance of mitral valve regurgitation (mild to moderate) is reduced cardiac function and valvular regurgitation".

[0162] Splitting the retrieval problem into sub-problems for retrieval can improve the coverage and precision of knowledge retrieval, and thus improve the accuracy of data reasoning.

[0163] Below, in combination with the reasoning process in the medical diagnosis and treatment reasoning scenario, the complete processing flow of the data reasoning method performed by the multi-agent reasoning system proposed in the embodiment of the present application is introduced.

[0164] See also Figure 5 As shown, the fifth agent pre-builds a medical knowledge base containing knowledge of various diseases. When the first agent receives the diagnosis and treatment reasoning problem and the patient's medical documents, it extracts key information from the patient's medical documents according to the diagnosis and treatment reasoning problem to obtain the patient's portrait information, wherein the patient's basic portrait information and the patient's time-series portrait information are protected, and then the first agent sends the diagnosis and treatment reasoning problem and the patient's portrait information to the second agent.

[0165] The second agent generates a reasoning path based on the diagnosis and treatment reasoning problem and the patient portrait information, and determines whether it is necessary to retrieve reference knowledge. If it is necessary to retrieve reference knowledge, a retrieval question is generated and sent to the fourth agent.

[0166] After receiving the search question, the fourth agent retrieves knowledge from the knowledge base constructed by the fifth agent, and determines whether the search question needs to be split. If differentiation is required, a splitting instruction is generated and sent to the second agent.

[0167] After receiving the splitting instruction, the second agent splits the search problem into multiple search sub-problems, and then sends each of the split search sub-problems to the fourth agent.

[0168] The fourth agent retrieves relevant knowledge from the knowledge base based on each retrieval sub-question, and generates a retrieval result based on the retrieved relevant knowledge. The retrieval result may also include evaluation information of the fourth agent on the retrieved relevant knowledge.

[0169] The reasoning path generated by the second agent and the retrieval results generated by the fourth agent are sent together with the diagnosis and treatment reasoning questions to the third agent, which performs integrated reasoning to obtain reasoning results and evaluates the reasoning results to obtain evaluation conclusions.

[0170] If the evaluation conclusion indicates that the reasoning result is correct, the third agent outputs the reasoning result.

[0171] If the evaluation conclusion indicates that the reasoning result is wrong, the third agent sends a re-reasoning instruction to the second agent, so that the second agent regenerates the reasoning path according to the diagnosis and treatment reasoning problem and the patient portrait information. Each agent re-executes the above reasoning process until the third agent believes that the reasoning result obtained is correct and outputs the reasoning result.

[0172] The above embodiment only introduces the processing flow of the data reasoning method proposed in the embodiment of the present application in the medical diagnosis and treatment scenario from a global perspective, so that those skilled in the art can more comprehensively understand the complete processing flow of the data reasoning method proposed in the embodiment of the present application. For the specific processing content of each step in the process, please refer to the specific processing content of the corresponding processing steps in the above embodiment, which will not be repeated here.

[0173] In the above embodiments, the data reasoning method proposed in the present application is exemplarily introduced by introducing the work content and collaborative processing process of each agent in the multi-agent reasoning system during the data reasoning process. At the same time, based on the introduction of the above embodiments, technical personnel in this field can know the composition structure of the multi-agent reasoning system proposed in the present application and the specific functions of each part of the structure.

[0174] For example, an embodiment of the present application proposes a multi-agent reasoning system, which includes multiple agents, and the multiple agents in the multi-agent reasoning system are configured to implement the data reasoning method introduced in any of the above embodiments.

[0175] In some embodiments, the multi-agent reasoning system includes a first agent, a second agent, and a third agent.

[0176] The first agent is used to extract key information from the reasoning reference data according to the reasoning problem to obtain the reasoning key information;

[0177] The second agent is used to generate a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem;

[0178] The third agent is used to perform integrated reasoning according to the reasoning problem and the reasoning path, and generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0179] In another embodiment, the third agent is further used to evaluate the reasoning result to obtain an evaluation conclusion;

[0180] When the evaluation conclusion indicates that the reasoning result is wrong, the third agent calls the second agent to regenerate the reasoning path, and the third agent performs integrated reasoning based on the reasoning problem and the reasoning path regenerated by the second agent to regenerate the reasoning result corresponding to the reasoning problem.

