Task processing method and task processing system

Abstract

The embodiment of the invention provides a task processing method and a task processing system.The task processing method is applied to a service unit in a task processing platform, the task processing platform comprises the service unit and a plurality of processing components, and the task processing method comprises the steps that a task processing request is received, and the task processing request carries to-be-processed data; the to-be-processed data and the component description information are input into a task processing model, at least one target processing component corresponding to the task processing request is determined, and the task processing model is a deep learning model; and calling at least one target processing component to process the to-be-processed data to obtain a task processing result corresponding to the task processing request. By integrating the multiple processing components on the task processing platform, at least one target processing component can be directly called to process the to-be-processed data, the task processing efficiency is improved, the processing capacity of the task processing model can be expanded through the multiple processing components, and task processing is more comprehensive.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and more particularly to a task processing method. One or more embodiments of this specification also relate to a task processing system, a task processing device, a computing device, a computer-readable storage medium, and a computer program. Background Art

[0002] With the development of computer technology, building an intelligent system that can conduct multi-modal conversational interactions like humans has always been a long-term goal for achieving artificial general intelligence. Although traditional intelligent systems have achieved a high level in text conversations, they still have shortcomings in interactions with more modalities such as voice, images, and videos.

[0003] Currently, the task input by the user can be decomposed into several subtasks, and the model corresponding to each subtask can be selected in the model library, so as to use the corresponding model to process the subtask and output the processing result. However, the above solution requires jumping to the model library to select the model corresponding to each subtask, resulting in extremely low task processing efficiency. Therefore, an efficient task processing solution is urgently needed. Summary of the invention

[0004] In view of this, an embodiment of this specification provides a task processing method. One or more embodiments of this specification also relate to a task processing system, a task processing device, a computing device, a computer-readable storage medium and a computer program to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a task processing method is provided, which is applied to a service unit in a task processing platform, wherein the task processing platform includes a service unit and a plurality of processing components, and the method includes:

[0006] receiving a task processing request, wherein the task processing request carries data to be processed;

[0007] Inputting the data to be processed and the component description information into a task processing model, and determining at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model;

[0008] At least one target processing component is called to process the data to be processed, and a task processing result corresponding to the task processing request is obtained.

[0009] According to a second aspect of an embodiment of this specification, a task processing device is provided, which is applied to a service unit in a task processing platform. The task processing platform includes a service unit and a plurality of processing components. The device includes:

[0010] A first receiving module is configured to receive a task processing request, wherein the task processing request carries data to be processed;

[0011] A first input module is configured to input the data to be processed and the component description information into a task processing model, and determine at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model;

[0012] The first calling module is configured to call at least one target processing component to process the data to be processed and obtain a task processing result corresponding to the task processing request.

[0013] According to a third aspect of the embodiments of this specification, there is provided a task processing system, including a communication interface and a calling unit;

[0014] A communication interface, used to receive a processing request sent by a client and forward the processing request to a calling unit, wherein the processing request carries data to be queried;

[0015] A calling unit is used to call a query model to process the data to be queried and the processing model description information in response to a processing request, determine at least one target processing model corresponding to the data to be queried, wherein the query model is a deep learning model; and send model information of at least one target processing model to the client through a communication interface.

[0016] According to a fourth aspect of an embodiment of this specification, a computing device is provided, including:

[0017] Memory and processor;

[0018] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method provided in the first aspect above are implemented.

[0019] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the method provided in the first aspect are implemented.

[0020] According to a sixth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the method provided in the first aspect above.

[0021] The task processing method provided in one embodiment of the present specification is applied to a service unit in a task processing platform, wherein the task processing platform includes a service unit and multiple processing components, receives a task processing request, wherein the task processing request carries data to be processed; inputs the data to be processed and component description information into a task processing model, determines at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model; calls at least one target processing component to process the data to be processed, and obtains a task processing result corresponding to the task processing request. By integrating multiple processing components on the task processing platform, at least one target processing component can be directly called to process the data to be processed, thereby improving the task processing efficiency, and through multiple processing components, on the one hand, the text generation capability of the task processing model in natural language tasks can be improved, and on the other hand, the multimodal generation capability missing in the task processing model can be compensated, making the processing capability of the task processing model more comprehensive. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart of a question-and-answer method;

[0023] Figure 2a is an architecture diagram of a first task processing system provided by an embodiment of this specification;

[0024] Figure 2b is an architecture diagram of a second task processing system provided by an embodiment of this specification;

[0025] Figure 3 is an architecture diagram of a third task processing system provided by an embodiment of this specification;

[0026] Figure 4 is a flowchart of a task processing method provided by an embodiment of this specification;

[0027] Figure 5 It is a schematic diagram of generating sample data in a task processing method provided in one embodiment of this specification;

[0028] Figure 6 is a flow chart of an automatic question-answering method provided by an embodiment of this specification;

[0029] Figure 7 is a flowchart of another task processing method provided by an embodiment of this specification;

[0030] Figure 8 is a processing flow chart of a task processing method provided by an embodiment of this specification;

[0031] Fig. 9 is a task example diagram in a task processing method provided by an embodiment of this specification;

[0032] Fig.10 This is a schematic diagram of an automatic question-and-answer interface provided by an embodiment of this specification;

[0033] Fig.11 is a structural diagram of a task processing device provided by an embodiment of this specification;

[0034] Fig.12 is a structural diagram of an automatic question-answering device provided by an embodiment of this specification;

[0035] Fig.13 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0036] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0037] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0038] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0039] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0040] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, which usually contains hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than 10 trillion model parameters. A large model can also be called a foundation model / foundation model. The large model is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as a large-scale language model (LLM, Large Language Model), a multi-modal pre-training model, etc.

[0041] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image description (IC, Image Caption), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0042] First, the terms involved in one or more embodiments of this specification are explained.

[0043] Large-scale language models: Large-scale language models refer to neural network models with billions or even trillions of parameters, which can be used for natural language processing tasks such as language generation, question answering, and text classification. These models are usually trained with large amounts of text data and can perform well on a variety of different tasks.

[0044] Generative AI: Generative AI refers to a system based on artificial intelligence technology that can generate natural language text, images, audio or video data. Unlike traditional tasks such as classification and regression, the models in generative AI are creative and imaginative, and can generate new and unseen content based on input conditions. Common generative AIs include natural language generation models, image generation models, etc.

[0045] ToolAPI integration: ToolAPI integration refers to an integration method that encapsulates the artificial intelligence model into an application programming interface (API) and provides it to developers or users. Through the application programming interface, developers or users can use the functions provided by the artificial intelligence model, such as language translation, image recognition, speech synthesis, etc., to add richer functions to the application.

[0046] General artificial intelligence: General artificial intelligence refers to an artificial intelligence system that has the same broad intelligence capabilities as humans and can continuously learn and adapt to various environments and tasks.

[0047] With the development of computer technology, building an intelligent system that can conduct multi-modal conversational interactions like humans has always been a long-term goal for achieving artificial general intelligence. Although traditional intelligent systems have achieved a high level in text conversations, they still have shortcomings in interactions with more modalities such as voice, images, and videos.

[0048] See also Figure 1 , Figure 1 A flowchart of a question-answering method is shown, which obtains the description of the model in the model library and integrates it into the prompt to establish a connection with the big model. After that, the big model can act as a brain to determine the answer to the user's question. Figure 1 As shown in Figure 1, the workflow of the question-answering method includes four stages: task planning, model selection, task execution, and response generation. The task planning stage can decompose the user input into several subtasks and determine the type and difficulty of each subtask. The model selection stage can search for a suitable model from the model library to perform the subtask. The task execution stage can use the selected model to predict the user's input. The response generation stage can collect the model processing results.

[0049] However, the above solution requires jumping to the model library to select the model corresponding to each subtask, and its work execution process is a tree structure, resulting in extremely low task processing efficiency.

