Matching method and device of AI model

Abstract

The invention relates to the technical field of computers, in particular to an AI model matching method and device. The method comprises the steps of obtaining text data of a question asked by a user; determining a text task difficulty coefficient and a text task category based on the text data; based on the text task category, determining a final proxy AI model; determining a model reward based on the satisfaction degree of a user feedback result and a text task difficulty coefficient under the final agent AI model; and the model reward is sent to the final creator of the proxy AI model, so that the matching efficiency of the AI model can be improved on the premise of realizing the market value of the AI creator.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and apparatus for matching AI models. Background Technology

[0002] In recent years, artificial intelligence technology has developed rapidly, with various powerful AI models (such as large-scale language models, image generation models, and policy analysis models) emerging in large numbers. However, existing technologies still face the following major problems in the application and promotion of AI services: User-unfriendly and high selection costs: The market offers a large number of AI tools and services with varying functions. Ordinary users often need to research, compare, and select suitable AI models themselves when faced with specific tasks, a cumbersome process with high technical barriers. Users need a platform that can intelligently understand tasks and match the best solutions, not a tool supermarket. Low value realization efficiency for AI service providers (agents): For developers or operators of AI models (i.e., AI agents), their service capabilities are difficult to market efficiently and reliably. AI agents lack a unified, automated market to "take orders" and generate revenue. Their service history, capabilities, and reputation are difficult to quantify and record, leading to high costs for value discovery and trust building. Under the existing model, AI agents cannot automatically and continuously create and capture value by providing services like "labor."

[0003] Based on this, the present invention proposes an AI model matching method and apparatus to address the problem of improving the matching efficiency of AI models while realizing the market value of AI creators. Summary of the Invention

[0004] To address the challenge of improving the matching efficiency of AI models while realizing the market value of AI creators, this invention provides a method and apparatus for matching AI models.

[0005] In a first aspect, embodiments of the present invention provide a matching method for an AI model, the method comprising:

[0006] Obtain the text data of the user's question;

[0007] Based on the text data, determine the text task difficulty coefficient and text task category;

[0008] Based on the text task category, the final agent AI model is determined;

[0009] The model reward is determined based on the user feedback satisfaction and the text task difficulty coefficient under the final agent AI model.

[0010] The model reward is sent to the creator of the final proxy AI model.

[0011] Secondly, embodiments of the present invention provide an AI model matching device, comprising:

[0012] The acquisition module is used to acquire the text data of user-asked questions;

[0013] The first data processing module is used to determine the text task difficulty coefficient and text task category based on the text data;

[0014] The second data processing module is used to determine the final agent AI model based on the text task category;

[0015] The third data processing module is used to determine the model reward based on the user feedback results under the final agent AI model and the text task difficulty coefficient.

[0016] The fourth data processing module is used to send the model reward to the creator of the final proxy AI model.

[0017] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of the present invention.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of the present invention.

[0019] This invention provides an AI model matching method and apparatus. First, it acquires user-generated text data in natural language through multi-terminal interactive interfaces (such as web pages, apps, and APIs). Based on the acquired text data, features are extracted using a pre-trained semantic parsing model (such as a BERT derivative model), and a task complexity evaluation algorithm is used to simultaneously determine the text task difficulty coefficient and task category. Then, based on the text task category, the final proxy AI model is determined. Combining user feedback satisfaction (converted from five-star ratings and text evaluations to 0-1 quantification values) with the text task difficulty coefficient, a model reward is determined. Higher text task difficulty and higher user satisfaction result in a more generous reward, incentivizing the proxy to continuously optimize service quality. Finally, the model reward is sent to the creator of the final proxy AI model. This invention not only builds a stable revenue channel for AI creators and realizes their market value, but also reduces the cost of model selection for users. Thus, this invention improves the matching efficiency of AI models while realizing the market value of AI creators. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of a matching method for an AI model according to one embodiment is shown;

[0022] Figure 2 This is a hardware architecture diagram of an electronic device provided in an embodiment of the present invention;

[0023] Figure 3 A structural diagram of a matching device for an AI model according to one embodiment is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] Please refer to Figure 1 This invention provides a matching method for AI models, the method comprising:

[0026] Step 100: Obtain the text data of the user's question;

[0027] Step 102: Based on the text data, determine the text task difficulty coefficient and text task category;

[0028] Step 104: Determine the final agent AI model based on the text task category;

[0029] Step 106: Determine the model reward based on the user feedback results and the text task difficulty coefficient under the final agent AI model;

[0030] Step 108: Send the model reward to the creator of the final proxy AI model.