[0181] In another embodiment, the multi-agent reasoning system further includes a fourth agent and a fifth agent;

[0182] In the process of generating the reasoning path according to the reasoning problem and the reasoning key information, the second agent also generates a retrieval problem, wherein the retrieval problem is used to retrieve the knowledge required to execute the reasoning path from the knowledge base;

[0183] The fourth agent is used to retrieve the search result corresponding to the search question from the knowledge base according to the search question; the search result includes relevant knowledge corresponding to the search question, and the knowledge base is constructed by the fifth agent;

[0184] The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, including:

[0185] The third agent performs integrated reasoning according to the reasoning problem, the retrieval results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

[0186] In another embodiment, the fifth agent constructs the knowledge base by performing the following processing:

[0187] Perform knowledge segmentation on the knowledge source file to obtain knowledge labels;

[0188] Generating a knowledge triple based on the knowledge tag, wherein the knowledge triple includes a knowledge entity, attribute information of the knowledge entity, and the knowledge entity and a source of the attribute information of the knowledge entity;

[0189] Each knowledge triple is integrated with knowledge entities as the center to obtain the knowledge graph.

[0190] In another embodiment, the attribute information of the knowledge entity includes the knowledge entity and the encoding of the chapter content where the attribute information of the knowledge entity is located in the source file of the knowledge entity and the source file of the knowledge entity.

[0191] In another embodiment, the search result also includes evaluation information of the relevant knowledge.

[0192] In another embodiment, the fourth agent is further used to determine whether the retrieval question needs to be split;

[0193] In the case where it is determined that the retrieval problem needs to be split, the fourth agent calls the second agent to split the retrieval problem into a plurality of retrieval sub-problems;

[0194] Furthermore, the fourth agent retrieves from the knowledge base according to the multiple retrieval sub-questions and obtains retrieval results corresponding to each retrieval sub-question.

[0195] In another embodiment, the reasoning problem comprises a diagnosis and treatment reasoning problem, and the reasoning reference data comprises a patient's medical documents.

[0196] In another embodiment, the patient's medical documents include a plurality of medical documents of the patient at different periods;

[0197] The first agent extracts key information from the reasoning reference data according to the reasoning problem to obtain key reasoning information, including:

[0198] The first intelligent agent extracts patient portrait information from the multiple medical documents based on the reasoning problem, and generates a basic information portrait and a time-series medical portrait corresponding to the patient.

[0199] For the specific processing process of each intelligent agent in the above embodiments, as well as the implementation form and training method of each intelligent agent, please refer to the above-mentioned embodiment introduction of the data reasoning method, which will not be repeated here.

[0200] Corresponding to the above-mentioned data reasoning method, another embodiment of the present application further proposes a data reasoning device, which is applied to a multi-agent reasoning system including multiple agents, and the device includes:

[0201] An information integration unit, configured to extract key information from the reasoning reference data according to the reasoning problem through the first agent, so as to obtain the reasoning key information;

[0202] An analysis and reasoning unit, configured to generate a reasoning path through a second agent according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning and answering the reasoning problem based on the reasoning key information;

[0203] The integrated reasoning unit is used to perform integrated reasoning according to the reasoning problem and the reasoning path through a third agent to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0204] In some embodiments, the integrated reasoning unit is further configured to:

[0205] The third agent evaluates the reasoning result to obtain an evaluation conclusion;

[0206] When the evaluation conclusion indicates that the reasoning result is wrong, the third agent calls the analysis reasoning unit to regenerate the reasoning path through the second agent, and the integration reasoning unit performs integrated reasoning through the third agent based on the reasoning problem and the reasoning path regenerated by the second agent to regenerate the reasoning result corresponding to the reasoning problem.

[0207] In some embodiments, the analysis and reasoning unit generates a reasoning path through the second agent according to the reasoning problem and the reasoning key information, and further generates a retrieval problem, wherein the retrieval problem is used to retrieve the knowledge required to execute the reasoning path from the knowledge base;

[0208] The device also includes:

[0209] a knowledge retrieval unit, configured to retrieve, through a fourth agent, a retrieval result corresponding to the retrieval question from a knowledge base according to the retrieval question; the retrieval result includes relevant knowledge corresponding to the retrieval question, and the knowledge base is constructed by a fifth agent;

[0210] The integrated reasoning unit performs integrated reasoning according to the reasoning problem and the reasoning path through the third agent to generate a reasoning result corresponding to the reasoning problem, including:

[0211] The third agent performs integrated reasoning according to the reasoning problem, the retrieval results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

[0212] In some embodiments, the fifth agent constructs the knowledge base by performing the following processing:

[0213] Perform knowledge segmentation on the knowledge source file to obtain knowledge labels;

[0214] Generating a knowledge triple based on the knowledge tag, wherein the knowledge triple includes a knowledge entity, attribute information of the knowledge entity, and the knowledge entity and a source of the attribute information of the knowledge entity;

[0215] Each knowledge triple is integrated with knowledge entities as the center to obtain the knowledge graph.

