Consultation task instance generation method and device, electronic equipment and storage medium

Abstract

The application discloses a consultation task instance generation method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining an initial consultation task instance set; randomly extracting M initial consultation task instructions from the initial consultation task instance set, and obtaining Q target consultation task instructions based on the M initial consultation task instructions and a language model; determining the target instruction types of the Q target consultation task instructions respectively; for each target consultation task instruction, generating a target consultation task instance in a target mode matched with the target instruction type of the current target consultation task instruction according to the target instruction type; updating the initial consultation task instance set based on the target consultation task instance, and executing the step of randomly extracting M initial consultation task instructions from the initial consultation task instance set until a preset termination condition is met. The scheme can continuously construct a consultation task instance set with accuracy, completeness and richness, and is helpful to improve the operation management of related industries.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a consultation task instance generation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the rapid development of technology and the increasing demand of the public for service quality, the operation management of various industries becomes crucial, such as the operation management of hospitals, companies and factories, etc. Taking the operation management of hospitals as an example, under the background of promoting the reform of hospital examination and payment system, the degree of specialization and refinement of various businesses of operation management is increasing, and the complexity makes the workload and pressure of hospital management personnel further increase. For professional problems in various dimensions of operation management, such as operation index interpretation, department cost allocation, performance scheme point calculation, medical and technical workload evaluation, etc., accurate, comprehensive and timely answers are urgently needed.

[0003] Currently, the problem can be solved by a consultation business manager or a consultant, but this method is time-consuming and labor-intensive. In addition, a conventional natural language processing technology generated operation management consultation model can be used to answer the user's question through the operation management consultation model. However, the accuracy of solving the consultation task based on the operation management consultation model is usually low and cannot deal with more complex problems; moreover, the training data required to build the operation management consultation model is large and the cost of manual annotation is extremely high, making it difficult to form a complete and continuous model training input. SUMMARY

[0004] The present application provides a consultation task instance generation method, device, electronic equipment and storage medium, which can continuously build a consultation task instance set with accuracy, completeness and richness, and help to improve the operation management of related industries.

[0005] According to an aspect of the present application, a consultation task instance generation method is provided, comprising:

[0006] obtaining an initial consultation task instance set; wherein the initial consultation task instance set comprises at least one group of initial consultation task instances; each group of initial consultation task instances comprises an initial consultation task instruction, input information, output information and an instruction type corresponding to the initial consultation task instruction;

[0007] randomly selecting M initial consultation task instructions from the initial consultation task instance set, and obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model;

[0008] determining the target instruction type of the Q target consultation task instructions respectively; wherein the target instruction type comprises a classification task instruction and a non-classification task instruction;

[0009] For each of the Q pieces of target consultation task instructions, a target consultation task instance corresponding to a current target consultation task instruction is generated according to a target instruction type of the current target consultation task instruction in a target manner matched with the target instruction type;

[0010] The initial consultation task instance set is updated based on the target consultation task instance, and the step of randomly extracting M pieces of initial consultation task instructions from the initial consultation task instance set is executed again until a preset termination condition is met.

[0011] According to another aspect of the present application, a consultation task instance generation device is provided, comprising:

[0012] An initial consultation task instance set acquisition module is configured to acquire an initial consultation task instance set; wherein the initial consultation task instance set comprises at least one group of initial consultation task instances; each group of initial consultation task instances comprises initial consultation task instructions, input information, output information and instruction types corresponding to the initial consultation task instructions;

[0013] A target consultation task instruction acquisition module is configured to randomly extract M pieces of initial consultation task instructions from the initial consultation task instance set, and acquire Q pieces of target consultation task instructions based on the M pieces of initial consultation task instructions and a pre-trained language model;

[0014] A target instruction type determination module is configured to determine target instruction types of the Q pieces of target consultation task instructions respectively; wherein the target instruction types comprise classification task instructions and non-classification task instructions;

[0015] A target consultation task instance generation module is configured to, for each of the Q pieces of target consultation task instructions, generate a target consultation task instance corresponding to a current target consultation task instruction according to a target instruction type of the current target consultation task instruction in a target manner matched with the target instruction type;

[0016] An initial consultation task instance set update module is configured to update the initial consultation task instance set based on the target consultation task instance, and execute the step of randomly extracting M pieces of initial consultation task instructions from the initial consultation task instance set again until a preset termination condition is met.

[0017] According to another aspect of the present application, an electronic device is provided, comprising:

[0018] at least one processor; and

[0019] a memory connected in communication with the at least one processor; wherein

[0020] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the consultation task instance generation method according to any one of the embodiments of the present application.

[0021] According to another aspect of the present application, a computer readable storage medium is provided, which stores computer instructions for enabling a processor to implement the consultation task instance generation method according to any one of the embodiments of the present application when executed by the processor.

