Intelligent interaction method, system, medium and equipment

Abstract

The invention provides an intelligent interaction method and system, a medium and equipment. The method comprises the following steps: acquiring a natural language of a user; inputting the natural language into an intelligent interaction model, and outputting an interaction result; the processing step of the intelligent interaction model comprises the following steps: complementing missing information in the natural language according to the obtained historical interaction data; creating an intention recognition sample library according to the acquired service data set and the open source data set; identifying the intention of the natural language according to the intention identification sample library; judging an execution scene of the natural language according to the intention of the natural language; when the natural language executes the card calling scene, identifying card parameters and calling a corresponding card to obtain an interaction result; and when the natural language executes a real-time interface calling scene, identifying interface parameters, scoring and sorting return results of interfaces, and creating background knowledge to obtain an interaction result.

Description

Technical Field

[0001] The present invention relates to the field of intelligent interaction technology, and in particular to an intelligent interaction method, system, medium and device. Background Art

[0002] Adhering to the principle of "actively using modern digital technology to empower and empower the power grid, and promoting the transformation and upgrading of traditional power grids and businesses to digitalization and intelligence", under the dual drive of the current energy revolution and digital revolution, intelligent operation and maintenance needs to be combined with AI to enter a new era. Intelligent operation and maintenance is a new mode of operation and maintenance with the goal of ensuring the safe operation of the power grid and improving the quality, efficiency and benefits of operation and maintenance, and taking the informatization, digitization and intelligence of equipment operation and maintenance, maintenance and production management as the means.

[0003] In the process of operation and maintenance of power grid control and protection systems, problems such as information islands, cumbersome manual operations, slow response speed, insufficient system flexibility and poor interactivity are often encountered. These limitations not only affect the operation and maintenance efficiency, but also restrict the intelligent development of the power grid. In addition, it is impossible to obtain the real-time operating status and alarm data of the equipment in the station, and the scope of questions and answers is relatively narrow. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide an intelligent interaction method, system, medium and device that can quickly obtain the real-time operating status, alarm data, etc. of the equipment in the station, and can continuously add new data to the system during use, thereby continuously expanding the scope of questions and answers.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In one aspect, the present invention provides an intelligent interaction method, comprising:

[0007] Obtain the user's natural language;

[0008] Inputting the natural language into an intelligent interaction model and outputting an interaction result;

[0009] The processing steps of the intelligent interaction model include:

[0010] Based on the historical interaction data obtained, the missing information in the natural language is completed; it can solve the problem of inability to accurately identify intent during multiple rounds of question and answer, so as to facilitate more accurate question and answer intent identification.

[0011] Create an intent recognition sample library based on the acquired business data sets and open source data sets;

[0012] Identify the intent of the natural language according to the intent recognition sample library;

[0013] According to the intention of the natural language, determining the execution scenario of the natural language; the execution scenario includes direct answer, card call and real-time interface call;

[0014] When the natural language executes a card calling scenario, the card parameters are identified and the corresponding card is called to obtain an interaction result;

[0015] When the natural language executes a real-time interface call scenario, the interface parameters are identified and the returned results of the interface are scored and sorted, background knowledge is created, and interaction results are obtained.

[0016] Optionally, the step of creating the intent recognition sample library includes:

[0017] Randomly sampling the business data set to obtain a seed sample, expanding the seed sample using the intelligent interaction model to obtain a candidate sample, and manually verifying the candidate sample to obtain a first sample library;

[0018] Converting the format of the open source data set to obtain a second sample library;

[0019] The first sample library and the second sample library constitute an intention recognition sample library.

[0020] Optionally, the card parameters include time, date, person's name and device name.

[0021] Optionally, the calling corresponding card includes:

[0022] Vectorize the natural language to obtain a language vector;

[0023] Calculating the cosine similarity between the language vector and the history card vector in the intelligent interaction model;

[0024] The card with the highest cosine similarity that exceeds the preset threshold is used as the called card, and the card calling instruction is constructed. The user can configure the threshold for similarity matching in the system. When the threshold is high, the card may not be matched; when the threshold is low, the wrong card may be returned.

