AI task processing method and device based on assembly line mode, computer equipment and readable storage medium

Abstract

The invention discloses an AI task processing method and device based on an assembly line mode, computer equipment and a readable storage medium, and the method comprises the steps: firstly, a server responds to a task demand, collects basic information, constructs an AI assembly line containing structure definition, role task setting and storage circulation configuration according to the basic information, and encapsulates an AI assistant by a cue word; receiving a query information triggering task, calling AI assistants according to an assembly line sequence, taking output of the AI assistants as follow-up input and storing the follow-up input in sequence, and finally displaying a generated result to a user. According to the method, the complex task processing capability can be effectively improved on the premise that the capability of a single large model is not enhanced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a pipeline-based AI task processing method, apparatus, computer equipment, and readable storage medium. Background Art

[0002] With the increasing application of artificial intelligence, efficiently handling complex AI tasks has become a pressing issue. Traditional AI task processing methods often lack systematicity and process-based approaches. When faced with diverse task demands, they suffer from low task execution efficiency and poor information flow, making it difficult to meet users' expectations for fast and accurate AI services. Therefore, a new approach is needed to optimize the AI ​​task processing process and improve its efficiency and quality. Summary of the Invention

[0003] The purpose of the present invention is to provide an AI task processing method, device, computer equipment and readable storage medium based on a pipeline mode.

[0004] In a first aspect, an embodiment of the present invention provides an AI task processing method based on a pipeline mode, comprising:

[0005] In response to a task requirement input by a user, collecting basic information related to the task requirement;

[0006] Building an AI pipeline based on the basic information, the AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow, wherein the AI ​​assistant is encapsulated by prompt words and is used to perform specific tasks, and the pipeline structure determines the steps and order of task processing;

[0007] Receive query information input by the user to trigger task execution, call each AI assistant to perform the task in sequence according to the pipeline order, the output of each AI assistant serves as the input of the next AI assistant, and the system stores the output content of each AI assistant;

[0008] The final output generated by the AI ​​assistant is presented to the user as the final result.

[0009] In a possible implementation, the task requirement includes generating an entrepreneur tag task, and the step of collecting basic information related to the task requirement in response to the task requirement input by the user includes:

[0010] According to the task of generating entrepreneur labels, a plurality of large model prompt word templates corresponding to the task of generating entrepreneur labels are obtained.

[0011] In one possible implementation, the setting of multiple AI assistant roles and tasks includes:

[0012] Set up three AI assistant roles: data searcher, content extractor and copywriter, among which:

[0013] The task of the data searcher is to complete data collection based on the provided enterprise information or entrepreneur information, and collect structured presentation data;

[0014] The task of the content refiner is to refine the content based on the given enterprise information and entrepreneur information, and highlight specific features;

[0015] The task of the copywriter is to output content in a specific format.

[0016] In a possible implementation, the plurality of large model prompt word templates include a data search prompt word template, a content extraction prompt word template, and a copywriting output prompt word template;

[0017] The system receives query information input by the user to trigger task execution, calls each AI assistant to perform the task in sequence according to the pipeline order, and the output of each AI assistant is used as the input of the next AI assistant. The system also stores the output content of each AI assistant, including:

[0018] Receive query information input by a user, wherein the query information includes a company name and an entrepreneur name;

[0019] The data searcher places the company name into the data search prompt word template to obtain a first complete search prompt word;

[0020] Searching the first complete search prompt word using a preset large model to obtain and store enterprise information;

[0021] The data searcher places the company name and the entrepreneur name into the data search prompt word template to obtain a second complete search prompt word;

[0022] Searching according to the second complete search prompt word through the preset large model to obtain and store the entrepreneur information;

[0023] The content refiner places the enterprise information and the entrepreneur information into the content refinement prompt word template to obtain a complete refined prompt word;

[0024] Refining the complete refined prompt words through the preset large model to obtain a refined result for the task of generating entrepreneur labels, and storing the refined result;

[0025] The copywriter places the extracted results into the copy output prompt word template to obtain a complete output prompt word;

[0026] The target entrepreneur label is obtained by generating the target entrepreneur label based on the complete output prompt word through the preset large model.

[0027] In a possible implementation, the receiving query information input by the user to trigger task execution, the method further includes:

[0028] Receive query information input by the user on the preset page to trigger task execution;

[0029] The method of presenting the output generated by the final AI assistant to the user as the final result includes:

[0030] The target entrepreneur tag is displayed on the preset page.

[0031] In one possible implementation, the method further includes:

[0032] Receive user feedback in detail and optimize the multiple large model prompt word templates or pipeline structures.

[0033] In a possible implementation, when searching, refining, or generating using a preset large model, the method further includes a step of performing a quality assessment on an output result of the large model, which step includes:

[0034] Each large model output is scored based on pre-set quality assessment indicators, including but not limited to information accuracy, completeness, language fluency, and fit with task requirements.

[0035] When the evaluation score is lower than the preset threshold, the parameters or content of the corresponding prompt word are automatically adjusted, and the large model is called again for processing until the evaluation score of the output result reaches or exceeds the preset threshold.

[0036] In a second aspect, an embodiment of the present invention provides an AI task processing device based on a pipeline mode, comprising:

[0037] A preparation module, configured to respond to a task requirement input by a user and collect basic information related to the task requirement; construct an AI pipeline based on the basic information, wherein the AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow. The AI ​​assistant is encapsulated by prompt words and is used to perform a specific task. The pipeline structure determines the steps and order of task processing;

[0038] The execution module is used to receive query information input by the user to trigger task execution, call each AI assistant in sequence to perform the task in a pipeline order, and the output of each AI assistant is used as the input of the next AI assistant. The system stores the output content of each AI assistant; the output generated by the final AI assistant is displayed to the user as the final result.

[0039] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor and a non-volatile memory storing computer instructions, wherein when the computer instructions are executed by the processor, the computer device executes the method described in the first aspect.

[0040] In a fourth aspect, an embodiment of the present invention provides a readable storage medium, wherein the readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method described in the first aspect.