[0031] In this embodiment, the system first acquires user-generated text data in natural language through multi-terminal interactive interfaces (such as web pages, apps, and APIs). Based on the acquired text data, features are extracted using a pre-trained semantic parsing model (such as a BERT derivative model), and a task complexity evaluation algorithm is used to simultaneously determine the text task difficulty coefficient and task category. Subsequently, based on the text task category, the final proxy AI model is determined. The model reward is determined by combining user feedback satisfaction (converted from five-star ratings and text evaluations to 0-1 quantification values) with the text task difficulty coefficient. Higher text task difficulty and higher user satisfaction result in a more generous reward, incentivizing the proxy to continuously optimize service quality. Finally, the model reward is sent to the creator of the final proxy AI model. This invention not only builds a stable revenue channel for AI creators and realizes their market value, but also reduces the cost of model selection for users. Thus, this invention improves the matching efficiency of AI models while realizing the market value of AI creators.

[0032] In one embodiment of the present invention, determining the text task difficulty coefficient and text task category based on text data includes:

[0033] Text features are obtained by extracting features from text data;

[0034] The text features are input into a pre-defined task matching model to obtain the text task difficulty coefficient and text task category;

[0035] The preset task matching model is obtained by training a preset deep learning network using labeled text data.

[0036] In this embodiment, firstly, text data is subjected to feature extraction to obtain text features. This requires preprocessing the original text data (including denoising, word segmentation, stop word removal, and text normalization), followed by the extraction of key features from multiple dimensions: Firstly, semantic features are obtained by using a pre-trained language model (such as BERT-base or RoBERTa) to map the text into a 768-dimensional vector, capturing deep semantic information such as "core requirements (e.g., generating equipment fault diagnosis reports)" and "domain attributes (e.g., industrial manufacturing)." Secondly, structural features are obtained by quantifying the number of sentences, the density of constraints (e.g., the number of limiting items such as "must include data from the last 3 months"), and the proportion of technical terms (e.g., the frequency of terms such as "fault tree analysis" and "sensor threshold"), forming a 12-dimensional structured feature vector. The above text feature vector is then input into a pre-defined task matching model. The training phase of this model requires a labeled text dataset containing 100,000 labeled samples. Each sample is labeled with a "task category label" and a "difficulty coefficient label". The model is jointly optimized by cross-entropy loss (category prediction) and mean squared error loss (difficulty coefficient prediction). After multiple rounds of iterative training, the preset task matching model is obtained.

[0037] In one embodiment of the present invention, determining the final agent AI model based on the text task category includes:

[0038] Based on the text task category, determine the category set of AI models; wherein, the category set of AI models includes multiple agent AI models corresponding to the text task category;

[0039] Calculate the overall score of each agent AI model in the set of AI model categories to obtain an overall score ranking table;

[0040] The final agent AI model is determined based on the comprehensive score ranking table.

[0041] In this embodiment, the first step is to select agent AI models with corresponding capabilities from the platform model library based on the determined text task categories (such as "text generation", "image creation", "strategy analysis"), and construct an AI model category set. All models in this set have passed the pre-qualification to ensure that they can meet the basic requirements of the current task and completely exclude models with mismatched capabilities. The second step is to calculate a comprehensive score for each agent AI model in the set by combining multi-dimensional indicators to quantify the model's capability reliability and service timeliness. The third step is to generate a ranking table based on the comprehensive score. After sorting the models from high to low scores, the model with the highest ranking is selected as the final agent AI model by default. If there are ties in scores, the model with the faster real-time response speed is selected first to improve the matching efficiency of AI models.

[0042] In one embodiment of the present invention, the categories of AI models include large language models, image generation models, and policy analysis models.

[0043] In this embodiment, the large language model, with natural language processing capabilities at its core, can complete tasks such as text generation, semantic understanding, and multi-turn dialogue. For example, it can be used to write product promotional copy, answer questions in professional fields, extract the core viewpoints of long documents, or build intelligent customer service dialogue logic. The image generation model focuses on visual content creation and can generate images based on text descriptions, sketches, or reference images. It also supports editing functions such as style transfer and detail optimization, making it suitable for visual needs such as e-commerce product image creation, scene restoration design, and creative illustration generation. The strategy analysis model is mainly based on data-driven decision-making and has the capabilities of trend prediction, risk assessment, and solution optimization. It can be applied to scenarios that require in-depth data analysis, such as sales forecasting, supply chain process optimization, medical diagnosis assistance, and financial risk analysis.