[0025] Optionally, the interface parameters include time, date, person's name, device name, device type, measurement name, device area name, and company name.

[0026] Optionally, the formula for scoring the returned result of the interface is:

[0027] ;

[0028] In the formula, Indicates that the BM25 algorithm is used to score the natural language and interface return results; Indicates using the Embedding algorithm to score natural language and interface return results; Indicates that the Zero-shot algorithm is used to score the natural language and interface return results; Represents the BM25 algorithm weight parameter; Represents the weight parameter of the Embedding algorithm; Indicates the Zero-shot algorithm weight parameter.

[0029] Optionally, the sorting of the returned results of the interface includes:

[0030] Sort the scores of the returned results of the interface from high to low to obtain the sorted interfaces;

[0031] The natural language and the sorted interfaces are subjected to correlation analysis, and are re-sorted from high to low according to the correlation, and the interface with the highest correlation is used as the background knowledge.

[0032] In a second aspect, the present invention provides an intelligent interaction system, comprising:

[0033] A natural language acquisition module, used to acquire the user's natural language;

[0034] Intelligent interaction model, used to obtain interaction results, complete missing information in natural language, and create an intent recognition sample library;

[0035] Text classification module, used to identify the intent of natural language;

[0036] Named entity recognition module, used to identify card parameters and interface parameters;

[0037] The calling module is used to call the corresponding card and score and sort the return results of the interface.

[0038] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent interaction method described in the first aspect.

[0039] In a fourth aspect, the present invention provides a computer device, comprising:

[0040] Memory, for storing computer instructions;

[0041] A processor is used to execute the computer instructions to implement the steps of the intelligent interaction method described in the first aspect.

[0042] Beneficial Effects

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention enables intelligent operation and maintenance through an intelligent interactive model. Users interact with the system directly through natural language, and quickly obtain the real-time operation status, alarm data, etc. of the equipment in the station through the system. The data is fully integrated without the need to understand the source of the data. The query and display of key data are automated and intelligent, which greatly reduces the operation of manual query and improves efficiency and accuracy. At the same time, it is an extensible architecture that can continuously add new data to the system during use, continuously expanding the scope of questions and answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Shown is a schematic diagram of a flow chart of an intelligent interaction method of the present invention in an embodiment;

[0046] Figure 2 The figure shows a schematic diagram of the creation process of the intention recognition sample library of the present invention in one embodiment;

[0047] Figure 3 The figure is a schematic diagram of a process of executing a card calling scenario in an embodiment of the present invention;

[0048] Figure 4 The figure is a flow chart of executing a real-time interface calling scenario in one embodiment of the present invention. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0050] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment introduces an intelligent interaction method, including the following steps:

[0053] Get the user's natural language.

[0054] The natural language is input into the intelligent interaction model, and the interaction result is output. The intelligent interaction model in this embodiment can be a large language model.

[0055] The processing steps of the intelligent interaction model include:

[0056] Based on the historical interaction data obtained, the missing information in the natural language is completed. This can solve the problem of inability to accurately identify intent during multiple rounds of question and answer, so as to facilitate more accurate question and answer intent identification.

[0057] Use the understanding ability of the large language model to complete the natural language, that is, the current question. When completing, the historical dialogue information of this conversation is used, and an appropriate prompt is constructed. The large language model changes the wording of this question based on the historical information and uses a more complete expression.

[0058] like Figure 2 As shown in the figure, an intent recognition sample library is created based on the acquired business data set and open source data set. Combining the business data set and the open source data set, a large language model and manual combination are used to build an intent recognition sample library that meets actual needs.

[0059] The business data set is randomly sampled to obtain a seed sample, the seed sample is expanded using the intelligent interaction model to obtain a candidate sample, and the candidate sample is manually verified to obtain a first sample library.

[0060] The open source data set is format converted to obtain a second sample library.

[0061] The first sample library and the second sample library constitute an intention recognition sample library.

