User demand processing method, system and computing device

Abstract

The embodiment of the present application discloses a method, system and computing device for processing user needs. The method includes: obtaining user needs input by the user; inputting the user needs into the planning model, the planning model determines the target response tool that matches the user needs, and calls the target response tool to process the user needs to obtain response data, and the planning model is obtained based on multiple rounds of multi-tool mixed sample training; the response data is fed back to the user. This solution can identify the target response tool that is currently most suitable for responding to user needs, so as to call the target response tool to respond to user needs to obtain response data, thereby improving the accuracy of the response; and it can flexibly switch between multiple response tools to improve the response efficiency; and the planning model is trained using multiple rounds of multi-tool mixed samples, which can improve the context understanding ability of the planning model and further improve the accuracy of the response.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically to a user demand processing method, system, computing device, computer storage medium and computer program product. Background Art

[0002] With the continuous development of science and technology and society, the emergence of intelligent question-answering systems has greatly facilitated people's work and life. Users can input relevant questions through intelligent question-answering systems and quickly get matching answer data.

[0003] However, the inventors found that there are at least the following problems in the prior art during implementation: After the user inputs the user question, the prior art system only uses a single processing method to process the user question. For example, only private library retrieval enhancement is used to process the user question, or only the frequently asked questions model is used to process the user question. The single processing method in the prior art cannot adapt to different types of user questions in different scenarios, the scope of user question processing is limited, and the accuracy of user question processing is insufficient. Summary of the invention

[0004] In view of the above problems, the present application is proposed to provide a user demand processing method, system, computing device, computer storage medium and computer program product that overcome the above problems or at least partially solve the above problems.

[0005] According to a first aspect of the present application, a method for processing user needs is provided, comprising:

[0006] Obtain user input for user needs;

[0007] The user demand is input into the planning model obtained after fine-tuning the preset large model, so that the planning model determines the target response tool matching the user demand from multiple response tools, and calls the target response tool to process the user demand to obtain response data; wherein the planning model is obtained based on multiple rounds of multi-tool mixed sample training;

[0008] The reply data is fed back to the user.

[0009] In an optional implementation, before the invoking of the target reply tool to process the user demand and obtain reply data, the method further comprises: a planning model extracting tool parameters from the user demand;

[0010] The step of calling the target answering tool to process the user demand and obtain the answering data includes: providing the tool parameters to the target answering tool.

[0011] In an optional implementation, after the planning model extracts tool parameters from the user requirements, the method further includes:

[0012] The planning model verifies whether the tool parameters meet the parameter requirements of the target response tool;

[0013] If not, a parameter prompt response is fed back to the user.

[0014] In an optional implementation manner, before feeding back the reply data to the user, the method further includes:

[0015] Determining whether the response data is natural language data;

[0016] If not, converting the response data into natural language response data;

[0017] Then the feeding back of the reply data to the user includes: feeding back the natural language reply data to the user.

[0018] In an optional implementation, the multi-round multi-tool mixed sample is generated by:

[0019] For each reply tool, generating description data of the reply tool and a tool dialogue sample set; wherein the tool dialogue sample set of the reply tool includes a plurality of dialogue samples of the reply tool;

[0020] Generate a general conversation sample set; wherein the general conversation sample set includes a plurality of non-tool conversation samples;

[0021] A multi-round multi-tool mixed sample is generated based on the tool conversation sample set of different response tools and the universal conversation sample set.

[0022] In an optional embodiment, the reply tool includes at least multiple of the following tools:

[0023] Task-based answering tools, FAQ tools, and knowledge base search enhancement tools.

[0024] According to a second aspect of the present application, a user demand processing system is provided, comprising:

[0025] A user interaction module is used to obtain user requirements input by the user, input the user requirements into the planning model obtained after fine-tuning the preset large model; and feed back the response data to the user;

[0026] A planning model, used to determine a target response tool that matches the user's needs from multiple response tools, and to obtain response data after invoking the target response tool to process the user's needs; wherein the planning model is obtained based on multiple rounds of multi-tool mixed sample training;

[0027] The response tool is used to obtain response data after processing the user demand under the call of the planning model.