[0216] In some embodiments, the attribute information of the knowledge entity includes the knowledge entity and an encoding of the chapter content where the attribute information of the knowledge entity is located in the source file of the knowledge entity and the source file of the knowledge entity.

[0217] In some embodiments, the search results also include evaluation information on the relevant knowledge.

[0218] In some embodiments, the fourth agent further determines whether the retrieval question needs to be split;

[0219] When it is determined that the retrieval problem needs to be split, the knowledge retrieval unit calls the second agent through the fourth agent to split the retrieval problem into multiple retrieval sub-problems;

[0220] Furthermore, the knowledge retrieval unit retrieves the retrieval results corresponding to each retrieval sub-question from the knowledge base through the fourth agent according to the multiple retrieval sub-questions.

[0221] In some embodiments, the reasoning problem comprises a diagnosis and treatment reasoning problem, and the reasoning reference data comprises a medical document of a patient.

[0222] In some embodiments, the patient's medical documents include multiple medical documents of the patient at different periods;

[0223] The information integration unit extracts key information from the reasoning reference data according to the reasoning problem through the first agent to obtain the reasoning key information, including:

[0224] The first intelligent agent extracts patient portrait information from the multiple medical documents based on the reasoning problem, and generates a basic information portrait and a time-series medical portrait corresponding to the patient.

[0225] The data reasoning device provided in this embodiment belongs to the same application concept as the data reasoning method provided in the above embodiments of this application, and can execute the data reasoning method provided in any of the above embodiments of this application, and has the corresponding functional units and beneficial effects of the execution method. For technical details not fully described in this embodiment, please refer to the specific processing content of the data reasoning method provided in the above embodiments of this application, and will not be repeated here.

[0226] Another embodiment of the present application further provides a data reasoning device, which includes an input device, an output device, and a multi-agent reasoning system;

[0227] The input device is connected to the processor and is used to input reasoning questions and reasoning reference data to the processor;

[0228] The output device is connected to the processor and is used to output the inference result generated by the processor;

[0229] The multi-agent reasoning system is configured to execute the data reasoning method described in any of the above embodiments.

[0230] The above-mentioned input device may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer or a gravity sensor.

[0231] The output devices mentioned above may include devices that allow information to be output to a user, such as a display screen, a printer, a speaker, etc.

[0232] The specific structure and functions of the above-mentioned multi-agent reasoning system, as well as its specific processing process when executing the data reasoning method described in any of the above-mentioned embodiments, can be found in the corresponding introduction of the above-mentioned method embodiments.

[0233] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, and the computer program instructions may include multiple segments or groups of computer program instructions. When these computer program instructions are executed separately by multiple agents in a multi-agent reasoning system, the multi-agent reasoning system executes the data reasoning method described in any of the above-mentioned embodiments.

[0234] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0235] In addition, an embodiment of the present application may also be a storage medium on which a computer program is stored. The computer program may include multiple segments or multiple groups of computer programs. When these computer programs are executed separately by multiple agents in a multi-agent reasoning system, the multi-agent reasoning system executes the data reasoning method described in any of the above embodiments.

[0236] Another embodiment of the present application also proposes another data reasoning method. The data reasoning process of this method is the same as the processing process of the data reasoning method described in the above embodiment. The difference is that the data reasoning method proposed in the embodiment of the present application can be executed by a single intelligent entity or on a single device, such as a processor, computer, server, smart terminal and other devices or equipment.

[0237] Exemplarily, the above-mentioned single agent or single device for implementing the data reasoning method proposed in the embodiment of the present application may be an agent obtained by integrating multiple agents in the above-mentioned multi-agent reasoning system, or a device that runs the above-mentioned multiple agents at the same time, such as a server, workstation, etc. Alternatively, multiple data reasoning models may be deployed on a single agent or device, each model corresponding to each agent in the above-mentioned multi-agent reasoning system, and respectively used to implement the functions of each agent in the multi-agent reasoning system and execute the processing of each agent.