[0062] The intent of the natural language is identified according to the intent identification sample library.

[0063] In this embodiment, the text classification module is trained to replace the process of training the large language model to achieve the same effect as training the large language model. The "intention recognition" here is to achieve the effect of "agent calling (Agent)" in the large language model. The large language model has a certain agent calling ability by default, and determines whether the user's question needs to call a certain "tool". However, for some unrecognizable agents, it is necessary to build a data set for model training, and this training process requires a graphics card with a large video memory and a large computing power.

[0064] By using the traditional text classification module to identify the intent of a question, we can think of the "intent" of the question as the "tool" in the call of the large language model agent. Therefore, we only need to train the text classification module without changing the model parameters of the large language model, and can achieve very fast iteration for new data.

[0065] According to the intention of the natural language, the execution scenario of the natural language is determined; the execution scenario includes three types: direct answer, card call and real-time interface call. The user's question recognition result is delivered to the corresponding operation and inspection scenario, and different processing strategies are adopted for different scenarios.

[0066] When the natural language executes the card calling scenario, the card parameters are identified and the corresponding card is called to obtain the interaction result.

[0067] The key parameters of the card are identified through the named entity recognition model, including time, date, person name and device name.

[0068] The cards corresponding to the call include:

[0069] The natural language card is vectorized to obtain a language vector. The question is discretized using vectorization. The natural language of the user's question is converted into a vector representation using a text vectorization model, and all the card information supported by the system has been converted into vectors in advance in the same way.

[0070] The cosine similarity between the language vector and the historical card vector in the intelligent interaction model is calculated. The cosine similarity between the user question vector and the existing card vector is calculated, and the card that exceeds the threshold and has the highest similarity is taken as the card for this call.

[0071] Users can configure the threshold for similarity matching in the system. When the threshold is high, cards may not be matched; when the threshold is low, incorrect cards may be returned.

[0072] The card with the highest cosine similarity that exceeds the preset threshold is used as the calling card, and the card calling instruction is constructed. All the information required to call the card is organized, and the card calling instruction is constructed. After receiving the instruction, the front-end framework will automatically pop up the corresponding card and pass in the parameters.

[0073] When the natural language executes a real-time interface call scenario, the interface parameters are identified and the returned results of the interface are scored and sorted, background knowledge is created, and interaction results are obtained.

[0074] The named entity recognition model is used to extract key information from the user's question as the input parameter of the interface. The interface parameters include time, date, person's name, device name, device type, measurement name, device region name, and company name.

[0075] Get the return results of each interface. When formatting, add the interface description to the return results as the basis for sorting.

[0076] Use a combination of methods to sort. Score all candidate interface return results using the formula:

[0077] ;

[0078] In the formula, Indicates that the BM25 algorithm is used to score the natural language and interface return results; Indicates that the Embedding algorithm is used to score the natural language and interface return results. The formatted return results of the natural language and interface are converted into dense vectors, and the scores are performed based on the similarity of the vectors. Indicates that the zero-shot algorithm is used to score natural language and interface return results. To optimize the retrieval effect in the zero-shot case, after the user asks a question, a large language model is used to generate a document containing the answer to the current question. The newly generated document and the formatted return results of the interface are vectorized into dense vectors and scored based on vector similarity. Represents the BM25 algorithm weight parameter; Represents the weight parameter of the Embedding algorithm; Indicates the Zero-shot algorithm weight parameter.

[0079] In this embodiment, use =0.25, =0.5, =0.25 to assign weights to each algorithm.

[0080] The sorting of the returned results of the interface includes:

[0081] Sort the scores of the returned results of the interface from high to low to obtain the sorted interfaces;

[0082] The natural language and the sorted interfaces are subjected to correlation analysis, and the interfaces are re-sorted from high to low according to the correlation, and the interfaces with the highest correlation are used as background knowledge. The interface with the highest re-sorting score is used as background knowledge. The algorithm accepts the natural language and the candidate interface information screened in the previous step as input, and outputs the correlation between them. Since the interface information contains less semantic information, the large language model's direct answer to the current question is first obtained. The answer will contain a detailed introduction to the current question, and the answer is used to replace the user's question.