[0028] In an optional implementation, the planning model is used to extract tool parameters from the user requirements and provide the tool parameters to the target answering tool.

[0029] In an optional embodiment, the planning model is used to: verify whether the tool parameters meet the parameter requirements of the target answer tool;

[0030] If not, a parameter prompt response is fed back to the user.

[0031] In an optional implementation, the planning model is used to: determine whether the reply data is natural language data; if not, convert the reply data into natural language reply data;

[0032] The user interaction module is used to feed back the natural language response data to the user.

[0033] In an optional embodiment, the system further includes a training module for generating, for each reply tool, description data of the reply tool and a tool dialogue sample set; wherein the tool dialogue sample set of the reply tool includes a plurality of dialogue samples of the reply tool;

[0034] Generate a general conversation sample set; wherein the general conversation sample set includes a plurality of non-tool conversation samples;

[0035] A multi-round multi-tool mixed sample is generated based on the tool conversation sample set of different response tools and the universal conversation sample set.

[0036] In an optional embodiment, the reply tool includes at least multiple of the following tools:

[0037] Task-based answering tools, FAQ tools, and knowledge base search enhancement tools.

[0038] According to a third aspect of the present application, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0039] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned user demand processing method.

[0040] According to a fourth aspect of the present application, a computer storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned user demand processing method.

[0041] According to a fifth aspect of the present application, a computer program product is provided, comprising at least one executable instruction, wherein the executable instruction enables a processor to perform operations corresponding to the above-mentioned user demand processing method.

[0042] The user demand processing method, system, computing device, computer storage medium and computer program product provided by the present application use the planning model fine-tuned by the large model to identify the target response tool that is currently most suitable for answering user needs, thereby calling the target response tool to answer user needs to obtain response data, improve the accuracy of user demand responses, and be able to flexibly switch between multiple response tools to improve response efficiency and shorten response time. In addition, multiple response tools are independent of each other, reducing knowledge interference between tools, achieving knowledge isolation between different response tools, and further improving response accuracy; and the embodiment of the present application uses multiple rounds of multi-tool mixed samples to train the planning model, thereby improving the context understanding ability of the planning model and further improving response accuracy.

[0043] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 A schematic diagram of a process for processing user needs provided in the first embodiment of the present application is shown;

[0046] Figure 2 A schematic diagram of a reply tool provided in Embodiment 1 of the present application is shown;

[0047] Figure 3 A schematic diagram of a user demand processing flow provided in Embodiment 1 of the present application is shown;

[0048] Figure 4 A schematic diagram of a user demand processing architecture provided in Embodiment 1 of the present application is shown;

[0049] Figure 5 A schematic diagram of a knowledge base building process provided in Example 1 of the present application is shown;

[0050] Figure 6 A schematic diagram of a planning model training method provided in Embodiment 2 of the present application is shown;

[0051] Figure 7 A schematic diagram of describing data provided in Embodiment 2 of the present application is shown;

[0052] Figure 8 A schematic diagram showing a conversation sample provided in Embodiment 2 of the present application is shown;

[0053] Fig. 9 A schematic diagram of the structure of a user demand processing system provided in Embodiment 3 of the present application is shown;

[0054] Fig.10 A schematic diagram of the structure of a computing device provided in Embodiment 4 of the present application is shown. DETAILED DESCRIPTION

[0055] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0056] Embodiment 1

[0057] Figure 1 A flow chart of a method for processing user demands provided in Embodiment 1 of the present application is shown.

[0058] Among them, the user demand processing method provided in the embodiment of the present application is applied to the intelligent question and answer scenario, and can be widely used in customer service, smart assistants, smart homes, online education, medical health and other fields.

[0059] like Figure 1 As shown, the method comprises the following steps:

[0060] Step S110, obtaining user requirements input by the user.

[0061] The embodiment of the present application provides a question input entrance for the user, through which the user can input corresponding user needs, which include various types of user questions and / or user requests, etc.

[0062] Step S120, input the user needs into the planning model obtained after fine-tuning the preset large model, so that the planning model can determine the target response tool that matches the user needs from multiple response tools, and call the target response tool to process the user needs to obtain response data; wherein, the planning model is obtained based on multiple rounds of multi-tool mixed sample training.