[0238] See also Figure 6 As shown, the data reasoning method proposed in the embodiment of the present application includes:

[0239] S201. Extract key information from the reasoning reference data according to the reasoning problem to obtain reasoning key information.

[0240] S202: Generate a reasoning path according to the reasoning problem and the reasoning key information, wherein the reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem.

[0241] S203: Perform integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0242] In another embodiment, the method further comprises:

[0243] Evaluating the reasoning result to obtain an evaluation conclusion;

[0244] In the case where the evaluation conclusion indicates that the reasoning result is wrong, regenerating the reasoning path according to the reasoning problem and the reasoning key information;

[0245] Integrated reasoning is performed according to the reasoning problem and the regenerated reasoning path to regenerate the reasoning result corresponding to the reasoning problem.

[0246] In another embodiment, the method further comprises:

[0247] Generate a retrieval question according to the reasoning question and the reasoning key information, wherein the retrieval question is used to retrieve knowledge required to execute the reasoning path from a knowledge base;

[0248] Retrieving from a knowledge base according to the retrieval question to obtain a retrieval result corresponding to the retrieval question; the retrieval result includes relevant knowledge corresponding to the retrieval question;

[0249] The performing integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem includes:

[0250] Integrated reasoning is performed according to the reasoning problem, the search results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

[0251] In another embodiment, the search result also includes evaluation information of the relevant knowledge.

[0252] In another embodiment, the method further comprises:

[0253] Determine whether the search question needs to be split;

[0254] When it is determined that the search problem needs to be split, split the search problem into multiple search sub-problems;

[0255] According to the multiple search sub-questions, search results corresponding to each search sub-question are obtained from the knowledge base.

[0256] In another embodiment, the reasoning problem comprises a diagnosis and treatment reasoning problem, and the reasoning reference data comprises a patient's medical documents.

[0257] In another embodiment, the patient's medical documents include a plurality of medical documents of the patient at different periods;

[0258] The key information of the reasoning reference data is extracted according to the reasoning problem to obtain the key information of the reasoning, including:

[0259] According to the reasoning problem, patient portrait information is extracted from the multiple medical documents to generate a basic information portrait and a time-series medical portrait corresponding to the patient.

[0260] The data reasoning method described in the above embodiment and the data reasoning method applied to the multi-agent reasoning system described above belong to the same application concept, and the specific processing process of each processing step can refer to the specific processing process of the corresponding processing step in the data reasoning method applied to the multi-agent reasoning system.

[0261] By using a single agent or device to execute the above data reasoning method, on the one hand, the flexibility and efficiency of data reasoning can be improved. Users can provide any reasoning questions and reasoning reference data, so that the agent or device can execute the above data reasoning method, realize the data reasoning process, and obtain the reasoning result corresponding to the reasoning question. The entire reasoning process does not need to specify reasoning rules, and its data reasoning is more flexible and can meet various data reasoning needs. The reasoning process can also be included in the reasoning result, so that the data reasoning conclusion has stronger interpretability, which can enable users to understand the origin of the reasoning conclusion more intuitively and efficiently, and improve the user's information acquisition efficiency.

[0262] On the other hand, the method can be executed by a single agent or device, which reduces the implementation cost of the method.

[0263] Corresponding to the above-mentioned data reasoning method, the present application embodiment also proposes a data reasoning device, see Figure 7 As shown, the device comprises:

[0264] The information integration unit 100 is used to extract key information from the reasoning reference data according to the reasoning problem to obtain the reasoning key information;

[0265] An analysis and reasoning unit 110 is used to generate a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning and answering the reasoning problem based on the reasoning key information;

[0266] The integrated reasoning unit 120 is used to perform integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

[0267] In some implementations, the integrated reasoning unit 120 is further configured to evaluate the reasoning result to obtain an evaluation conclusion.

[0268] In the case where the evaluation conclusion indicates that the reasoning result is wrong, the integrated reasoning unit 120 calls the analytical reasoning unit 110 to regenerate the reasoning path according to the reasoning problem and the reasoning key information;

[0269] Furthermore, the integrated reasoning unit 120 performs integrated reasoning according to the reasoning problem and the regenerated reasoning path, and regenerates the reasoning result corresponding to the reasoning problem.