[0083] The intelligent interaction model answers based on the background knowledge to obtain an interaction result.

[0084] Example 2

[0085] This embodiment introduces an intelligent interaction system, including:

[0086] A natural language acquisition module for acquiring the natural language of the user.

[0087] An intelligent interaction model for obtaining interaction results, complementing missing information in natural language, and creating an intention recognition sample library;

[0088] A text classification module for identifying card parameters and interface parameters.

[0089] A named entity recognition module for identifying card parameters and interface parameters.

[0090] A call module for calling the corresponding card and scoring and sorting the return results of the interface.

[0091] Example 3

[0092] This example introduces a specific test example of an intelligent interaction method, including the following steps:

[0093] Step 1, Q&A completion. Use a large language model to automatically complete the current question according to historical data to solve the problem of inaccurate intention recognition in multi-round Q&A.

[0094] For example: The user's conversation history is as follows:

[0095] User: What is the current power of Pole 1?

[0096] Large language model: The current power of Pole 1 is 0.00586 MW

[0097] User: What about Pole 2?

[0098] In this multi-round conversation scenario, since the user's second question omits some useful information, the system is likely to be unable to determine what information the user is asking about "Pole 2". Therefore, question completion is required.

[0099] Question completion requires constructing a suitable prompt to prompt the model.

[0100] Step 2, Intention recognition sample construction. Combine the business data set and the open source data set to construct a sample library that meets the business conditions. The sample library construction process is shown in Figure 2 .

[0101] Construct a large amount of corpus data required for the intention recognition model, some of which come from the open source database, such as ShareGPT, and the remaining data are all obtained from the business data. For example, the existing device information shown in Table 1 and the existing card information shown in Table 2:

[0102] Table 1 Existing device information

[0103]

[0104] Table 2 Existing card information

[0105]

[0106] The business data set is randomly sampled to obtain a seed sample, the seed sample is expanded using the intelligent interaction model to obtain a candidate sample, and the candidate sample is manually verified to obtain a first sample library. Seed samples are shown in Table 3, and candidate samples are shown in Table 4. The first sample library is then obtained through manual screening.

[0107] The open source data set is format converted to obtain a second sample library.

[0108] The first sample library and the second sample library constitute an intention recognition sample library.

[0109] Table 3 Seed samples

[0110]

[0111] Table 4 Candidate samples

[0112]

[0113] Step 3: Question intent identification: Use the text classification module to identify the user's question intent, which is used for different scenarios. Different processing strategies are required for different scenarios.

[0114] The execution scenarios include direct answer, card call and real-time interface call.

[0115] For example, a user asks: "Please open the light-up statistics card?" Intent recognition needs to make a judgment: call the card at this time and decide the path to be executed next.

[0116] Step 4: Execute scene path

[0117] Step 4.1: Execute the card calling process. Figure 3 As shown, if it is determined in step 3 that the current question needs to be "called a card", the current process is executed, which includes three sub-steps:

[0118] Step 4.1.1, card parameter identification. The named entity recognition module is used to identify the key parameters of the card. For example, "I want to see today's fault briefing information", the named entity recognition module identifies the time information from the question: "today", and the card name information: "fault briefing".

[0119] Step 4.1.2, card selection. Determine which card to display based on the similarity between the user and the card name. Based on the card name "Fault Briefing" identified in the previous step, select the card with the highest similarity from the existing card list as the candidate card. To calculate the similarity of the cards, you need to first vectorize all the card names, and then calculate the similarity using cosine similarity.

[0120] Step 4.1.3, card call. Use the card parameters and card name identified in the first two steps to pop up the corresponding card interface through interaction with the framework. Use the "fault briefing" and "today" time parameters obtained from the first two steps to construct the card call instruction.