[0050] In order to solve the above problems, the embodiment of this specification proposes a scalable and learnable large model enhancement framework based on the model as a service sharing platform (Model Scope), which integrates multiple artificial intelligence models on the task processing platform in the form of processing components to form a generative artificial intelligence and a general artificial intelligence system, thereby realizing task processing. Among them, the model as a service platform can provide developers with rich, easy-to-use, low-cost one-stop model services, making model application simpler. Model services include rich pre-trained SOTA models, diverse and open data sets, one line of code to use model reasoning capabilities, ten lines of code to quickly build exclusive industry models, ready-to-use online development platforms, flexible model frameworks and deployment methods, and rich teaching content and technical resources, so that knowledge questions and answers, more detailed understanding of needs, and recommendation of more suitable models can be achieved, while also providing innovative experiences, such as simple tasks, complex tasks, and multiple application programming interface planning (API planning) calls.

[0051] Specifically, the embodiment of this specification proposes a task processing method for a service unit in a task processing platform, receiving a task processing request, wherein the task processing request carries data to be processed; inputting the data to be processed and component description information into a task processing model, determining at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model; calling at least one target processing component to process the data to be processed, and obtaining a task processing result corresponding to the task processing request. Applying the above scheme, by integrating multiple processing components on the task processing platform, at least one target processing component can be directly called to process the data to be processed, thereby improving the task processing efficiency, and, through multiple processing components, on the one hand, the text generation capability of the task processing model in natural language tasks can be improved, and on the other hand, the multimodal generation capability missing in the task processing model can be compensated, such as video generation, picture generation, graphic-text dialogue, speech synthesis, knowledge question and answer, address structuring, etc., making the processing capability of the task processing model more comprehensive.

[0052] In this specification, a task processing method is provided. This specification also involves a task processing system, a task processing apparatus, a computing device, a computer-readable storage medium and a computer program, which are described in detail one by one in the following embodiments.

[0053] See also Figure 2a , Figure 2a The first task processing system provided by an embodiment of the present specification is shown in the architecture diagram. The task processing system may include a client 100 and a task processing platform 200. The task processing platform 200 includes a service unit 202 and a plurality of processing components 204.

[0054] The client 100 is used to send a task processing request to the task processing platform 200, wherein the task processing request carries data to be processed;

[0055] The service unit 202 is used to input the data to be processed and the component description information into the task processing model, determine at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model; call at least one target processing component to process the data to be processed, and obtain the task processing result corresponding to the task processing request;

[0056] In an optional embodiment of the present specification, the service unit 200 is further used to send the task processing result to the client 100 ; the client 100 is further used to receive the task processing result sent by the service unit 202 .

[0057] It should be noted that multiple processing components can handle different types of tasks, such as video generation tasks, image generation tasks, text-to-picture dialogue tasks, speech synthesis tasks, knowledge question-and-answer tasks, address structuring tasks, English-Chinese translation tasks, named entity recognition tasks, Chinese-English translation tasks, general information extraction tasks, etc.

[0058] By applying the solution of the embodiments of this specification, multiple processing components are integrated on the task processing platform, so that at least one target processing component can be directly called to process the data to be processed, thereby improving the task processing efficiency. In addition, through multiple processing components, on the one hand, the text generation capability of the task processing model in natural language tasks can be improved, and on the other hand, the missing multimodal generation capability of the task processing model can be compensated, making the processing capability of the task processing model more comprehensive.

[0059] See also Figure 2b , Figure 2b The architecture diagram of the second task processing system provided by an embodiment of the present specification is shown. The task processing system may include a communication interface 206 and a calling unit 208;

[0060] The communication interface 206 is used to receive a processing request sent by a client and forward the processing request to the calling unit 208, wherein the processing request carries the data to be queried;

[0061] The calling unit 208 is used to call the query model to process the data to be queried and the processing model description information in response to the processing request, and determine at least one target processing model corresponding to the data to be queried, wherein the query model is a deep learning model; and send the model information of at least one target processing model to the client through the communication interface 206.

[0062] Specifically, the model information may be a model file of the target processing model, and the client may locally construct the target processing model based on the model file of the target processing model. The model information may also be model call notification information of the target processing model, and the client may call the target processing model through the model call interface in response to the model call notification information. The model information may also be model introduction information of the target processing model, and the model introduction information includes but is not limited to model description information, model usage method, model usage scope, model operating environment, model code example, and model training data introduction.

[0063] The data to be queried includes model query demand information. The query model is a deep learning model. The query model can be trained based on sample data, or a large model (such as a large-scale language model) obtained by pre-training can be used as a query model. The processing model can be a model with different functions, such as a video generation model, a picture generation model, a picture-text dialogue model, a speech synthesis model, a knowledge question-and-answer model, an address structuring model, an English-Chinese translation model, a named entity recognition model, a Chinese-English translation model, a general information extraction model, and the like. The processing model description information is used to describe the processing model, including but not limited to the model name, model function, and model type, which are selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0064] In an optional embodiment of the present specification, after the calling unit sends the model information of at least one target processing model to the client through the communication interface, the user can directly perform task processing based on the model information. Furthermore, the user can also send model adjustment parameters through the client based on actual needs, and the calling unit can perform secondary development adjustment on at least one target processing component based on the model adjustment parameters, that is, the above-mentioned calling unit is also used to adjust at least one target processing model according to the model adjustment parameters, and send the model information of at least one adjusted target processing model to the client through the communication interface. Among them, the model adjustment parameters include but are not limited to model adjustment sample sets, model adjustment weights, etc., and the adjustment sample set includes multiple adjustment sample data, and the adjustment sample data carries adjustment sample labels.

[0065] See also Figure 3 , Figure 3 The architecture diagram of the third task processing system provided by an embodiment of the present specification is shown, and the task processing system may include multiple clients 100 and a task processing platform 200. Multiple clients 100 may establish communication connections through the task processing platform 200. In the task processing scenario, the task processing platform 200 is used to provide task processing services between multiple clients 100. Multiple clients 100 may respectively serve as sending ends or receiving ends, and realize communication through the task processing platform 200.

[0066] The user can interact with the task processing platform 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In the task processing scenario, the user can publish a data stream to the task processing platform 200 through the client 100, and the task processing platform 200 generates a task processing result based on the data stream, and pushes the task processing result to other clients that have established communication. Among them, the client 100 and the task processing platform 200 are connected through a network. The network provides a medium for the communication link between the client 100 and the task processing platform 200. The network can include various connection types, such as wired, wireless communication links or optical fiber cables, etc. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, etc. before being published to the task processing platform 200.

[0067] The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5, Hypertext Markup Language Version 5) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The client 100 can be based on the software development kit (SDK, Software Development Kit) of the corresponding service provided by the task processing platform 200, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition. The client 100 can be deployed in an electronic device, and needs to rely on the device to run or some APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, mailbox clients, social platform software, etc.

[0068] The task processing platform 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers for background training that provide support for models used on clients, and servers that process data sent by clients. It should be noted that the task processing platform 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0069] It is worth noting that the task processing method provided in the embodiments of this specification is generally executed by the task processing platform, but in other embodiments of this specification, the client may also have similar functions to the task processing platform, thereby executing the task processing method provided in the embodiments of this specification. In other embodiments, the task processing method provided in the embodiments of this specification may also be jointly executed by the client and the task processing platform.

[0070] See also Figure 4 , Figure 4 A flowchart of a task processing method provided by an embodiment of the present specification is shown. The task processing method is applied to a service unit in a task processing platform. The task processing platform includes a service unit and multiple processing components. Specifically, the following steps are included:

[0071] Step 402: Receive a task processing request, wherein the task processing request carries data to be processed.

[0072] In one or more embodiments of the present specification, data to be processed may be obtained, and task processing may be performed based on the data to be processed to obtain a task processing result.