[0063] In the embodiment of the present application, each reply model, reply function, reply tool, etc. that can independently handle user needs is encapsulated into a corresponding reply tool. Each reply tool can independently handle user needs and provide a corresponding calling interface to the planning model. The reply tools are independent of each other. Optionally, the reply tool includes at least a plurality of the following tools: a task-based reply tool, an FAQ tool, and a knowledge base retrieval enhancement tool. Among them, the task-based reply tool usually obtains relevant reply data through parameter matching, etc.; the FAQ tool is a frequently asked questions tool (Frequently Asked Questions), which usually obtains relevant reply data through similarity detection; the knowledge base retrieval enhancement tool usually retrieves relevant knowledge from the constructed knowledge base first and uses it as the data basis for the large model to output the reply. In the embodiment of the present application, the knowledge base retrieval enhancement tool is specifically a RAG (Retrieval Augmented Generation) module. The RAG of the present application and the planning model use the same base large model, thereby streamlining the system structure and improving resource utilization. The above-mentioned tool division is only a feasible division method. During the implementation process, the corresponding reply tool can be selected according to the actual situation.

[0064] In addition, the embodiment of the present application is based on a preset large model, and based on a multi-round multi-tool mixed sample, a large model fine-tuning algorithm is used to train the preset large model to obtain a planning model. Among them, the embodiment of the present application does not limit the preset large model and the large model fine-tuning algorithm. The multi-round multi-tool mixed sample is a sample that contains multiple rounds of dialogue, and the multiple rounds of dialogue correspond to multiple reply tools.

[0065] After the user demand is input into the planning model, the planning model will determine the answer tool that best matches the user demand at present, and the answer tool that best matches is called the target answer tool. Then the target answer tool is called to process the user demand to obtain the corresponding answer data. It should be understood here that the user can conduct multiple rounds of questions and answers, and the planning model determines the target answer tool that matches this round according to the user demand in each round of questions and answers. The target answer tool of each round in multiple rounds of questions and answers can be the same or different. Optionally, after the user inputs the user demand, when conducting the first round of questions and answers, the planning model determines the target answer tool that matches this round according to the user demand; when conducting non-first round of questions and answers, the planning model determines the target answer tool corresponding to this round of questions and answers according to the user demand, the user input in the historical questions and answers, the answer data, and the answer tool information of the source of the answer data. That is, this application is different from the traditional large model that analyzes the historical question and answer data as the context data of this round of answers, but further adds the target answer tool used in the historical answer process on the basis of the historical question and answer data, so as to improve the determination accuracy of the subsequent target answer tool.

[0066] In an optional implementation, the planning model of the present application is obtained based on fine-tuning of the large model, so that the planning model has the question-answering capability of the large model. When the planning model determines that the chat module rather than the various docked reply tools matches the user's needs, it uses its own chat module to generate reply data. The chat module can use the general knowledge, understanding and reasoning ability of the model itself to complete the user's instructions, thereby improving the reply efficiency, so that the planning module can also provide functions such as general instruction compliance.

[0067] refer to Figure 2 The planning model itself has a chat module, and is connected to tools 1-7 respectively, among which tools 1-3 belong to task-based answering tools, tools 4-5 belong to FAQ tools, and tools 6-7 belong to knowledge base search enhancement tools, and each tool is independent of each other. The chat module of the planning model itself can also be used as a reply tool. The planning model determines the tool that best matches the user's needs from the connected tools (including the chat module) to process the user's needs.

[0068] In an optional implementation, the planning model extracts tool parameters from user requirements, and provides the tool parameters to the target response tool, so that the target response tool generates response data according to the tool parameters.

[0069] Further optionally, after the planning model extracts the tool parameters from the user requirements, the planning model verifies whether the extracted tool parameters meet the parameter requirements of the target reply tool; if not, a parameter prompt reply is fed back to the user; if so, a subsequent target reply tool call operation is performed. Specifically, the planning model has the parameter requirements of each reply tool. If the extracted tool parameters do not meet the parameter requirements (such as a certain necessary parameter is missing, or the parameter format does not meet the set requirements), the planning model returns a parameter prompt reply to the user to guide the user to enter the correct tool parameters, etc.