[0022] The consultation task instance generation scheme of the embodiments of the present application obtains an initial consultation task instance set; wherein the initial consultation task instance set includes at least one group of initial consultation task instances; each group of initial consultation task instances includes initial consultation task instructions, input information, output information and instruction types corresponding to the initial consultation task instructions; M initial consultation task instructions are randomly extracted from the initial consultation task instance set, and Q target consultation task instructions are obtained based on the M initial consultation task instructions and a pre-trained language model; target instruction types of the Q target consultation task instructions are determined respectively; wherein the target instruction types include classification task instructions and non-classification task instructions; for each target consultation task instruction in the Q target consultation task instructions, a target consultation task instance corresponding to the current target consultation task instruction is generated in a target manner matched with the target instruction type of the current target consultation task instruction according to the target instruction type of the current target consultation task instruction; the initial consultation task instance set is updated based on the target consultation task instance, and the step of randomly extracting M initial consultation task instructions from the initial consultation task instance set is performed again until a preset termination condition is met. Through the technical scheme provided by the embodiments of the present application, the problems of large amount of training data required for constructing an operation and management consulting model, high cost of manual annotation, insufficient training input integrity and continuity can be solved, and consultation task instances with accuracy, integrity and richness can be continuously constructed, which is helpful to improve the operation and management of related industries.

[0023] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0025] Figure 1 is a flow chart of a consulting task instance generation method according to an embodiment of the present application;

[0026] Figure 2 is a structural schematic diagram of a consulting task instance generation device according to an embodiment of the present application;

[0027] Figure 3 is a structural schematic diagram of an electronic device implementing the consulting task instance generation method according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the personnel in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by the person of ordinary skill in the art without creative labor should belong to the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0030] The consulting task instance generation method provided by the embodiments of the present application can be applied to various fields such as hospital operation management, company operation management and factory operation management. It should be noted that the embodiments of the present application do not limit the application scenarios of the consulting task instance generation method. The following will take the application of the consulting task instance generation method in hospital operation management as an example for explanation and description.

[0031] Embodiment one

[0032] Figure 1A flowchart of a consultation task instance generation method is provided for the first embodiment of the present application. The first embodiment can be applied to the generation of a consultation task instance. The method can be executed by a consultation task instance generation device, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device. As shown in FIG. 8, the method comprises the following steps. Figure 1

[0033] In S110, an initial consultation task instance set is obtained. The initial consultation task instance set comprises at least one group of initial consultation task instances. Each group of initial consultation task instances comprises initial consultation task instructions, input information, output information, and instruction types corresponding to the initial consultation task instructions.

[0034] In the embodiment of the present application, an initial consultation task instance set is obtained. The initial consultation task instance set can comprise N groups of initial consultation task instances, where N is an integer greater than or equal to 1. For example, each group of initial consultation task instances can be a task instruction instance with comprehensiveness and diversity, which is generated based on natural language processing technology for hospital operation management consultation tasks. The subject content involved in the initial consultation task instance can include performance evaluation, budget, cost, DRG / DIP, operation resource allocation, human resource management, and other related content. The natural language processing technology can include keyword extraction, text summarization, entity recognition, question answering, text generation, text correction, natural language reasoning, and other technologies.

[0035] Each group of initial consultation task instances comprises initial consultation task instructions, input information, output information, and instruction types corresponding to the initial consultation task instructions. It can be understood that each group of initial consultation task instances is a four-tuple, which can be represented as (P i , I i , O i , C i ), where P i represents the initial consultation task instruction, I i represents the input information (which can be empty), O i represents the output information, and C i represents the instruction type corresponding to the initial consultation task instruction P i . C i = 1 indicates that P i is a classification task instruction; and C i = 0 indicates that P i is a non-classification task instruction. Therefore, the initial consultation task instance set S0 can be represented as S0 = {(P1, I1, O1, C1), (P2, I2, O2, C2),..., (P N , I N , O N ​, C N}。

[0036] For example, a set of possible initial consultation task instances (P i , I i , O i , C i ) can be: P i = "What is the formula for calculating the average length of stay after CMI correction?"; I i = " "; O i = "The formula for calculating the average length of stay after CMI correction is: (total bed days occupied by discharged patients) / (number of discharged patients * CMI) * 100%. After correcting the average length of stay using CMI, the average bed days occupied by discharged patients per unit CMI can be obtained, which can better reflect the complexity of the patients admitted and the utilization of beds."; C i = 0.

[0037] For another example, another set of possible initial consultation task instances (P i , I i , O i , C i ) can be: P i = "Summarize the content of the paragraph in one sentence."; I i = "Fine management is the only way for public hospital management modernization, and fine human resource management is the foundation of hospital fine management. Beijing Tsinghua Changgeng Hospital has built a fine human resource management system based on the job and level system, integrating development system, evaluation system and compensation system. This system not only better reflects the particularity of different types and levels of positions in the hospital, but also balances the relationship between different types of positions, and explicitly establishes promotion relationships for different levels of positions, striving to achieve the best use of human resources. This system can provide certain reference for public hospital personnel system reform."; O i = "The main content of this paragraph is: Beijing Tsinghua Changgeng Hospital has built a fine human resource management system that embodies fairness, balance and practicality, which can provide reference for public hospital personnel system reform."; C i = 0.

