A Method and System for Realizing the Operational Ecosystem of MaaS Platform Based on Large Model Evaluation Technology

Abstract

This invention relates to the field of large model application technology, specifically to a method and system for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology. This application uses large model evaluation technology to perform semantic parsing and risk preference quantification of user needs, achieving dynamic segmentation of target users. Based on development information, different target models are evaluated. Then, based on user segmentation results and model evaluation labels, a suitable target model is accurately matched for each level of target user. Combined with model evaluation results, a multi-subscription mode is dynamically optimized, establishing a dynamically coupled relationship library of "user segmentation - model label - subscription mode." Finally, a transfer learning algorithm is used to map this relationship to the target user segmentation system, pushing personalized subscription modes to users in real time. This application achieves precise supply and demand matching by combining a cosine similarity algorithm and optimizes multi-subscription modes through reinforcement learning, realizing proactive adaptation of subscription modes.

Description

Technical Field

[0001] This invention relates to the field of large model application technology, and more specifically, to a method and system for realizing the operation and ecosystem of MaaS platform based on large model evaluation technology. Background Technology

[0002] MaaS platforms provide AI capabilities to various users through a Model-as-a-Service (MAaS) model. However, different users vary significantly in terms of technical capabilities, business scenarios, budgets, and data sensitivity. Furthermore, the models on the platform differ in performance, capacity, and applicable domains. Traditional MaaS platforms often use static packages or uniform API call billing, making it difficult to dynamically match user needs with model characteristics. This results in poor user experience and limited platform revenue.

[0003] Chinese patent CN119311333B discloses a method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology. The method includes: determining the HTTP service interface and functional requirements of the model to be evaluated, and preparing a dataset related to the model's functions; configuring the HTTP service address and OpenAPI interface documentation of the model to be evaluated; using a large model tool to generate request parameter templates through prompt words, replacing file parameters with placeholders; setting up the test environment using JMeter; running the JMeter load testing process; writing Groovy scripts or generating code using the large model tool; and using the script to read the generated VOC format messages and comparing them with the original dataset. However, the existing technology does not consider the precise guidance of the platform's subscription model for different target users, making it difficult to provide a subscription model that satisfies the target customers, resulting in low customer retention on the platform. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for realizing the operation and ecosystem of MaaS platform based on large model evaluation technology, so as to solve the problems mentioned in the background.

[0005] To address the aforementioned technical problems, one objective of this invention is to provide a method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology, comprising: Collect multi-dimensional user information from multiple target users, and use large model evaluation technology to perform semantic analysis of user needs and quantify risk preferences to achieve dynamic segmentation of target users; Collect full development information of multiple target models, and conduct comprehensive capability assessment and labeling of target models through a large-scale multi-dimensional evaluation system; Based on user segmentation results and model evaluation tags, a supply and demand matching model is constructed using the cosine similarity algorithm to accurately match the appropriate target model for each level of target users. Obtain real-time usage feedback and behavioral data from target users, combine model evaluation results to dynamically optimize the multi-subscription model, and establish a dynamic coupling relationship library of "user segmentation - model tagging - subscription model"; Real-time information from real-time users is collected and mapped to a target user hierarchy through a transfer learning algorithm. Combined with a dynamically coupled relationship library and predictions of real-time user usage scenarios, personalized subscription modes are pushed to real-time users, and real-time iterative adjustments to the subscription modes are supported.

[0006] Preferably, the user information includes user type, user data, user source channel, and user needs.

[0007] Preferably, the step of collecting multi-dimensional user information from multiple target users, performing semantic analysis of user needs and quantifying risk preferences based on large-scale model evaluation technology, and completing the dynamic segmentation of target users includes the following steps: Select multi-dimensional user information of a target user and input it into a pre-trained large model; The Prompt project guides a large model to perform semantic analysis of user needs, extracting key terms and scenario features, and quantifying the complexity of the needs. Based on user business data and risk preferences, a risk preference quantitative model is constructed using large-scale model assessment technology to output a risk preference score. Set a tiered threshold, combine the demand complexity score and risk preference score to calculate the user's comprehensive score, and divide the user into tiers; the user tiers include high-end enterprise tier, professional developer tier, ordinary individual tier, and trial level tier. Return the multi-dimensional user information of the selected target user until all target users have been selected, and obtain the user level and level feature tags of each target user.