[0041] Compared with the existing technology, the beneficial effects provided by the present invention include: using the AI ​​task processing method, device, computer equipment and readable storage medium based on the pipeline mode disclosed in the present invention, the server first responds to the task requirements and collects basic information, and then builds an AI pipeline containing structure definition, role task setting and storage flow configuration. The AI ​​assistant is encapsulated by prompt words. Then, the query information is received to trigger the task, and the AI ​​assistant is called in the pipeline sequence. Its output is sequentially used as subsequent input and stored, and the generated results are finally displayed to the user. This method can effectively improve the processing capabilities of complex tasks without enhancing the capabilities of a single large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly describes the drawings required for use in the embodiments. It should be understood that the following drawings illustrate only certain embodiments of the present invention and should not be construed as limiting the scope of the present invention. Those skilled in the art can, without inventive effort, derive other relevant drawings from these drawings.

[0043] Figure 1 A schematic diagram of the steps of the pipeline-based AI task processing method provided by an embodiment of the present invention;

[0044] Figure 2 A schematic block diagram of the structure of an AI task processing device based on a pipeline mode provided by an embodiment of the present invention;

[0045] Figure 3 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the described embodiments are only a portion of the embodiments of the present invention, not all of them. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

[0047] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0048] In order to solve the technical problems in the above background technology, Figure 1 The following is a flow chart of the pipeline-based AI task processing method provided in an embodiment of the present disclosure. The pipeline-based AI task processing method is introduced in detail below.

[0049] Step S201, in response to a task requirement input by a user, collecting basic information related to the task requirement;

[0050] Step S202: Building an AI pipeline based on the basic information. The AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow. The AI ​​assistant is encapsulated by prompt words and is used to perform specific tasks. The pipeline structure determines the steps and order of task processing.

[0051] Step S203: Receive query information input by the user to trigger task execution, call each AI assistant to perform the task in sequence according to the pipeline order, and use the output of each AI assistant as the input of the next AI assistant. The system also stores the output of each AI assistant.

[0052] Step S204: Display the final output generated by the AI ​​assistant to the user as the final result.

[0053] In this embodiment of the present invention, for example, assume that the server is an AI task processing platform serving enterprises, and the user is an enterprise operator. After logging into the platform, the operator enters the task requirement "Generate Entrepreneur Tag Task" on the front-end page. After receiving this task requirement, the server retrieves multiple large model prompt word templates corresponding to the Generate Entrepreneur Tag Task based on the task requirement.

[0054] For example, the server retrieves pre-set templates for data search prompts, content extraction prompts, and copywriting output prompts from its storage system. These templates are designed and optimized by professionals and are used to guide subsequent AI assistants in performing corresponding tasks. The server also collects other relevant information. For example, when entering task requirements, users may include special requirements for entrepreneur tags, such as word limits and highlighting specific aspects. This information is recorded as foundational information for subsequent processing.

[0055] The server starts building the AI ​​pipeline based on the collected basic information.

[0056] Define the pipeline structure: The server determines the steps and order for processing the task of generating entrepreneur tags. For example, the first step is data search, the second step is content extraction, and the third step is copywriting output. This clarifies the order of task processing and ensures that each step proceeds in an orderly manner.

[0057] Set up multiple AI assistant roles and tasks:

[0058] Data Searcher: The server sets up a data searcher AI assistant role, whose task is to complete data collection based on the provided company information or entrepreneur information and present the collected data in a structured manner. For example, when the server subsequently receives the company name "ABC Technology Co., Ltd." and the entrepreneur name "Zhang San" input by the user, the data searcher will search for information related to the company and entrepreneur through various data sources (such as corporate databases, news information websites, etc.), such as the company's business scope, development history, market share, and the entrepreneur's education background, professional experience, and honors received, and organize this information into a structured format for subsequent processing.

[0059] Content Distiller: A content distiller role is created, tasked with extracting content from given company and entrepreneur information to highlight specific characteristics. For example, based on the company and entrepreneur profiles provided by the data searcher, the content distiller should identify information that best reflects the entrepreneur's unique characteristics and core competitiveness, such as whether entrepreneur Zhang San was the first to propose a certain innovative technological concept in the industry or whether ABC Technology Co., Ltd. holds a leading market share in a specific niche. This key information should be extracted.

[0060] Output Copywriter: Create a Output Copywriter role, tasked with outputting content according to a specific format. In this case, the format might require a label of 20 characters or less, including information such as the main business, region, achievements, and status. Based on the refined results provided by the Content Extractor, the Output Copywriter creates the final Entrepreneur Label in this format.

[0061] Configuring system storage and data flow: The server is configured with system storage to store the output of each AI assistant. For example, the company and entrepreneur information searched by the data searcher will be stored in a specific storage area, and the results extracted by the content refiner will also be stored to facilitate subsequent AI assistant calls. At the same time, the server sets data flow rules to ensure that the output of each AI assistant can accurately serve as the input for the next AI assistant. For example, the content refiner obtains the output of the data searcher from the storage area for processing, and the copywriter obtains the output of the content refiner from the storage area to generate labels.

[0062] The user enters query information on a preset page (such as the task submission page of the platform), assuming that the query information is the company name "ABC Technology Co., Ltd." and the entrepreneur name "Zhang San". After the server receives the query information on the preset page, it triggers the task execution.

[0063] Invoking a data searcher: The server, through the data searcher, places the company name "ABC Technology Co., Ltd." into the data search prompt template, generating a first complete search prompt. For example, "You are a data searcher. Your task is to collect data based on the provided company information 'ABC Technology Co., Ltd.'. The collected data must be presented in a structured manner: the company's business scope, development history, market share, etc." The server then searches according to the first complete search prompt using a preset macro model (such as ChatGPT), obtaining company data from various data sources and storing this data in a system-defined storage area. Next, the server, through the data searcher, again places the company name "ABC Technology Co., Ltd." and the entrepreneur name "Zhang San" into the data search prompt template, generating a second complete search prompt. For example, "You are a data searcher. Your task is to collect data based on the provided company and entrepreneur information 'ABC Technology Co., Ltd., Zhang San'. The collected data must be presented in a structured manner: the entrepreneur's education background, professional experience, honors, etc." The server searches according to the second complete search prompt using the preset macro model, obtains the entrepreneur's data, and stores it.

