Recommendation methods, devices, processors, and electronic equipment for the model

Abstract

This application discloses a model recommendation method, apparatus, processor, and electronic device. Relating to the field of big data, the method includes: obtaining a user's model requirement target for product recommendation models; retrieving model information for M candidate models from a database; processing the model information of each candidate model to obtain M sets of model data; calculating the initial similarity between the M sets of model data and preset model data in the model requirement target; identifying candidate models associated with initial similarity data greater than a preset similarity threshold as source-related models, resulting in N source-related models, where N is a positive integer; determining a target model from the N source-related models based on the requirement target in the model requirement target, and recommending the target model to the user. This application solves the problems of high cost and low efficiency in providing users with the required models in related technologies.

Description

Technical Field

[0001] This application relates to the field of big data, and more specifically, to a recommendation method, apparatus, processor, and electronic device for a model. Background Technology

[0002] With the development of low-code big data platforms, when users attempt to perform modeling operations using these platforms, they typically need to go through processes such as building an initial model, data cleaning, analytical modeling, result analysis, and report display. These processes are performed on the user interface canvas using operators dragging and dropping, concatenating, and configuring parameters. The task is then submitted to the backend computing nodes for execution to obtain the desired model. However, as the platform's functionality becomes increasingly rich, low-code big data platforms suffer from problems such as consuming significant computing resources, requiring huge resource investments, low model output rates, difficulty in providing new models, and increased maintenance costs.

[0003] There is currently no effective solution to the problem of high cost and low efficiency in providing users with the required models in related technologies. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, processor, and electronic device for recommending models, in order to solve the problems of high cost and low efficiency in providing users with the models they need in related technologies.

[0005] To achieve the above objectives, according to one aspect of this application, a model recommendation method is provided. The method includes: obtaining a user's model requirement target for a product recommendation model; retrieving model information for M candidate models from a database; processing the model information of each candidate model to obtain M sets of model data, where M is a positive integer; the model information includes model configuration information and model structure information; and the candidate models are product recommendation models. The method further includes: calculating the initial similarity between the M sets of model data and preset model data in the model requirement target; identifying candidate models associated with initial similarity data greater than a preset similarity threshold as source-related models, resulting in N source-related models, where N is a positive integer; determining a target model from the N source-related models based on the requirement target in the model requirement target; and recommending the target model to the user.

[0006] Optionally, the demand objective includes at least one of the following: a first demand for requesting a model recommendation for the user, a second demand for requesting adjustment of the original model provided by the user, and a third demand for requesting a model different from the original model. Determining the target model from N source-related models based on the demand objective includes: if the demand objective is the first demand, parsing the preset model from the first demand, calculating the model similarity data between the N source-related models and the preset model, and selecting the source-related model corresponding to the highest model similarity data as the target model; if the demand objective is the second demand, parsing the original model from the second demand, calculating the model similarity data between the original model and the N source-related models, selecting the source-related model corresponding to the highest model similarity data as a candidate source-related model, and adjusting the original model based on the candidate source-related model to obtain the target model; if the demand objective is the third demand, parsing the original model from the third demand, calculating the dissimilarity data between the original model and the N source-related models, and selecting the source-related model corresponding to the highest dissimilarity data as the target model.

[0007] Optionally, adjusting the original model based on the candidate source-related models to obtain the target model includes: obtaining the model scores associated with the candidate source-related models during the time period of iteratively updating the model structure of the candidate source-related models; determining whether the model scores associated with the candidate source-related models change by a factor greater than a preset factor during the time period; if the model scores associated with the candidate source-related models change by a factor greater than a preset factor during the time period, obtaining the model structure difference information between the changed candidate source-related models and the original candidate source-related models; and adjusting the original model based on the model structure difference information to obtain the target model.

[0008] Optionally, calculating the initial similarity between the M sets of model data and the model data in the model requirement target includes: determining whether the M sets of model data are data in a first preset format, where the first preset format refers to formatted data; if the M sets of model data contain model data in the first preset format, obtaining the fields of the model data in the first preset format, and using a first similarity algorithm to calculate the initial similarity between the fields in each set of model data and the fields in the preset model data to obtain first initial similarity data; if the M sets of model data contain model data in a non-first preset format, obtaining the fields of the model data in the non-first preset format, and using a second similarity algorithm to calculate the initial similarity between the fields in each set of model data and the fields in the preset model data to obtain second initial similarity data, where the first similarity algorithm and the second similarity algorithm are different; and combining the first initial similarity data and the second initial similarity data to obtain initial similarity data.