[0073] Specifically, the data to be processed may be data in different scenarios, such as data to be processed in conference scenarios, data to be processed in e-commerce scenarios, etc. The tasks corresponding to the data to be processed may be different types of tasks, such as video generation tasks, image generation tasks, graphic dialogue tasks, speech synthesis tasks, knowledge question-answering tasks, address structuring tasks, English-Chinese translation tasks, named entity recognition tasks, Chinese-English translation tasks, general information extraction tasks, etc. The data to be processed include but are not limited to data content, task execution time, task identifier, etc., which are selected according to actual conditions, and the embodiments of this specification do not make any restrictions on this.

[0074] In practical applications, there are many ways to receive task processing requests, which are selected according to actual conditions, and the embodiments of this specification do not limit this. In one possible implementation of this specification, a task processing request sent by a user through a client can be received. In another possible implementation of this specification, a task processing request sent by a client at a fixed time can be received.

[0075] Step 404: Input the data to be processed and the component description information into the task processing model, and determine at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model.

[0076] In one or more embodiments of the present specification, after receiving a task processing request, the data to be processed and component description information can be further input into a task processing model to determine at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model.

[0077] Specifically, the task processing model is a deep learning model, which can be trained based on sample data, or a pre-trained large model (such as a large-scale language model) can be used as a task processing model. The multiple processing components in the task processing platform can be an artificial intelligence model encapsulated into an API interface in the form of ToolAPI integration and integrated into the task processing platform, or they can be pluggable components pre-configured on the task processing platform by the configuration party. The processing components can be components with different functions, such as components that support translation, components that support video generation, components that support image generation, and so on. The component description information is used to describe the processing component, including but not limited to the component name, component function, and component type. The specific selection is based on the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0078] It should be noted that there are multiple ways to input the data to be processed and the component description information into the task processing model and determine at least one target processing component corresponding to the task processing request. The specific selection is made according to the actual situation, and the embodiments of this specification do not impose any limitations on this.

[0079] In a possible implementation of the present specification, the data to be processed and the component description information of each processing component in the task processing platform can be directly input into the task processing model to determine at least one target processing component corresponding to the task processing request.

[0080] In another possible implementation of the present specification, in order to reduce the processing pressure of the task processing model, multiple processing components in the task processing platform can be screened to determine at least one candidate processing component, and the data to be processed and the component description information of at least one candidate processing component are input into the task processing model to determine at least one target processing component corresponding to the task processing request. There are many ways to screen multiple processing components, for example, it can be to compare the task type of the task processing request and the component type of each processing component, and screen out the candidate processing components that match the task type from the multiple processing components; it can also be to compare the task description information of the task processing request and the component function information of each processing component, and screen out the candidate processing components that match the task description information from the multiple processing components.

[0081] Furthermore, the data to be processed and the component description information are input into the task processing model, and the task processing model can determine at least one target processing component corresponding to the task processing request through semantic understanding capability.

[0082] In an optional embodiment of the present specification, the data to be processed and the local knowledge may be roughly sorted to quickly determine the task processing result. In the case where the rough sorting fails, the data to be processed and the component description data of the processing component may be further finely sorted. That is, before the data to be processed and the component description information are input into the task processing model and at least one target processing component corresponding to the task processing request is determined, the following steps may be further included:

[0083] Obtain reference text data;

[0084] Matching the data to be processed with the reference text data to obtain a matching result;

[0085] Inputting the data to be processed and the component description information into the task processing model, and determining at least one target processing component corresponding to the task processing request, including:

[0086] When the matching result is a matching failure, the data to be processed and the component description information are input into the task processing model to determine at least one target processing component corresponding to the task processing request.

[0087] Specifically, the reference text data is also local knowledge, and the reference text data includes but is not limited to task processing prompt text, answer template, etc. There are many ways to obtain the reference text data, which can be receiving the reference text data input by the user, or reading the reference text data from other data acquisition devices or databases. The reference text data can be updated in real time.

[0088] In practical applications, there are many ways to match the data to be processed with the reference text data to obtain matching results, which can be selected according to the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, sparse retrieval can be performed, that is, the similarity between the data to be processed and the reference text data is calculated on the word dimension. In another possible implementation of this specification, vector retrieval can be performed, that is, contextual semantic information is incorporated, and the similarity between the data to be processed and the reference text data is calculated on the sentence dimension. In another possible implementation of this specification, the data to be processed and the reference text data can be input into the task processing model to obtain matching results.

[0089] It should be noted that the data to be processed and the reference text data are matched to obtain a matching result. If the matching result is a matching failure, it means that there is no data related to the data to be processed in the reference text data, that is, the task processing result corresponding to the task processing request cannot be quickly determined through the reference text data, and it is necessary to use the task processing model to determine at least one target processing component corresponding to the task processing request, so as to use the target processing component to process the data to be processed; if the matching result is a matching success, it means that there is data related to the data to be processed in the reference text data, that is, the task processing result corresponding to the task processing request can be quickly determined through the reference text data without calling the processing component.

[0090] Apply the scheme of the embodiment of this specification to obtain reference text data; match the data to be processed with the reference text data to obtain a matching result; when the matching result is a matching failure, input the data to be processed and the component description information into the task processing model to determine at least one target processing component corresponding to the task processing request. An extensible knowledge retrieval and component selection are proposed, and the extended local knowledge and extended processing components are integrated into the output of the task processing model through a rough sorting based on retrieval and a fine sorting based on processing components, thereby improving the extensibility of task processing. Moreover, since the processing components and local knowledge can be updated, expanded and enhanced in real time, the task processing process can be updated, further improving the real-time, adaptability and compatibility of task processing.

[0091] Furthermore, after matching the data to be processed with the reference text data and obtaining the matching result, the following steps may be further included:

[0092] When the matching result is successful, the data to be processed and the reference text data are input into the task processing model to determine the task processing result corresponding to the task processing request.

[0093] It should be noted that, when the matching result is a successful match, the task processing result corresponding to the task processing request can be determined according to the data to be processed and the reference text data.

[0094] In actual applications, there are many ways to determine the task processing result corresponding to the task processing request based on the data to be processed and the reference text data. The specific method is selected according to the actual situation, and the embodiments of this specification do not impose any limitation on this.

[0095] In a possible implementation of the present specification, first, multiple indexes can be created for reference text data, and one index can correspond to multiple reference text data. After creating multiple indexes, a retriever can be further created based on the multiple indexes, wherein the type of the retriever includes at least one of sparse retrieval and vector retrieval. Then, the data to be processed can be input into the retriever, and the retriever can return retrieval knowledge related to the data to be processed. Finally, the retrieval knowledge returned by the retriever is used as the task processing result.

[0096] When performing sparse retrieval and vector retrieval, the retriever can calculate the similarity between the data to be processed and the reference text data, thereby determining the task processing result corresponding to the task processing request based on the similarity. The methods for calculating the similarity between the data to be processed and the reference text data include but are not limited to cosine similarity, term frequency-inverse document frequency (TF-IDF) calculation, and edit distance calculation.

[0097] In another possible implementation of the present specification, the data to be processed and the reference text data may be input into the task processing model to determine the task processing result corresponding to the task processing request.

[0098] By applying the solution of the embodiment of this specification, when the matching result is a successful match, the data to be processed and the reference text data are input into the task processing model to determine the task processing result corresponding to the task processing request, so that the task processing result can be determined quickly, thereby improving the task processing efficiency.

[0099] In an optional embodiment of the present specification, the training method of the above-mentioned task processing model may include the following steps:

[0100] Obtain a sample set, wherein the sample set includes a plurality of sample data, the sample data carries a sample label, and the sample data is constructed based on at least one of sample text data, sample question-answer pair data, and sample component call data;

[0101] Extracting first sample data from the sample set, wherein the first sample data is any one of the plurality of sample data;

[0102] Inputting the first sample data into the task processing model to obtain a first prediction result of the first sample data;

[0103] Calculate a first loss value according to the first prediction result and the first sample label carried by the first sample data;

[0104] The model parameters of the task processing model are adjusted according to the first loss value, and the step of extracting the first sample data from the sample set is returned to be executed until the first preset stop condition is reached, so as to obtain a trained task processing model.