[0038] For another example, another set of possible initial consultation task instances (P i , I i , O i , C i ) can be: P i = "Can the cost be allocated proportionally according to the number of medical staff?"; I i = " "; O i= "The cost is not recommended to be allocated according to the number of people. From the perspective of intuitive income, doctors are much higher than nurses in creating benefits, but the development of a department cannot rely only on doctors, and the existence of nursing is inevitable and indispensable. Moreover, the actual use of cost control is up to everyone. If it is divided equally, saving 1 yuan is equivalent to increasing everyone's performance by 0.5 yuan. Doctors are few, and nurses are many. After actual use and cost control, the performance of doctors is still relatively favorable." i = 1.

[0039] In the embodiment of the application, the initial consultation task instance set S0 can contain 10 groups of initial consultation task instances, and |S0| = 10, wherein 8 groups of initial consultation task instances are of the instruction type of non-classification tasks, and 2 groups of initial consultation task instances are of the instruction type of classification tasks. For example, Table 1 is an initial consultation task instance table provided by the embodiment of the application:

[0040] Table 1: Initial consultation task instance table

[0041]

[0042]

[0043] S120, randomly extracting M initial consultation task instructions from the initial consultation task instance set, and obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model.

[0044] For example, M initial consultation task instructions can be randomly extracted from the initial consultation task instance set, and the M initial consultation task instructions are input into the pre-trained language model, and Q target consultation task instructions are obtained according to the output result of the language model. The language model can include GPT, LLaMA, BLOOM, Vicuna, ChatGLM, BELLE, etc. or a fine-tuned model of the above model fine-tuned in the hospital operation management field. Q can be any integer greater than or equal to 1.

[0045] Optionally, obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model comprises: generating a first template instruction based on the M initial consultation task instructions; wherein the first template instruction is an instruction for generating Q new instructions; inputting the first template instruction into the language model, and obtaining Q target consultation task instructions according to the output result of the language model.

[0046] For example, the first template instruction is an instruction for generating 1 new instruction, and the first template instruction (P gen , Igen ) can be represented as: "P gen = generate one new instruction according to the following instructions." gen = "P1; P2; …; P M ". For example, M = 5, the first template instruction (P gen , I gen ) for generating one new instruction can be: P gen = "generate one new instruction according to the following instructions."

[0047] I gen = "Can the cost be allocated proportionally according to the number of medical staff?"

[0048] "Can the cost be allocated proportionally according to the number of medical staff?"

[0049] "Can the cost be allocated proportionally according to the number of medical staff?"

[0050] "Can the cost be allocated proportionally according to the number of medical staff?"

[0051] "Can the cost be allocated proportionally according to the number of medical staff?"

[0052] "Can the cost be allocated proportionally according to the number of medical staff?"

[0053] "Can the cost be allocated proportionally according to the number of medical staff?"

[0054] In the embodiments of the present application, the first template instruction (P gen , I gen ) is input into the language model GPT3.5 model to generate one target consultation task instruction P out . For example, the target consultation task instruction P out = "In hospital financial management, how to determine and allocate various costs to achieve financial revenue and expenditure balance and maximize economic benefits?"

[0055] S130, respectively determine the target instruction type of the Q target consultation task instructions; wherein the target instruction type includes a classification task instruction and a non-classification task instruction.

[0056] In the embodiments of the present application, each target consultation task instruction in the Q target consultation task instructions is analyzed to determine the target instruction type of each target consultation task instruction. For example, each target consultation task instruction can be input into a pre-trained instruction type judgment model, and the instruction type of each target consultation task instruction is determined according to the output result of the instruction type judgment model.

[0057] Optionally, the target instruction type of the Q target consultation task instructions is determined respectively, including: randomly extracting L groups of initial consultation task instances from the initial consultation task instance set, wherein the L groups of initial consultation task instances include L1 groups of initial consultation task instances with instruction type of classification task instruction and L2 groups of initial consultation task instances with instruction type of non-classification task instruction; generating a second template instruction according to the L groups of initial consultation task instances; wherein the second template instruction is an instruction for judging the instruction type of the Q target consultation task instructions; inputting the second template instruction into the language model, and determining the target instruction type of the Q target consultation task instructions according to the output result of the language model.

[0058] For example, L1 groups of initial consultation task instances with instruction type of classification task instruction and L2 groups of initial consultation task instances with instruction type of non-classification task instruction are randomly extracted from the initial consultation task instance set, wherein L1+L2=L, the L1 groups of initial consultation task instances with instruction type of classification task instruction can be understood as C i =1 initial consultation task instances, and the L2 groups of initial consultation task instances with instruction type of non-classification task instruction can be understood as C i =0 initial consultation task instances. The second template instruction generated according to the L groups of initial consultation task instances can be represented as (P cls , I cls ), wherein the second template instruction (P cls , I cls ) can be understood as an instruction for judging the instruction type of the Q target consultation task instructions P out . For example, the second template instruction (P cls , I cls ) can be represented as:

[0059] P cls =“According to the following instructions and whether they are classification tasks (0: non-classification task; 1: classification task), determine whether the target consultation task instruction P out is a classification task. If it is a non-classification task, output 0; if it is a classification task, output 1.”