[0009] Optionally, data processing is performed on the model information of each candidate model to obtain M sets of model data, including: obtaining the format of the model information of each candidate model and determining whether the format of the model information of each candidate model is a second preset format; for each candidate model, if the model information of the candidate model contains model information in the second preset format, the format of the model information in the second preset format is converted into a first preset format, and the model information in the first preset format and the model information in the non-second preset format constitute the model data of the candidate model; for each candidate model, if the model information of the candidate model does not contain model information in the second preset format, the model information in the non-second preset format constitutes the model data of the candidate model.

[0010] Optionally, calculating the model similarity data between the original model and the N source-related models includes: using a graph neural network model to convert the model topology of each source-related model and the model topology of the original model into graph vectors, obtaining N+1 vectors; calculating the model similarity data between the vectors of each source-related model and the vectors of the original model, obtaining N model similarity data; calculating the dissimilarity data between the original model and the N source-related models includes: calculating the model similarity data between the vectors of each source-related model and the vectors of the original model, obtaining N model similarity data; calculating the reciprocal of the N model similarity data, obtaining N dissimilarity data.

[0011] Optionally, obtaining the model score of the candidate source-related model includes: obtaining X scoring factors for each source-related model and determining the weight of each scoring factor to obtain X weights, where X is a positive integer; and performing a weighted summation using the X scoring factors and X weights of each source-related model to obtain the model score of each source-related model.

[0012] To achieve the above objectives, according to another aspect of this application, a model recommendation apparatus is provided. The apparatus includes: an acquisition unit, configured to acquire a user's model requirement target for a product recommendation model, and acquire model information of M candidate models from a database, performing data processing on the model information of each candidate model to obtain M sets of model data, where M is a positive integer, the model information includes model configuration information and model structure information, and the candidate models are product recommendation models; a calculation unit, configured to calculate the initial similarity between the M sets of model data and preset model data in the model requirement target, and determine candidate models associated with initial similarity data greater than a preset similarity threshold as source-related models, obtaining N source-related models, where N is a positive integer; and a determination unit, configured to determine a target model from the N source-related models according to the requirement target in the model requirement target, and recommend the target model to the user.

[0013] According to another aspect of the present invention, a processor is also provided, which is used to run a program, wherein the program controls a device containing a non-volatile storage medium to perform a recommended method of a model during runtime.

[0014] According to another aspect of the present invention, an electronic device is also provided, comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is configured to execute the computer-readable instructions, wherein the computer-readable instructions, when executed, perform a recommended method of a model.

[0015] This application employs the following steps: First, it obtains the user's model requirement target for a product recommendation model and retrieves model information for M candidate models from a database. Then, it processes the model information of each candidate model to obtain M sets of model data, where M is a positive integer. The model information includes model configuration information and model structure information, and the candidate models are product recommendation models. Next, it calculates the initial similarity between the M sets of model data and the preset model data in the model requirement target. Candidate models associated with initial similarity data exceeding a preset similarity threshold are identified as source-related models, resulting in N source-related models, where N is a positive integer. Finally, it determines the target model from the N source-related models based on the requirement target in the model requirement target and recommends the target model to the user. This solves the problems of high cost and low efficiency in providing the model required by users in related technologies. By calculating the similarity of the model data of candidate models in the database, source-related models are selected based on the calculation results. Then, personalized recommendations are made based on the source-related models and the user's model requirements, thereby achieving accurate model recommendation and improving model recommendation efficiency. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a flowchart of the recommendation method for the model provided in the embodiments of this application;

[0018] Figure 2 This is a schematic diagram illustrating the determination of a target model based on required objectives, according to an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of a recommended device based on the model provided in the embodiments of this application;

[0020] Figure 4 This is a schematic diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] It should be noted that all information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are information and data authorized by the user or fully authorized by all parties.

[0025] The present invention will now be described in conjunction with preferred implementation steps. Figure 1 This is a flowchart of the recommended method for the model provided in the embodiments of this application, such as... Figure 1 As shown, the method includes the following steps:

[0026] Step S101: Obtain the user's model requirement target for the product recommendation model, and obtain the model information of M candidate models from the database. Process the model information of each candidate model to obtain M sets of model data, where M is a positive integer. The model information includes model configuration information and model structure information. The candidate model is the product recommendation model.

[0027] Specifically, when users need to use models to perform product recommendation operations, they can submit modeling requirements to the low-code big data platform after adding the modeling system. The updated low-code big data platform will then recommend new models to meet the user's needs. The added modeling system can be a modeling lifecycle support system, which is used to identify user modeling requirements and provide accurate model recommendations based on those requirements.