[0070] Step S130, feeding back the reply data to the user.

[0071] The final response data is fed back to the user. The process of the planning model calling the target response tool is invisible to the user, thus avoiding interference to the user and improving the user experience.

[0072] In an optional implementation, in order to improve the user experience, after obtaining the reply data fed back by the target reply tool, it is determined whether the reply data is natural language data; if so, the natural language reply data is directly returned to the user; if not, the reply data is converted into natural language reply data, and the natural language reply data is fed back to the user.

[0073] refer to Figure 3 , the front-end obtains the question and processes it through the planning model. If it is determined that the target question answering model matching the question is a chat module, the chat module of the planning model itself is used to process the user's needs to obtain a chat answer and return it to the user; if it is determined that the target question answering model matching the question is a task-based tool, the task-based tool is called, and parameter verification and other clarification replies are performed during the calling process, and finally the tool's reply data is obtained and returned to the user; if it is determined that the target question answering model matching the question is a FAQ tool, the FAQ performs content recall and calculates the similarity between the user's needs and the recalled content, and then returns a reply that meets the similarity threshold or returns a fallback reply; if it is determined that the target question answering model matching the question is a document question and answer tool (knowledge base retrieval enhancement tool), keyword recall and / or vector recall are performed from the corresponding database, and the question and answer large model is input through recall fusion, reordering, text expansion, de-duplication merging, numbering, prompt word splicing, etc., and the output of the question and answer large model is returned to the user.

[0074] refer to Figure 4, the architecture adopted by the embodiment of the present application includes: data and storage layer, model layer, basic capability layer, application layer, and business layer. Among them, the data and storage layer contains various types of data, such as session data, FAQ library, etc.; the model layer contains various models, such as LLM planning model, LLM question and answer model that knowledge base retrieval enhancement tool relies on, etc.; the basic capability layer contains various basic processing capabilities, such as data indexing capability, tool calling capability (Function Calling), RAG, etc.; the application layer contains various tool abstractions, such as chatting, FAQ question and answer, task-based question and answer / tool ​​calling, document question and answer, etc.; the business layer contains various adapted businesses, such as customer service business, intelligent assistant, e-commerce shopping guide, smart home and other business scenarios.

[0075] The knowledge base construction process can refer to Figure 5 , the knowledge base can be applied to the knowledge base retrieval enhancement tool or to FAQ. When used in the knowledge base retrieval enhancement tool, relevant document data is submitted and stored. Further, document parsing, document segmentation, vectorization and storage, as well as word segmentation and storage are performed. When used in FAQ, FAQ data can be uploaded, then vectorized and a vectorized index is established.

[0076] It can be seen that the user demand processing method provided by the embodiment of the present application uses the planning model fine-tuned by the large model to identify the target response tool that is most suitable for answering user needs at the moment, thereby calling the target response tool to answer user needs to obtain response data, improve the accuracy of user demand responses, and be able to flexibly switch between multiple response tools to improve response efficiency and shorten response time. Moreover, multiple response tools are independent of each other, which reduces knowledge interference between tools, realizes knowledge isolation between different response tools, and further improves response accuracy; and the embodiment of the present application uses multiple rounds of multi-tool mixed samples to train the planning model, thereby improving the context understanding ability of the planning model and further improving response accuracy.

[0077] Embodiment 2

[0078] Figure 6 The flowchart of a planning model training method provided in the second embodiment of the present application is shown. The planning module is specifically the planning model described in other embodiments.

[0079] Specifically, Figure 6 As shown, the method comprises the following steps:

[0080] Step S610, for each reply tool, generating description data of the reply tool and a tool dialogue sample set; wherein the tool dialogue sample set of the reply tool includes a plurality of dialogue samples of the reply tool.

[0081] For each reply tool, the description data of the reply tool and the tool dialogue sample set are generated. The description data includes tool description data and parameter description data, such as tool description data including tool identification, tool calling interface, the range of replies that the tool can provide, etc.; parameter description data is used to describe parameter requirements, etc.