[0008] Preferably, the development information includes model standards, model version, model performance, and model capacity.

[0009] Preferably, the subscription model includes a metered subscription model, a subscription model, a buyout model, and a customized model.

[0010] Preferably, the step of obtaining real-time usage feedback and behavioral data of target users, dynamically optimizing the multi-subscription model in combination with model evaluation results, and establishing a dynamically coupled relationship library of "user segmentation - model tagging - subscription model" includes the following steps: Based on the hierarchical feature labels of target users and model evaluation labels, the K-means clustering algorithm is used to perform cluster analysis on all target users and select representative users covering each level and scenario. Collect real-time usage feedback from users; Based on the model evaluation results, an initial subscription pattern is matched to representative users, and the usage effect is tracked; Based on usage data, the subscription model's rate structure and service content are optimized using a gradient descent algorithm to form a "user segmentation-model tag-subscription model" coupling relationship that adapts to represent users. All coupling relationships representing users are then integrated to build a dynamic coupling relationship library.

[0011] Preferably, the real-time information of real-time users is collected, mapped to the target user hierarchical system through a transfer learning algorithm, and combined with a dynamically coupled relationship library and real-time user usage scenario prediction to push personalized subscription modes to real-time users, and supports real-time iterative adjustment of subscription modes, including the following steps: Collect real-time information from users and input it into the transfer learning model; Based on the hierarchical features of the target user hierarchical system, a transfer learning algorithm is used to map real-time users to the corresponding level and generate temporary feature labels for real-time users. The cosine similarity algorithm is used to calculate the similarity between the temporary feature labels of real-time users and the feature labels of representative users in the dynamic coupling relation library, and the optimal representative user is matched. Obtain the optimal coupling relationship corresponding to the representative user, and combine it with the real-time user usage scenario prediction to initially screen suitable subscription modes; Based on real-time user price sensitivity and usage frequency estimates, the initially screened subscription models are prioritized, and the preferred subscription model is pushed to users.

[0012] Preferably, the step of obtaining the optimal coupling relationship corresponding to the representative user, and combining it with real-time user usage scenario prediction to initially screen suitable subscription modes includes the following steps: By analyzing real-time user demand texts and historical behaviors using large-scale model evaluation techniques, price sensitivity coefficients can be quantified. Based on real-time user scenario characteristics and industry benchmark data, the average monthly usage frequency and call volume under different subscription models are estimated. Based on the rates of each subscription model, calculate the estimated average monthly payment amount for real-time users under different subscription models; Set the platform's profit target coefficient and user value coefficient, and construct a comprehensive scoring model: Comprehensive score = α × (1 - price sensitivity coefficient) + β × expected average monthly payment amount + γ × potential user value; where α + β + γ = 1; The subscription modes are ranked from highest to lowest based on their overall scores to determine the recommendation priority, and the optimal subscription mode is pushed to the user.

[0013] On the other hand, this application also provides a system for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology, applied to the method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology as described in any of the foregoing. The system includes a data collection component and an operation component. The data collection component collects user information of target users and development information of target models. The operation component is communicatively connected to the data collection component. All data collected by the data collection component is input to the operation component. The operation component performs user segmentation of target users and model evaluation of target models, thereby matching target users with corresponding target models and establishing subscription patterns between different users and different models, thus providing suitable subscription patterns for different types of users.

[0014] Preferably, the acquisition component includes a user information acquisition module and a model information acquisition module. The user information acquisition module acquires user information of the target user, and the model information acquisition module acquires development information of the target model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This application collects multi-dimensional user information from multiple target users, performs semantic parsing and risk preference quantification of user needs based on large model evaluation technology, achieves dynamic segmentation of target users, and collects development information of multiple target models. Based on the development information, different target models are evaluated. Then, based on the user segmentation results and model evaluation labels, a suitable target model is accurately matched for each level of target users. Combined with the model evaluation results, the application dynamically optimizes the multi-subscription mode and establishes a dynamic coupling relationship library of "user segmentation - model label - subscription mode". Finally, the application maps the relationship to the target user segmentation system through transfer learning algorithm. Combined with the dynamic coupling relationship library and real-time user usage scenario prediction, the application pushes personalized subscription modes to real-time users. This application solves the problems of single subscription mode and low matching accuracy of traditional platforms by segmenting target users and matching corresponding target models to users at different levels, achieving accurate supply and demand matching with cosine similarity algorithm, and optimizing multi-subscription modes with reinforcement learning. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the method for realizing the operational ecosystem of a MaaS platform based on large-scale model evaluation technology; Figure 2 A schematic diagram of the system framework for realizing the operational ecosystem of the MaaS platform based on large model evaluation technology; The reference numerals are: 100, Data Acquisition Component; 101, User Information Acquisition Module; 102, Model Information Acquisition Module. 200. Operational components. Detailed Implementation