[0028] The requirements specify that when a user submits a model request to the low-code big data platform, the platform analyzes the model request objectives and selects suitable candidate models from its associated database based on the analysis results. These candidate models are then filtered to obtain suitable recommendable models, which are then recommended to the user. Specifically, when selecting recommendable models on the low-code big data platform, the platform first processes the model information of the candidate models selected from the database, and then filters them based on the processed model information. Since the low-code big data platform combines low-code development and big data processing capabilities, it defines data flow and processing logic through a visual graphical interface, transforming the developer's graphical definitions into actual code, and then using minimal coding to jointly build the application. Therefore, the model information for each candidate model can include a visual graph of each candidate model and configuration parameter information within each graph.

[0029] Step S102: Calculate the initial similarity between the M sets of model data and the preset model data in the model requirement target respectively. Determine the candidate models associated with the initial similarity data that are greater than the preset similarity threshold as source-related models, and obtain N source-related models, where N is a positive integer.

[0030] Specifically, before selecting the model needed by the user, it is necessary to filter out suitable models from the candidate models, and then select the desired model from among them. When filtering candidate models, the initial similarity between the model data of each candidate model and the model data of the model needed by the user can be calculated to obtain multiple initial similarity data. Each initial similarity data is then compared with a preset similarity threshold. Based on the comparison results, the candidate model with the larger initial similarity data is selected as the source related model. The source related model refers to the model with a high similarity. A high similarity means that there are few differences in the structure, configuration and other information of the two candidate models, and the two can be recommended as models to each other.

[0031] Step S103: Determine the target model from N source-related models based on the demand objectives in the model demand objectives, and recommend the target model to the user.

[0032] Specifically, by analyzing the model requirements, the similarity and dissimilarity data are calculated between the analyzed model information and related source models. Then, the recommended model, or target model, is determined using the calculated similarity and dissimilarity data. It's important to note that dissimilarity data is an indicator of the differences between the model data of two models, while model similarity data is an indicator of the similarity between the model data of two models. When a user's model requirements differ, the indicators used for model recommendation will vary.

[0033] The model recommendation method provided in this application embodiment obtains the user's model requirement target for product recommendation models and retrieves model information of M candidate models from the database. Data processing is performed on the model information of each candidate model to obtain M sets of model data, where M is a positive integer. The model information includes model configuration information and model structure information, and the candidate models are product recommendation models. The initial similarity between the M sets of model data and the preset model data in the model requirement target is calculated. Candidate models associated with initial similarity data greater than a preset similarity threshold are identified as source-related models, resulting in N source-related models, where N is a positive integer. Based on the requirement target in the model requirement target, a target model is determined from the N source-related models and recommended to the user. This solves the problems of high cost and low efficiency in providing the model required by the user in related technologies. By calculating the similarity of the model data of candidate models in the database, source-related models are selected based on the calculation results, and then personalized recommendations are made based on the source-related models and the user's model requirements, thereby achieving the effect of accurate model recommendation and improved model recommendation efficiency.

[0034] Figure 2 This is a schematic diagram illustrating the determination of a target model based on required objectives, provided in an embodiment of this application. Figure 2 As shown, the target model required by the user is recommended by the low-code big data platform based on model requirements. Optionally, in the model recommendation method provided in this application embodiment, the requirement target includes at least one of the following: a first requirement for requesting model recommendations for the user, a second requirement for requesting adjustments to the original model provided by the user, and a third requirement for requesting recommendations of a model different from the original model. Determining the target model from N source-related models based on the requirement target includes:

[0035] Step S201: When the target requirement is the first requirement, the preset model is parsed from the first requirement, and the model similarity data between the N source related models and the preset model is calculated respectively. The source related model corresponding to the highest model similarity data is selected as the target model.

[0036] Step S202: When the target requirement is the second requirement, the original model is parsed from the second requirement, the model similarity data between the original model and N source related models is calculated, and the source related model corresponding to the highest model similarity data is selected as the candidate source related model. The original model is then adjusted according to the candidate source related model to obtain the target model.

[0037] Step S203: When the target requirement is the third requirement, parse the original model from the third requirement, calculate the dissimilarity data between the original model and N source related models, and select the source related model corresponding to the highest dissimilarity data as the target model.

[0038] Specifically, when selecting a model based on the model requirements and objectives, it is first necessary to determine the type of the model requirements and objectives. For example, when a user has only determined the required model data but has not determined the model, they need to select a new model through the low-code big data platform, and at this time, the user can issue the first requirement. If the user needs to make partial innovations to the existing model, that is, to improve the existing model, they can issue the second requirement to the low-code big data platform. When the user is not satisfied with the processing results of the existing model and wants to change their approach to find a new model that is different from the existing model, they can issue the third requirement to the low-code big data platform.