[0064] Calling a content refiner: The server uses the content refiner to place the company and entrepreneur profiles into a content refinement prompt template to generate a complete refinement prompt. For example, "You are a content refiner. Your task is to extract the best features of the provided company information {company profile content} and entrepreneur information {entrepreneur profile content}, and to showcase their unique characteristics as much as possible." The server then uses a preset large model to extract the complete refinement prompts, generating a refinement result for the task of generating entrepreneur labels. For example, "ABC Technology Co., Ltd. is a leader in artificial intelligence technology, and Zhang San is an industry innovation pioneer." The server then stores the refined result in the system storage area.

[0065] Calling a copywriter: The server uses the copywriter to place the extracted results into the copywriting output prompt template to obtain the complete output prompt. For example, "You are a copywriter. Your task is to output the provided company and entrepreneur information, 'ABC Technology Co., Ltd. is a leader in artificial intelligence technology, and Zhang San is an industry innovation pioneer,' in the format of the first label. Please first learn what the 'first label' is. The definition of the first label is: Use the most concise language to accurately refine your core capabilities and exchangeable value into unique and exclusive characteristics. The requirement is to generate a 'first label' of 20 words or less, which must include 'main business, region, achievements and status'." Based on the complete output prompt, the server generates the target entrepreneur label using a preset large model, for example, "artificial intelligence field (region), Zhang San is a pioneer (achievements and status)."

[0066] The server displays the target entrepreneur label "Zhang Sanchuang is a pioneer in the field (region) of artificial intelligence (achievements and status)" on the preset page, which is the page where the user submits the task. The user can see the final entrepreneur label after logging into the platform.

[0067] In addition, during the entire process, if the server receives user feedback, such as if the user believes that the generated label does not highlight the aspects he wants, the server will optimize the multiple large model prompt word templates or pipeline structures. At the same time, when searching, refining or generating through the preset large model, the server will score the content output by the large model each time based on preset quality assessment indicators (such as information accuracy, completeness, language fluency, and fit with task requirements, etc.). When the evaluation score is lower than the preset threshold, the parameters or content of the corresponding prompt word are automatically adjusted, and the large model is called again for processing until the evaluation score of the output result reaches or exceeds the preset threshold to ensure that the final generated result can meet the user's needs.

[0068] In an embodiment of the present invention, the task requirement includes generating an entrepreneur tag task, and the task requirement input in response to the user is collected with basic information related to the task requirement, which can be implemented through the following examples.

[0069] According to the task of generating entrepreneur labels, a plurality of large model prompt word templates corresponding to the task of generating entrepreneur labels are obtained.

[0070] In an embodiment of the present invention, for example, the server, as the core carrier of AI task processing, will perform a series of orderly operations to collect relevant basic information when faced with the task of generating entrepreneur labels input by the user.

[0071] On an enterprise service platform, a user logged into the task submission interface and explicitly entered the task requirement as "Generate an entrepreneur tag task." The server keenly captured this input and quickly took action.

[0072] The server first searches its vast and organized storage system. This storage system is like a vast knowledge repository, pre-stored with large-scale model prompt word templates corresponding to various tasks. Since the task is to generate entrepreneur tags, the server accurately locates the storage area related to this task.

[0073] In this specific area, there are multiple large-scale prompt word templates with different functions. The first one is the data search prompt word template, which is like a detailed work instruction given to the data searcher. For example, the template can contain the following content structure: "You are a data searcher, and your task is to complete comprehensive data collection based on the provided company name and entrepreneur name. The collected content must cover the basic information of the company, such as the establishment time, registered capital, and registered address; the business scope of the company, including the main products or services, target customer groups; the market performance of the company, such as market share, industry ranking, etc. For entrepreneurs, their personal resumes should be collected, including educational background, work experience; personal achievements, such as awards and honorary titles received; information related to their influence in the industry, etc. The collected information is required to be presented in a structured form to facilitate subsequent processing."

[0074] Next is the content distillation prompt template. This is a tailored work guide for content distillers. It might read: "You are a content distiller. Based on the provided company and entrepreneur information, your task is to extract the most representative and unique content. Focus on highlighting the company's competitive advantages in the industry, such as unique technology and innovative business models; and highlight the entrepreneur's personal strengths, such as outstanding leadership and forward-looking strategic vision. The distillation should be concise and clear, accurately grasping the core points."

[0075] Finally, there's the output copywriting prompt template, which provides the final work direction for the output copywriter: "You are a output copywriter. Your task is to extract content based on the provided companies and entrepreneurs and output the entrepreneur's label according to a specific format. The format requires that the label be no more than 25 words and must clearly reflect the entrepreneur's main business area, region (if there are significant regional characteristics), and outstanding achievements or status. When outputting, the language should be concise, vivid, and appealing."

[0076] The server retrieves these large-scale model prompt word templates corresponding to the entrepreneur tag generation task, serving as essential foundational information for subsequent AI pipeline construction and task execution. These templates detail the work content and requirements of different AI assistant roles, laying a solid foundation for the smooth implementation of the entire entrepreneur tag generation task.

[0077] In an embodiment of the present invention, the setting of multiple AI assistant roles and tasks can be implemented through the following examples.

[0078] Set up three AI assistant roles: data searcher, content extractor and copywriter, among which:

[0079] The task of the data searcher is to complete data collection based on the provided enterprise information or entrepreneur information, and collect structured presentation data;

[0080] The task of the content refiner is to refine the content based on the given enterprise information and entrepreneur information, and highlight specific features;

[0081] The task of the copywriter is to output content in a specific format.