[0270] In some implementations, the analysis and reasoning unit 110 is further used to generate a retrieval question based on the reasoning question and the reasoning key information, and the retrieval question is used to retrieve the knowledge required to execute the reasoning path from the knowledge base;

[0271] The device also includes a knowledge retrieval unit, which is used to retrieve a retrieval result corresponding to the retrieval question from a knowledge base according to the retrieval question; the retrieval result includes relevant knowledge corresponding to the retrieval question;

[0272] The integrated reasoning unit 120 performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, including:

[0273] Integrated reasoning is performed according to the reasoning problem, the search results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

[0274] In some implementations, the search results also include evaluation information on the relevant knowledge.

[0275] In some implementations, the knowledge retrieval unit is further used to determine whether the retrieval question needs to be split;

[0276] In the case where it is determined that the retrieval question needs to be split, the knowledge retrieval unit calls the analysis and reasoning unit 110 to split the retrieval question into multiple retrieval sub-questions;

[0277] The knowledge retrieval unit retrieves from the knowledge base according to the multiple retrieval sub-questions to obtain retrieval results corresponding to each retrieval sub-question.

[0278] In some implementations, the reasoning problem includes a diagnosis and treatment reasoning problem, and the reasoning reference data includes a patient's medical documents.

[0279] In some implementations, the patient's medical documents include multiple medical documents of the patient at different periods;

[0280] The information integration unit 100 extracts key information from the reasoning reference data according to the reasoning problem to obtain the reasoning key information, including:

[0281] According to the reasoning problem, patient portrait information is extracted from the multiple medical documents to generate a basic information portrait and a time-series medical portrait corresponding to the patient.

[0282] The data reasoning device provided in this embodiment belongs to the same application concept as the data reasoning method provided in the above embodiments of this application, and can execute the data reasoning method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not fully described in this embodiment, please refer to the specific processing content of the data reasoning method provided in the above embodiments of this application, and will not be repeated here.

[0283] The functions implemented by the above-mentioned units may be implemented by the same or different processors respectively, which is not limited in the embodiments of the present application.

[0284] It should be understood that the units in the above devices can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the units in the device can be implemented in the form of hardware circuits, and the functions of some or all units can be realized by designing the hardware circuits. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above units are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by the configuration file, so as to realize the functions of some or all of the above units. All units of the above devices can be implemented in the form of a processor calling software, or in the form of hardware circuits, or in part by a processor calling software, and the remaining part is implemented in the form of hardware circuits.

[0285] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0286] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0287] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0288] Another embodiment of the present application also provides a data inference device, see Figure 8 As shown, the device includes:

[0289] Memory 200 and processor 210;

[0290] The memory 200 is connected to the processor 210 and is used to store programs;

[0291] The processor 210 is used to implement the data reasoning method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0292] Specifically, the above-mentioned data inference device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0293] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are connected to each other via a bus.

[0294] A bus may include a pathway that transfers information between components of a computer system.

[0295] The processor 210 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0296] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0297] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 200 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0298] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0299] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0300] The communication interface 220 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0301] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any data inference method provided in the above embodiments of the present application.

[0302] An embodiment of the present application also proposes a chip, which includes a processor and a data interface. The processor reads and runs a program stored in a memory through the data interface to execute the data reasoning method introduced in any of the above embodiments. The specific processing process and its beneficial effects can be found in the introduction to the embodiment of the above data reasoning method.

[0303] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps in the data reasoning method described in any of the above-mentioned embodiments of this specification.

[0304] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0305] In addition, an embodiment of the present application may also be a storage medium on which a computer program is stored, and the computer program is executed by a processor to execute the steps of the data reasoning method described in any of the above embodiments of this specification.

[0306] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0307] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0308] The steps in the methods of each embodiment of the present application can be adjusted in order, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0309] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0310] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0311] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0312] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0313] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may 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.