[0105] Specifically, the training method of the task processing model is supervised training. The task processing model is trained based on multiple sample data and the sample labels carried by each sample data. The sample label is the real label of the sample data, which can also be understood as the processing target of the task processing model, which is used to guide the training process of the task processing model.

[0106] In a possible implementation of the present specification, the first preset stop condition includes that the first loss value is less than or equal to a first preset threshold value. After calculating the first loss value according to the first prediction result and the first sample label carried by the first sample data, the first loss value can be compared with the first preset threshold value.

[0107] Specifically, if the first loss value is greater than the first preset threshold, it means that the difference between the first prediction result and the first sample label is large, and the task processing model has poor prediction ability for the first prediction result. At this time, the parameters of the task processing model can be adjusted, and the step of extracting the first sample data from the sample set is returned to continue training the task processing model until the first loss value is less than or equal to the first preset threshold. This means that the difference between the first prediction result and the first sample label is small, and the prediction ability of the task processing model is strong. The first preset stop condition is met, and the training can be stopped to obtain a trained task processing model.

[0108] In another possible implementation of the present specification, in addition to comparing the first loss value and the first preset threshold, it is also possible to determine whether the current task processing model has been trained in combination with the first number of iterations.

[0109] Specifically, if the first loss value is greater than the first preset threshold, the parameters of the task processing model are adjusted, and the step of extracting the first sample data from the sample set is returned to continue training the task processing model until the first preset number of iterations is reached, and the iteration is stopped to obtain a task processing model that has completed the training, wherein the first preset threshold and the first preset number of iterations are selected according to actual conditions, and the embodiments of this specification do not impose any limitations on this.

[0110] In practical applications, there are many functions for calculating the first loss value, such as cross entropy loss function, L1 norm loss function, maximum loss function, mean square error loss function, logarithmic loss function, etc. The specific selection is based on the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0111] The solution of the embodiment of this specification is applied, and a first loss value is calculated based on the first prediction result and the first sample label carried by the first sample data, and the first loss value is compared with the first preset stop condition. If the first preset stop condition is not met, the task processing model is continuously trained until the first preset stop condition is met, and the task processing model is obtained after the training is completed. By continuously adjusting the parameters of the task processing model, the task processing model finally obtained can be made more accurate.

[0112] It should be noted that there are many ways to obtain a sample set, which can be selected according to actual conditions, and the embodiments of this specification do not limit this. A large amount of sample data carrying sample labels can be read from other data acquisition devices or databases to form a training set. A large amount of sample data carrying sample labels input by a user can also be received to form a sample set.

[0113] In practical applications, high-quality annotated data (sample data) is a key factor in determining the upper limit of model performance. As the number of model parameters increases, the demand for annotated data will also increase. However, the sample data acquisition process is usually time-consuming and laborious, and generates considerable annotation costs, which seriously affects the iteration efficiency of the model. Therefore, in the embodiments of this specification, sample data can be constructed based on at least one of sample text data, sample question-answer pair data, and sample component call data.

[0114] Specifically, sample text data, sample question-answer pair data, and sample component call data are obtained, and a seed file is constructed based on the sample text data, sample question-answer pair data, and sample component call data. Then, the seed file is processed through a data processing strategy, and more sample data is generated through the powerful generalization ability of the task processing model, which is used for fine-tuning the task processing model, greatly improving the scale and accuracy of data annotation. Among them, the data processing strategy includes but is not limited to at least one of zero-shot learning, few-shot learning, and cot prompt engineering.

[0115] See also Figure 5 , Figure 5 FIG. 1 shows a schematic diagram of generating sample data in a task processing method provided in an embodiment of the present specification, such as Figure 5 As shown, sample text data, sample question and answer pair data, and sample component call data are obtained, sample question and answer pair data are obtained through zero-shot learning and few-shot learning, the sample question and answer pair data are fused to construct sample conversation data, and the sample component call data can also be processed to construct sample conversation data, and the sample conversation data is used as sample data.

[0116] In practical applications, when calculating the first loss value according to the first prediction result and the first sample label carried by the first sample data, the calculation may be performed directly according to the first prediction result and the first sample label, or the first prediction result may be parsed based on model knowledge distillation to incorporate the weight to calculate the first loss value. That is, the calculation of the first loss value according to the first prediction result and the first sample label carried by the first sample data may include the following steps:

[0117] Parsing the first prediction result to obtain a plurality of first prediction sub-results;

[0118] Determining a weight corresponding to each first prediction sub-result according to the type of each first prediction sub-result;

[0119] A first loss value is calculated according to the multiple first prediction sub-results, the weights corresponding to the first prediction sub-results, and the first sample labels carried by the first sample data.

[0120] It should be noted that there are multiple ways to parse the first prediction result and obtain multiple first prediction sub-results, which are selected according to actual conditions, and the embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, the first prediction result can be matched with a preset parsing dictionary, so as to split the first prediction result and obtain multiple first prediction sub-results. In another possible implementation of this specification, the first prediction result can be parsed using a preset word segmentation tool to obtain multiple first prediction sub-results.

[0121] Furthermore, after obtaining multiple first prediction sub-results, corresponding weights can be set for each first prediction sub-result. In the process of setting the weights, the weights can be set based on the types of the first prediction sub-results, and the weights corresponding to the first prediction sub-results can also be determined according to actual needs. Finally, the first loss value is weightedly calculated based on the multiple first prediction sub-results, the weights corresponding to the first prediction sub-results, and the first sample labels carried by the first sample data.

[0122] For example, since there may be problems with the generation and calling correctness of component names in the process of calling and processing components, and in the process of multiple rounds of task processing, the incoming component calling parameters may need to be found in the previous rounds of task processing, resulting in parameter passing errors, the first prediction result can be parsed to obtain multiple first prediction sub-results, such as the component name, component calling parameters, component content and component reply content in the component reply, weights are set for each of these first prediction sub-results, and the loss value is calculated by weight.

[0123] By applying the solution of the embodiment of this specification, the first prediction result is parsed to obtain multiple first prediction sub-results; the weight corresponding to each first prediction sub-result is determined according to the type of each first prediction sub-result; the first loss value is calculated according to the multiple first prediction sub-results, the weight corresponding to each first prediction sub-result and the first sample label carried by the first sample data, thereby adjusting the proportion of the first prediction sub-result in the model training process according to the weight, which can improve the task completion rate on the one hand, and improve the correctness of parameter input on the other hand.

[0124] Step 406: Call at least one target processing component to process the data to be processed, and obtain a task processing result corresponding to the task processing request.

[0125] In one or more embodiments of the present specification, after receiving a task processing request, inputting the data to be processed and component description information into the task processing model, and determining at least one target processing component corresponding to the task processing request, further, at least one target processing component can be called to process the data to be processed to obtain a task processing result corresponding to the task processing request.

[0126] By applying the solution of the embodiments of this specification, multiple processing components are integrated on the task processing platform, so that at least one target processing component can be directly called to process the data to be processed, thereby improving the task processing efficiency. In addition, through multiple processing components, on the one hand, the text generation capability of the task processing model in natural language tasks can be improved, and on the other hand, the missing multimodal generation capability of the task processing model can be compensated, making the processing capability of the task processing model more comprehensive.

[0127] In an optional embodiment of the present specification, since the data to be processed may not meet the input parameter requirements of the target processing component, the format of the data to be processed may be converted according to the component call parameters of the target processing component, that is, before the above-mentioned calling of at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the following steps may also be included:

[0128] Obtaining component call parameters of at least one target processing component;

[0129] Convert the format of the data to be processed according to the component calling parameters to obtain updated data to be processed;

[0130] Calling at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request includes:

[0131] At least one target processing component is called to process the updated data to be processed, and a task processing result corresponding to the task processing request is obtained.