[0060] I cls =“

[0061] Is P1 a classification task? 0

[0062] Is P2 a classification task? 1

[0063] …

[0064] Is P L a classification task? 0

[0065] ”

[0066] For example, L = 6, L0 = 4, then the instruction (P cls , I cls ) for judging the instruction type of Q target consultation task instructions can be:

[0067] P cls = "Judge whether the instruction 'How to determine and allocate various costs in hospital financial management to achieve financial revenue and expenditure balance and maximize economic benefits?' is a classification task according to the following instructions and whether they are classification tasks (0: non-classification task; 1: classification task). If it is not a classification task, output 0; if it is a classification task, output 1."

[0068] I cls = "Please explain the following index."

[0069] "Is it a classification task?"

[0070] 0

[0071] "How to handle the following situation when performance bonuses are paid?"

[0072] Is it a classification task? 0

[0073] "Does the patient perform multiple operations during hospitalization? When counting the number of operating tables for discharged patients, is it counted by the number of people?"

[0074] Is it a classification task? 1

[0075] "What does this content inspire for the implementation of the performance evaluation scheme?"

[0076] Is it a classification task? 0

[0077] "When calculating the complication rate, do you need to exclude newborn patients?"

[0078] Is it a classification task? 1

[0079] "What is the formula for calculating the average length of stay after CMI value correction?"

[0080] Is it a classification task? 0

[0081] "

[0082] In the embodiments of the present application, the second template instruction (P cls , I cls) input into the language model GPT3.5, and the judgment result obtained is: "the target consultation task instruction 'how to determine and allocate various costs in hospital financial management to achieve financial revenue and expenditure balance and maximize economic benefits?' is a non-classification task, and therefore the output is 0." It can be understood that inputting the second template instruction into the language model enables the language model to learn the instruction types of each initial consultation task instruction in the L group of initial consultation task instances, so as to accurately judge the target instruction type of the target consultation task instruction.

[0083] In the embodiments of the present application, according to the target instruction type of each target consultation task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in a target manner matched with the target instruction type.

[0084] In the embodiments of the present application, according to the target instruction type of each target consultation task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in a target manner matched with the target instruction type. Optionally, for each target consultation task instruction in the Q target consultation task instructions, according to the target instruction type of the current target consultation task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in a target manner matched with the target instruction type, including: for each target consultation task instruction in the Q target consultation task instructions, when the target instruction type of the current target consultation task instruction is a non-classification task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in an input priority manner; and when the target instruction type of the current target consultation task instruction is a classification task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in an output priority manner.

[0085] For example, when the target consultation task instruction is a non-classification task instruction, the target consultation task instruction is input into a first instance generation model, and a target consultation task instance corresponding to the target consultation task instruction is determined according to an output result of the first instance generation model. The first instance generation model is a machine learning model trained based on a plurality of input-priority first consultation task instance samples, and thus, the target consultation task instance is determined by the first instance generation model in an input-priority manner. The input-priority manner can be understood as a manner in which input information in the first consultation task instance sample is placed before output information, or a manner in which the weight of the input information in the first consultation task instance sample is greater than the weight of the output information. For another example, when the target consultation task instruction is a classification task instruction, the target consultation task instruction is input into a second instance generation model, and a target consultation task instance corresponding to the target consultation task instruction is determined according to an output result of the second instance generation model. The second instance generation model is a machine learning model trained based on a plurality of output-priority second consultation task instance samples, and thus, the target consultation task instance is determined by the second instance generation model in an output-priority manner. The output-priority manner can be understood as a manner in which output information in the second consultation task instance sample is placed before input information, or a manner in which the weight of the output information in the second consultation task instance sample is greater than the weight of the input information.

[0086] Optionally, when the target instruction type of the current target consultation task instruction is a non-classification task instruction, a target consultation task instance corresponding to the current target consultation task instruction is generated in an input-priority manner, including: when the target instruction type of the current target consultation task instruction is a non-classification task instruction, K0 groups of first consultation task instances with a non-classification task instruction type are randomly extracted from the initial consultation task instance set; a third template instruction is generated according to the K0 groups of first consultation task instances, where the third template instruction is an instruction for generating a corresponding instance of the current target consultation task instruction in an input-priority manner; and the third template instruction is input into the language model, and a target consultation task instance corresponding to the current target consultation task instruction is determined according to an output result of the language model.

[0087] For example, when the target consultation task instruction is a non-classification task instruction, K0 groups of initial consultation task instances with a non-classification task instruction type are randomly extracted from the initial consultation task instance set, that is, K0 groups of Ci The initial consultation task instance is set to 0, and this K0 group of initial consultation task instances is used as the first consultation task instance. A third template instruction (P) is generated based on the K0 group of first consultation task instances. exp I exp The third template instruction can be understood as an instruction used to generate instances of target consultation task instructions of non-classified task instruction type in an input-first manner. For example, the third template instruction (P exp I exp ) can be represented as: P exp = "Based on the following instruction examples, for instruction P..." out Generate multiple instances, including inputs and outputs. If instruction P... out No input is required; output will be generated directly.

[0088] I exp =“

[0089] Instruction P1

[0090] Input I1

[0091] Output O1

[0092] Instruction P2

[0093] Input I2

[0094] Output O2

[0095] …

[0096] instruction

[0097] enter

[0098] Output

[0099] "

[0100] The third template instruction (P) exp I exp The instructions are input into the language model, and the output is the instruction P for the target consultation task. out The corresponding n target consultation task instances {(P out (I1, O1, C1 = 0), (P) out ,I2,O2,C2=0),...,(P out I n O n C n =0)}.