[0121] Step 4.2: Execute the real-time interface calling process. Figure 4 As shown, if it is determined in step 3 that the current question needs to be "called in real-time interface", the current process is executed. This step indicates that the user's question has high real-time nature and needs to obtain a value from the real-time interface. It is divided into the following steps:

[0122] Step 4.2.1, interface parameter identification. Use the named entity recognition module to extract key information from the user's question as the input parameter of the interface. For example, "What is the gear of the extremely high-end YYA?" The named entity recognition module identifies "extremely high-end YYA" and "gear" as keywords from the question.

[0123] Step 4.2.2: Format the interface information. Add the description of the current interface to the interface and convert the interface information into a document format.

[0124] Step 4.2.3: Mixed sorting. Use a combination of methods to sort. The score composition is as follows:

[0125] ;

[0126] =0.25, =0.5, =0.25;

[0127] For example, for "What is the gear position of the current high-end YYA phase of pole 1", use BM25 to score and get =0.64; using embedding to calculate, we get =0.5266;

[0128] for , you need to build prompt first.

[0129] Current question: What is the gear position of the current extreme 1 high-end YYA phase?

[0130] The model generates the result, which is obtained after vectorization =0.7535, the final score is:

[0131] S=0.25*0.64+0.5*0.5266+0.25*0.7535= 0.450258446.

[0132] The same calculation is performed on all backend interfaces and they are sorted by the final score.

[0133] Step 4.2.4, re-ranking. Use the reranker algorithm for re-ranking. The re-ranking algorithm is a type of algorithm that scores the relevance of questions and candidate paragraphs input by the user. In order to improve the accuracy of the re-ranking algorithm, this patent uses the content generated by the large language model to replace the user's question for relevance scoring. Specifically, for the question "What is the current gear position of the extreme 1 high-end YYA phase?" First, build a prompt to let the large language model pre-answer the current question.

[0134] Tips:

[0135] Please briefly answer the following questions based on your understanding of electrical knowledge:

[0136] What is the gear position of the current extreme 1 high-end YYA phase?

[0137] Large language model answers:

[0138] The question you asked, "What is the gear position of the high-end YYA phase of pole 1 at present?" involves a specific configuration or operating state in the power system, but the details and answer to this question may need to be determined based on the specific power equipment, system configuration, and operating procedures. In the power system, "pole" usually refers to an operating unit in the DC transmission system, which is used to describe the voltage level, operating mode, or control strategy in the system. For example, a DC transmission system may have multiple operating "poles", each with a positive pole and a negative pole.

[0139] The terms "high side" and "low side" usually refer to the higher and lower voltage levels of a DC transmission system. In some systems, these terms may be used to describe the control strategy, for example, in a bipolar system, there may be two separate control areas, the "high side" and the "low side".

[0140] The "YYA" in "YYA phase" may be an identifier of a specific device or system. For example, if the phase sequence of a specific inverter or converter is being described, then "YYA" represents a specific phase of the device.

[0141] "Gear" may refer to the system operating state, the control settings of the equipment, the power transfer level, or other related parameter settings. In a DC transmission system, the gear is usually related to the power transfer capability, the operating mode (such as unipolar, bipolar, independent control area, etc.) or the control settings of the equipment.

[0142] To answer this question accurately, you need to know the specific system configuration, equipment type, and operating procedures. For example, if it is a DC transmission system, you may need to check the real-time monitoring data of the control center, the system configuration file, or the operation manual to determine "the current gear position of the high-end YYA phase of pole 1".

[0143] In actual operation, power system operators will adjust these parameters according to system requirements, safety regulations and operation strategies to ensure system stability and efficiency. Therefore, obtaining specific answers usually requires contacting system operators or consulting system-related technical documents and operation guides.

[0144] The Reranker model calculates the score of the large language model answer and the formatted return result of the interface together to obtain the score of the re-ranking part. After performing the same operation on all interfaces, all scores are obtained and sorted from high to low.

[0145] Step 4.2.5: The large language model answers. The large language model uses the result obtained in step c as background knowledge to answer. The large language model uses the retrieved information as background knowledge and answers the user's question based on the background knowledge. For the alarm information retrieved in the previous step, a prompt is constructed.