[0044] In one embodiment of the present invention, the overall score is determined by the following formula:

[0045] S i = a.f1-β.f2-γ.f3

[0046] In the formula, S i For the overall score, 'a' is the first weighting coefficient, 'β' is the second weighting coefficient, 'γ' is the third weighting coefficient, 'f1' is the historical reputation score, 'f2' is the average of historical quotes, and 'f3' is the workload status.

[0047] In this embodiment, existing technologies employ single-dimensional evaluation (such as considering only reputation or only price). This invention incorporates historical reputation score (long-term credibility), average historical quotes (cost-effectiveness), and workload status (real-time service capability) into a single evaluation system, achieving a comprehensive assessment of the "overall value" of the agent AI model. Simultaneously, by introducing a first weighting coefficient, a second weighting coefficient, and a third weighting coefficient, the influence of each factor on the "overall score" can be flexibly adjusted according to the needs of actual business scenarios (such as scenarios prioritizing reputation while others prioritize cost or load pressure), enabling the evaluation to have dynamic, scenario-based adaptability and improving the accuracy of the overall score calculation.

[0048] In one embodiment of the present invention, the model reward is determined by the following formula:

[0049] R new =R old +(ΔR base ×M out ×M ra )

[0050] In the formula, R new For model rewards, R old As the preset base reward, ΔR base M is the preset base variation value. out M represents the difficulty level of the text task. ra For satisfaction.

[0051] In this embodiment, the higher the difficulty coefficient of the text task and the better the satisfaction, the greater the increase in reward. The abstract business logic of "task difficulty and service quality affecting rewards" is transformed into a quantitative relationship of "difficulty coefficient multiplied by satisfaction multiplied by basic change value". This upgrades the reward from experience-based judgment to a mathematical rule that can be precisely controlled, making the reward distribution more objective and efficiently guiding creators to optimize the model to improve the user experience.

[0052] In one embodiment of the present invention, sending the model reward to the creator of the final proxy AI model includes:

[0053] The model reward will be sent to the creator of the final proxy AI model via a blockchain network;

[0054] The creator of the proxy AI model has a unique address on the blockchain network.

[0055] In this embodiment, once the model reward amount is calculated and determined, a preset blockchain smart contract is automatically triggered. The contract will transfer the reward funds to the creator's exclusive digital address on the blockchain network according to predetermined rules. This digital address is unique; it is an exclusive identifier generated after the creator completes platform identity verification (such as KYC compliance review). It is bound to the creator's main information to ensure accurate attribution, and uses encryption technology to shield real identity information to protect privacy. Simultaneously, the entire reward transfer process is recorded in real-time on the blockchain's distributed nodes. Leveraging the blockchain's immutable nature, each transaction is traceable and unforgeable. Creators can check the fund arrival status at any time, and the platform can complete subsequent reconciliation based on the on-chain records, effectively avoiding mispayment or omissions in the reward distribution process and improving the efficiency and trustworthiness of fund transfers.

[0056] like Figure 2 , Figure 3 As shown, this embodiment of the invention provides an AI model matching device. The device embodiment can be implemented through software, hardware, or a combination of both. From a hardware perspective, as... Figure 2 The diagram shown is a hardware architecture diagram of an electronic device containing an AI model matching device according to an embodiment of the present invention. (Except for...) Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of the electronic device in which it is located reading the corresponding computer program from the non-volatile memory into the memory for execution.

[0057] like Figure 3 As shown in the figure, this embodiment provides an AI model matching device, including:

[0058] Module 300 is used to acquire the text data of user-asked questions.

[0059] The first data processing module 302 is used to determine the text task difficulty coefficient and text task category based on the text data;

[0060] The second data processing module 304 is used to determine the final agent AI model based on the text task category;

[0061] The third data processing module 306 is used to determine the model reward based on the satisfaction of user feedback results under the final agent AI model and the difficulty coefficient of the text task.

[0062] The fourth data processing module 308 is used to send the model reward to the creator of the final proxy AI model.

[0063] In one embodiment of the present invention, the first data processing module 302 is configured to perform the following operations:

[0064] The text data is subjected to feature extraction to obtain text features;

[0065] The text features are input into a preset task matching model to obtain the text task difficulty coefficient and the text task category;

[0066] The preset task matching model is obtained by training a preset deep learning network using labeled text data.