[0017] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 like Figure 1 As shown, one of the objectives of this invention is to provide a method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology, including: S100 collects multi-dimensional user information from multiple target users, performs semantic analysis of user needs and quantifies risk preferences based on large model evaluation technology, and completes dynamic segmentation of target users; S200 collects development information from multiple target models and evaluates different target models based on the development information. S300, based on user segmentation results and model evaluation labels, constructs a supply and demand matching model through the cosine similarity algorithm to accurately match the appropriate target model for each level of target users. S400 acquires real-time usage feedback and behavioral data from target users, dynamically optimizes multi-subscription models based on model evaluation results, and establishes a dynamic coupling relationship library of "user segmentation - model tags - subscription models". The S500 collects real-time information from real-time users, maps it to the target user hierarchy through a transfer learning algorithm, and combines a dynamically coupled relational database with real-time user usage scenario prediction to push personalized subscription modes to real-time users, while supporting real-time iterative adjustments to the subscription modes.

[0019] It should be noted that this application collects multi-dimensional user information from multiple target users, performs semantic parsing and risk preference quantification of user needs based on large model evaluation technology, completes dynamic stratification of target users, and collects development information of multiple target models. Based on the development information, different target models are evaluated. Then, based on the user stratification results and model evaluation labels, a suitable target model is accurately matched for each level of target users. Combined with the model evaluation results, the multi-subscription mode is dynamically optimized, and a dynamic coupling relationship library of "user stratification - model label - subscription mode" is established. Finally, the transfer learning algorithm is used to map to the target user stratification system. Combined with the dynamic coupling relationship library and the prediction of real-time user usage scenarios, a personalized subscription mode is pushed to real-time users. This application solves the problems of single subscription mode and low matching accuracy of traditional platforms by stratifying target users and matching corresponding target models for different levels of users, combining cosine similarity algorithm to achieve accurate supply and demand matching, and reinforcement learning to optimize multi-subscription mode, thereby achieving proactive adaptation of subscription mode.

[0020] In one embodiment of this application, the user information includes user type, user data, user source channel, and user needs.

[0021] It should be noted that by collecting user information to define user profiles, the target users in this application are not limited to individual users, but also include enterprise users, etc. Therefore, different types of target models and subscription models can be recommended for different types of target users, thereby ensuring that customers are retained as much as possible after acquisition and improving the retention rate of the MaaS platform.

[0022] In one embodiment of this application, S100 includes: S110: Select multi-dimensional user information of a target user and input it into the pre-trained large model; S120 uses the Prompt project to guide a large model to perform semantic parsing of user needs, extract key terms and scenario features, and quantify the complexity score of the needs. S130, based on user business data and risk preferences, constructs a risk preference quantitative model through large model assessment technology and outputs a risk preference score; S140, Set a stratification threshold, combine the demand complexity score and risk preference score to calculate the user's comprehensive score, and divide the user into strata; the user strata include high-end enterprise strata, professional developer strata, ordinary individual strata and trial level strata. S150 returns the multi-dimensional user information of the selected target user until all target users have been selected, and obtains the user level and level feature label of each target user.

[0023] It should be noted that the needs of different types of target users are generally different. For example, for ordinary individuals, the target model they need is generally "free trial + low-code tools - providing free quota", while for high-end enterprises or professional developers, the target model they need generally needs to have features such as "customized model, data security, and private deployment". Therefore, by dividing different target users into different user levels, the MaaS platform can quickly determine the type of target model required by the target user based on the user level.

[0024] In one embodiment of this application, the development information includes model standards, model version, model performance, and model capacity.

[0025] It should be noted that since the operational goals of the MaaS platform are "easy model acquisition, high service reliability, and commercial sustainability", in addition to collecting user information of target users, it is also necessary to collect development information of multiple target models, so as to set different model levels of target models based on the development information.