[0082] In this embodiment of the present invention, for example, after receiving a user's request to generate an entrepreneur tag and obtaining the relevant large-scale model prompt word template, the server will then set up multiple AI assistant roles and tasks to build a complete AI pipeline to handle the task. The following example uses a company called "Xingchen Technology" and its entrepreneur "Li Hua" as an example.

[0083] The mission of the AI ​​assistant role of server setting data searcher is to comprehensively and accurately collect data related to enterprises and entrepreneurs and present it in a structured form.

[0084] The server first places the company name "Xingchen Technology" entered by the user into the data search prompt template and generates the first complete search prompt: "You are a data searcher. Your task is to complete the data collection based on the provided company information "Xingchen Technology". The collected data must be presented in a structured manner. Specifically, it includes the company's basic information, such as establishment time, registered capital, and registered address; the company's business information, such as main products or services, R&D investment, and core technology; the company's market information, such as market share, main competitors, industry rankings, etc." Then, based on this complete search prompt, the server calls the preset large model and searches by connecting to various data sources such as corporate databases, industry information platforms, and news websites.

[0085] After searching, the server obtained information that "Xingchen Technology" was established in 2015 with a registered capital of 50 million yuan and a registered address in a certain science and technology park; its main products are smart sensors, and its R&D investment accounts for 15% of its annual revenue. It has a number of core patented technologies; it occupies a 10% share in the smart sensor market, and its main competitors are "Qiming Technology" and "Langyue Electronics", and it ranks among the top five in the industry. The information is then stored in the designated storage area of ​​the system in a structured table format.

[0086] Subsequently, the server placed the company name "Xingchen Technology" and the entrepreneur's name "Li Hua" together in the data search prompt template to generate a second complete search prompt: "You are a data searcher. Your task is to complete the data collection based on the provided company and entrepreneur information "Xingchen Technology, Li Hua". The collected data must be presented in a structured manner. Specifically, it covers the entrepreneur's personal information, such as educational background and work experience; personal achievements, such as scientific and technological awards and published academic papers; influence in the industry, such as industry positions held and participation in the formulation of industry standards." The server called the preset large model again for a search and obtained information such as Li Hua graduated from a well-known university with a degree in electronic engineering, worked in many large technology companies, won the XX Science and Technology Progress First Prize, published several academic papers in the field of sensors, and served as vice president of an industry association, and the information was also stored in a structured form.

[0087] The server sets up the role of content extractor, whose task is to extract key content from the enterprise and entrepreneur information collected by the data searcher and highlight specific features.

[0088] The server retrieves the collected and stored information about the company "Xingchen Technology" and entrepreneur Li Hua from the system storage area, places this information into the content extraction prompt template, and generates a complete extraction prompt: "You are a content extractor. Your task is to extract the content that best reflects the company's core competitiveness and the entrepreneur's unique value based on the provided information about the company "Xingchen Technology" and the entrepreneur "Li Hua." Focus on highlighting the company's characteristics in terms of technological innovation and market advantages, and the entrepreneur's characteristics in terms of professional ability and industry influence. The extracted content must be concise and accurate."

[0089] Based on this complete and refined prompt, the server invokes a pre-set large model for extraction. The resulting extraction results are: Regarding the company, Xingchen Technology, with its advanced core intelligent sensor technology, holds a leading position in the market and enjoys a high market share; regarding the entrepreneur, Li Hua, an electronic engineering professional with extensive academic expertise and industry influence in the sensor field, is a key figure in driving industry development. The server stores these extraction results in the system's storage area for subsequent use by the copywriter.

[0090] The server sets up the role of copywriter, whose task is to convert the content refined by the content refiner into the final entrepreneur label according to a specific format.

[0091] The server retrieves the content extractor's output from the system storage area and places it into the copy output prompt template, generating a complete output prompt: "You are a copy output operator. Your task is to extract the content 'Xingchen Technology has a leading position in the market with its advanced intelligent sensor core technology, and Li Hua, as an electronic engineering professional, has profound academic accumulation and industry influence in the sensor field' based on the provided company and entrepreneur, and output the entrepreneur label in a specific format. The format requires the generated label to be no more than 20 characters and must include key information such as the main business, region (which can be omitted if there is no significant regional feature), achievements, and status."

[0092] Based on the complete output prompt, the server invokes a pre-set large model to generate the target entrepreneur's label: "Li Hua, a technical leader in the field of smart sensors." This label is then prepared for display to the user.

[0093] Through the above setting and execution of the three AI assistant roles and their tasks, the server gradually advances the task of generating entrepreneur labels and ultimately provides users with results that meet their requirements.

[0094] In the embodiment of the present invention, the plurality of large model prompt word templates include a data search prompt word template, a content extraction prompt word template and a copy output prompt word template;

[0095] The query information received from the user is used to trigger the execution of the task, and each of the AI ​​assistants is called in sequence to perform the task in the order of the pipeline. The output of each AI assistant is used as the input of the next AI assistant, and the system stores the output content of each AI assistant. It can be implemented through the following examples.

[0096] Receive query information input by a user, wherein the query information includes a company name and an entrepreneur name;

[0097] The data searcher places the company name into the data search prompt word template to obtain a first complete search prompt word;

[0098] Searching the first complete search prompt word using a preset large model to obtain and store enterprise information;

[0099] The data searcher places the company name and the entrepreneur name into the data search prompt word template to obtain a second complete search prompt word;

[0100] Searching according to the second complete search prompt word through the preset large model to obtain and store the entrepreneur information;

[0101] The content refiner places the enterprise information and the entrepreneur information into the content refinement prompt word template to obtain a complete refined prompt word;

[0102] Refining the complete refined prompt words through the preset large model to obtain a refined result for the task of generating entrepreneur labels, and storing the refined result;

[0103] The copywriter places the extracted results into the copy output prompt word template to obtain a complete output prompt word;

[0104] The target entrepreneur label is obtained by generating the target entrepreneur label based on the complete output prompt word through the preset large model.