[0039] Furthermore, when a user's request is the primary requirement, the platform parses the received primary requirement to obtain the model data needed by the user. It then calculates the model similarity between the model data and the source-related models that have been selected from the candidate models. From the multiple calculated model similarity data, it selects the model with the highest value and recommends the source-related model associated with the model with the highest value as the target model.

[0040] Furthermore, when the user's requested target is the second requirement, the received second requirement is parsed to obtain the user's existing original model. The original model is then compared with each source-related model to calculate model similarity data. The model with the highest value among the calculated model similarity data is determined, and the source-related model associated with the model with the highest value is identified. Then, the relevant model structure of the original model is adjusted according to the model structure of the source-related model, and the adjusted model is recommended to the user as the target model.

[0041] Furthermore, when a user's requested requirement is a third requirement, the user's existing original model is parsed. Since the user desires a new model different from the existing one, the dissimilarity between the user's existing model and the source-related models needs to be calculated. The source-related model with the largest dissimilarity value is selected as the target model for recommendation. This embodiment analyzes model requirements and recommends specific models to users, thereby improving model matching accuracy, reducing model maintenance time, and lowering operating costs.

[0042] When the user's target requirement is the second requirement, the original model needs to be adjusted based on the selected source-related models. Optionally, in the model recommendation method provided in this application embodiment, adjusting the original model based on the candidate source-related models to obtain the target model includes: obtaining the model scores associated with the candidate source-related models during the time period of iteratively updating the model structure of the candidate source-related models; determining whether the model scores associated with the candidate source-related models change by a factor greater than a preset factor during the time period; if the model scores associated with the candidate source-related models change by a factor greater than a preset factor during the time period, obtaining the model structure difference information between the changed candidate source-related models and the original candidate source-related models; and adjusting the original model based on the model structure difference information to obtain the target model.

[0043] It should be noted that since the user's model requirement is to make partial adjustments to the original model, the model score of the source model can be obtained first within a certain period of time. Due to the iterative updates of the model, the user's evaluation of different versions of the model will be different, which will cause the model scores of different versions of the model to be in a state of change. If the user needs to adjust the original model, the update status of the model can be judged based on the model score, and then the model parameters or model configuration of the required version can be obtained.

[0044] Specifically, a change exceeding a preset range can be considered a sudden change in the model score of the source-related models. By determining whether such a sudden change exists in the model scores of the selected source-related models, a suitable model structure is selected for local adjustments to the original model. For example, if the model score of a source-related model is 1 point, and it suddenly jumps to 5 points within a certain time period, the process proceeds. Further, the model structure of the source-related models at the time of the sudden change is compared to the model structure before the change, and the differences are identified. These differences are then analyzed, and the configuration and parameter information are fine-tuned to conform to the original model's structural configuration. The adjusted model structure is then added back to the original model, thus achieving a local adjustment of the original model.

[0045] It should be noted that if the selected source-related models do not exhibit abrupt changes in model scores, another source-related model with high model similarity data to the original model can be selected, until a source-related model with abrupt changes in model scores is found. This embodiment analyzes the model scores of source-related models and selects a better model structure based on the changes in model scores, thereby optimizing the original model and improving model processing speed.

[0046] When data formats differ, calculating the initial similarity requires the use of multiple similarity algorithms. Optionally, in the model recommendation method provided in this application embodiment, calculating the initial similarity between M sets of model data and the model data in the model requirement target includes: determining whether the M sets of model data are data in a first preset format, where the first preset format refers to formatted data; if the M sets of model data contain model data in the first preset format, obtaining the fields of the model data in the first preset format, and using a first similarity algorithm to calculate the initial similarity between the fields in each set of model data and the fields in the preset model data to obtain first initial similarity data; if the M sets of model data contain model data in a non-first preset format, obtaining the fields of the model data in the non-first preset format, and using a second similarity algorithm to calculate the initial similarity between the fields in each set of model data and the fields in the preset model data to obtain second initial similarity data, where the first similarity algorithm and the second similarity algorithm are different; and combining the first initial similarity data and the second initial similarity data to obtain initial similarity data.

[0047] Specifically, when screening source-related models from candidate models, it is necessary to calculate the initial similarity for each candidate model. However, the model format of each model is different, and the algorithm for calculating the initial similarity is also different depending on the model format.