[0132] Specifically, the component calling parameters are used to describe the calling rules of the target processing component, and the component calling parameters include but are not limited to the language type, data format, etc. supported by the target processing component. There are many ways to obtain the component calling parameters of at least one target processing component, which may be the component calling parameters of at least one target processing component sent by the receiving configuration party, or the component calling parameters of at least one target processing component may be read from other data acquisition devices or databases.

[0133] It should be noted that after obtaining the component calling parameters of at least one target processing component, it can be further determined whether the data to be processed conforms to the component calling parameters. If so, it means that the target processing component can directly process the data to be processed without converting the format of the data to be processed, and at least one target processing component is called to process the data to be processed to obtain the task processing result corresponding to the task processing request. If not, it means that the target processing component cannot directly process the data to be processed, and the format of the data to be processed can be converted according to the component calling parameters to obtain updated data to be processed, and at least one target processing component is called to process the updated data to be processed to obtain the task processing result corresponding to the task processing request.

[0134] In actual applications, there are many ways to convert the format of the processed data according to the component call parameters to obtain the updated data to be processed, and the specific selection is based on the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, a format conversion tool can be used to convert the processed data, such as using a translation tool to translate the Chinese data to be processed into English. In another possible implementation of this specification, a conversion processing component that can convert the format of the processed data so that the processed data conforms to the component call parameters can be found from multiple processing components, and the conversion processing component is used to convert the format of the processed data based on the component call parameters to obtain the updated data to be processed, such as using a translation model to translate the processed data at noon into English.

[0135] By applying the solution of the embodiment of this specification, the component calling parameters of at least one target processing component are obtained; the format of the data to be processed is converted according to the component calling parameters to obtain the updated data to be processed; at least one target processing component is called to process the updated data to be processed to obtain the task processing result corresponding to the task processing request. It is ensured that at least one target processing component can be successfully called, and the success rate of task processing is guaranteed.

[0136] In an optional embodiment of the present specification, the task processing model may be optimized through the call information of at least one target processing component to implement reinforcement learning based on processing component feedback, that is, after the above-mentioned calling of at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the following steps may also be included:

[0137] The calling information of at least one target processing component is obtained, and the task processing model is optimized based on the calling information.

[0138] Specifically, the call information of the target processing component includes but is not limited to the call process status, call result and task processing result.

[0139] It should be noted that, in the process of calling at least one target processing component to process the data to be processed, at least one target processing component can be monitored to obtain the calling information of at least one target processing component. Furthermore, the calling information can be used as a reward signal to optimize the task processing model through the Actor-Critic algorithm, thereby optimizing the decision-making process.

[0140] By applying the solution of the embodiments of this specification, the calling information of at least one target processing component is obtained, and the task processing model is optimized based on the calling information, so that the task processing model can determine at least one target processing component corresponding to the task processing request more accurately, and can also accurately determine the processing order of at least one target processing component, so as to accurately call at least one target processing component to realize task processing, thereby improving the generalization ability of the task processing model.

[0141] In an optional embodiment of the present specification, the task processing request also carries task requirement information; after the above-mentioned calling at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the following steps may also be included:

[0142] The task requirement information and the task processing result are input into the task processing model to obtain an updated task processing result, wherein the updated task processing result meets the task requirement information.

[0143] Specifically, the task requirement information is used to limit the task processing results according to the actual task requirements. The task requirement information includes but is not limited to the word count requirement and the language requirement of the task processing results, which are selected according to the actual situation, and this specification embodiment does not make any limitation on this.

[0144] For example, if the task processing result is in English and the task requirement information is in Chinese, the task requirement information and the task processing result can be input into the task processing model, and the task processing model will translate the English task processing result into the Chinese task processing result. If the task processing result has too many words, the task requirement information and the task processing result can be input into the task processing model, and the task processing model will extract the main idea of ​​the task processing result to obtain an updated task processing result.

[0145] By applying the solution of the embodiment of this specification, the task requirement information and the task processing result are input into the task processing model to obtain the updated task processing result, thereby ensuring that the task processing result meets the task requirement information and improving the accuracy of task processing and user experience.

[0146] In an optional embodiment of the present specification, after calling at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the following steps may also be included:

[0147] Send the task processing result to the client so that the client can display the task processing result to the user.

[0148] It should be noted that after calling at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the task processing result can be sent to the client, and the processing method corresponding to the task processing result can also be sent to the client, wherein the processing method can be obtained based on the reference text data processing or by calling the target processing component for processing. If the processing method is obtained by calling the target processing component for processing, the processing method can also include component information of the target processing component, such as the number of components, the source of components, the source address of components, etc. Furthermore, if the task processing result is an updated task processing result, the task processing result before the update and the updated processing result can be sent to the client at the same time.

[0149] In actual applications, there are many ways for the client to display the task processing results to the user, which can be selected according to the actual situation. The embodiments of this specification do not impose any restrictions on this. In one possible implementation of this specification, the client can only display the task processing results to the user. In another possible implementation of this specification, the client can display the task processing results and the processing methods to the user at the same time, so that the user can accurately know the task processing results and the corresponding processing methods. In another possible implementation of this specification, the task processing results before the update and the processing results after the update can be displayed to the user at the same time.

[0150] It is worth noting that after obtaining the task processing result corresponding to the task processing request, the task processing result can be quality checked. If the quality check passes, the task processing result is sent to the client. If the quality check fails, the task processing is performed again.

[0151] By applying the solution of the embodiment of this specification, the task processing result is sent to the client, so that the client can display the task processing result to the user, so that the user can accurately obtain the task processing result, increase the interaction with the user, and improve user satisfaction.

[0152] In an optional embodiment of the present specification, the task processing request also carries display requirement information; and the sending of the task processing result to the client may include the following steps:

[0153] Send the task processing results to the client based on the display requirement information.

[0154] Specifically, the display requirement information is used to describe the user's need to view the task processing results. The display requirement information includes but is not limited to displaying only the task processing results, displaying the task processing results and the corresponding processing methods, displaying the task processing results obtained by the target processing component and the component information of the target processing component, displaying the task processing results obtained based on the reference text data, etc., which are specifically set according to the actual needs of the user, and the embodiments of this specification do not impose any restrictions on this.

[0155] It should be noted that, when the processing method corresponding to the data to be processed includes both processing based on reference text data and processing based on the target processing component, the first task processing result can be generated based on the reference text data, and the second task processing result can be generated using the target processing component. Then, the first task processing result and the second task processing result can be displayed to the user. Furthermore, after the first task processing result and the second task processing result are generated, the user can be provided with display options of "displaying the results obtained by processing using the target processing component" and "displaying the results obtained based on the reference text data processing", and the target display option selected by the user is received, and the task processing result corresponding to the target display option is displayed.

[0156] By applying the solution of the embodiment of this specification, the task processing result is sent to the client according to the display requirement information, so that the client can display the task processing result to the user, thereby increasing the interaction with the user and improving the user satisfaction.

[0157] In an optional embodiment of the present specification, after the client displays the task processing result to the user, the user can perform data processing according to the task processing result, or send a post-processing request based on the task processing result displayed by the client to perform multiple rounds of task processing. That is, after sending the task processing result to the client, the following steps may also be included:

[0158] Receive post-processing requests sent by users based on task processing results, process the post-processing requests, and obtain post-processing results corresponding to the post-processing requests.

[0159] Specifically, a post-processing request refers to a task processing request sent by a user based on a task processing result after obtaining the task processing result. The data requested to be processed by the post-processing request can be new data to be processed or data to be processed carried by the task processing request. The selection is made according to the actual situation, and the embodiments of this specification do not make any limitation on this.

[0160] It should be noted that if the data requested to be processed by the post-processing request is the pending data carried by the task processing request, it means that the user is not satisfied with the task processing result and needs to reprocess the pending data. Furthermore, the post-processing request sent by the user based on the task processing result can carry the processing method specified by the user. Exemplarily, if the task display result is generated based on the reference text data, the post-processing request sent by the user based on the task processing result can request the use of the target processing component to process the pending data; if the task display result is generated by the target processing component, the post-processing request sent by the user based on the task processing result can request the processing of the pending data based on the reference text data. In actual applications, the implementation method of "processing the post-processing request and obtaining the post-processing result corresponding to the post-processing request" is the same as the implementation method of the task processing method, and the embodiments of this specification will not be repeated.