[0026] In one embodiment of this application, the subscription model includes a metered system, a subscription system, a buyout system, and a customized system.

[0027] It should be noted that since different target users have different needs for using the target model, the Maas platform provides various subscription models to meet the needs of different types of target users.

[0028] The metered pricing model charges based on the number of user calls, the amount of data processed, and the duration of inference. The subscription model charges fees on a fixed period (monthly / yearly). The buyout model means that customers pay a one-time fee to obtain permanent long-term services. Customization is for customers' private deployments, model customization, and dedicated computing power needs, and uses a project-based pricing model.

[0029] In one embodiment of this application, S400 includes: S410, based on the hierarchical feature labels of target users and model evaluation labels, uses the K-means clustering algorithm to perform cluster analysis on all target users and select representative users covering each level and scenario; S420 collects real-time usage feedback representing users; S430, based on the model evaluation results, matches an initial subscription pattern to a representative user and tracks the usage effect; S440 optimizes the subscription model's rate structure and service content based on usage performance data using a gradient descent algorithm, forming a "user segmentation-model tag-subscription model" coupling relationship that adapts to represent users, and integrates all user-representing coupling relationships to build a dynamic coupling relationship library.

[0030] It should be noted that in this application, representative users are selected using the K-means clustering algorithm to ensure coverage of all levels and scenarios and avoid sample bias. Based on the user feedback and model evaluation results of the representative users, a one-to-one correspondence between "user stratification - model label - subscription mode" is constructed to form a reusable coupled relationship library.

[0031] K-means clustering is used for both target users and real-time users to select representative users. The default coupling relationship for this group is based on their actual subscription mode and target model. For example, for three individual users, A, B, and C, after clustering, A is the representative user. A uses the target model A and the subscription method is subscription. Then, the coupling relationship of individual user A-A-subscription method can be constructed.

[0032] In one embodiment of this application, S500 includes: S510 collects real-time information from users and inputs it into the transfer learning model; S520, based on the hierarchical features of the target user hierarchical system, uses a transfer learning algorithm to map real-time users to the corresponding level and generate temporary feature labels for real-time users; S530 uses the cosine similarity algorithm to calculate the similarity between the real-time user temporary feature labels and the representative user feature labels in the dynamic coupling relationship database, and matches the optimal representative user. S540: Obtain the optimal coupling relationship corresponding to the representative user, and combine it with the real-time user usage scenario prediction to initially screen suitable subscription modes; The S550 prioritizes initially selected subscription models based on real-time user price sensitivity and usage frequency estimates, and pushes the preferred subscription model.

[0033] It should be noted that by using transfer learning algorithms, real-time users can be mapped to the existing target user hierarchy without retraining, which improves the adaptation efficiency. Combined with the cosine similarity algorithm to match the optimal representative user, the subscription mode of the push notification has scenario adaptability. Finally, the use scenario prediction is introduced, and real-time needs are analyzed based on the semantics of the large model to achieve accurate matching of "need-model-subscription", avoid blind push, and improve user experience.

[0034] Since the coupling relationships corresponding to different target users have been obtained in the aforementioned embodiments, these coupling relationships are actually a reference standard for real-time users, thereby enabling the rapid provision of target models and subscription models suitable for real-time users. This ensures the profitability of the MaaS platform while maintaining customer retention, allowing the MaaS platform to operate better.

[0035] In one embodiment of this application, S540 includes: S541 uses large-scale model evaluation technology to analyze real-time user demand texts and historical behaviors to quantify price sensitivity coefficients. S542, based on real-time user scenario characteristics and industry benchmark data, estimates the average monthly usage frequency and call volume under different subscription models; S543, combining the rates of each subscription model, calculates the estimated average monthly payment amount for real-time users under different subscription models; S544, set the platform profit target coefficient and user value coefficient, and construct a comprehensive scoring model: Comprehensive score = α × (1 - price sensitivity coefficient) + β × expected average monthly payment amount + γ × potential user value; where α + β + γ = 1; S545, sort by comprehensive score from high to low, determine the recommendation priority of subscription mode, and push the optimal subscription mode.

[0036] It should be noted that, for real-time users, since the representative users they are matched with may have multiple coupling relationships, the expected payment for real-time users can be calculated separately for the target model in each coupling relationship, so as to select the coupling relationship that best fits the profit model from multiple expected payments.