[0082] In an optional implementation, in order to improve the generalization ability of the planning model, the present application obtains basic description data (including basic tool description data and basic parameter description data), and the basic description data can be obtained by configuration and the like. The basic description data is further processed to obtain extended description data, and the basic description data and the extended description data together constitute the description data in the embodiment of the present application. Among them, the extended description data can be obtained by the following methods: expanding the basic description data with similar words (such as replacing and inserting with synonyms and near-synonyms), switching the character positions of the basic description data, back-translating the data, etc. Alternatively, pre-train an extended model, and use the extended model to expand the basic description data to obtain extended description data, and so on.

[0083] The tool dialogue sample set of the reply tool includes multiple dialogue samples of the reply tool. The dialogue sample of the reply tool may include one or more dialogues for the reply tool. The dialogue sample may cover various stages such as tool call request, parameter clarification, tool call result feedback, etc. For example, Figure 7 A description sample of a reply tool is shown, which describes the name, purpose, etc. of the reply tool; Figure 8 A sample diagram of a dialogue of a reply tool is shown.

[0084] Step S620, generating a general conversation sample set; wherein the general conversation sample set includes a plurality of non-tool conversation samples.

[0085] The general data set includes at least one other instruction data other than the data involved in the reply tool, such as small talk dialogue data, general instructions, etc. In an optional implementation, historical non-task dialogue data of a preset large model can be collected and added to the general dialogue sample set to realize automatic generation of the sample set.

[0086] Step S630, generating a multi-round multi-tool mixed sample based on the tool dialogue sample sets of different response tools and the general dialogue sample set.

[0087] In an optional implementation, each response tool can be used as a corresponding tool element to generate all permutations and combinations of each tool element and a common element (corresponding to a common dialogue sample set). For example, the planning model is connected to three response tools A, B, and C, and the common element is called T, then the permutations and combinations obtained include: A, B, C, T, AB, AC, AT, BC, BT, CT, ABC, ABT, ACT, BCT, ABCT. For each combination, at least one sample set corresponding to the combination is determined, and dialogue samples are randomly extracted from each corresponding sample set to obtain a multi-round multi-tool mixed sample. This method can make the mixed samples cover different combinations, improve sample balance, and improve the prediction accuracy of the trained planning model.

[0088] Further optionally, the usage frequency of different response tools in the historical period can be obtained in advance, and corresponding weight coefficients can be assigned to different response tools according to the usage frequency, wherein the weight coefficient is positively correlated with the usage frequency. Then, for each combination, a combined weight coefficient is generated according to the weight coefficient of the tool corresponding to the combination (for example, the normalized value of the product of the weight coefficients of the tools corresponding to the combination can be used as the combined weight coefficient), and the number of mixed samples corresponding to the combination is determined according to the combined weight coefficient. The number of mixed samples corresponding to the combination is positively correlated with the combined weight coefficient, that is, the larger the combined weight coefficient, the more mixed samples generated by the combination.

[0089] In another optional implementation, conversation samples may be randomly extracted from different conversation sample sets and randomly inserted to generate multi-round multi-tool mixed samples. This method can improve the efficiency of generating mixed samples.

[0090] Step S640, a planning model is obtained by fine-tuning and training the preset large model using multiple rounds of multi-tool mixed samples.

[0091] For example, large language model fine-tuning techniques such as LoRA (Low-Rank Adaptation) and full parameter fine-tuning algorithms can be used to train the preset large model based on multiple rounds of multi-tool mixed samples. The model obtained when the training end conditions are met is the planning model.

[0092] In an optional implementation, when a new reply tool needs to be connected, the description data of the new reply tool is constructed, and the planning model is made to learn the tool attribute information of the new reply tool by means of prompt word engineering, etc. Further, optionally, a tool dialogue sample set of the new reply tool can also be constructed, and the tool dialogue sample set is combined with other dialogue sample sets to generate a new mixed sample to train the planning model.

[0093] In an optional implementation, when a certain reply tool needs to be removed from the shelves, it is only necessary to exclude the removed reply tool from the matching tool range through prompt word engineering.

[0094] In an optional implementation, after the planning model is obtained, the planning model may be updated based on information such as user feedback when the planning model processes user needs, so as to further improve the prediction accuracy of the planning model.