[0105] In this embodiment of the present invention, suppose a user logs in to the platform and enters query information on the task submission page, where the company name is "Yunhai Software Co., Ltd." and the entrepreneur name is "Zhang Feng." After the server receives this query information entered by the user on a preset page, the task is officially triggered.

[0106] The server first activates the AI ​​assistant role of data searcher. It places the company name "Yunhai Software Co., Ltd." entered by the user into the data search prompt template. The content of the data search prompt template may be as follows: "You are a data searcher. Your task is to complete the data collection based on the provided company information. The collected data needs to be presented in a structured manner. Specifically, it includes basic information of the company, such as the year of establishment, registration location, and registered capital; the business scope of the company, including the main software product types and service objects; the market performance of the company, such as the revenue data of the past three years, market share, and industry ranking; the R&D capabilities of the company, such as the size of the R&D team, core technology achievements, and the number of patents."

[0107] After placing the company name, the first complete search prompt appears: "You are a data searcher. Your task is to collect data based on the provided company information, 'Yunhai Software Co., Ltd.'." The collected data must be presented in a structured manner. This includes basic company information, such as year of establishment, registration location, and registered capital; business scope, including the main software product types and service targets; market performance, such as revenue data, market share, and industry ranking over the past three years; and R&D capabilities, such as the size of the R&D team, core technological achievements, and number of patents."

[0108] Based on the first complete search query, the server invoked a pre-set large model and conducted a search by connecting to multiple data sources, including enterprise information databases, industry reporting platforms, and news websites. The search yielded the following company information: Yunhai Software Co., Ltd. was founded in 2010 and registered in a certain city's high-tech industrial park, with registered capital of 80 million yuan. Its primary software products are enterprise management and data analysis software, primarily serving large manufacturing companies and financial institutions. Its revenue over the past three years was 120 million yuan, 150 million yuan, and 180 million yuan, respectively, accounting for 8% of the enterprise management software market share and ranking among the top ten in the industry. Its R&D team consists of 200 people and holds 20 software copyrights and 5 invention patents. The server then stored this company information in a structured table format in a designated storage area within the system.

[0109] Next, the server again uses the data searcher to place the company name "Yunhai Software Co., Ltd." and the entrepreneur name "Zhang Feng" into the data search prompt template, generating a second complete search prompt: "You are a data searcher. Your task is to complete the data collection based on the provided company and entrepreneur information 'Yunhai Software Co., Ltd., Zhang Feng'. The collected data must be presented in a structured manner. Specifically, it includes the entrepreneur's personal background information, such as the school they graduated from, their major, and their educational level; their work experience, including previous companies and positions, and their tenure; their personal achievements, such as industry awards they have received, important academic papers they have published, and industry standards they have participated in; and their social positions within the industry, such as those they hold in industry associations."

[0110] The server used the pre-set large model to search based on the second complete search prompt. The resulting entrepreneur profile included: Zhang Feng graduated with a master's degree in computer science from a renowned domestic university; previously worked as a senior R&D manager and product director at two well-known software companies for three and four years, respectively; won a provincial software industry innovation award, published five academic papers on software development and data analysis, and served as vice president of the municipal software industry association. The server also stored this entrepreneur profile in a structured format in the system storage area.

[0111] The server then activates the AI ​​assistant role of content refiner. It retrieves previously stored company and entrepreneur information from the system storage area and places it into a content refinement prompt template. The content of the content refinement prompt template roughly reads: "You are a content refiner. Your task is to refine content based on the provided company and entrepreneur information, highlighting the specific characteristics of the company in terms of technological innovation and market competitiveness, and the specific characteristics of the entrepreneur in terms of professional ability and industry influence. The refined results must be concise, accurate, and able to intuitively reflect the core points."

[0112] After placing the data, I received a complete extraction prompt: "You are a content extractor. Your task is to extract the content based on the provided company information, 'Yunhai Software Co., Ltd. was established in 2010 and is registered in a certain city's high-tech industrial park...' and the entrepreneur information, 'Zhang Feng graduated from a well-known domestic university with a degree in computer science...'. This will highlight the company's specific characteristics in terms of technological innovation and market competitiveness, and the entrepreneur's specific characteristics in terms of professional ability and industry influence. The extraction results must be concise, accurate, and able to intuitively reflect the core points."

[0113] Based on the complete extraction prompt, the server invokes a pre-set large model for extraction. The final extraction results are: Regarding the enterprise, Yunhai Software demonstrates strong technological innovation and a significant market share in the enterprise management software field; regarding the entrepreneur, Zhang Feng, a highly educated computer science graduate, possesses outstanding expertise in the software industry and considerable industry influence. The server stores the extraction results in the system storage area.

[0114] Finally, the server activates the Copywriter role. It retrieves the content extractor's output from the system storage area and places it into the copywriter output prompt template. The copywriter output prompt template reads: "You are a copywriter. Your task is to extract content from the provided companies and entrepreneurs and output entrepreneur tags in a specific format. The format requires that the generated tags be no more than 20 words and include key information such as the main business, achievements, and status."

[0115] After placing the extracted results, a complete output prompt appears: "You are a copywriter. Your task is to extract the content based on the provided company and entrepreneur: 'Yunhai Software has strong technological innovation capabilities in the field of enterprise management software and a considerable market share; Zhang Feng, as a highly educated computer science graduate, has outstanding professional capabilities in the software industry and has a significant influence in the industry,' and output the entrepreneur's label in a specific format. The format requires the generated label to be no more than 20 characters and must include key information such as the main business, achievements, and status."

[0116] Based on the complete output prompt, the server calls the pre-set large model to generate the target entrepreneur's label: "Zhang Feng, an influential figure in the enterprise management software field." This label is stored and displayed to the user.

[0117] During the entire task execution process, the server calls each AI assistant in sequence in strict accordance with the pipeline order to perform the task. The output of each AI assistant is accurately used as the input of the next AI assistant, and the system completely stores the output content of each AI assistant to ensure the consistency and traceability of task processing.