[0037] By quantifying user price sensitivity through large-scale model evaluation technology and estimating usage frequency based on scenario characteristics, the projected payment amount is accurately calculated. A comprehensive scoring model of "price sensitivity - payment amount - potential value" is constructed to ensure that the recommended subscription model meets users' cost expectations and satisfies the platform's profit goals. At the same time, it explores users' potential value, achieves a balance between short-term adaptation and long-term value, and improves recommendation conversion rate and user lifetime value.

[0038] For example, given a real-time user A and its corresponding representative user A, since the coupling relationship of representative user A is individual user A - A-type model - subscription and individual user A - B-type model - buyout, then for real-time user A, the expected cost for each subscription model can be calculated. The expected cost for real-time user A under the A-type model is: ,in, Let p be the estimated cost for real-time user A under model A, where p is the cost of model A under a subscription model, and t is the subscription duration. For model B, the estimated cost for real-time user A is: ,in, The estimated cost for real-time user A under model B. It's the buyout fee for model B, then we'll determine... and If the profit threshold is met, the target model that meets the profit threshold will be recommended to real-time user A. If both the profit threshold and the target model meet the threshold, both the A-type model and the B-type model will be pushed to real-time user A for the real-time user to choose from.

[0039] Optionally, in addition to the profit threshold, you can also set the profit duration, which includes long-term and short-term. The main difference between long-term and short-term is the length of time the MaaS platform wants to make a profit. For the long-term profit model, the subscription time has a greater weight, while for the short-term profit model, the profit has a greater weight. like Figure 2 As shown, in one embodiment of this application, a system for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology is also provided, applied to the method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology as described in any of the foregoing. The system includes a data collection component 100 and an operation component 200. The data collection component 100 collects user information of target users and development information of target models. The operation component 200 is communicatively connected to the data collection component 100. All data information collected by the data collection component 100 is input to the operation component 200. The operation component 200 completes user segmentation of target users and model evaluation of target models, thereby matching target users with corresponding target models and establishing subscription modes between different users and different models, thus providing suitable subscription modes for different types of users.

[0040] It should be noted that the data collection component 100 collects user information of the target users and development information of the target models, and then the operation component 200 calculates user segmentation of the target users and model evaluation of the target models, thereby matching the corresponding target models to the corresponding target users to meet user needs. Based on user feedback, a coupling relationship is created so that matching can be done quickly based on the coupling relationship in subsequent matching, thereby improving the operational efficiency of the platform.

[0041] In one embodiment of this application, the acquisition component 100 includes a user information acquisition module 101 and a model information acquisition module 102. The user information acquisition module 101 acquires user information of the target user, and the model information acquisition module 102 acquires development information of the target model.

[0042] It should be noted that by separating the user information collection module 101 and the model information collection module 102, data can be collected and processed in a classified and professional manner, avoiding the difficulties and errors caused by the mixing of different types of data. Different types of collection modules collect different data information, thereby avoiding cross-leakage between data information and improving the security of user information of target users.

[0043] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology, characterized in that, include: Collect multi-dimensional user information from multiple target users, and use large model evaluation technology to perform semantic analysis of user needs and quantify risk preferences to achieve dynamic segmentation of target users; Collect full development information of multiple target models, and conduct comprehensive capability assessment and labeling of target models through a large-scale multi-dimensional evaluation system; Based on user segmentation results and model evaluation tags, a supply and demand matching model is constructed using the cosine similarity algorithm to accurately match the appropriate target model for each level of target users. Obtain real-time usage feedback and behavioral data from target users, combine model evaluation results to dynamically optimize the multi-subscription model, and establish a dynamic coupling relationship library of "user segmentation - model tagging - subscription model"; Real-time information from real-time users is collected and mapped to a target user hierarchy through a transfer learning algorithm. Combined with a dynamically coupled relationship library and predictions of real-time user usage scenarios, personalized subscription modes are pushed to real-time users, and real-time iterative adjustments to the subscription modes are supported.

2. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 1, characterized in that: The user information includes user type, user data, user source channels, and user needs.

3. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 2, characterized in that: The process of collecting multi-dimensional user information from multiple target users, performing semantic analysis of user needs and quantifying risk preferences based on large-scale model evaluation technology, and completing the dynamic segmentation of target users includes the following steps: Select multi-dimensional user information of a target user and input it into a pre-trained large model; The Prompt project guides a large model to perform semantic analysis of user needs, extracting key terms and scenario features, and quantifying the complexity of the needs. Based on user business data and risk preferences, a risk preference quantitative model is constructed using large-scale model assessment technology to output a risk preference score. Set a stratification threshold, combine the demand complexity score and risk preference score to calculate the user's comprehensive score, and divide the user into strata; The user tiers include a high-end enterprise tier, a professional developer tier, a general individual tier, and a trial-level tier. Return the multi-dimensional user information of the selected target user until all target users have been selected, and obtain the user level and level feature tags of each target user.

4. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 3, characterized in that: The development information includes model standards, model version, model performance, and model capacity.

5. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 4, characterized in that: The subscription models include metered subscription, subscription-based subscription, buyout subscription, and customized subscription.

6. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 5, characterized in that: The process of obtaining real-time usage feedback and behavioral data from target users, dynamically optimizing the multi-subscription model based on model evaluation results, and establishing a dynamically coupled relationship library of "user segmentation - model tags - subscription model" includes the following steps: Based on the hierarchical feature labels of target users and model evaluation labels, the K-means clustering algorithm is used to perform cluster analysis on all target users and select representative users covering each level and scenario. Collect real-time usage feedback from users; Based on the model evaluation results, an initial subscription pattern is matched to representative users, and the usage effect is tracked; Based on usage data, the subscription model's rate structure and service content are optimized using a gradient descent algorithm to form a "user segmentation-model tag-subscription model" coupling relationship that adapts to represent users. All coupling relationships representing users are then integrated to build a dynamic coupling relationship library.

7. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 6, characterized in that: The process involves collecting real-time user information, mapping it to a target user hierarchy using a transfer learning algorithm, and combining this with a dynamically coupled relationship database and real-time user usage scenario prediction to push personalized subscription models to real-time users. This also supports real-time iterative adjustments to the subscription models, including the following steps: Collect real-time information from users and input it into the transfer learning model; Based on the hierarchical features of the target user hierarchical system, a transfer learning algorithm is used to map real-time users to the corresponding level and generate temporary feature labels for real-time users. The cosine similarity algorithm is used to calculate the similarity between the temporary feature labels of real-time users and the feature labels of representative users in the dynamic coupling relation library, and the optimal representative user is matched. Obtain the optimal coupling relationship corresponding to the representative user, and combine it with the real-time user usage scenario prediction to initially screen suitable subscription modes; Based on real-time user price sensitivity and usage frequency estimates, the initially screened subscription models are prioritized, and the higher-priority subscription models are pushed to users.

8. The method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 7, characterized in that: The process of obtaining the optimal coupling relationship corresponding to the representative user, combined with real-time user usage scenario prediction, and initially screening suitable subscription modes includes the following steps: By analyzing real-time user demand texts and historical behaviors using large-scale model evaluation techniques, price sensitivity coefficients can be quantified. Based on real-time user scenario characteristics and industry benchmark data, the average monthly usage frequency and call volume under different subscription models are estimated. Based on the rates of each subscription model, calculate the estimated average monthly payment amount for real-time users under different subscription models; Set the platform's profit target coefficient and user value coefficient, and construct a comprehensive scoring model: Comprehensive score = α × (1 - price sensitivity coefficient) + β × expected average monthly payment amount + γ × potential user value; where α + β + γ = 1; The subscription modes are ranked from highest to lowest based on their overall scores to determine the recommendation priority, and the optimal subscription mode is pushed to the user.

9. A system for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology, applied to the method for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology as described in any one of claims 1 to 8, characterized in that, include: Data acquisition components; The acquisition components collect user information of the target user and development information of the target model, respectively. Operational components; The operation component is communicatively connected to the collection component. All data information collected by the collection component is input into the operation component. The operation component completes the user segmentation of the target user and the model evaluation of the target model, thereby completing the matching of the target user with the corresponding target model and establishing the subscription mode between different users and different models.

10. The system for realizing the operational ecosystem of a MaaS platform based on large model evaluation technology according to claim 9, characterized in that, The acquisition components include a user information acquisition module and a model information acquisition module. The user information acquisition module acquires user information of the target user, and the model information acquisition module acquires development information of the target model.

Images