[0161] By applying the solution of the embodiment of this specification, a post-processing request sent by a user based on a task processing result is received, and the post-processing request is processed to obtain a post-processing result corresponding to the post-processing request, thereby increasing interaction with the user and improving user satisfaction.

[0162] The following combination Figure 6 , taking the application of the task processing method provided in this specification in the field of intelligent question answering as an example, the task processing method is further explained. Figure 6 A flowchart of an automatic question-answering method provided by an embodiment of the present specification is shown. The automatic question-answering method is applied to a service unit in a task processing platform. The task processing platform includes a service unit and multiple processing components, and specifically includes the following steps:

[0163] Step 602: Obtain questions to be answered.

[0164] Step 604: Input the question to be answered and the component description information into the task processing model to determine at least one target processing component corresponding to the question to be answered, wherein the task processing model is a deep learning model of natural language.

[0165] Step 606: Call at least one target processing component to process the question to be answered, obtain a reply result corresponding to the question to be answered, and send the reply result to the user.

[0166] Step 608: Receive the updated unanswered questions sent by the user based on the reply result, and return to execute the step of inputting the unanswered questions and component description information into the task processing model, and determining at least one target processing component corresponding to the unanswered questions, until the second preset stop condition is reached to obtain the reply result.

[0167] It should be noted that the question to be answered is the data to be processed, and the answer result is the task processing result corresponding to the task processing request. The implementation of steps 602 to 606 is the same as the implementation of steps 402 to 406 above, and will not be repeated in this embodiment of the specification.

[0168] In actual applications, the automatic question-answering process can be a single round of question-answering or a multi-round question-answering process. The questions to be answered can be general information retrieval questions, image generation questions, or text combined with video generation questions. The questions to be answered are selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this. The second preset stop condition can be that the user stops sending updated questions to be answered, or it can be that the second preset number of iterations is reached. The second preset stop condition and the second preset number of iterations are selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0169] For example, assume that the automatic question-answering process is a four-round question-answering process. Question 1: Write a Chinese advertising recommendation for product A; Answer 1: Product A makes your life more colorful; Question 2 (first question update): Help me translate it into English; Answer 2: A commodity, make your life more colorful; Question 3 (second question update): Based on the above advertising words, help me generate a picture; Answer 3: OK, I will call the picture generation component to draw for you, (show the picture); Question 4 (third question update): Based on the above advertising words, help me generate a video; Answer 4: Currently, the task processing platform only supports English video generation components, so I will first convert your question into English using the Chinese-English translation model, and then call the English video generation model. The translated advertising words are A commodity, make your life more colorful, and the video generation model is used to generate the video, (show the video).

[0170] Assume that the automatic question answering is a two-round knowledge question answering. Question 1: Is there a model that supports open domain dialogue? Answer 1: Yes, we have a model for open domain dialogue, named X. It can receive user dialogue history and external knowledge related to dialogue, user persona, model persona information, and output appropriate text replies. You can use the code examples we provide to use the model; Question 2: How to use this model? Answer 2: First, you need to install the latest version of our app, and then you can use the X model. You can use the code examples we provide to use the model.

[0171] It is worth noting that the above-mentioned automatic question-answering method can be applied to conversational robots. Conversational robots have intelligent operation solutions and can provide users with human-like conversational experience. Conversational robots can communicate by accessing knowledge without the need for knowledge processing.

[0172] Specifically, conversations can be conducted by accessing knowledge without knowledge processing: the types of knowledge supported by conversational robots include but are not limited to documents, data tables, and project processes. Document knowledge can be obtained by directly connecting to enterprise-level websites, enterprise knowledge bases, etc., or by importing documents. Conversational robots can redefine Frequent Asked Questions and provide users with intelligent operations.

[0173] Human-like conversation experience: The language comprehension ability of the conversation robot has been greatly improved. It can achieve joint understanding of different types of knowledge, and has the ability of multi-round conversation and reasoning analysis. The conversation process of the conversation robot is more natural and anthropomorphic, and the generated dialogue is more coherent and natural. In the process of knowledge question and answer, the original knowledge fragments can be cited as a reference.

[0174] Intelligent operation solution: The conversational robot has complete beginner guides and glossaries, and provides a one-stop usage introduction, so as to provide step-by-step guidance; the conversational robot can support the annotation of the robot's replies, and support the use of high-frequency documents and other methods to ensure the effect, so that the effect can be intervened; the conversational robot can also provide concise and practical conversation overview data, project analysis and other data through data reports, so as to provide feedback on the effect.

[0175] By applying the solution of the embodiments of this specification, multiple processing components are integrated on the task processing platform, so that at least one target processing component can be directly called to process the questions to be answered, thereby improving the efficiency of automatic question answering. In addition, multiple processing components can, on the one hand, enhance the text generation capability of the task processing model in natural language tasks, and on the other hand, make up for the lack of multimodal generation capability of the task processing model, making the automatic question answering more comprehensive.

[0176] See also Figure 7 , Figure 7A flowchart of another task processing method provided by an embodiment of this specification is shown. Figure 7 As shown, the solution provided in the embodiment of this specification includes three parts, namely a front end, a task processing platform and a model-as-a-service sharing platform. The front end is used to provide a task processing window for users to communicate with the task processing platform; the task processing platform is used to provide reference text data and component services, component lists and component descriptions provided by the model-as-a-service sharing platform for users to use through dialogue; the model-as-a-service sharing platform is used to provide processing component services for calls to the learnable large model enhancement framework, and provide reference text data for generating task processing results.

[0177] The task processing platform processing process can be divided into knowledge search and component selection, and the corresponding reference text data and target processing components can be searched according to the data to be processed input by the user. Specifically, knowledge search can select the knowledge fragments to be used from the reference text data according to the data to be processed input by the user; component selection can select the target processing component to be used from multiple processing components according to the data to be processed input by the user.

[0178] The learnable large model enhancement framework can complete the task processing request and give the task processing result based on the data to be processed, the reference text data, and multiple processing components. Specifically, the learnable large model enhancement framework can be divided into three parts: task planning, using the task processing model to analyze the user request to understand its intention, and determine the target processing component and component call parameters to be used. Task execution, calling and executing the target processing component, and returning the execution result output by the target processing component to the task processing model. Answer generation, generating the processing result based on the execution result, and generating a new task plan until the task is completed.

[0179] The solution of the embodiments of this specification provides a generative big model based on processing components and retrieval enhancement, which acquires domain knowledge and updated processing components in real time through knowledge search and component selection, thereby achieving real-time and updatableness; based on the learnable big model enhancement framework, the generalization ability of the task processing model is improved by constructing sample data, model knowledge distillation, and reinforcement learning based on component feedback, and the task processing efficiency is improved through the task processing flow of sequence results, thereby realizing knowledge question and answer, understanding of more detailed requirements, and model recommendation, and further realizing simple tasks, complex tasks, and multi-application programming interface planning and calling, and can also be used by developers to integrate processing components developed by themselves and provide them to users.

[0180] See also Figure 8 , Figure 8A processing flow chart of a task processing method provided by an embodiment of the present specification is shown, receiving a task processing request, wherein the task processing request carries data to be processed; inputting the data to be processed into a task processing model, and determining whether to process the task by component selection or by knowledge search:

[0181] If task processing is performed by component selection: input the data to be processed and component description information into the task processing model, determine at least one target processing component corresponding to the task processing request and the component calling parameters of the target processing component; call at least one target processing component, input the data to be processed that meets the component calling parameters into at least one target processing component, and input the execution result output by the target processing component into the task processing model to generate the task processing result corresponding to the task processing request, wherein the target processing component can obtain the pre-stored local knowledge from the database and process the data to be processed based on the local knowledge. Local knowledge includes but is not limited to pictures, audio and video;

[0182] If task processing is performed through knowledge search: encode the data to be processed, generate a processing vector corresponding to the data to be processed, obtain a reference vector from the vector repository, calculate the similarity between the processing vector and the reference vector, further generate a prompt text based on the similarity, input the prompt text into the task processing model, and obtain the task processing result corresponding to the data to be processed, wherein the reference vector is obtained by encoding the reference text data in the local knowledge.