[0048] Specifically, the format of each candidate model needs to be analyzed first to determine whether the model data contained in the candidate model is formatted, that is, whether the model data is in a topological structure format. When there is model data in a topological structure format in the candidate model, similarity processing is directly performed on the fields in the data and the fields in the user's preset model data. This can be calculated using the Jaccard similarity coefficient or relative entropy to obtain the first initial similarity data. The Jaccard similarity coefficient, also known as the Jaccard index, is a commonly used indicator to calculate the similarity between two sets. It is defined as the size of the intersection of the two sets divided by the size of their union, and the Jaccard similarity coefficient ranges from 0 to 1, with a higher value indicating a higher similarity between the two sets. Relative entropy, also known as Kullback-Leibler (KL) divergence, is an indicator used to measure the difference between two probability distributions.

[0049] When the format of the candidate model's data is not a topological structure format, an anomaly detection algorithm can be used to calculate the initial similarity of fields in the model data and the user's preset model data, thereby obtaining second initial similarity data. This embodiment improves the efficiency of determining model similarity by applying different algorithms to model data of different formats, thus laying the foundation for model recommendation.

[0050] To obtain model data, the model information needs to be formatted. Optionally, in the model recommendation method provided in this application embodiment, data processing is performed on the model information of each candidate model to obtain M sets of model data, including: obtaining the format of the model information of each candidate model and determining whether the format of the model information of each candidate model is a second preset format; for each candidate model, if the model information of the candidate model contains model information in the second preset format, the format of the model information in the second preset format is converted into a first preset format, and the model information in the first preset format and the model information in the non-second preset format constitute the model data of the candidate model; for each candidate model, if the model information of the candidate model does not contain model information in the second preset format, the model information in the non-second preset format constitutes the model data of the candidate model.

[0051] Specifically, before calculating similarity among the model data, the model information of each candidate model needs to be processed. First, a program is written to extract model data from the candidate data. The second preset format can be Extensible Markup Language (XML). When converting models saved in XML format, it is necessary to extract the configuration parameters of each candidate model, the data flow lines of the operators connecting the candidate models, operator parameters, and data flow relationships, etc., and then convert them to the first preset format, i.e., the topology structure format, and update the model data stored in the database. If the stored format is not XML, no format modification is required. This embodiment lays the foundation for calculating model similarity using the same similarity algorithm by converting the model format, thereby reducing data processing costs.

[0052] Model recommendation requires the use of model similarity data and dissimilarity data. Optionally, in the model recommendation method provided in this application embodiment, calculating the model similarity data between the original model and N source-related models includes: using a graph neural network model to convert the model topology of each source-related model and the model topology of the original model into graph vectors, obtaining N+1 vectors; calculating the model similarity data between the vectors of each source-related model and the vectors of the original model, obtaining N model similarity data; calculating the dissimilarity data between the original model and N source-related models includes: calculating the model similarity data between the vectors of each source-related model and the vectors of the original model, obtaining N model similarity data; calculating the reciprocal of the N model similarity data, obtaining N dissimilarity data.

[0053] When using source-related models for model recommendation, model similarity data and model dissimilarity data are needed. These two data are calculated based on the vectors of each source-related model. Specifically, firstly, a graph neural network model is used to obtain the model topology relationship between each source-related model and the original model provided by the user, and convert it into graph vector format. Then, the similarity of each graph vector is calculated using the "cross-graph attention matching degree" (i.e., cross-graph matching mechanism) in graph matching network theory, thus obtaining the model similarity data between each source-related model and the original model.

[0054] Furthermore, since the dissimilarity data is used to characterize the differences between models, it is calculated from the model similarity data. That is, by calculating the reciprocal of the similarity data of each model, the dissimilarity data between each source-related model and the original model can be obtained. In this embodiment, by calculating the model similarity data and the dissimilarity data, data support is provided for the recommendation model.

[0055] To demonstrate the model usage of each source-related model, optionally, in the model recommendation method provided in this application embodiment, obtaining the associated model score of the candidate source-related model includes: obtaining X scoring factors for each source-related model, and determining the weight of each scoring factor to obtain X weights, where X is a positive integer; using the X scoring factors and X weights of each source-related model to perform a weighted summation calculation to obtain the model score of each source-related model.

[0056] Specifically, since the user's model requirement is the second requirement, a time-series test is needed to evaluate the model scores of the source-related models. Therefore, before recommending a model, each source-related model needs to be scored. Specifically, the scoring factors can include model fit score, single-user usage score, multi-user usage and generalization score, and model iteration update activity score. By obtaining the scoring factors and determining the weight of each scoring factor according to their importance, for example, the multi-user usage and generalization score and the model iteration update activity score are more important than the model fit score. Therefore, the weights of the multi-user usage and generalization score and the model iteration update activity score can be set to 0.35, and the weights of the single-user usage score and the model fit score can be set to 0.15.