[0067] In one embodiment of the present invention, the second data processing module 304 is configured to perform the following operations:

[0068] Based on the text task category, a set of AI model categories is determined; wherein, the set of AI model categories includes multiple proxy AI models corresponding to the text task category;

[0069] Calculate the comprehensive score of each agent AI model in the category set of the AI ​​models, and obtain a comprehensive score ranking table;

[0070] Based on the comprehensive score ranking table, the final agent AI model is determined.

[0071] In one embodiment of the present invention, the categories of the AI ​​models include large language models, image generation models, and policy analysis models.

[0072] In one embodiment of the present invention, the comprehensive score is determined by the following formula:

[0073] S i = a.f1-β.f2-γ.f3

[0074] In the formula, S i The comprehensive score is defined as follows: a is the first weighting coefficient, β is the second weighting coefficient, γ is the third weighting coefficient, f1 is the historical reputation score, f2 is the average of historical quotes, and f3 is the workload status.

[0075] In one embodiment of the present invention, the model reward is determined by the following formula:

[0076] R new =R old +(ΔR base ×M out ×M ra )

[0077] In the formula, R new For the model reward, R old As the preset base reward, ΔR base M is the preset base variation value. out M represents the difficulty coefficient of the text task. ra The satisfaction level is described above.

[0078] In one embodiment of the present invention, sending the model reward to the creator of the final proxy AI model includes:

[0079] The model reward will be sent to the creator of the final proxy AI model via a blockchain network;

[0080] The creator of the proxy AI model has a unique address on the blockchain network.

[0081] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on a matching device for an AI model. In other embodiments of the present invention, a matching device for an AI model may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0082] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0083] This invention also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements an AI model matching method according to any embodiment of this invention.

[0084] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform a matching method for an AI model according to any embodiment of this invention.

[0085] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0086] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0087] Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0088] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0089] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion module connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion module execute some and all of the actual operations, thereby realizing the functions of any of the embodiments described above.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0091] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A matching method for an AI model, characterized in that, include: Obtain the text data of the user's question; Based on the text data, determine the text task difficulty coefficient and text task category; Based on the text task category, the final agent AI model is determined; The model reward is determined based on the user feedback satisfaction and the text task difficulty coefficient under the final agent AI model. The model reward is sent to the creator of the final proxy AI model.

2. The method according to claim 1, characterized in that, The process of determining the text task difficulty coefficient and text task category based on the text data includes: The text data is subjected to feature extraction to obtain text features; The text features are input into a preset task matching model to obtain the text task difficulty coefficient and the text task category; The preset task matching model is obtained by training a preset deep learning network using labeled text data.

3. The method according to claim 1, characterized in that, The process of determining the final agent AI model based on the text task category includes: Based on the text task category, a set of AI model categories is determined; wherein, the set of AI model categories includes multiple proxy AI models corresponding to the text task category; Calculate the comprehensive score of each agent AI model in the category set of the AI ​​models, and obtain a comprehensive score ranking table; Based on the comprehensive score ranking table, the final agent AI model is determined.

4. The method according to claim 3, characterized in that, The categories of AI models include large language models, image generation models, and policy analysis models.

5. The method according to claim 4, characterized in that, The overall score is determined using the following formula: S i =a.f1-b.f2-c.f3 In the formula, S i The comprehensive score is defined as follows: a is the first weighting coefficient, β is the second weighting coefficient, γ is the third weighting coefficient, f1 is the historical reputation score, f2 is the average of historical quotes, and f3 is the workload status.

6. The method according to claim 1, characterized in that, The model reward is determined by the following formula: R new =R old +(ΔR base ×M out ×M ra ) In the formula, R new For the model reward, R old As the preset base reward, ΔR base M is the preset base variation value. out M represents the difficulty coefficient of the text task. ra The satisfaction level is described above.

7. The method according to claim 6, characterized in that, Sending the model reward to the creator of the final proxy AI model includes: The model reward will be sent to the creator of the final proxy AI model via a blockchain network; The creator of the proxy AI model has a unique address on the blockchain network.

8. A matching device for an AI model, characterized in that, include: The acquisition module is used to acquire the text data of user-asked questions; The first data processing module is used to determine the text task difficulty coefficient and text task category based on the text data; The second data processing module is used to determine the final agent AI model based on the text task category; The third data processing module is used to determine the model reward based on the user feedback results under the final agent AI model and the text task difficulty coefficient. The fourth data processing module is used to send the model reward to the creator of the final proxy AI model.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images