[0183] By applying the solution of the embodiments of this specification, a scalable and learnable large model enhancement framework with multiple component capabilities is proposed. On the basis of the task processing model, multiple artificial intelligence models are integrated by processing components. On the one hand, the text generation ability of the task processing model in natural language tasks can be improved. On the other hand, it can make up for the lack of multimodal generation ability of the task processing model, making the processing capability of the task processing model more comprehensive.

[0184] See also Fig. 9 , Fig. 9 An example diagram of a task in a task processing method provided by an embodiment of this specification is shown. Fig. 9 As shown, the task processing platform proposed in the embodiment of this specification supports different types of tasks, including but not limited to video generation tasks, image generation tasks, graphic dialogue tasks, speech synthesis tasks, knowledge question and answer tasks, address structuring tasks, English-Chinese translation tasks, named entity recognition tasks, Chinese-English translation tasks, and general information extraction tasks. Next, the question and answer process of each task is described in detail one by one.

[0185] Video generation task: Question 1: Generate a video describing two puppies playing; Answer 1: Currently, the task processing platform only supports English video generation components, so I will first convert your question into English using the Chinese-English translation model, and then call the English video generation model. The translated question is Two puppies playing, and the video is generated using the video generation model (display video).

[0186] Image generation task: Question 1: Generate a high-definition picture of a puppy lying on the sofa; Answer 1: Calling the image generation component to generate (displaying the picture).

[0187] Picture-text dialogue task: Question 1: What color is the man’s clothes in picture A? Answer 1: Yellow.

[0189] Knowledge quiz task: Question 1: What are the characteristics of model A? Answer 1: It supports efficient input of multiple pieces of external knowledge, character information and other texts for knowledge enhancement.

[0190] Address structuring task: Question 1: Parse the following address and extract specific elements: City A, District B, Street C, D, Home, E, Visit, No. 56, 401, 11111111111x; Answer 1: Address decomposition json representation: {"city":"City A","district":"District B","houseno":"No. 56","poi":"D, Home","road":"E","town":"C, Street"}. The model I used is XX model. This is fine-tuned based on the multi-task multi-modal address pre-training base model.

[0191] Named Entity Recognition Task: Question 1: Find the entities in the following sentence: The 2023 Artificial Intelligence Conference was held in M ​​City; Answer 1: Entity extraction json representation: {"GPE":["M City"]}. The model I used is the XXX model.

[0192] English-Chinese translation task: Question 1: Translate the following sentence into Chinese: Ilovelearning; Answer: I love learning.

[0193] Chinese-English translation task: Question 1: Translate the following sentence into English: The weather is sonice today; Answer: The weather is sonice today.

[0194] General information extraction task: Question 1: Extract the information corresponding to the following text according to the given schema, schema: {"person":null,"geographic location":null,"organization":null,"time":null}, In 2019, University A held the 6th Forum in B, and more than 200 young scientists from all over the world participated in the forum, including Li Si, Wang Wu, etc.; Answer 1: Information extraction results: {"person":["Li Si","Wang Wu"],"geographic location":["B"],"time":["2019"],"organization":["University A"]}. The model I used is XX model. This is a general information extraction model.

[0195] See also Fig.10 , Fig.10 The following is a schematic diagram of an automatic question-and-answer interface provided by an embodiment of the present specification. The automatic question-and-answer interface is divided into a question-handling request input interface and a reply result display interface. The question-handling request input interface includes a question-handling request input box, an "OK" control, and a "Cancel" control. The reply result display interface includes a reply result display box.

[0196] The user inputs a question processing request through the question processing request input box displayed on the client, wherein the question processing request carries the question to be answered, clicks the "OK" control, and the task processing platform receives the question processing request sent by the client, inputs the question to be answered and the component description information into the task processing model, determines at least one target processing component corresponding to the question processing request, wherein the task processing model is a deep learning model; calls at least one target processing component to process the question to be answered, obtains the reply result corresponding to the question processing request, and sends the reply result to the client. The client displays the reply result in the reply result display box. In actual applications, the user operates the control in any of the following ways: clicking, double-clicking, touching, hovering the mouse, sliding, long pressing, voice control, or shaking, etc. The specific selection is based on the actual situation, and the embodiments of this specification do not impose any restrictions on this.

[0197] Corresponding to the above-mentioned task processing method embodiment, this specification also provides a task processing device embodiment. Fig.11 FIG. 1 shows a schematic diagram of a task processing device provided by an embodiment of the present specification. Fig.11 As shown, the task processing device is applied to the service unit in the task processing platform. The task processing platform includes a service unit and multiple processing components. The device includes:

[0198] The first receiving module 1102 is configured to receive a task processing request, wherein the task processing request carries data to be processed;

[0199] A first input module 1104 is configured to input the data to be processed and the component description information into a task processing model, and determine at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model;

[0200] The first calling module 1106 is configured to call at least one target processing component to process the data to be processed and obtain a task processing result corresponding to the task processing request.

[0201] Optionally, the device also includes: a matching module, configured to obtain reference text data; match the data to be processed and the reference text data to obtain a matching result; a first input module 1104, further configured to input the data to be processed and the component description information into the task processing model when the matching result is a matching failure, and determine at least one target processing component corresponding to the task processing request.

[0202] Optionally, the matching module is further configured to input the data to be processed and the reference text data into the task processing model to determine the task processing result corresponding to the task processing request when the matching result is a successful match.

[0203] Optionally, the device also includes: a training module, configured to obtain a sample set, wherein the sample set includes multiple sample data, the sample data carries a sample label, and the sample data is constructed based on at least one of sample text data, sample question-answer pair data, and sample component call data; extracting first sample data from the sample set, wherein the first sample data is any one of the multiple sample data; inputting the first sample data into the task processing model to obtain a first prediction result of the first sample data; calculating a first loss value based on the first prediction result and the first sample label carried by the first sample data; adjusting the model parameters of the task processing model based on the first loss value, and returning to execute the step of extracting the first sample data from the sample set until the first preset stop condition is reached to obtain a trained task processing model.

[0204] Optionally, the training module is further configured to parse the first prediction result to obtain multiple first prediction sub-results; determine the weight corresponding to each first prediction sub-result according to the type of each first prediction sub-result; and calculate the first loss value according to the multiple first prediction sub-results, the weight corresponding to each first prediction sub-result and the first sample label carried by the first sample data.

[0205] Optionally, the device also includes: a conversion module, configured to obtain component calling parameters of at least one target processing component; convert the format of the data to be processed according to the component calling parameters to obtain updated data to be processed; a first calling module 1106, further configured to call at least one target processing component to process the updated data to be processed, and obtain a task processing result corresponding to the task processing request.

[0206] Optionally, the device further includes: an optimization module configured to obtain call information of at least one target processing component and optimize the task processing model based on the call information.

[0207] Optionally, the task processing request also carries task requirement information; the device also includes: a third input module, configured to input the task requirement information and the task processing result into the task processing model to obtain an updated task processing result, wherein the updated task processing result conforms to the task requirement information.

[0208] Optionally, the device further includes: a sending module configured to send the task processing result to the client, so that the client displays the task processing result to the user.

[0209] Optionally, the device further includes: a second receiving module, configured to receive a post-processing request sent by a user based on the task processing result, and process the post-processing request to obtain a post-processing result corresponding to the post-processing request.