[0057] Furthermore, by weighting and summing each score and its respective score factor, a model score for each source-related model can be obtained. This embodiment constructs a model score using multiple scores, thereby laying the foundation for model recommendation.

[0058] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0059] This application also provides a model recommendation apparatus. It should be noted that the model recommendation apparatus of this application can be used to execute the model recommendation method provided in this application. The model recommendation apparatus provided in this application is described below.

[0060] Figure 3 This is a schematic diagram of a recommended device based on the model provided in the embodiments of this application, such as... Figure 3 As shown, the device includes: an acquisition unit 30, a calculation unit 31, and a determination unit 32.

[0061] The acquisition unit 30 is used to acquire the user's model requirement target for the product recommendation model, and to acquire the model information of M candidate models from the database. The model information of each candidate model is processed to obtain M sets of model data, where M is a positive integer. The model information includes model configuration information and model structure information, and the candidate model is the product recommendation model.

[0062] The calculation unit 31 is used to calculate the initial similarity between the M sets of model data and the preset model data in the model requirement target, respectively, and to determine the candidate models associated with the initial similarity data that are greater than the preset similarity threshold as source related models, thereby obtaining N source related models, where N is a positive integer;

[0063] The determination unit 32 is used to determine the target model from N source-related models based on the demand objectives in the model demand objectives, and recommend the target model to the user.

[0064] Optionally, in the model recommendation device provided in this application embodiment, the determining unit 32 includes: a first parsing model, used to parse a preset model from the first requirement when the requirement target is a first requirement, and calculate the model similarity data between the N source related models and the preset model respectively, and select the source related model corresponding to the highest model similarity data as the target model; a second parsing module, used to parse the original model from the second requirement when the requirement target is a second requirement, calculate the model similarity data between the original model and the N source related models, and select the source related model corresponding to the highest model similarity data as a candidate source related model, and adjust the original model according to the candidate source related model to obtain the target model; a third parsing module, used to parse the original model from the third requirement when the requirement target is a third requirement, calculate the dissimilarity data between the original model and the N source related models, and select the source related model corresponding to the highest dissimilarity data as the target model.

[0065] Optionally, in the model recommendation device provided in this application embodiment, the determining unit 32 includes: a first acquisition module, used to acquire the model score associated with the candidate source-related model during the time period of iteratively updating the model structure of the candidate source-related model; a first judgment module, used to judge whether there is a change in the model score associated with the candidate source-related model within the time period that is greater than a preset range; a second acquisition module, used to acquire the model structure difference information between the changed candidate source-related model and the original candidate source-related model if there is a change in the model score associated with the candidate source-related model within the time period that is greater than a preset range; and an adjustment module, used to adjust the original model according to the model structure difference information to obtain the target model.

[0066] Optionally, in the model recommendation device provided in this application embodiment, the calculation unit 31 includes: a second judgment module, used to judge whether the M groups of model data are data in a first preset format, wherein the data in the first preset format refers to formatted data; a third acquisition module, used to acquire the fields of the model data in the first preset format when the M groups of model data contain model data in the first preset format, and use a first similarity algorithm to perform initial similarity calculation on the fields in each group of model data and the fields in the preset model data to obtain first initial similarity data; a fourth acquisition module, used to acquire the fields of the model data in the non-first preset format when the M groups of model data contain model data in the non-first preset format, and use a second similarity algorithm to perform initial similarity calculation on the fields in each group of model data and the fields in the preset model data to obtain second initial similarity data, wherein the first similarity algorithm and the second similarity algorithm are different; and a combination module, used to combine the first initial similarity data and the second initial similarity data to obtain initial similarity data.

[0067] Optionally, in the model recommendation device provided in this application embodiment, the acquisition unit 30 includes: a fifth acquisition module, used to acquire the format of the model information of each candidate model and determine whether the format of the model information of each candidate model is a second preset format; a conversion module, used to, for each candidate model, if the model information of the candidate model contains model information of the second preset format, convert the format of the model information of the second preset format into a first preset format, and the model information of the first preset format and the model information of the non-second preset format constitute the model data of the candidate model; and a composition module, used to, for each candidate model, if the model information of the candidate model does not contain model information of the second preset format, construct the model data of the candidate model from the model information of the non-second preset format.

[0068] Optionally, in the model recommendation device provided in this application embodiment, the calculation unit 31 includes: a conversion module, used to convert the model topology of each source-related model and the model topology of the original model into graph vectors using a graph neural network model, to obtain N+1 vectors; a first calculation module, used to calculate the model similarity data of the vectors of each source-related model and the vectors of the original model, to obtain N model similarity data; the calculation of the dissimilarity data between the original model and the N source-related models includes: a second calculation module, used to calculate the model similarity data of the vectors of each source-related model and the vectors of the original model, to obtain N model similarity data; and a third calculation module, used to calculate the reciprocal of the N model similarity data, to obtain N dissimilarity data.