[0314] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0315] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0316] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data reasoning method, characterized in that: Applied to a multi-agent reasoning system including a plurality of agents, the method comprises: The first agent extracts key information from the reasoning reference data according to the reasoning problem to obtain the key information of reasoning; The second agent generates a reasoning path according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning based on the reasoning key information to answer the reasoning problem; The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

2. The method according to claim 1, characterized in that The method further comprises: The third agent evaluates the reasoning result to obtain an evaluation conclusion; When the evaluation conclusion indicates that the reasoning result is wrong, the third agent calls the second agent to regenerate the reasoning path, and the third agent performs integrated reasoning based on the reasoning problem and the reasoning path regenerated by the second agent to regenerate the reasoning result corresponding to the reasoning problem.

3. The method according to claim 1, characterized in that In the process of generating the reasoning path according to the reasoning problem and the reasoning key information, the second agent also generates a retrieval problem, wherein the retrieval problem is used to retrieve the knowledge required to execute the reasoning path from the knowledge base; The method further comprises: The fourth agent retrieves a search result corresponding to the search question from a knowledge base according to the search question; the search result includes relevant knowledge corresponding to the search question, and the knowledge base is constructed by the fifth agent; The third agent performs integrated reasoning according to the reasoning problem and the reasoning path to generate a reasoning result corresponding to the reasoning problem, including: The third agent performs integrated reasoning according to the reasoning problem, the retrieval results and the reasoning path to generate a reasoning result corresponding to the reasoning problem.

4. The method according to claim 3, characterized in that The fifth agent constructs the knowledge base by executing the following processing: Perform knowledge segmentation on the knowledge source file to obtain knowledge labels; Generating a knowledge triple based on the knowledge tag, wherein the knowledge triple includes a knowledge entity, attribute information of the knowledge entity, and the knowledge entity and a source of the attribute information of the knowledge entity; Each knowledge triple is integrated with knowledge entities as the center to obtain the knowledge graph.

5. The method according to claim 4, characterized in that The attribute information of the knowledge entity includes the knowledge entity and the encoding of the chapter content where the attribute information of the knowledge entity is located in the source file of the knowledge entity and the source file of the knowledge entity.

6. The method according to claim 3, characterized in that The search result also includes evaluation information on the relevant knowledge.

7. The method according to claim 3, characterized in that The fourth agent also determines whether the retrieval question needs to be split; In the case where it is determined that the retrieval problem needs to be split, the fourth agent calls the second agent to split the retrieval problem into a plurality of retrieval sub-problems; Furthermore, the fourth agent retrieves from the knowledge base according to the multiple retrieval sub-questions and obtains retrieval results corresponding to each retrieval sub-question.

8. The method according to any one of claims 1 to 7, characterized in that The reasoning problem includes a diagnosis and treatment reasoning problem, and the reasoning reference data includes a patient's medical documents.

9. The method according to claim 8, characterized in that The medical documents of the patient include multiple medical documents of the patient at different periods; The first agent extracts key information from the reasoning reference data according to the reasoning problem to obtain key reasoning information, including: The first intelligent agent extracts patient portrait information from the multiple medical documents based on the reasoning problem, and generates a basic information portrait and a time-series medical portrait corresponding to the patient.

10. A multi-agent reasoning system, characterized in that: The multi-agent system comprises a plurality of agents, wherein the plurality of agents in the multi-agent system are configured to implement the data reasoning method as described in any one of claims 1 to 9.

11. A data inference device, characterized in that: Applied to a multi-agent reasoning system including a plurality of agents, the device comprises: An information integration unit, configured to extract key information from the reasoning reference data according to the reasoning problem through the first agent, so as to obtain the reasoning key information; An analysis and reasoning unit, configured to generate a reasoning path through a second agent according to the reasoning problem and the reasoning key information; the reasoning path includes prompt information for reasoning and answering the reasoning problem based on the reasoning key information; The integrated reasoning unit is used to perform integrated reasoning according to the reasoning problem and the reasoning path through a third agent to generate a reasoning result corresponding to the reasoning problem, wherein the reasoning result includes a reasoning process and / or a reasoning conclusion.

12. A data inference device, characterized in that: include: Input devices, output devices, and multi-agent reasoning systems; The input device is connected to the processor and is used to input reasoning questions and reasoning reference data to the processor; The output device is connected to the processor and is used to output the inference result generated by the processor; The multi-agent reasoning system is configured to execute the data reasoning method as described in any one of claims 1 to 9.

13. A computer program product, characterized in that It comprises computer program instructions which, when executed by a plurality of agents in a multi-agent reasoning system, enable the multi-agent reasoning system to execute the data reasoning method as described in any one of claims 1 to 9.

Images