[0118] In the embodiment of the present invention, the method of receiving query information input by the user to trigger task execution further includes:

[0119] Receive query information input by the user on the preset page to trigger task execution;

[0120] The output generated by the final AI assistant is presented to the user as the final result, which can be implemented through the following example.

[0121] The target entrepreneur tag is displayed on the preset page.

[0122] In an embodiment of the present invention, for example, in the AI ​​task processing platform for enterprise services, the server has built a complete AI pipeline to handle the task of generating entrepreneur tags. The next step is the key link of interacting with users - receiving query information and displaying results.

[0123] The platform features a dedicated preset page, where users submit task requests and retrieve results. A business consulting employee logged into the platform and entered the preset page. He accurately entered the company name, "Xinghe Network Technology Company," and the entrepreneur's name, "Chen Li," into the corresponding input boxes and clicked the Submit button. The server, constantly monitoring user input on the preset page, instantly detected this action and received the query containing "Xinghe Network Technology Company" and "Chen Li," officially triggering the task execution.

[0124] The server then follows a pre-defined AI pipeline, sequentially invoking AI assistants such as data searchers, content extractors, and copywriters. Data searchers, using relevant prompt templates, search for and store detailed information on companies and entrepreneurs across various data sources. Content extractors retrieve and extract information from storage, highlighting key features before storing it again. Copywriters use the extracted information to generate a tag for the target entrepreneur, "Chen Li, an innovation leader in the field of internet technology," in a specific format.

[0125] After the task is completed, the server will provide feedback to the user. It will generate the target entrepreneur tag, "Chen Li, an innovation leader in the field of Internet technology," and send it to a pre-set page. The person who submitted the task will then see the tag displayed in a designated area of ​​the page, likely a section dedicated to presenting the task results, without any additional steps.

[0126] The entire process is smooth and efficient. The pre-set page serves as a window for user-server interaction, clearly bridging the gap between task input and output. The server accurately receives query information on the pre-set page, triggering the task and promptly displays the target entrepreneur's tag there, enabling users to quickly access the content they need, meeting their needs in practical scenarios like business analysis and personalization. This demonstrates the convenience and practicality of this AI task processing method in real-world applications.

[0127] In the embodiments of the present invention, the following implementation modes are also provided.

[0128] Receive user feedback in detail and optimize the multiple large model prompt word templates or pipeline structures.

[0129] In an embodiment of the present invention, illustratively, after the server completes the task of generating entrepreneur tags and displays the target entrepreneur tags on a preset page, user feedback becomes a key basis for further optimization.

[0130] Suppose a user sees the server-generated tag "Chen Li, an innovative leader in the internet technology field" for "Xinghe Network Technology Company" and entrepreneur "Chen Li" and feels that it fails to adequately reflect the company's unique advantages in a specific market segment, nor does it highlight the entrepreneur's leading position in a key technology. The user then finds the feedback portal on the pre-set page and fills in their detailed feedback: "The company holds a large market share in social network data analysis software, and the algorithm developed by Chen Li is groundbreaking in the industry. I hope the tag can reflect this."

[0131] The server continuously monitors the feedback information on the preset page and quickly receives the feedback content submitted by the user. Next, the server will analyze and process the feedback information to determine whether to optimize multiple large model prompt word templates or pipeline structures.

[0132] If the server determines that the primary issue is inaccurate information extraction, it prioritizes optimizing the content extraction prompt template. It will add guidance related to user feedback to the existing content extraction prompt template, such as: "You are a content extractor. Your task is to extract content based on the provided company and entrepreneur information, specifically highlighting specific characteristics such as the company's strengths in a specific market segment and the entrepreneur's leading position in key technologies. The extracted results must be concise, accurate, and clearly reflect the core points." This allows the extractor to more accurately extract content that meets user expectations based on the optimized prompts for similar tasks.

[0133] If the server analysis reveals that, from the overall process perspective, the data search phase is insufficiently comprehensive in collecting information related to market segments and key technologies, resulting in subsequent extraction and output failing to meet user needs, the pipeline structure and data search prompt template will be optimized. For example, within the pipeline structure, the data search phase will increase the search weight for information in specific fields; at the same time, the data search prompt template will be supplemented with the following: "It is necessary to focus on collecting business data and market share in the market segments, as well as the entrepreneur's contributions to key technology research and development, such as leading projects and innovative algorithms."

[0134] By receiving user feedback and performing targeted optimization on the large model prompt word template or pipeline structure, the server can continuously improve the processing quality of the task of generating entrepreneur labels, better meet the diverse and personalized needs of users, and make the entire AI task processing system more intelligent and efficient in practical applications.

[0135] In an embodiment of the present invention, when searching, refining or generating through a preset large model, the method further includes a step of performing a quality assessment on the output result of the large model, which can be implemented through the following example.

[0136] Each large model output is scored based on pre-set quality assessment indicators, including but not limited to information accuracy, completeness, language fluency, and fit with task requirements.

[0137] When the evaluation score is lower than the preset threshold, the parameters or content of the corresponding prompt word are automatically adjusted, and the large model is called again for processing until the evaluation score of the output result reaches or exceeds the preset threshold.

[0138] In an embodiment of the present invention, exemplarily, when the server performs the task of generating entrepreneur tags, when searching, refining or generating operations through a preset large model, it will strictly perform the quality assessment steps of the output results of the large model.

[0139] Take, for example, the task of generating labels for "Xuri Media Co., Ltd." and its entrepreneur, Zhao Gang. During the data search, the server invokes a pre-set large model to retrieve company data based on the first complete search prompt. The large model outputs information such as the company's founding date and business scope. The server then scores this output based on pre-set quality assessment metrics.