[0069] Optionally, in the model recommendation device provided in the embodiments of this application, the determining unit 32 includes: a sixth acquisition module, used to acquire X rating factors for each source-related model and determine the weight of each rating factor to obtain X weights, where X is a positive integer; and a fourth calculation module, used to perform weighted summation calculation using the X rating factors and X weights of each source-related model to obtain the model score of each source-related model.

[0070] The model recommendation device provided in this application embodiment includes an acquisition unit 30, which acquires the user's model requirement target for product recommendation models and obtains model information of M candidate models from the database. The model information of each candidate model is processed to obtain M sets of model data, where M is a positive integer. The model information includes model configuration information and model structure information, and the candidate models are product recommendation models. A calculation unit 31 is used to calculate the initial similarity between the M sets of model data and the preset model data in the model requirement target. Candidate models associated with initial similarity data greater than a preset similarity threshold are identified as source-related models, resulting in N source-related models, where N is a positive integer. A determination unit 32 is used to determine the target model from the N source-related models based on the requirement target in the model requirement target and recommend the target model to the user. This solves the problems of high cost and low efficiency in providing the model required by the user in related technologies. By calculating the similarity of the model data of the candidate models in the database, source-related models are selected based on the calculation results, and then personalized recommendations are made based on the source-related models and the user's model requirements, thereby achieving the effect of accurate model recommendation and improved model recommendation efficiency.

[0071] The recommended device for the model includes a processor and a memory. The aforementioned acquisition unit 30, calculation unit 31, determination unit 32, etc., are all stored in the memory as program units. The processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0072] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the high cost and low efficiency issues in providing the user-required models in related technologies.

[0073] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0074] This invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements a recommended method for a model.

[0075] This invention provides a processor for running programs, wherein the recommended method of the execution model is used during program execution.

[0076] Figure 4 This is a schematic diagram of an electronic device provided according to an embodiment of this application, such as... Figure 4 As shown, this embodiment of the invention provides an electronic device 40, which includes a processor, a memory, and a program stored in the memory and executable on the processor. The processor is used to execute computer-readable instructions, wherein the computer-readable instructions execute a recommended method of a model. The device in this document can be a server, PC, PAD, mobile phone, etc.

[0077] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing a recommended method for a model.

[0078] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0082] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0083] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0084] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0085] It should also be noted that 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 process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A model recommendation method characterized by comprising: The method comprises the following steps: obtaining a model demand target of a product recommendation model of a user, and obtaining model information of M candidate models from a database, performing data processing on the model information of each candidate model to obtain M sets of model data, wherein M is a positive integer, the model information comprises model configuration information and model structure information, and the candidate model is a product recommendation model; calculating the initial similarity between the M sets of model data and the preset model data in the model demand target respectively, determining the candidate model associated with the initial similarity data greater than the preset similarity threshold as a source related model, and obtaining N source related models, wherein N is a positive integer; determining a target model from the N source related models according to the demand target in the model demand target, and recommending the target model to the user; wherein the demand target comprises at least one of the following: a first demand for requesting a model for the user, a second demand for requesting adjustment of an original model provided by the user, and a third demand for requesting a model different from the original model for the user; determining a target model from the N source related models according to the demand target in the model demand target, comprising: in the case that the demand target is the first demand, parsing a preset model from the first demand, and calculating model similarity data of the N source related models and the preset model respectively, and selecting the source related model corresponding to the highest model similarity data as the target model; in the case that the demand target is the second demand, parsing an original model from the second demand, calculating model similarity data of the original model and the N source related models, and selecting the source related model corresponding to the highest model similarity data as a candidate source related model, and adjusting the original model according to the candidate source related model to obtain the target model; in the case that the demand target is the third demand, parsing an original model from the third demand, calculating separation degree data of the original model and the N source related models, and selecting the source related model corresponding to the highest separation degree data as the target model; adjusting the original model according to the candidate source related model to obtain the target model, comprising: obtaining model scores associated with the candidate source related model in a time period for iteratively updating the model structure of the candidate source related model; determining whether the model scores associated with the candidate source related model in the time period have a change amplitude greater than a preset amplitude; if the model scores associated with the candidate source related model in the time period have a change amplitude greater than a preset amplitude, obtaining model structure difference information of the changed candidate source related model and the changed candidate source related model; and adjusting the original model according to the model structure difference information to obtain the target model.