[0101] Optionally, when the target instruction type of the current target consultation task instruction is a classification task instruction, the target consultation task instance corresponding to the current target consultation task instruction is generated in an output-priority manner, including: when the target instruction type of the current target consultation task instruction is a classification task instruction, K1 groups of second consultation task instances with instruction types of classification task instructions are randomly extracted from the initial consultation task instance set; a fourth template instruction is generated according to the K1 groups of second consultation task instances, wherein the fourth template instruction is an instruction for generating the corresponding instance of the current target consultation task instruction in an output-priority manner; the fourth template instruction is input into the language model, and according to the output result of the language model, the target consultation task instance corresponding to the current target consultation task instruction is determined.

[0102] For example, when the target consultation task instruction is a classification task instruction, K1 groups of initial consultation task instances with instruction types of classification task instructions are randomly extracted from the initial consultation task instance set, that is, K1 groups of initial consultation task instances with C i =1 are randomly extracted from the initial consultation task instance set, and the K1 groups of initial consultation task instances are taken as second consultation task instances. A fourth template instruction (P′ exp , I′ exp ) is generated based on the K1 groups of second consultation task instances. The fourth template instruction can be understood as an instruction for generating the corresponding instance of the target consultation task instruction with the instruction type of the classification task instruction in an output-priority manner. For example, the fourth template instruction (P′ exp , I′ exp ) can be represented as: P′ exp =“According to the following instruction instance, generate multiple instances of instruction P out , including input and output. If instruction P out does not require input, directly generate output.”

[0103] I′ exp =“

[0104] Instruction P′1

[0105] Output O′1

[0106] Input I′1

[0107] Instruction P′2

[0108] Output O′2

[0109] Input I′2

[0110] …

[0111] Instruction

[0112] Output

[0113] Input

[0114] ”

[0115] The fourth template instruction (P′ exp , I′ exp ) is input into the language model, and the output obtained is the target consultation task instruction P out and the corresponding n′ target consultation task instances {(P out , I1, O1, C1=1), (P out , I2, O2, C2=1),..., (P out , I n ′, O n ′, C n ′=1}.

[0116] For example, the instruction type judgment result of the target consultation task instruction generated in the embodiment of the application is 0 (non-classification task instruction), and the input priority is used to generate the target consultation task instance corresponding to the target consultation task instruction. The first consultation task instance of K0 non-classification task instructions in the initial consultation task instance set is randomly extracted, and the third template instruction (P exp , I exp ) can be formed as follows:

[0117] P exp =“According to the following instruction instance, generate multiple instances for the instruction “How to determine and allocate various costs in hospital financial management to achieve financial revenue and expenditure balance and maximize economic benefits?” including input and output. If the instruction does not require input, directly generate output.”

[0118] I exp =“

[0119] Instruction: “How to handle the following situation when performance bonuses are paid?”

[0120] Input: “The workload of the nursing team is heavy and miscellaneous, and it is impossible to accurately record the workload of each person.”

[0121] Output: “The performance project team can provide secondary distribution guidance, or continue to pay according to the original mature distribution scheme.”

[0122] Instruction: “Please explain the following index.”

[0123] Input: “Increase in average cost per hospitalization”

[0124] Output: "The hospitalization average cost increase is the difference between the average medical cost of discharged patients in the current year and the previous year, divided by the average medical cost of discharged patients in the previous year. This indicator is a national testing indicator, measuring the level of patient cost burden and its growth, with the goal of gradually reducing it."

[0125] Instruction: "Summarize the paragraph in one sentence."

[0126] Input: "Fine management is the only way for public hospital management modernization, and fine human resource management is the foundation of hospital fine management. Beijing Tsinghua Changgeng Hospital has built a fine human resource management system based on the job and level system, integrating development, evaluation, and compensation systems. This system not only reflects the uniqueness of different types and levels of positions in the hospital, but also balances the relationship between different types of positions and explicitly establishes promotion relationships for different levels of positions, striving to realize the best use of human resources. This system can provide some reference for public hospital personnel system reform."

[0127] Output: "The main content of this paragraph is that Beijing Tsinghua Changgeng Hospital has built a fine human resource management system that embodies fairness, balance, and practicality, which can provide a reference for public hospital personnel system reform."

[0128] "

[0129] In the embodiment of the present application, the third template instruction (P exp ,I exp ) is input into the GPT3.5 model, and the output obtained is:

[0130] "

[0131] Instruction: "In hospital financial management, how to determine and allocate various costs to achieve financial balance and maximize economic benefits?"

[0132] Input: None

[0133] Output: "Hospitals can adopt the cost center system to allocate costs to corresponding cost centers, thereby determining and allocating various costs. In addition, hospitals can also use the ABC cost management method to manage and allocate various costs in detail, in order to achieve financial balance and maximize economic benefits."

[0134] Instruction: "In hospital financial management, how to determine and allocate various costs to achieve financial balance and maximize economic benefits?"

[0135] Input: None

[0136] Output: "Hospital financial management can be achieved through the establishment of cost calculation, cost control and cost analysis means, to determine and allocate costs, to achieve financial balance and maximize economic benefits."