[0146] Please answer the questions:

[0147] What is the gear position of the A phase of the high-end YY commutation transformer of pole 1?

[0148] Based on the above information, the large language model will conclude: "Based on the existing knowledge, the gear position of the A phase (i.e., YYA phase) of the high-end YY commutator of pole 1 is 21."

[0149] Example 4

[0150] This embodiment introduces a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the intelligent interaction method described in Embodiment 1 or 3 are implemented.

[0151] Example 5

[0152] This embodiment introduces a computer device, including:

[0153] Memory, for storing computer instructions;

[0154] A processor is used to execute the computer instructions to implement the steps of the intelligent interaction method described in Example 1 or 3.

[0155] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0156] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0157] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0159] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. An intelligent interaction method, characterized in that: include: Obtain the user's natural language; Inputting the natural language into an intelligent interaction model and outputting an interaction result; The processing steps of the intelligent interaction model include: Completing the missing information in the natural language according to the acquired historical interaction data; Create an intent recognition sample library based on the acquired business data sets and open source data sets; Identify the intent of the natural language according to the intent recognition sample library; According to the intention of the natural language, determining the execution scenario of the natural language; the execution scenario includes direct answer, card call and real-time interface call; When the natural language executes a card calling scenario, the card parameters are identified and the corresponding card is called to obtain an interaction result; When the natural language executes a real-time interface call scenario, the interface parameters are identified and the returned results of the interface are scored and sorted, background knowledge is created, and interaction results are obtained.

2. The intelligent interaction method according to claim 1, characterized in that: The steps of creating the intent recognition sample library include: Randomly sampling the business data set to obtain a seed sample, expanding the seed sample using the intelligent interaction model to obtain a candidate sample, and manually verifying the candidate sample to obtain a first sample library; Converting the format of the open source data set to obtain a second sample library; The first sample library and the second sample library constitute an intention recognition sample library.

3. The intelligent interaction method according to claim 1, characterized in that: The card parameters include time, date, person's name and device name.

4. The intelligent interaction method according to claim 1, characterized in that: The cards corresponding to the call include: Vectorize the natural language card to obtain a language vector; Calculating the cosine similarity between the language vector and the history card vector in the intelligent interaction model; The card that exceeds the preset threshold and has the highest cosine similarity is used as the called card, and a card calling instruction is constructed.

5. The intelligent interaction method according to claim 1, characterized in that: The interface parameters include time, date, person's name, device name, device type, measurement name, device area name, and company name.

6. The intelligent interaction method according to claim 1, characterized in that: The formula for scoring the returned results of the interface is: ； In the formula, Indicates that the BM25 algorithm is used to score the natural language and interface return results; Indicates using the Embedding algorithm to score natural language and interface return results; Indicates that the Zero-shot algorithm is used to score the natural language and interface return results; Represents the BM25 algorithm weight parameter; Represents the weight parameter of the Embedding algorithm; Indicates the Zero-shot algorithm weight parameter.

7. The intelligent interaction method according to claim 1, characterized in that: The sorting of the returned results of the interface includes: Sort the scores of the returned results of the interface from high to low to obtain the sorted interfaces; The natural language and the sorted interfaces are subjected to correlation analysis, and are re-sorted from high to low according to the correlation, and the interface with the highest correlation is used as the background knowledge.

8. An intelligent interactive system, characterized in that: include: A natural language acquisition module, used to acquire the user's natural language; Intelligent interaction model, used to obtain interaction results, complete missing information in natural language, and create an intent recognition sample library; Text classification module, used to identify the intent of natural language; Named entity recognition module, used to identify card parameters and interface parameters; The calling module is used to call the corresponding card and score and sort the return results of the interface.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent interaction method described in any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: include: Memory, for storing computer instructions; A processor, configured to execute the computer instructions to implement the steps of the intelligent interaction method according to any one of claims 1 to 7.

Images