2. The method of claim 1, wherein, calculating the initial similarity between the M sets of model data and the model data in the model demand target respectively comprises: determining whether the M sets of model data are data in a first preset format, wherein the data in the first preset format refers to formatted data; In the case that the first preset format of model data exists in the M groups of model data, fields of the first preset format of model data are acquired, initial similarity calculation is performed on the fields in each group of model data and the fields in the preset model data by using a first similarity algorithm to obtain first initial similarity data; In the case that the first preset format of model data does not exist in the M groups of model data, fields of the model data other than the first preset format are acquired, initial similarity calculation is performed on the fields in each group of model data and the fields in the preset model data by using a second similarity algorithm to obtain second initial similarity data, wherein the first similarity algorithm and the second similarity algorithm are different; The first initial similarity data and the second initial similarity data are combined to obtain the initial similarity data.

3. The method of claim 2, wherein, The model information of each candidate model is subjected to data processing to obtain M groups of model data, including: The format of the model information of each candidate model is acquired, and it is determined whether the format of the model information of each candidate model is a second preset format; For each candidate model, in the case that the second preset format of model information exists in the model information of the candidate model, the format of the second preset format of model information is converted into the first preset format, and the model data of the candidate model is composed of the first preset format of model information and the model information other than the second preset format; For each candidate model, in the case that the second preset format of model information does not exist in the model information of the candidate model, the model data of the candidate model is composed of the model information other than the second preset format.

4. The method of claim 1, wherein, The model similarity data of the original model and the N source-related models includes: The model topology structure of each source-related model and the model topology structure of the original model are converted into graph vectors by using a graph neural network model to obtain N+1 vectors; The model similarity data of each source-related model and the original model is calculated respectively to obtain N model similarity data; The model similarity data of the original model and the N source-related models includes: The model similarity data of each source-related model and the original model is calculated respectively to obtain N model similarity data; The reciprocal of the N model similarity data is calculated to obtain N divergence data.

5. The method of claim 1, wherein, The associated model score of the candidate source-related model includes: X score factors of each source-related model are acquired, and the weight of each score factor is determined to obtain X weights, wherein X is a positive integer; Weighted summation calculation is performed by using the X score factors of each source-related model and the X weights to obtain the model score of each source-related model.

6. A model recommendation apparatus characterized by comprising: It includes: An acquisition unit is configured to acquire a model demand target of a product recommendation model by a user, and acquire model information of M candidate models from a database, and perform data processing on the model information of each candidate model to obtain M groups of model data, wherein M is a positive integer, the model information includes model configuration information and model structure information, and the candidate model is a product recommendation model. The computing unit is configured to calculate initial similarity between the M groups of model data and preset model data in the model demand target respectively, determine a candidate model associated with initial similarity data greater than a preset similarity threshold as a source-related model, and obtain N source-related models, where N is a positive integer. The determining unit is configured to determine a target model from the N source-related models according to a demand target in the model demand target, and recommend the target model to the user. The demand target at least includes one of the following: a first demand for requesting to recommend a model for the user, a second demand for requesting to adjust an original model provided by the user, and a third demand for requesting to recommend a model different from the original model for the user. The determining unit includes: a first analysis model configured to, when the demand target is the first demand, analyze a preset model from the first demand, calculate model similarity data of the N source-related models and the preset model respectively, and select a source-related model corresponding to the highest model similarity data as the target model; a second analysis model configured to, when the demand target is the second demand, analyze an original model from the second demand, calculate model similarity data of the original model and the N source-related models, select a source-related model corresponding to the highest model similarity data as a candidate source-related model, and adjust the original model according to the candidate source-related model to obtain the target model; and a third analysis model configured to, when the demand target is the third demand, analyze an original model from the third demand, calculate separation degree data of the original model and the N source-related models, and select a source-related model corresponding to the highest separation degree data as the target model. The determining unit further includes: a first acquisition module configured to acquire a model score associated with the candidate source-related model in a time period in which the model structure of the candidate source-related model is iteratively updated; a first judgment module configured to judge whether the model score associated with the candidate source-related model in the time period has a change amplitude greater than a preset amplitude; a second acquisition module configured to, if the model score associated with the candidate source-related model in the time period has a change amplitude greater than the preset amplitude, acquire model structure difference information of the candidate source-related model before the change and the candidate source-related model after the change; and an adjustment module configured to adjust the original model according to the model structure difference information to obtain the target model.

7. A processor, comprising: The processor is configured to run a program, where the program, when executed, performs the model recommendation method of any one of claims 1 to 5.

8. An electronic device, comprising: The device includes one or more processors and a memory configured to store one or more programs, where the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the model recommendation method of any one of claims 1 to 5.

Images