[0137] ”

[0138] S150, update the initial consultation task instance set based on the target consultation task instance, and return to execute random extraction of M initial consultation task instructions from the initial consultation task instance set until the preset termination condition is met.

[0139] In the embodiments of the application, the target consultation task instance is added to the initial consultation task instance set to update the initial consultation task instance set, and the execution of S120-S140 is returned until the updated initial consultation task instance set meets the preset termination condition, such as the number of consultation task instances in the updated initial consultation task instance set reaching the preset number threshold.

[0140] Optionally, in the process of cyclically executing S120-S140, M0 of the M initial consultation task instructions are from the initial consultation task instance set at the beginning, and M1 are from the target consultation task instructions generated by the last execution of S120-S140. The advantage of such setting is that the initial consultation task instruction at the beginning can be used as a reference standard, which helps to improve the accuracy of subsequent generation of target consultation task instructions.

[0141] Optionally, before updating the initial consultation task instance set based on the target consultation task instance, it further includes: for each target consultation task instruction in the Q target consultation task instructions, calculating the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set; when the similarity is greater than a preset similarity threshold, filtering the target consultation task instance corresponding to the current target consultation task instruction.

[0142] For example, for each target consultation task instruction in the Q target consultation task instructions, the similarity between the current target consultation task instruction and each initial consultation task instruction in the initial consultation task instance set is calculated, for example, the maximum value of the similarity or the average value of all similarities can be taken as the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set. The similarity between the current target consultation task instruction and each initial consultation task instruction in the initial consultation task instance set can be calculated based on cosine similarity or ROUGE. When the similarity is greater than a preset similarity threshold, the target consultation task instance corresponding to the current target consultation task instruction is filtered out. In the embodiment of the present application, the similarity between the target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set is evaluated based on the ROUGE-L index, and if the maximum ROUGE-L is greater than 0.7, the target consultation task instruction is filtered out. For example, the maximum ROUGE-L value of the target consultation task instruction "In hospital financial management, how to determine and allocate various costs to achieve financial revenue and expenditure balance and maximize economic benefits?" and the initial consultation task instruction in the initial consultation task instance set is 0.21, and the target consultation task instruction is retained. It can be understood that after the audit, the target consultation task instance corresponding to the target consultation task instruction can be put into the initial consultation task instance set.

[0143] For example, the updated initial consultation task instance set that meets the preset termination condition can be referred to as a target consultation task instance set. For example, the target consultation task instance set is represented by S1, |S1| = 100, and Table 2 is a target consultation task instance table provided by the embodiment of the present application:

[0144]

[0145]

[0146] The consultation task instance generation method of the embodiment of the application comprises the following steps: obtaining an initial consultation task instance set; wherein the initial consultation task instance set comprises at least one group of initial consultation task instances; each group of initial consultation task instances comprises initial consultation task instructions, input information, output information and instruction types corresponding to the initial consultation task instructions; randomly extracting M initial consultation task instructions from the initial consultation task instance set, and obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model; determining target instruction types of the Q target consultation task instructions respectively; wherein the target instruction types comprise classification task instructions and non-classification task instructions; for each target consultation task instruction in the Q target consultation task instructions, generating a target consultation task instance corresponding to the current target consultation task instruction in a target mode matched with the target instruction type of the current target consultation task instruction according to the target instruction type of the current target consultation task instruction; updating the initial consultation task instance set based on the target consultation task instance, and executing the step of randomly extracting M initial consultation task instructions from the initial consultation task instance set until a preset termination condition is met. The technical scheme provided by the embodiment of the application can not only solve the problems of large amount of training data required when constructing an operation and management consultation model, high cost of manual annotation, insufficient training input integrity and continuity, but also continuously construct consultation task instances with accuracy, integrity and richness, which is helpful to improve the operation and management of related industries.

[0147] Embodiment two

[0148] Figure 2 A structural schematic diagram of a consultation task instance generation device provided by the second embodiment of the application is shown in FIG. 2. As shown in FIG. 2, the device comprises: Figure 2

[0149] An initial consultation task instance set obtaining module 210 is configured to obtain an initial consultation task instance set; wherein the initial consultation task instance set comprises at least one group of initial consultation task instances; each group of initial consultation task instances comprises initial consultation task instructions, input information, output information and instruction types corresponding to the initial consultation task instructions;

[0150] A target consultation task instruction obtaining module 220 is configured to randomly extract M initial consultation task instructions from the initial consultation task instance set, and obtain Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model;

[0151] A target instruction type determining module 230 is configured to determine target instruction types of the Q target consultation task instructions respectively; wherein the target instruction types comprise classification task instructions and non-classification task instructions;

[0152] ​The target consultation task instance generation module 240 is configured to, for each target consultation task instruction in the Q target consultation task instructions, generate a target consultation task instance corresponding to a current target consultation task instruction according to a target instruction type of the current target consultation task instruction in a target manner matched with the target instruction type.

[0153] The initial consultation task instance set updating module 250 is configured to update the initial consultation task instance set based on the target consultation task instance, and perform random extraction of M initial consultation task instructions from the initial consultation task instance set until a preset termination condition is met.