[0210] By applying the solution of the embodiments of this specification, multiple processing components are integrated on the task processing platform, so that at least one target processing component can be directly called to process the data to be processed, thereby improving the task processing efficiency. In addition, through multiple processing components, on the one hand, the text generation capability of the task processing model in natural language tasks can be improved, and on the other hand, the missing multimodal generation capability of the task processing model can be compensated, making the processing capability of the task processing model more comprehensive.

[0211] The above is a schematic scheme of a task processing device of this embodiment. It should be noted that the technical scheme of the task processing device and the technical scheme of the above task processing method belong to the same concept, and the details of the technical scheme of the task processing device that are not described in detail can all be referred to the description of the technical scheme of the above task processing method.

[0212] Corresponding to the above-mentioned automatic question-answering method embodiment, this specification also provides an automatic question-answering device embodiment, Fig.12 FIG. 1 shows a schematic diagram of the structure of an automatic question-answering device provided by an embodiment of this specification. Fig.12 As shown, the automatic question-answering device is applied to a service unit in a task processing platform. The task processing platform includes a service unit and multiple processing components. The device includes:

[0213] A first acquisition module 1202 is configured to acquire a question to be answered;

[0214] The second input module 1204 is configured to input the question to be answered and the component description information into the task processing model to determine at least one target processing component corresponding to the question to be answered, wherein the task processing model is a deep learning model of natural language;

[0215] The second calling module 1206 is configured to call at least one target processing component to process the question to be answered, obtain a reply result corresponding to the question to be answered, and send the reply result to the user;

[0216] Determination module 1208 is configured to receive updated unanswered questions sent by the user based on the reply result, and return to execute the step of inputting the unanswered questions and component description information into the task processing model, and determining at least one target processing component corresponding to the unanswered questions, until the second preset stop condition is reached to obtain the reply result.

[0217] By applying the solution of the embodiments of this specification, multiple processing components are integrated on the task processing platform, so that at least one target processing component can be directly called to process the questions to be answered, thereby improving the efficiency of automatic question answering. In addition, multiple processing components can, on the one hand, enhance the text generation capability of the task processing model in natural language tasks, and on the other hand, make up for the lack of multimodal generation capability of the task processing model, making the automatic question answering more comprehensive.

[0218] The above is a schematic scheme of an automatic question-answering device of this embodiment. It should be noted that the technical scheme of the automatic question-answering device and the technical scheme of the automatic question-answering method described above are of the same concept, and the details not described in detail in the technical scheme of the automatic question-answering device can be found in the description of the technical scheme of the automatic question-answering method described above.

[0219] Fig.13 13 shows a block diagram of a computing device provided by an embodiment of the present specification. The components of the computing device 1300 include but are not limited to a memory 1310 and a processor 1320. The processor 1320 is connected to the memory 1310 via a bus 1330, and the database 1350 is used to store data.

[0220] The computing device 1300 also includes an access device 1340 that enables the computing device 1300 to communicate via one or more networks 1360. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1340 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0221] In one embodiment of the present specification, the above components of the computing device 1300 and Fig.13 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig.13 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0222] The computing device 1300 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 1300 may also be a mobile or stationary server.

[0223] The processor 1320 is used to execute the following computer executable instructions, which implement the steps of the above-mentioned task processing method when executed by the processor.

[0224] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the above-mentioned task processing method belong to the same concept, and the details not described in detail in the technical scheme of the computing device can be referred to the description of the technical scheme of the above-mentioned task processing method.

[0225] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which can implement the steps of the above-mentioned task processing method when executed by a processor.

[0226] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the above-mentioned task processing method belong to the same concept, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the above-mentioned task processing method.

[0227] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned task processing method.

[0228] The above is a schematic scheme of a computer program of this embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the above task processing method belong to the same concept, and the details not described in detail in the technical scheme of the computer program can be referred to the description of the technical scheme of the above task processing method.

[0229] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0230] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0231] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0232] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0233] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A task processing method, applied to a service unit in a task processing platform, wherein the task processing platform comprises the service unit and a plurality of processing components, the method comprising: receiving a task processing request, wherein the task processing request carries data to be processed; Inputting the data to be processed and the component description information into a task processing model, and determining at least one target processing component corresponding to the task processing request, wherein the task processing model is a deep learning model; The at least one target processing component is called to process the data to be processed to obtain a task processing result corresponding to the task processing request.

2. The method according to claim 1, before inputting the to-be-processed data and component description information into the task processing model and determining at least one target processing component corresponding to the task processing request, further comprising: Obtain reference text data; Matching the data to be processed with the reference text data to obtain a matching result; The step of inputting the to-be-processed data and component description information into a task processing model and determining at least one target processing component corresponding to the task processing request includes: When the matching result is a matching failure, the data to be processed and the component description information are input into a task processing model to determine at least one target processing component corresponding to the task processing request.

3. The method according to claim 2, after matching the data to be processed with the reference text data and obtaining a matching result, further comprises: When the matching result is a successful match, the data to be processed and the reference text data are input into a task processing model to determine a task processing result corresponding to the task processing request.

4. According to the method of claim 1, the training method of the task processing model comprises: Acquire a sample set, wherein the sample set includes a plurality of sample data, the sample data carries a sample label, and the sample data is constructed based on at least one of sample text data, sample question-answer pair data, and sample component call data; Extracting first sample data from the sample set, wherein the first sample data is any one of the multiple sample data; Inputting the first sample data into a task processing model to obtain a first prediction result of the first sample data; Calculating a first loss value according to the first prediction result and a first sample label carried by the first sample data; The model parameters of the task processing model are adjusted according to the first loss value, and the step of extracting the first sample data from the sample set is returned to be executed until a first preset stop condition is reached, thereby obtaining a trained task processing model.

5. The method according to claim 4, wherein calculating the first loss value according to the first prediction result and the first sample label carried by the first sample data comprises: Parsing the first prediction result to obtain a plurality of first prediction sub-results; Determining, according to the type of each first prediction sub-result, a weight corresponding to each first prediction sub-result; A first loss value is calculated according to the multiple first prediction sub-results, the weights corresponding to the first prediction sub-results, and the first sample label carried by the first sample data.

6. The method according to claim 1, before calling the at least one target processing component to process the to-be-processed data and obtaining the task processing result corresponding to the task processing request, further comprises: Obtaining component call parameters of the at least one target processing component; Performing format conversion on the data to be processed according to the component calling parameters to obtain updated data to be processed; The calling of the at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request includes: The at least one target processing component is called to process the updated data to be processed to obtain a task processing result corresponding to the task processing request.

7. The method according to claim 1, wherein after calling the at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the method further comprises: Acquire call information of the at least one target processing component, and optimize the task processing model based on the call information.

8. According to the method of claim 1, the task processing request also carries task requirement information; After calling the at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the method further includes: The task requirement information and the task processing result are input into the task processing model to obtain an updated task processing result, wherein the updated task processing result meets the task requirement information.

9. The method according to any one of claims 1 to 8, wherein after calling the at least one target processing component to process the data to be processed and obtaining the task processing result corresponding to the task processing request, the method further comprises: The task processing result is sent to the client, so that the client displays the task processing result to the user.

10. The method according to claim 9, after sending the task processing result to the client, further comprising: A post-processing request sent by a user based on the task processing result is received, and the post-processing request is processed to obtain a post-processing result corresponding to the post-processing request.

11. A task processing system, comprising a communication interface and a calling unit; The communication interface is used to receive a processing request sent by a client and forward the processing request to the calling unit, wherein: The processing request carries the data to be queried; The calling unit is used to call the query model to process the data to be queried and the processing model description information in response to the processing request, and determine at least one target processing model corresponding to the data to be queried, wherein the query model is a deep learning model; and send the model information of the at least one target processing model to the client through the communication interface.

12. The system according to claim 11, wherein the communication interface is further used to receive the model adjustment parameters sent by the client and forward the model adjustment parameters to the calling unit; The calling unit is further used to adjust the at least one target processing model according to the model adjustment parameter, and send the adjusted model information of the at least one target processing model to the client through the communication interface.

13. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 10 are implemented.

14. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

Images