[0140] Regarding information accuracy, the server compares the acquired information with authoritative databases to determine its accuracy. Regarding completeness, the server checks whether the required information, including basic company information, business scope, and market performance, is covered. Regarding language fluency, the server assesses whether the output is coherent and natural. Furthermore, the server's alignment with task requirements determines whether the relevant information provided is accurate and relevant to the company data collection task. Assuming a maximum score of 10 for each indicator, the server would receive an 8 for accuracy, a 7 for completeness, an 8 for language fluency, and a 7 for alignment with task requirements, for an overall score of 7.5. If the preset threshold is 8 points, and 7.5 is below the threshold, the server automatically adjusts the parameters of the data search prompt, adding detailed requirements for collecting market performance information. The prompt is modified to read, "You are a data searcher. Your task is to collect data based on the provided company information, 'Sunrise Media Co., Ltd.' This data must be presented in a structured manner. This includes basic company information, such as year of establishment, registered location, and registered capital; business scope, including the primary media business type and service targets; market performance, such as revenue data, market share, and key competitors over the past five years; and R&D capabilities, such as whether it has innovative media technologies or models." The server then calls the preset large model for another search, which re-outputs more detailed and accurate company information. The server then conducts another quality assessment, and this time, the scores for all indicators have improved, reaching an overall score of 8.5, meeting the preset threshold. The output of this step is accepted and stored.

[0141] During the content extraction phase, the large model extracts the company and entrepreneur profiles based on the complete extraction prompts and outputs the results. The server also scores the results based on the aforementioned quality assessment metrics. If the score falls below the preset threshold, such as due to poor language fluency or insufficient alignment with the task requirements for highlighting specific features, the server adjusts the content extraction prompts, such as increasing language requirements or clarifying the scope of specific features, and then calls the large model again for extraction until the output meets the quality requirements.

[0142] The same applies to the copywriting output process. After the large model generates the target entrepreneur label, the server conducts a quality assessment. If the label fails to accurately reflect key information required by the task, such as the core business, achievements, and status, resulting in an evaluation score below the threshold, the server adjusts the copywriting output prompts, such as refining the formatting requirements or emphasizing the presentation of key information. The large model is then regenerated until the generated label meets the preset quality standards. Finally, the qualified target entrepreneur label is displayed to the user. Through this rigorous quality assessment and adjustment mechanism, the server ensures the high quality of the output results of the entrepreneur label generation task.

[0143] In order to more clearly describe the solution provided by the embodiment of the present invention, a relatively complete implementation method is provided below.

[0144] 1. The user enters {company name} and {name} on the front-end page;

[0145] 2. The system adds {Company Name} to the data searcher prompt template to form the complete prompt for the data searcher. The complete prompt is: You are a data searcher. Your task is to complete the data collection based on the provided company information. The collected data must be structured to present the characteristics of the company. The company name is: {Company Name};

[0146] 3. After the content refiner inputs the prompt words, the big model outputs the content {Company Introduction}, and the system stores this content;

[0147] 4. The system adds {Company + Name} to the data searcher prompt template to form the complete data searcher prompt. The complete prompt is: You are a data searcher. Your task is to complete the data collection based on the provided entrepreneur information. The collected data must be structured to present the characteristics of the entrepreneur. The entrepreneur's name is: {Name}.

[0148] 5. After the content refiner inputs the prompt words, the big model outputs the content {Entrepreneur Introduction}, and the system stores this content;

[0149] 6. The system adds the output of the large model to the content refiner's prompt template to form the content refiner's complete prompt. The complete prompt is: You are a content refiner. Your task is to extract the best features of each company and entrepreneur based on the provided information, and to showcase the characteristics of the industry leaders as much as possible. The company introduction is: {Company Introduction}, and the entrepreneur introduction is: {Entrepreneur Introduction}.

[0150] 7. After the big model receives the content refiner's prompt words, it outputs the content {Enterprise and Entrepreneur Refined Content}, and the system stores this content;

[0151] 8. The system places the content generated by the master model (the extracted content from the company and entrepreneur) into the copywriter's prompt template, forming the complete prompt for the copywriter. The complete prompt is: You are a copywriter. Your task is to output the provided company and entrepreneur information in the format of the first label. Please first learn what the "first label" is. The definition of the first label is: Use the most concise language to accurately distill your core capabilities and exchangeable value into unique and exclusive characteristics. The first label contains three elements: 1. What is your main business? 2. In which of the following regions does your main business have the highest industry status or market share? 3. Based on your company's achievements and status, what is your favorite and most accurate description?

[0152] Examples of good "first labels":

[0153] (1) Xia X: A global fashion spokesperson for Chinese elements who helped thousands of embroiderers in the mountains escape poverty and become rich;

[0154] (2) Li XX: The first private car maker in China who was born into a family of cattle herders;

[0155] (3) Cao XX: The world's automotive glass king known as China's top philanthropist;

[0156] Based on the above learning, generate the following company profile into a "first label" of less than 20 words, which must include "main business, region, achievements and status";

[0157] The company profile is as follows: {Enterprise and Entrepreneur Extracted Content};

[0158] 9. After the large model receives the prompt word input from the copywriter, it outputs the content {first label}, and the system stores this content;

[0159] 10. Show the {first label} generated by the large model to the customer.

[0160] Please refer to Figure 2 , Figure 2 An embodiment of the present invention provides an AI task processing device 110 based on a pipeline mode, including:

[0161] Preparation module 1101 is configured to respond to a task requirement input by a user and collect basic information related to the task requirement; construct an AI pipeline based on the basic information, wherein the AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow. The AI ​​assistant is encapsulated by prompt words and is used to perform a specific task. The pipeline structure determines the steps and order of task processing;

[0162] Execution module 1102 is used to receive query information input by the user to trigger task execution, call each of the AI ​​assistants to perform the task in sequence according to the pipeline order, and the output of each AI assistant is used as the input of the next AI assistant. The system stores the output content of each AI assistant; the output generated by the final AI assistant is displayed to the user as the final result.