[0154] Optionally, the target consultation task instruction acquisition module is configured to:

[0155] generate a first template instruction based on the M initial consultation task instructions; the first template instruction is an instruction for generating Q new instructions;

[0156] input the first template instruction into the language model, and acquire the Q target consultation task instructions according to an output result of the language model.

[0157] Optionally, the target instruction type determination module is configured to:

[0158] randomly extract L groups of initial consultation task instances from the initial consultation task instance set, wherein the L groups of initial consultation task instances include L1 groups of initial consultation task instances of which instruction types are classification task instructions and L2 groups of initial consultation task instances of which instruction types are non-classification task instructions;

[0159] generate a second template instruction according to the L groups of initial consultation task instances; the second template instruction is an instruction for determining instruction types of the Q target consultation task instructions;

[0160] input the second template instruction into the language model, and respectively determine target instruction types of the Q target consultation task instructions according to an output result of the language model.

[0161] Optionally, the target consultation task instance generation module includes:

[0162] The first task instance generation unit is configured to, for each target consultation task instruction in the Q target consultation task instructions, when a target instruction type of the current target consultation task instruction is a non-classification task instruction, generate a target consultation task instance corresponding to the current target consultation task instruction in an input-priority manner;

[0163] The second task instance generation unit is configured to generate, in a priority of output, a target consultation task instance corresponding to the current target consultation task instruction when the target instruction type of the current target consultation task instruction is a classification task instruction.

[0164] Optionally, the first task instance generation unit is configured to:

[0165] When the target instruction type of the current target consultation task instruction is a non-classification task instruction, randomly select K0 groups of first consultation task instances with a non-classification task instruction from the initial consultation task instance set;

[0166] generate a third template instruction according to the K0 groups of first consultation task instances, wherein the third template instruction is an instruction for generating an instance corresponding to the current target consultation task instruction in a priority of input;

[0167] input the third template instruction into the language model, and determine the target consultation task instance corresponding to the current target consultation instruction according to an output result of the language model.

[0168] Optionally, the second task instance generation unit is configured to:

[0169] When the target instruction type of the current target consultation task instruction is a classification task instruction, randomly select K1 groups of second consultation task instances with a classification task instruction from the initial consultation task instance set;

[0170] generate a fourth template instruction according to the K1 groups of second consultation task instances, wherein the fourth template instruction is an instruction for generating an instance corresponding to the current target consultation task instruction in a priority of output;

[0171] input the fourth template instruction into the language model, and determine the target consultation task instance corresponding to the current target consultation instruction according to an output result of the language model.

[0172] Optionally, the apparatus further comprises:

[0173] The similarity calculation module is configured to, before updating the initial consultation task instance set based on the target consultation task instance, calculate, for each target consultation task instruction in the Q target consultation task instructions, a similarity between the current target consultation task instruction and an initial consultation task instruction in the initial consultation task instance set.

[0174] The task instance filtering module is configured to, when the similarity is greater than a preset similarity threshold, filter the target consultation task instance corresponding to the current target consultation task instruction.

[0175] The consultation task instance generation device provided by the embodiment of the present application can execute the consultation task instance generation method provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0176] Embodiment three

[0177] Figure 3 A structural schematic diagram of an electronic device 10 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices (e.g., headsets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present application described and / or claimed in this document.

[0178] As shown in Figure 3 The electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11, wherein the memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0179] A plurality of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunications networks.

[0180] The processor 11 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, and the like. The processor 11 performs various methods and processes described above, such as the counseling task instance generation method.

[0181] In some embodiments, the counseling task instance generation method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded onto the RAM 13 and executed by the processor 11, one or more steps of the counseling task instance generation method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the counseling task instance generation method by any other suitable means, such as by means of firmware.

[0182] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0183] Computer programs used to implement the methods of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program, when executed by the processor of the machine, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, and partially on a machine or a remote machine or a server.

[0184] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0185] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0186] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), blockchain network, and the Internet.

[0187] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0188] It should be understood that the various forms of flow shown above can be reordered, added to, or have steps deleted. For example, the steps described in the present application can be performed in parallel, in series, or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, and this is not limited herein.

[0189] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for generating a consulting task instance, characterized in that, The method comprises the following steps: obtaining an initial consultation task instance set; wherein the initial consultation task instance set comprises at least one group of initial consultation task instances; each group of initial consultation task instances comprises initial consultation task instructions, input information, output information, and instruction types corresponding to the initial consultation task instructions; randomly selecting M initial consultation task instructions from the initial consultation task instance set, and obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model; determining the target instruction types of the Q target consultation task instructions respectively; wherein the target instruction types comprise classification task instructions and non-classification task instructions; for each target consultation task instruction in the Q target consultation task instructions, generating a target consultation task instance corresponding to the current target consultation task instruction in a target manner matched with the target instruction type of the current target consultation task instruction; for each target consultation task instruction in the Q target consultation task instructions, calculating the similarity between the current target consultation task instruction and each initial consultation task instruction in the initial consultation task instance set, and taking the maximum similarity value or the average of all similarity values as the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set; when the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set is greater than a preset similarity threshold, filtering the target consultation task instance corresponding to the current target consultation task instruction; updating the initial consultation task instance set based on the target consultation task instance, and returning to the step of randomly selecting M initial consultation task instructions from the initial consultation task instance set until a preset termination condition is met; wherein the step of generating a target consultation task instance corresponding to the current target consultation task instruction in a target manner matched with the target instruction type of the current target consultation task instruction comprises: for each target consultation task instruction in the Q target consultation task instructions, when the target instruction type of the current target consultation task instruction is a non-classification task instruction, generating the target consultation task instance corresponding to the current target consultation task instruction in an input-priority manner; when the target instruction type of the current target consultation task instruction is a classification task instruction, generating the target consultation task instance corresponding to the current target consultation task instruction in an output-priority manner.