[0095] It can be seen that the planning model training method provided in the embodiment of the present application mixes different response tools and conversation data of general conversations to obtain mixed samples, and uses the mixed samples to train the planning model, which can improve the planning model's ability to analyze context and improve the processing accuracy of the planning model in multi-round conversations; and it can flexibly increase or decrease the docking response tools to improve the scalability of the planning model.

[0096] Embodiment 3

[0097] Fig. 9 FIG. 1 shows a schematic diagram of a user demand processing system provided in Embodiment 3 of the present application. Fig. 9 As shown, the system 900 includes: a user interaction module 910 , a planning model 920 , and multiple response tools 930 .

[0098] The user interaction module 910 is used to obtain user requirements input by the user, input the user requirements into the planning model obtained after fine-tuning the preset large model; and feed back the response data to the user;

[0099] A planning model 920, used to determine a target response tool that matches the user's needs from multiple response tools, and to obtain response data after invoking the target response tool to process the user's needs; wherein the planning model is obtained based on multiple rounds of multi-tool mixed sample training;

[0100] The reply tool 930 is used to obtain reply data after processing the user demand under the call of the planning model.

[0101] In an optional implementation, the planning model 920 is used to extract tool parameters from the user requirements and provide the tool parameters to the target answer tool.

[0102] In an optional implementation, the planning model 920 is used to: verify whether the tool parameters meet the parameter requirements of the target answer tool;

[0103] If not, a parameter prompt response is fed back to the user.

[0104] In an optional implementation, the planning model 920 is used to: determine whether the reply data is natural language data; if not, convert the reply data into natural language reply data;

[0105] The user interaction module 910 is used to feed back the natural language response data to the user.

[0106] In an optional embodiment, the system further includes a training module (not shown in the figure), which is used to generate description data of the reply tool and a tool dialogue sample set for each reply tool; wherein the tool dialogue sample set of the reply tool includes a plurality of dialogue samples of the reply tool;

[0107] Generate a general conversation sample set; wherein the general conversation sample set includes a plurality of non-tool conversation samples;

[0108] A multi-round multi-tool mixed sample is generated based on the tool conversation sample set of different response tools and the universal conversation sample set.

[0109] In an optional embodiment, the reply tool includes at least multiple of the following tools:

[0110] Task-based answering tools, FAQ tools, and knowledge base search enhancement tools.

[0111] It can be seen that the user demand processing system provided by the embodiment of the present application uses the planning model fine-tuned by the large model to identify the target response tool that is most suitable for answering user needs at the moment, thereby calling the target response tool to answer user needs to obtain response data, improve the accuracy of user demand responses, and can flexibly switch between multiple response tools to improve response efficiency and shorten response time. Moreover, multiple response tools are independent of each other, which reduces knowledge interference between tools, realizes knowledge isolation between different response tools, and further improves response accuracy; and the embodiment of the present application uses multiple rounds of multi-tool mixed samples to train the planning model, thereby improving the context understanding ability of the planning model and further improving response accuracy.

[0112] Embodiment 4

[0113] Fig.10 A schematic diagram of the structure of a computing device provided in Embodiment 4 of the present application is shown. The specific embodiments of the present application do not limit the specific implementation of the computing device.

[0114] like Fig.10 As shown, the computing device may include: a processor (processor) 1002 , a communication interface (Communications Interface) 1004 , a memory (memory) 1006 , and a communication bus 1008 .

[0115] The processor 1002, the communication interface 1004, and the memory 1006 communicate with each other via the communication bus 1008. The communication interface 1004 is used to communicate with other devices such as a client or other server network elements. The processor 1002 is used to execute the program 1010, which can specifically execute the relevant steps in the above-mentioned embodiment of the method for processing user requirements of a computing device.

[0116] Specifically, the program 1010 may include program codes, which include computer operation instructions.

[0117] The processor 1002 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0118] The memory 1006 is used to store the program 1010. The memory 1006 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. The program 1010 may be specifically used to enable the processor 1002 to perform the operations in any of the above embodiments.

[0119] Embodiment 5

[0120] Embodiment 5 of the present application provides a non-volatile computer storage medium, which stores at least one executable instruction or computer program, and the executable instruction or computer program can enable the processor to perform operations corresponding to the user demand processing method in any of the above method embodiments.