[0163] It should be noted that the implementation principles of the aforementioned pipeline-based AI task processing device 110 can be referenced to the implementation principles of the aforementioned pipeline-based AI task processing method, and will not be elaborated upon here. It should be understood that the division of the various modules of the above device is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a single physical entity, or physically separated. Furthermore, these modules can be implemented entirely in the form of software called by a processing element; or entirely in the form of hardware; or some modules can be implemented in the form of software called by a processing element, and some modules can be implemented in the form of hardware. For example, the pipeline-based AI task processing device 110 can be a separate processing element, or it can be integrated into a chip of the aforementioned device. Furthermore, it can be stored in the form of program code in the memory of the aforementioned device, and called by a processing element of the aforementioned device to execute the functions of the aforementioned pipeline-based AI task processing device 110. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together, or implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each module above may be completed by an integrated logic circuit of hardware in a processor element or by instructions in the form of software.

[0164] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by scheduling program code on a processing element, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0165] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned AI task processing device 110 based on the pipeline mode. Figure 3 As shown, Figure 3 This is a block diagram of the structure of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes an AI task processing device 110 based on a pipeline mode, a memory 111, a processor 112, and a communication unit 113.

[0166] In order to realize the transmission or interaction of data, the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other directly or indirectly. For example, the electrical connection between these elements can be achieved through one or more communication buses or signal lines. The AI ​​task processing device 110 based on the pipeline mode includes at least one software function module that can be stored in the memory 111 in the form of software or firmware or solidified in the operating system (OS) of the computer device 100. The processor 112 is used to execute the AI ​​task processing device 110 based on the pipeline mode stored in the memory 111, such as the software function modules and computer programs included in the AI ​​task processing device 110 based on the pipeline mode.

[0167] An embodiment of the present invention provides a readable storage medium, which includes a computer program. When the computer program is running, it controls the computer device where the readable storage medium is located to execute the aforementioned AI task processing device 110 based on the pipeline mode.

[0168] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. These embodiments have been selected and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the present disclosure and to utilize various embodiments with various modifications as appropriate for the specific application contemplated.

Claims

1. A pipeline-based AI task processing method, characterized in that: include: In response to a task requirement input by a user, collecting basic information related to the task requirement; Building an AI pipeline based on the basic information, the AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow, wherein the AI ​​assistant is encapsulated by prompt words and is used to perform specific tasks, and the pipeline structure determines the steps and order of task processing; Receive query information input by the user to trigger task execution, call each AI assistant to perform the task in sequence according to the pipeline order, the output of each AI assistant serves as the input of the next AI assistant, and the system stores the output content of each AI assistant; The final output generated by the AI ​​assistant is presented to the user as the final result.

2. The method according to claim 1, characterized in that The task requirement includes generating an entrepreneur tag task, and the task requirement input by the user is responded to and basic information related to the task requirement is collected, including: According to the task of generating entrepreneur labels, a plurality of large model prompt word templates corresponding to the task of generating entrepreneur labels are obtained.

3. The method according to claim 2, characterized in that The aforementioned setting of multiple AI assistant roles and tasks includes: Set up three AI assistant roles: data searcher, content extractor and copywriter, among which: The task of the data searcher is to complete data collection based on the provided enterprise information or entrepreneur information, and collect structured presentation data; The task of the content refiner is to refine the content based on the given enterprise information and entrepreneur information, and highlight specific features; The task of the copywriter is to output content in a specific format.

4. The method according to claim 2, characterized in that Multiple large-scale model prompt word templates include data search prompt word templates, content extraction prompt word templates, and copywriting output prompt word templates; The system receives query information input by the user to trigger task execution, calls each AI assistant to perform the task in sequence according to the pipeline order, and the output of each AI assistant is used as the input of the next AI assistant. The system also stores the output content of each AI assistant, including: Receive query information input by a user, wherein the query information includes a company name and an entrepreneur name; The data searcher places the company name into the data search prompt word template to obtain a first complete search prompt word; Searching the first complete search prompt word using a preset large model to obtain and store enterprise information; The data searcher places the company name and the entrepreneur name into the data search prompt word template to obtain a second complete search prompt word; Searching according to the second complete search prompt word through the preset large model to obtain and store the entrepreneur information; The content refiner places the enterprise information and the entrepreneur information into the content refinement prompt word template to obtain a complete refined prompt word; Refining the complete refined prompt words through the preset large model to obtain a refined result for the task of generating entrepreneur labels, and storing the refined result; The copywriter places the extracted results into the copy output prompt word template to obtain a complete output prompt word; The target entrepreneur label is obtained by generating the target entrepreneur label based on the complete output prompt word through the preset large model.

5. The method according to claim 4, characterized in that The receiving of query information input by the user to trigger task execution, the method further includes: Receive query information input by the user on the preset page to trigger task execution; The method of presenting the output generated by the final AI assistant to the user as the final result includes: The target entrepreneur tag is displayed on the preset page.

6. The method according to claim 4, characterized in that The method further comprises: Receive user feedback in detail and optimize the multiple large model prompt word templates or pipeline structures.

7. The method according to claim 4, characterized in that When searching, refining or generating by using a preset large model, the method further includes a step of performing a quality assessment on the output result of the large model, which step includes: Each large model output is scored based on pre-set quality assessment indicators, including but not limited to information accuracy, completeness, language fluency, and fit with task requirements. When the evaluation score is lower than the preset threshold, the parameters or content of the corresponding prompt word are automatically adjusted, and the large model is called again for processing until the evaluation score of the output result reaches or exceeds the preset threshold.

8. An AI task processing device based on pipeline mode, characterized in that: include: A preparation module, configured to collect basic information related to a task requirement in response to a task requirement input by a user; Building an AI pipeline based on the basic information, the AI ​​pipeline includes defining a pipeline structure, setting multiple AI assistant roles and tasks, and configuring system storage and data flow, wherein the AI ​​assistant is encapsulated by prompt words and is used to perform specific tasks, and the pipeline structure determines the steps and order of task processing; The execution module is used to receive query information input by the user to trigger task execution, call each AI assistant in sequence to perform the task in a pipeline order, and the output of each AI assistant is used as the input of the next AI assistant. The system stores the output content of each AI assistant; the output generated by the final AI assistant is displayed to the user as the final result.

9. A computer device, characterized in that: The computer device includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device executes the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

Images