2. The method of claim 1, wherein, obtaining Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model comprises: generating a first template instruction based on the M initial consultation task instructions; wherein the first template instruction is an instruction for generating Q new instructions; inputting the first template instruction into the language model, and obtaining Q target consultation task instructions according to the output result of the language model.

3. The method of claim 1, wherein, determining the target instruction types of the Q target consultation task instructions respectively comprises: randomly select L groups of initial consultation task instances from the initial consultation task instance set, wherein the L groups of initial consultation task instances include L1 groups of initial consultation task instances of which the instruction types are classification task instructions and L2 groups of initial consultation task instances of which the instruction types are non-classification task instructions; generate a second template instruction according to the L groups of initial consultation task instances; wherein the second template instruction is an instruction for judging the instruction types of the Q target consultation task instructions; input the second template instruction into the language model, and determine the target instruction types of the Q target consultation task instructions respectively according to the output result of the language model.

4. The method of claim 1, wherein, when the target instruction type of the current target consultation task instruction is a non-classification task instruction, generate the target consultation task instance corresponding to the current target consultation task instruction in an input-priority manner, including: when the target instruction type of the current target consultation task instruction is a non-classification task instruction, randomly select K0 groups of first consultation task instances of which the instruction types are non-classification task instructions from the initial consultation task instance set; generate a third template instruction according to the K0 groups of first consultation task instances, wherein the third template instruction is an instruction for generating the instance corresponding to the current target consultation task instruction in an input-priority manner; input the third template instruction into the language model, and determine the target consultation task instance corresponding to the current target consultation task instruction according to the output result of the language model.

5. The method of claim 1, wherein, when the target instruction type of the current target consultation task instruction is a classification task instruction, generate the target consultation task instance corresponding to the current target consultation task instruction in an output-priority manner, including: when the target instruction type of the current target consultation task instruction is a classification task instruction, randomly select K1 groups of second consultation task instances of which the instruction types are classification task instructions from the initial consultation task instance set; generate a fourth template instruction according to the K1 groups of second consultation task instances, wherein the fourth template instruction is an instruction for generating the instance corresponding to the current target consultation task instruction in an output-priority manner; input the fourth template instruction into the language model, and determine the target consultation task instance corresponding to the current target consultation task instruction according to the output result of the language model.

6. A consulting task instance generation apparatus characterized by comprising: including: an initial consultation task instance set acquisition module, configured to acquire an initial consultation task instance set; wherein the initial consultation task instance set includes at least one group of initial consultation task instances; each group of initial consultation task instances includes an initial consultation task instruction, input information, output information and an instruction type corresponding to the initial consultation task instruction; a target consultation task instruction acquisition module, configured to randomly select M initial consultation task instructions from the initial consultation task instance set, and acquire Q target consultation task instructions based on the M initial consultation task instructions and a pre-trained language model; a target instruction type determination module, configured to determine target instruction types of the Q target consultation task instructions respectively; wherein the target instruction types include classification task instructions and non-classification task instructions; The target consultation task instance generation module is configured to, for each of the Q target consultation task instructions, generate a target consultation task instance corresponding to a current target consultation task instruction according to a target instruction type of the current target consultation task instruction in a target manner matched with the target instruction type; The similarity calculation module is configured to, for each of the Q target consultation task instructions, calculate a similarity between the current target consultation task instruction and each initial consultation task instruction in the initial consultation task instance set, and take a maximum value of the similarity or an average value of all the similarities as the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set; The task instance filtering module is configured to filter the target consultation task instance corresponding to the current target consultation task instruction when the similarity between the current target consultation task instruction and the initial consultation task instruction in the initial consultation task instance set is greater than a preset similarity threshold. The initial consultation task instance set updating module is configured to update the initial consultation task instance set based on the target consultation task instance, and perform the random extraction of the M initial consultation task instructions from the initial consultation task instance set until a preset termination condition is met. The target consultation task instance generation module includes: The first task instance generation unit is configured to, for each of the Q target consultation task instructions, generate a target consultation task instance corresponding to the current target consultation task instruction in an input-priority manner when the target instruction type of the current target consultation task instruction is a non-classification task instruction. The second task instance generation unit is configured to generate a target consultation task instance corresponding to the current target consultation task instruction in an output-priority manner when the target instruction type of the current target consultation task instruction is a classification task instruction.

7. An electronic device, comprising: The electronic device includes: at least one processor; and a memory connected with the at least one processor in communication; wherein The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the consultation task instance generation method in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling the processor to execute the consultation task instance generation method in any one of claims 1-5 when executed.

Images