[0121] Embodiment 6

[0122] Embodiment 6 of the present application provides a computer program product, which includes at least one executable instruction or computer program, and the executable instruction or computer program can enable a processor to execute operations corresponding to the user demand processing method in any of the above method embodiments.

[0123] In summary, according to the computing device, computer storage medium and computer program product provided by this embodiment, the planning model fine-tuned by the large model is used to identify the target response tool that is currently most suitable for answering user needs, so as to call the target response tool to answer user needs to obtain response data, improve the accuracy of responses to user needs, and be able to flexibly switch between multiple response tools, improve response efficiency, and shorten response time. Moreover, multiple response tools are independent of each other, which reduces knowledge interference between tools, realizes knowledge isolation between different response tools, and further improves response accuracy; and the embodiment of this application uses multiple rounds of multi-tool mixed samples to train the planning model, thereby improving the context understanding ability of the planning model and further improving response accuracy.

[0124] The algorithm or display provided here are not inherently related to any specific computer, virtual system or other equipment. Various general systems can also be used together with the teaching based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application described here, and the above description of specific languages ​​is to disclose the best mode of implementation of the present application.

[0125] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0126] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0127] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0128] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0129] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some or all components according to the embodiments of the present application. The application can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing a part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0130] It should be noted that the above embodiments illustrate the present application rather than limit the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be constructed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The present application may be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

Claims

1. A method for processing user needs, characterized in that: include: Obtain user input for user needs; The user demand is input into the planning model obtained after fine-tuning the preset large model, so that the planning model determines the target response tool matching the user demand from multiple response tools, and calls the target response tool to process the user demand to obtain response data; wherein the planning model is obtained based on multiple rounds of multi-tool mixed sample training; The reply data is fed back to the user.

2. The user demand processing method according to claim 1, characterized in that: Before the target answering tool is called to process the user demand and obtain answering data, the method further comprises: a planning model extracting tool parameters from the user demand; The step of calling the target answering tool to process the user demand and obtain the answering data includes: providing the tool parameters to the target answering tool.

3. The user demand processing method according to claim 2, characterized in that: After the planning model extracts tool parameters from the user requirements, the method further includes: The planning model verifies whether the tool parameters meet the parameter requirements of the target response tool; If not, a parameter prompt reply is fed back to the user.

4. The user demand processing method according to claim 1, characterized in that: Before feeding back the reply data to the user, the method further includes: Determining whether the response data is natural language data; If not, converting the response data into natural language response data; Then the feeding back of the reply data to the user includes: feeding back the natural language reply data to the user.

5. The method for processing user needs according to any one of claims 1 to 4, characterized in that: The multi-round multi-tool mixed sample is generated in the following way: For each reply tool, generating description data of the reply tool and a tool dialogue sample set; wherein the tool dialogue sample set of the reply tool includes a plurality of dialogue samples of the reply tool; Generate a general conversation sample set; wherein the general conversation sample set includes a plurality of non-tool conversation samples; A multi-round multi-tool mixed sample is generated based on the tool conversation sample set of different response tools and the general conversation sample set.

6. The method for processing user demand according to any one of claims 1 to 4, characterized in that: The reply tool includes at least a plurality of the following tools: Task-based answering tools, FAQ tools, and knowledge base search enhancement tools.

7. A user demand processing system, characterized in that: include: A user interaction module is used to obtain user requirements input by users, and input the user requirements into a planning model obtained after fine-tuning the preset large model; and feeding back the response data to the user; A planning model, used to determine a target response tool that matches the user's needs from multiple response tools, and to obtain response data after invoking the target response tool to process the user's needs; wherein the planning model is obtained based on multiple rounds of multi-tool mixed sample training; The response tool is used to obtain response data after processing the user demand under the call of the planning model.

8. A computing device, characterized in that include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the user demand processing method as described in any one of claims 1-6.

9. A computer storage medium, characterized in that The storage medium stores at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the user demand processing method as described in any one of claims 1-6.

10. A computer program product, characterized in that It includes at least one executable instruction, and the executable instruction enables the processor to execute the operation corresponding to the user demand processing method as described in any one of claims 1-6.

Images